diff --git a/contrib/elftoolchain/readelf/readelf.c b/contrib/elftoolchain/readelf/readelf.c index e5a6fa0de2e6..28daeede16ae 100644 --- a/contrib/elftoolchain/readelf/readelf.c +++ b/contrib/elftoolchain/readelf/readelf.c @@ -1,7810 +1,7816 @@ /*- * Copyright (c) 2009-2015 Kai Wang * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "_elftc.h" ELFTC_VCSID("$Id: readelf.c 3769 2019-06-29 15:15:02Z emaste $"); /* Backwards compatability for older FreeBSD releases. */ #ifndef STB_GNU_UNIQUE #define STB_GNU_UNIQUE 10 #endif #ifndef STT_SPARC_REGISTER #define STT_SPARC_REGISTER 13 #endif /* * readelf(1) options. */ #define RE_AA 0x00000001 #define RE_C 0x00000002 #define RE_DD 0x00000004 #define RE_D 0x00000008 #define RE_G 0x00000010 #define RE_H 0x00000020 #define RE_II 0x00000040 #define RE_I 0x00000080 #define RE_L 0x00000100 #define RE_NN 0x00000200 #define RE_N 0x00000400 #define RE_P 0x00000800 #define RE_R 0x00001000 #define RE_SS 0x00002000 #define RE_S 0x00004000 #define RE_T 0x00008000 #define RE_U 0x00010000 #define RE_VV 0x00020000 #define RE_WW 0x00040000 #define RE_W 0x00080000 #define RE_X 0x00100000 /* * dwarf dump options. */ #define DW_A 0x00000001 #define DW_FF 0x00000002 #define DW_F 0x00000004 #define DW_I 0x00000008 #define DW_LL 0x00000010 #define DW_L 0x00000020 #define DW_M 0x00000040 #define DW_O 0x00000080 #define DW_P 0x00000100 #define DW_RR 0x00000200 #define DW_R 0x00000400 #define DW_S 0x00000800 #define DW_DEFAULT_OPTIONS (DW_A | DW_F | DW_I | DW_L | DW_O | DW_P | \ DW_R | DW_RR | DW_S) /* * readelf(1) run control flags. */ #define DISPLAY_FILENAME 0x0001 /* * Internal data structure for sections. */ struct section { const char *name; /* section name */ Elf_Scn *scn; /* section scn */ uint64_t off; /* section offset */ uint64_t sz; /* section size */ uint64_t entsize; /* section entsize */ uint64_t align; /* section alignment */ uint64_t type; /* section type */ uint64_t flags; /* section flags */ uint64_t addr; /* section virtual addr */ uint32_t link; /* section link ndx */ uint32_t info; /* section info ndx */ }; struct dumpop { union { size_t si; /* section index */ const char *sn; /* section name */ } u; enum { DUMP_BY_INDEX = 0, DUMP_BY_NAME } type; /* dump type */ #define HEX_DUMP 0x0001 #define STR_DUMP 0x0002 int op; /* dump operation */ STAILQ_ENTRY(dumpop) dumpop_list; }; struct symver { const char *name; int type; }; /* * Structure encapsulates the global data for readelf(1). */ struct readelf { const char *filename; /* current processing file. */ int options; /* command line options. */ int flags; /* run control flags. */ int dop; /* dwarf dump options. */ Elf *elf; /* underlying ELF descriptor. */ Elf *ar; /* archive ELF descriptor. */ Dwarf_Debug dbg; /* DWARF handle. */ Dwarf_Half cu_psize; /* DWARF CU pointer size. */ Dwarf_Half cu_osize; /* DWARF CU offset size. */ Dwarf_Half cu_ver; /* DWARF CU version. */ GElf_Ehdr ehdr; /* ELF header. */ int ec; /* ELF class. */ size_t shnum; /* #sections. */ struct section *vd_s; /* Verdef section. */ struct section *vn_s; /* Verneed section. */ struct section *vs_s; /* Versym section. */ uint16_t *vs; /* Versym array. */ int vs_sz; /* Versym array size. */ struct symver *ver; /* Version array. */ int ver_sz; /* Size of version array. */ struct section *sl; /* list of sections. */ STAILQ_HEAD(, dumpop) v_dumpop; /* list of dump ops. */ uint64_t (*dw_read)(Elf_Data *, uint64_t *, int); uint64_t (*dw_decode)(uint8_t **, int); }; enum options { OPTION_DEBUG_DUMP }; static struct option longopts[] = { {"all", no_argument, NULL, 'a'}, {"arch-specific", no_argument, NULL, 'A'}, {"archive-index", no_argument, NULL, 'c'}, {"debug-dump", optional_argument, NULL, OPTION_DEBUG_DUMP}, {"dynamic", no_argument, NULL, 'd'}, {"file-header", no_argument, NULL, 'h'}, {"full-section-name", no_argument, NULL, 'N'}, {"headers", no_argument, NULL, 'e'}, {"help", no_argument, 0, 'H'}, {"hex-dump", required_argument, NULL, 'x'}, {"histogram", no_argument, NULL, 'I'}, {"notes", no_argument, NULL, 'n'}, {"program-headers", no_argument, NULL, 'l'}, {"relocs", no_argument, NULL, 'r'}, {"sections", no_argument, NULL, 'S'}, {"section-headers", no_argument, NULL, 'S'}, {"section-groups", no_argument, NULL, 'g'}, {"section-details", no_argument, NULL, 't'}, {"segments", no_argument, NULL, 'l'}, {"string-dump", required_argument, NULL, 'p'}, {"symbols", no_argument, NULL, 's'}, {"syms", no_argument, NULL, 's'}, {"unwind", no_argument, NULL, 'u'}, {"use-dynamic", no_argument, NULL, 'D'}, {"version-info", no_argument, 0, 'V'}, {"version", no_argument, 0, 'v'}, {"wide", no_argument, 0, 'W'}, {NULL, 0, NULL, 0} }; struct eflags_desc { uint64_t flag; const char *desc; }; struct flag_desc { uint64_t flag; const char *desc; }; struct mips_option { uint64_t flag; const char *desc; }; struct loc_at { Dwarf_Attribute la_at; Dwarf_Unsigned la_off; Dwarf_Unsigned la_lowpc; Dwarf_Half la_cu_psize; Dwarf_Half la_cu_osize; Dwarf_Half la_cu_ver; }; static void add_dumpop(struct readelf *re, size_t si, const char *sn, int op, int t); static const char *aeabi_adv_simd_arch(uint64_t simd); static const char *aeabi_align_needed(uint64_t an); static const char *aeabi_align_preserved(uint64_t ap); static const char *aeabi_arm_isa(uint64_t ai); static const char *aeabi_cpu_arch(uint64_t arch); static const char *aeabi_cpu_arch_profile(uint64_t pf); static const char *aeabi_div(uint64_t du); static const char *aeabi_enum_size(uint64_t es); static const char *aeabi_fp_16bit_format(uint64_t fp16); static const char *aeabi_fp_arch(uint64_t fp); static const char *aeabi_fp_denormal(uint64_t fd); static const char *aeabi_fp_exceptions(uint64_t fe); static const char *aeabi_fp_hpext(uint64_t fh); static const char *aeabi_fp_number_model(uint64_t fn); static const char *aeabi_fp_optm_goal(uint64_t fog); static const char *aeabi_fp_rounding(uint64_t fr); static const char *aeabi_hardfp(uint64_t hfp); static const char *aeabi_mpext(uint64_t mp); static const char *aeabi_optm_goal(uint64_t og); static const char *aeabi_pcs_config(uint64_t pcs); static const char *aeabi_pcs_got(uint64_t got); static const char *aeabi_pcs_r9(uint64_t r9); static const char *aeabi_pcs_ro(uint64_t ro); static const char *aeabi_pcs_rw(uint64_t rw); static const char *aeabi_pcs_wchar_t(uint64_t wt); static const char *aeabi_t2ee(uint64_t t2ee); static const char *aeabi_thumb_isa(uint64_t ti); static const char *aeabi_fp_user_exceptions(uint64_t fu); static const char *aeabi_unaligned_access(uint64_t ua); static const char *aeabi_vfp_args(uint64_t va); static const char *aeabi_virtual(uint64_t vt); static const char *aeabi_wmmx_arch(uint64_t wmmx); static const char *aeabi_wmmx_args(uint64_t wa); static const char *elf_class(unsigned int class); static const char *elf_endian(unsigned int endian); static const char *elf_machine(unsigned int mach); static const char *elf_osabi(unsigned int abi); static const char *elf_type(unsigned int type); static const char *elf_ver(unsigned int ver); static const char *dt_type(unsigned int mach, unsigned int dtype); static void dump_ar(struct readelf *re, int); static void dump_arm_attributes(struct readelf *re, uint8_t *p, uint8_t *pe); static void dump_attributes(struct readelf *re); static uint8_t *dump_compatibility_tag(uint8_t *p, uint8_t *pe); static void dump_dwarf(struct readelf *re); static void dump_dwarf_abbrev(struct readelf *re); static void dump_dwarf_aranges(struct readelf *re); static void dump_dwarf_block(struct readelf *re, uint8_t *b, Dwarf_Unsigned len); static void dump_dwarf_die(struct readelf *re, Dwarf_Die die, int level); static void dump_dwarf_frame(struct readelf *re, int alt); static void dump_dwarf_frame_inst(struct readelf *re, Dwarf_Cie cie, uint8_t *insts, Dwarf_Unsigned len, Dwarf_Unsigned caf, Dwarf_Signed daf, Dwarf_Addr pc, Dwarf_Debug dbg); static int dump_dwarf_frame_regtable(struct readelf *re, Dwarf_Fde fde, Dwarf_Addr pc, Dwarf_Unsigned func_len, Dwarf_Half cie_ra); static void dump_dwarf_frame_section(struct readelf *re, struct section *s, int alt); static void dump_dwarf_info(struct readelf *re, Dwarf_Bool is_info); static void dump_dwarf_macinfo(struct readelf *re); static void dump_dwarf_line(struct readelf *re); static void dump_dwarf_line_decoded(struct readelf *re); static void dump_dwarf_loc(struct readelf *re, Dwarf_Loc *lr); static void dump_dwarf_loclist(struct readelf *re); static void dump_dwarf_pubnames(struct readelf *re); static void dump_dwarf_ranges(struct readelf *re); static void dump_dwarf_ranges_foreach(struct readelf *re, Dwarf_Die die, Dwarf_Addr base); static void dump_dwarf_str(struct readelf *re); static void dump_eflags(struct readelf *re, uint64_t e_flags); static void dump_elf(struct readelf *re); static void dump_flags(struct flag_desc *fd, uint64_t flags); static void dump_dyn_val(struct readelf *re, GElf_Dyn *dyn, uint32_t stab); static void dump_dynamic(struct readelf *re); static void dump_liblist(struct readelf *re); static void dump_mips_abiflags(struct readelf *re, struct section *s); static void dump_mips_attributes(struct readelf *re, uint8_t *p, uint8_t *pe); static void dump_mips_odk_reginfo(struct readelf *re, uint8_t *p, size_t sz); static void dump_mips_options(struct readelf *re, struct section *s); static void dump_mips_option_flags(const char *name, struct mips_option *opt, uint64_t info); static void dump_mips_reginfo(struct readelf *re, struct section *s); static void dump_mips_specific_info(struct readelf *re); static void dump_notes(struct readelf *re); static void dump_notes_content(struct readelf *re, const char *buf, size_t sz, off_t off); static void dump_notes_data(struct readelf *re, const char *name, uint32_t type, const char *buf, size_t sz); static void dump_svr4_hash(struct section *s); static void dump_svr4_hash64(struct readelf *re, struct section *s); static void dump_gnu_hash(struct readelf *re, struct section *s); static void dump_gnu_property_type_0(struct readelf *re, const char *buf, size_t sz); static void dump_hash(struct readelf *re); static void dump_phdr(struct readelf *re); static void dump_ppc_attributes(uint8_t *p, uint8_t *pe); static void dump_section_groups(struct readelf *re); static void dump_symtab(struct readelf *re, int i); static void dump_symtabs(struct readelf *re); static uint8_t *dump_unknown_tag(uint64_t tag, uint8_t *p, uint8_t *pe); static void dump_ver(struct readelf *re); static void dump_verdef(struct readelf *re, int dump); static void dump_verneed(struct readelf *re, int dump); static void dump_versym(struct readelf *re); static const char *dwarf_reg(unsigned int mach, unsigned int reg); static const char *dwarf_regname(struct readelf *re, unsigned int num); static struct dumpop *find_dumpop(struct readelf *re, size_t si, const char *sn, int op, int t); static int get_ent_count(struct section *s, int *ent_count); static int get_mips_register_size(uint8_t flag); static char *get_regoff_str(struct readelf *re, Dwarf_Half reg, Dwarf_Addr off); static const char *get_string(struct readelf *re, int strtab, size_t off); static const char *get_symbol_name(struct readelf *re, int symtab, int i); static uint64_t get_symbol_value(struct readelf *re, int symtab, int i); static void load_sections(struct readelf *re); static int loc_at_comparator(const void *la1, const void *la2); static const char *mips_abi_fp(uint64_t fp); static const char *note_type(const char *note_name, unsigned int et, unsigned int nt); static const char *note_type_freebsd(unsigned int nt); static const char *note_type_freebsd_core(unsigned int nt); static const char *note_type_linux_core(unsigned int nt); static const char *note_type_gnu(unsigned int nt); static const char *note_type_netbsd(unsigned int nt); static const char *note_type_openbsd(unsigned int nt); static const char *note_type_unknown(unsigned int nt); static const char *note_type_xen(unsigned int nt); static const char *option_kind(uint8_t kind); static const char *phdr_type(unsigned int mach, unsigned int ptype); static const char *ppc_abi_fp(uint64_t fp); static const char *ppc_abi_vector(uint64_t vec); static void readelf_usage(int status); static void readelf_version(void); static void search_loclist_at(struct readelf *re, Dwarf_Die die, Dwarf_Unsigned lowpc, struct loc_at **la_list, size_t *la_list_len, size_t *la_list_cap); static void search_ver(struct readelf *re); static const char *section_type(unsigned int mach, unsigned int stype); static void set_cu_context(struct readelf *re, Dwarf_Half psize, Dwarf_Half osize, Dwarf_Half ver); static const char *st_bind(unsigned int sbind); static const char *st_shndx(unsigned int shndx); static const char *st_type(unsigned int mach, unsigned int os, unsigned int stype); static const char *st_vis(unsigned int svis); static const char *top_tag(unsigned int tag); static void unload_sections(struct readelf *re); static uint64_t _read_lsb(Elf_Data *d, uint64_t *offsetp, int bytes_to_read); static uint64_t _read_msb(Elf_Data *d, uint64_t *offsetp, int bytes_to_read); static uint64_t _decode_lsb(uint8_t **data, int bytes_to_read); static uint64_t _decode_msb(uint8_t **data, int bytes_to_read); static int64_t _decode_sleb128(uint8_t **dp, uint8_t *dpe); static uint64_t _decode_uleb128(uint8_t **dp, uint8_t *dpe); static struct eflags_desc arm_eflags_desc[] = { {EF_ARM_RELEXEC, "relocatable executable"}, {EF_ARM_HASENTRY, "has entry point"}, {EF_ARM_SYMSARESORTED, "sorted symbol tables"}, {EF_ARM_DYNSYMSUSESEGIDX, "dynamic symbols use segment index"}, {EF_ARM_MAPSYMSFIRST, "mapping symbols precede others"}, {EF_ARM_BE8, "BE8"}, {EF_ARM_LE8, "LE8"}, {EF_ARM_INTERWORK, "interworking enabled"}, {EF_ARM_APCS_26, "uses APCS/26"}, {EF_ARM_APCS_FLOAT, "uses APCS/float"}, {EF_ARM_PIC, "position independent"}, {EF_ARM_ALIGN8, "8 bit structure alignment"}, {EF_ARM_NEW_ABI, "uses new ABI"}, {EF_ARM_OLD_ABI, "uses old ABI"}, {EF_ARM_SOFT_FLOAT, "software FP"}, {EF_ARM_VFP_FLOAT, "VFP"}, {EF_ARM_MAVERICK_FLOAT, "Maverick FP"}, {0, NULL} }; static struct eflags_desc mips_eflags_desc[] = { {EF_MIPS_NOREORDER, "noreorder"}, {EF_MIPS_PIC, "pic"}, {EF_MIPS_CPIC, "cpic"}, {EF_MIPS_UCODE, "ugen_reserved"}, {EF_MIPS_ABI2, "abi2"}, {EF_MIPS_OPTIONS_FIRST, "odk first"}, {EF_MIPS_ARCH_ASE_MDMX, "mdmx"}, {EF_MIPS_ARCH_ASE_M16, "mips16"}, {0, NULL} }; static struct eflags_desc powerpc_eflags_desc[] = { {EF_PPC_EMB, "emb"}, {EF_PPC_RELOCATABLE, "relocatable"}, {EF_PPC_RELOCATABLE_LIB, "relocatable-lib"}, {0, NULL} }; static struct eflags_desc riscv_eflags_desc[] = { {EF_RISCV_RVC, "RVC"}, {EF_RISCV_RVE, "RVE"}, {EF_RISCV_TSO, "TSO"}, {0, NULL} }; static struct eflags_desc sparc_eflags_desc[] = { {EF_SPARC_32PLUS, "v8+"}, {EF_SPARC_SUN_US1, "ultrasparcI"}, {EF_SPARC_HAL_R1, "halr1"}, {EF_SPARC_SUN_US3, "ultrasparcIII"}, {0, NULL} }; static const char * elf_osabi(unsigned int abi) { static char s_abi[32]; switch(abi) { case ELFOSABI_NONE: return "NONE"; case ELFOSABI_HPUX: return "HPUX"; case ELFOSABI_NETBSD: return "NetBSD"; case ELFOSABI_GNU: return "GNU"; case ELFOSABI_HURD: return "HURD"; case ELFOSABI_86OPEN: return "86OPEN"; case ELFOSABI_SOLARIS: return "Solaris"; case ELFOSABI_AIX: return "AIX"; case ELFOSABI_IRIX: return "IRIX"; case ELFOSABI_FREEBSD: return "FreeBSD"; case ELFOSABI_TRU64: return "TRU64"; case ELFOSABI_MODESTO: return "MODESTO"; case ELFOSABI_OPENBSD: return "OpenBSD"; case ELFOSABI_OPENVMS: return "OpenVMS"; case ELFOSABI_NSK: return "NSK"; case ELFOSABI_CLOUDABI: return "CloudABI"; case ELFOSABI_ARM_AEABI: return "ARM EABI"; case ELFOSABI_ARM: return "ARM"; case ELFOSABI_STANDALONE: return "StandAlone"; default: snprintf(s_abi, sizeof(s_abi), "", abi); return (s_abi); } }; static const char * elf_machine(unsigned int mach) { static char s_mach[32]; switch (mach) { case EM_NONE: return "Unknown machine"; case EM_M32: return "AT&T WE32100"; case EM_SPARC: return "Sun SPARC"; case EM_386: return "Intel i386"; case EM_68K: return "Motorola 68000"; case EM_IAMCU: return "Intel MCU"; case EM_88K: return "Motorola 88000"; case EM_860: return "Intel i860"; case EM_MIPS: return "MIPS R3000 Big-Endian only"; case EM_S370: return "IBM System/370"; case EM_MIPS_RS3_LE: return "MIPS R3000 Little-Endian"; case EM_PARISC: return "HP PA-RISC"; case EM_VPP500: return "Fujitsu VPP500"; case EM_SPARC32PLUS: return "SPARC v8plus"; case EM_960: return "Intel 80960"; case EM_PPC: return "PowerPC 32-bit"; case EM_PPC64: return "PowerPC 64-bit"; case EM_S390: return "IBM System/390"; case EM_V800: return "NEC V800"; case EM_FR20: return "Fujitsu FR20"; case EM_RH32: return "TRW RH-32"; case EM_RCE: return "Motorola RCE"; case EM_ARM: return "ARM"; case EM_SH: return "Hitachi SH"; case EM_SPARCV9: return "SPARC v9 64-bit"; case EM_TRICORE: return "Siemens TriCore embedded processor"; case EM_ARC: return "Argonaut RISC Core"; case EM_H8_300: return "Hitachi H8/300"; case EM_H8_300H: return "Hitachi H8/300H"; case EM_H8S: return "Hitachi H8S"; case EM_H8_500: return "Hitachi H8/500"; case EM_IA_64: return "Intel IA-64 Processor"; case EM_MIPS_X: return "Stanford MIPS-X"; case EM_COLDFIRE: return "Motorola ColdFire"; case EM_68HC12: return "Motorola M68HC12"; case EM_MMA: return "Fujitsu MMA"; case EM_PCP: return "Siemens PCP"; case EM_NCPU: return "Sony nCPU"; case EM_NDR1: return "Denso NDR1 microprocessor"; case EM_STARCORE: return "Motorola Star*Core processor"; case EM_ME16: return "Toyota ME16 processor"; case EM_ST100: return "STMicroelectronics ST100 processor"; case EM_TINYJ: return "Advanced Logic Corp. TinyJ processor"; case EM_X86_64: return "Advanced Micro Devices x86-64"; case EM_PDSP: return "Sony DSP Processor"; case EM_FX66: return "Siemens FX66 microcontroller"; case EM_ST9PLUS: return "STMicroelectronics ST9+ 8/16 microcontroller"; case EM_ST7: return "STmicroelectronics ST7 8-bit microcontroller"; case EM_68HC16: return "Motorola MC68HC16 microcontroller"; case EM_68HC11: return "Motorola MC68HC11 microcontroller"; case EM_68HC08: return "Motorola MC68HC08 microcontroller"; case EM_68HC05: return "Motorola MC68HC05 microcontroller"; case EM_SVX: return "Silicon Graphics SVx"; case EM_ST19: return "STMicroelectronics ST19 8-bit mc"; case EM_VAX: return "Digital VAX"; case EM_CRIS: return "Axis Communications 32-bit embedded processor"; case EM_JAVELIN: return "Infineon Tech. 32bit embedded processor"; case EM_FIREPATH: return "Element 14 64-bit DSP Processor"; case EM_ZSP: return "LSI Logic 16-bit DSP Processor"; case EM_MMIX: return "Donald Knuth's educational 64-bit proc"; case EM_HUANY: return "Harvard University MI object files"; case EM_PRISM: return "SiTera Prism"; case EM_AVR: return "Atmel AVR 8-bit microcontroller"; case EM_FR30: return "Fujitsu FR30"; case EM_D10V: return "Mitsubishi D10V"; case EM_D30V: return "Mitsubishi D30V"; case EM_V850: return "NEC v850"; case EM_M32R: return "Mitsubishi M32R"; case EM_MN10300: return "Matsushita MN10300"; case EM_MN10200: return "Matsushita MN10200"; case EM_PJ: return "picoJava"; case EM_OPENRISC: return "OpenRISC 32-bit embedded processor"; case EM_ARC_A5: return "ARC Cores Tangent-A5"; case EM_XTENSA: return "Tensilica Xtensa Architecture"; case EM_VIDEOCORE: return "Alphamosaic VideoCore processor"; case EM_TMM_GPP: return "Thompson Multimedia General Purpose Processor"; case EM_NS32K: return "National Semiconductor 32000 series"; case EM_TPC: return "Tenor Network TPC processor"; case EM_SNP1K: return "Trebia SNP 1000 processor"; case EM_ST200: return "STMicroelectronics ST200 microcontroller"; case EM_IP2K: return "Ubicom IP2xxx microcontroller family"; case EM_MAX: return "MAX Processor"; case EM_CR: return "National Semiconductor CompactRISC microprocessor"; case EM_F2MC16: return "Fujitsu F2MC16"; case EM_MSP430: return "TI embedded microcontroller msp430"; case EM_BLACKFIN: return "Analog Devices Blackfin (DSP) processor"; case EM_SE_C33: return "S1C33 Family of Seiko Epson processors"; case EM_SEP: return "Sharp embedded microprocessor"; case EM_ARCA: return "Arca RISC Microprocessor"; case EM_UNICORE: return "Microprocessor series from PKU-Unity Ltd"; case EM_AARCH64: return "AArch64"; case EM_RISCV: return "RISC-V"; default: snprintf(s_mach, sizeof(s_mach), "", mach); return (s_mach); } } static const char * elf_class(unsigned int class) { static char s_class[32]; switch (class) { case ELFCLASSNONE: return "none"; case ELFCLASS32: return "ELF32"; case ELFCLASS64: return "ELF64"; default: snprintf(s_class, sizeof(s_class), "", class); return (s_class); } } static const char * elf_endian(unsigned int endian) { static char s_endian[32]; switch (endian) { case ELFDATANONE: return "none"; case ELFDATA2LSB: return "2's complement, little endian"; case ELFDATA2MSB: return "2's complement, big endian"; default: snprintf(s_endian, sizeof(s_endian), "", endian); return (s_endian); } } static const char * elf_type(unsigned int type) { static char s_type[32]; switch (type) { case ET_NONE: return "NONE (None)"; case ET_REL: return "REL (Relocatable file)"; case ET_EXEC: return "EXEC (Executable file)"; case ET_DYN: return "DYN (Shared object file)"; case ET_CORE: return "CORE (Core file)"; default: if (type >= ET_LOPROC) snprintf(s_type, sizeof(s_type), "", type); else if (type >= ET_LOOS && type <= ET_HIOS) snprintf(s_type, sizeof(s_type), "", type); else snprintf(s_type, sizeof(s_type), "", type); return (s_type); } } static const char * elf_ver(unsigned int ver) { static char s_ver[32]; switch (ver) { case EV_CURRENT: return "(current)"; case EV_NONE: return "(none)"; default: snprintf(s_ver, sizeof(s_ver), "", ver); return (s_ver); } } static const char * phdr_type(unsigned int mach, unsigned int ptype) { static char s_ptype[32]; if (ptype >= PT_LOPROC && ptype <= PT_HIPROC) { switch (mach) { case EM_ARM: switch (ptype) { case PT_ARM_ARCHEXT: return "ARM_ARCHEXT"; case PT_ARM_EXIDX: return "ARM_EXIDX"; } break; } snprintf(s_ptype, sizeof(s_ptype), "LOPROC+%#x", ptype - PT_LOPROC); return (s_ptype); } switch (ptype) { case PT_NULL: return "NULL"; case PT_LOAD: return "LOAD"; case PT_DYNAMIC: return "DYNAMIC"; case PT_INTERP: return "INTERP"; case PT_NOTE: return "NOTE"; case PT_SHLIB: return "SHLIB"; case PT_PHDR: return "PHDR"; case PT_TLS: return "TLS"; case PT_GNU_EH_FRAME: return "GNU_EH_FRAME"; case PT_GNU_STACK: return "GNU_STACK"; case PT_GNU_RELRO: return "GNU_RELRO"; case PT_OPENBSD_RANDOMIZE: return "OPENBSD_RANDOMIZE"; case PT_OPENBSD_WXNEEDED: return "OPENBSD_WXNEEDED"; case PT_OPENBSD_BOOTDATA: return "OPENBSD_BOOTDATA"; default: if (ptype >= PT_LOOS && ptype <= PT_HIOS) snprintf(s_ptype, sizeof(s_ptype), "LOOS+%#x", ptype - PT_LOOS); else snprintf(s_ptype, sizeof(s_ptype), "", ptype); return (s_ptype); } } static const char * section_type(unsigned int mach, unsigned int stype) { static char s_stype[32]; if (stype >= SHT_LOPROC && stype <= SHT_HIPROC) { switch (mach) { case EM_ARM: switch (stype) { case SHT_ARM_EXIDX: return "ARM_EXIDX"; case SHT_ARM_PREEMPTMAP: return "ARM_PREEMPTMAP"; case SHT_ARM_ATTRIBUTES: return "ARM_ATTRIBUTES"; case SHT_ARM_DEBUGOVERLAY: return "ARM_DEBUGOVERLAY"; case SHT_ARM_OVERLAYSECTION: return "ARM_OVERLAYSECTION"; } break; case EM_X86_64: switch (stype) { case SHT_X86_64_UNWIND: return "X86_64_UNWIND"; default: break; } break; case EM_MIPS: case EM_MIPS_RS3_LE: switch (stype) { case SHT_MIPS_LIBLIST: return "MIPS_LIBLIST"; case SHT_MIPS_MSYM: return "MIPS_MSYM"; case SHT_MIPS_CONFLICT: return "MIPS_CONFLICT"; case SHT_MIPS_GPTAB: return "MIPS_GPTAB"; case SHT_MIPS_UCODE: return "MIPS_UCODE"; case SHT_MIPS_DEBUG: return "MIPS_DEBUG"; case SHT_MIPS_REGINFO: return "MIPS_REGINFO"; case SHT_MIPS_PACKAGE: return "MIPS_PACKAGE"; case SHT_MIPS_PACKSYM: return "MIPS_PACKSYM"; case SHT_MIPS_RELD: return "MIPS_RELD"; case SHT_MIPS_IFACE: return "MIPS_IFACE"; case SHT_MIPS_CONTENT: return "MIPS_CONTENT"; case SHT_MIPS_OPTIONS: return "MIPS_OPTIONS"; case SHT_MIPS_DELTASYM: return "MIPS_DELTASYM"; case SHT_MIPS_DELTAINST: return "MIPS_DELTAINST"; case SHT_MIPS_DELTACLASS: return "MIPS_DELTACLASS"; case SHT_MIPS_DWARF: return "MIPS_DWARF"; case SHT_MIPS_DELTADECL: return "MIPS_DELTADECL"; case SHT_MIPS_SYMBOL_LIB: return "MIPS_SYMBOL_LIB"; case SHT_MIPS_EVENTS: return "MIPS_EVENTS"; case SHT_MIPS_TRANSLATE: return "MIPS_TRANSLATE"; case SHT_MIPS_PIXIE: return "MIPS_PIXIE"; case SHT_MIPS_XLATE: return "MIPS_XLATE"; case SHT_MIPS_XLATE_DEBUG: return "MIPS_XLATE_DEBUG"; case SHT_MIPS_WHIRL: return "MIPS_WHIRL"; case SHT_MIPS_EH_REGION: return "MIPS_EH_REGION"; case SHT_MIPS_XLATE_OLD: return "MIPS_XLATE_OLD"; case SHT_MIPS_PDR_EXCEPTION: return "MIPS_PDR_EXCEPTION"; case SHT_MIPS_ABIFLAGS: return "MIPS_ABIFLAGS"; default: break; } break; default: break; } snprintf(s_stype, sizeof(s_stype), "LOPROC+%#x", stype - SHT_LOPROC); return (s_stype); } switch (stype) { case SHT_NULL: return "NULL"; case SHT_PROGBITS: return "PROGBITS"; case SHT_SYMTAB: return "SYMTAB"; case SHT_STRTAB: return "STRTAB"; case SHT_RELA: return "RELA"; case SHT_HASH: return "HASH"; case SHT_DYNAMIC: return "DYNAMIC"; case SHT_NOTE: return "NOTE"; case SHT_NOBITS: return "NOBITS"; case SHT_REL: return "REL"; case SHT_SHLIB: return "SHLIB"; case SHT_DYNSYM: return "DYNSYM"; case SHT_INIT_ARRAY: return "INIT_ARRAY"; case SHT_FINI_ARRAY: return "FINI_ARRAY"; case SHT_PREINIT_ARRAY: return "PREINIT_ARRAY"; case SHT_GROUP: return "GROUP"; case SHT_SYMTAB_SHNDX: return "SYMTAB_SHNDX"; case SHT_SUNW_dof: return "SUNW_dof"; case SHT_SUNW_cap: return "SUNW_cap"; case SHT_GNU_HASH: return "GNU_HASH"; case SHT_SUNW_ANNOTATE: return "SUNW_ANNOTATE"; case SHT_SUNW_DEBUGSTR: return "SUNW_DEBUGSTR"; case SHT_SUNW_DEBUG: return "SUNW_DEBUG"; case SHT_SUNW_move: return "SUNW_move"; case SHT_SUNW_COMDAT: return "SUNW_COMDAT"; case SHT_SUNW_syminfo: return "SUNW_syminfo"; case SHT_SUNW_verdef: return "SUNW_verdef"; case SHT_SUNW_verneed: return "SUNW_verneed"; case SHT_SUNW_versym: return "SUNW_versym"; default: if (stype >= SHT_LOOS && stype <= SHT_HIOS) snprintf(s_stype, sizeof(s_stype), "LOOS+%#x", stype - SHT_LOOS); else if (stype >= SHT_LOUSER) snprintf(s_stype, sizeof(s_stype), "LOUSER+%#x", stype - SHT_LOUSER); else snprintf(s_stype, sizeof(s_stype), "", stype); return (s_stype); } } static const char * dt_type(unsigned int mach, unsigned int dtype) { static char s_dtype[32]; switch (dtype) { case DT_NULL: return "NULL"; case DT_NEEDED: return "NEEDED"; case DT_PLTRELSZ: return "PLTRELSZ"; case DT_PLTGOT: return "PLTGOT"; case DT_HASH: return "HASH"; case DT_STRTAB: return "STRTAB"; case DT_SYMTAB: return "SYMTAB"; case DT_RELA: return "RELA"; case DT_RELASZ: return "RELASZ"; case DT_RELAENT: return "RELAENT"; case DT_STRSZ: return "STRSZ"; case DT_SYMENT: return "SYMENT"; case DT_INIT: return "INIT"; case DT_FINI: return "FINI"; case DT_SONAME: return "SONAME"; case DT_RPATH: return "RPATH"; case DT_SYMBOLIC: return "SYMBOLIC"; case DT_REL: return "REL"; case DT_RELSZ: return "RELSZ"; case DT_RELENT: return "RELENT"; case DT_PLTREL: return "PLTREL"; case DT_DEBUG: return "DEBUG"; case DT_TEXTREL: return "TEXTREL"; case DT_JMPREL: return "JMPREL"; case DT_BIND_NOW: return "BIND_NOW"; case DT_INIT_ARRAY: return "INIT_ARRAY"; case DT_FINI_ARRAY: return "FINI_ARRAY"; case DT_INIT_ARRAYSZ: return "INIT_ARRAYSZ"; case DT_FINI_ARRAYSZ: return "FINI_ARRAYSZ"; case DT_RUNPATH: return "RUNPATH"; case DT_FLAGS: return "FLAGS"; case DT_PREINIT_ARRAY: return "PREINIT_ARRAY"; case DT_PREINIT_ARRAYSZ: return "PREINIT_ARRAYSZ"; case DT_MAXPOSTAGS: return "MAXPOSTAGS"; case DT_SUNW_AUXILIARY: return "SUNW_AUXILIARY"; case DT_SUNW_RTLDINF: return "SUNW_RTLDINF"; case DT_SUNW_FILTER: return "SUNW_FILTER"; case DT_SUNW_CAP: return "SUNW_CAP"; case DT_SUNW_ASLR: return "SUNW_ASLR"; case DT_CHECKSUM: return "CHECKSUM"; case DT_PLTPADSZ: return "PLTPADSZ"; case DT_MOVEENT: return "MOVEENT"; case DT_MOVESZ: return "MOVESZ"; case DT_FEATURE: return "FEATURE"; case DT_POSFLAG_1: return "POSFLAG_1"; case DT_SYMINSZ: return "SYMINSZ"; case DT_SYMINENT: return "SYMINENT"; case DT_GNU_HASH: return "GNU_HASH"; case DT_TLSDESC_PLT: return "DT_TLSDESC_PLT"; case DT_TLSDESC_GOT: return "DT_TLSDESC_GOT"; case DT_GNU_CONFLICT: return "GNU_CONFLICT"; case DT_GNU_LIBLIST: return "GNU_LIBLIST"; case DT_CONFIG: return "CONFIG"; case DT_DEPAUDIT: return "DEPAUDIT"; case DT_AUDIT: return "AUDIT"; case DT_PLTPAD: return "PLTPAD"; case DT_MOVETAB: return "MOVETAB"; case DT_SYMINFO: return "SYMINFO"; case DT_VERSYM: return "VERSYM"; case DT_RELACOUNT: return "RELACOUNT"; case DT_RELCOUNT: return "RELCOUNT"; case DT_FLAGS_1: return "FLAGS_1"; case DT_VERDEF: return "VERDEF"; case DT_VERDEFNUM: return "VERDEFNUM"; case DT_VERNEED: return "VERNEED"; case DT_VERNEEDNUM: return "VERNEEDNUM"; case DT_AUXILIARY: return "AUXILIARY"; case DT_USED: return "USED"; case DT_FILTER: return "FILTER"; case DT_GNU_PRELINKED: return "GNU_PRELINKED"; case DT_GNU_CONFLICTSZ: return "GNU_CONFLICTSZ"; case DT_GNU_LIBLISTSZ: return "GNU_LIBLISTSZ"; } if (dtype >= DT_LOPROC && dtype <= DT_HIPROC) { switch (mach) { case EM_ARM: switch (dtype) { case DT_ARM_SYMTABSZ: return "ARM_SYMTABSZ"; default: break; } break; case EM_MIPS: case EM_MIPS_RS3_LE: switch (dtype) { case DT_MIPS_RLD_VERSION: return "MIPS_RLD_VERSION"; case DT_MIPS_TIME_STAMP: return "MIPS_TIME_STAMP"; case DT_MIPS_ICHECKSUM: return "MIPS_ICHECKSUM"; case DT_MIPS_IVERSION: return "MIPS_IVERSION"; case DT_MIPS_FLAGS: return "MIPS_FLAGS"; case DT_MIPS_BASE_ADDRESS: return "MIPS_BASE_ADDRESS"; case DT_MIPS_CONFLICT: return "MIPS_CONFLICT"; case DT_MIPS_LIBLIST: return "MIPS_LIBLIST"; case DT_MIPS_LOCAL_GOTNO: return "MIPS_LOCAL_GOTNO"; case DT_MIPS_CONFLICTNO: return "MIPS_CONFLICTNO"; case DT_MIPS_LIBLISTNO: return "MIPS_LIBLISTNO"; case DT_MIPS_SYMTABNO: return "MIPS_SYMTABNO"; case DT_MIPS_UNREFEXTNO: return "MIPS_UNREFEXTNO"; case DT_MIPS_GOTSYM: return "MIPS_GOTSYM"; case DT_MIPS_HIPAGENO: return "MIPS_HIPAGENO"; case DT_MIPS_RLD_MAP: return "MIPS_RLD_MAP"; case DT_MIPS_DELTA_CLASS: return "MIPS_DELTA_CLASS"; case DT_MIPS_DELTA_CLASS_NO: return "MIPS_DELTA_CLASS_NO"; case DT_MIPS_DELTA_INSTANCE: return "MIPS_DELTA_INSTANCE"; case DT_MIPS_DELTA_INSTANCE_NO: return "MIPS_DELTA_INSTANCE_NO"; case DT_MIPS_DELTA_RELOC: return "MIPS_DELTA_RELOC"; case DT_MIPS_DELTA_RELOC_NO: return "MIPS_DELTA_RELOC_NO"; case DT_MIPS_DELTA_SYM: return "MIPS_DELTA_SYM"; case DT_MIPS_DELTA_SYM_NO: return "MIPS_DELTA_SYM_NO"; case DT_MIPS_DELTA_CLASSSYM: return "MIPS_DELTA_CLASSSYM"; case DT_MIPS_DELTA_CLASSSYM_NO: return "MIPS_DELTA_CLASSSYM_NO"; case DT_MIPS_CXX_FLAGS: return "MIPS_CXX_FLAGS"; case DT_MIPS_PIXIE_INIT: return "MIPS_PIXIE_INIT"; case DT_MIPS_SYMBOL_LIB: return "MIPS_SYMBOL_LIB"; case DT_MIPS_LOCALPAGE_GOTIDX: return "MIPS_LOCALPAGE_GOTIDX"; case DT_MIPS_LOCAL_GOTIDX: return "MIPS_LOCAL_GOTIDX"; case DT_MIPS_HIDDEN_GOTIDX: return "MIPS_HIDDEN_GOTIDX"; case DT_MIPS_PROTECTED_GOTIDX: return "MIPS_PROTECTED_GOTIDX"; case DT_MIPS_OPTIONS: return "MIPS_OPTIONS"; case DT_MIPS_INTERFACE: return "MIPS_INTERFACE"; case DT_MIPS_DYNSTR_ALIGN: return "MIPS_DYNSTR_ALIGN"; case DT_MIPS_INTERFACE_SIZE: return "MIPS_INTERFACE_SIZE"; case DT_MIPS_RLD_TEXT_RESOLVE_ADDR: return "MIPS_RLD_TEXT_RESOLVE_ADDR"; case DT_MIPS_PERF_SUFFIX: return "MIPS_PERF_SUFFIX"; case DT_MIPS_COMPACT_SIZE: return "MIPS_COMPACT_SIZE"; case DT_MIPS_GP_VALUE: return "MIPS_GP_VALUE"; case DT_MIPS_AUX_DYNAMIC: return "MIPS_AUX_DYNAMIC"; case DT_MIPS_PLTGOT: return "MIPS_PLTGOT"; case DT_MIPS_RLD_OBJ_UPDATE: return "MIPS_RLD_OBJ_UPDATE"; case DT_MIPS_RWPLT: return "MIPS_RWPLT"; default: break; } break; case EM_SPARC: case EM_SPARC32PLUS: case EM_SPARCV9: switch (dtype) { case DT_SPARC_REGISTER: return "DT_SPARC_REGISTER"; default: break; } break; default: break; } } snprintf(s_dtype, sizeof(s_dtype), "", dtype); return (s_dtype); } static const char * st_bind(unsigned int sbind) { static char s_sbind[32]; switch (sbind) { case STB_LOCAL: return "LOCAL"; case STB_GLOBAL: return "GLOBAL"; case STB_WEAK: return "WEAK"; case STB_GNU_UNIQUE: return "UNIQUE"; default: if (sbind >= STB_LOOS && sbind <= STB_HIOS) return "OS"; else if (sbind >= STB_LOPROC && sbind <= STB_HIPROC) return "PROC"; else snprintf(s_sbind, sizeof(s_sbind), "", sbind); return (s_sbind); } } static const char * st_type(unsigned int mach, unsigned int os, unsigned int stype) { static char s_stype[32]; switch (stype) { case STT_NOTYPE: return "NOTYPE"; case STT_OBJECT: return "OBJECT"; case STT_FUNC: return "FUNC"; case STT_SECTION: return "SECTION"; case STT_FILE: return "FILE"; case STT_COMMON: return "COMMON"; case STT_TLS: return "TLS"; default: if (stype >= STT_LOOS && stype <= STT_HIOS) { if ((os == ELFOSABI_GNU || os == ELFOSABI_FREEBSD) && stype == STT_GNU_IFUNC) return "IFUNC"; snprintf(s_stype, sizeof(s_stype), "OS+%#x", stype - STT_LOOS); } else if (stype >= STT_LOPROC && stype <= STT_HIPROC) { if (mach == EM_SPARCV9 && stype == STT_SPARC_REGISTER) return "REGISTER"; snprintf(s_stype, sizeof(s_stype), "PROC+%#x", stype - STT_LOPROC); } else snprintf(s_stype, sizeof(s_stype), "", stype); return (s_stype); } } static const char * st_vis(unsigned int svis) { static char s_svis[32]; switch(svis) { case STV_DEFAULT: return "DEFAULT"; case STV_INTERNAL: return "INTERNAL"; case STV_HIDDEN: return "HIDDEN"; case STV_PROTECTED: return "PROTECTED"; default: snprintf(s_svis, sizeof(s_svis), "", svis); return (s_svis); } } static const char * st_shndx(unsigned int shndx) { static char s_shndx[32]; switch (shndx) { case SHN_UNDEF: return "UND"; case SHN_ABS: return "ABS"; case SHN_COMMON: return "COM"; default: if (shndx >= SHN_LOPROC && shndx <= SHN_HIPROC) return "PRC"; else if (shndx >= SHN_LOOS && shndx <= SHN_HIOS) return "OS"; else snprintf(s_shndx, sizeof(s_shndx), "%u", shndx); return (s_shndx); } } static struct { const char *ln; char sn; int value; } section_flag[] = { {"WRITE", 'W', SHF_WRITE}, {"ALLOC", 'A', SHF_ALLOC}, {"EXEC", 'X', SHF_EXECINSTR}, {"MERGE", 'M', SHF_MERGE}, {"STRINGS", 'S', SHF_STRINGS}, {"INFO LINK", 'I', SHF_INFO_LINK}, {"OS NONCONF", 'O', SHF_OS_NONCONFORMING}, {"GROUP", 'G', SHF_GROUP}, {"TLS", 'T', SHF_TLS}, {"COMPRESSED", 'C', SHF_COMPRESSED}, {NULL, 0, 0} }; static const char * note_type(const char *name, unsigned int et, unsigned int nt) { if ((strcmp(name, "CORE") == 0 || strcmp(name, "LINUX") == 0) && et == ET_CORE) return note_type_linux_core(nt); else if (strcmp(name, "FreeBSD") == 0) if (et == ET_CORE) return note_type_freebsd_core(nt); else return note_type_freebsd(nt); else if (strcmp(name, "GNU") == 0 && et != ET_CORE) return note_type_gnu(nt); else if (strcmp(name, "NetBSD") == 0 && et != ET_CORE) return note_type_netbsd(nt); else if (strcmp(name, "OpenBSD") == 0 && et != ET_CORE) return note_type_openbsd(nt); else if (strcmp(name, "Xen") == 0 && et != ET_CORE) return note_type_xen(nt); return note_type_unknown(nt); } static const char * note_type_freebsd(unsigned int nt) { switch (nt) { case 1: return "NT_FREEBSD_ABI_TAG"; case 2: return "NT_FREEBSD_NOINIT_TAG"; case 3: return "NT_FREEBSD_ARCH_TAG"; case 4: return "NT_FREEBSD_FEATURE_CTL"; default: return (note_type_unknown(nt)); } } static const char * note_type_freebsd_core(unsigned int nt) { switch (nt) { case 1: return "NT_PRSTATUS"; case 2: return "NT_FPREGSET"; case 3: return "NT_PRPSINFO"; case 7: return "NT_THRMISC"; case 8: return "NT_PROCSTAT_PROC"; case 9: return "NT_PROCSTAT_FILES"; case 10: return "NT_PROCSTAT_VMMAP"; case 11: return "NT_PROCSTAT_GROUPS"; case 12: return "NT_PROCSTAT_UMASK"; case 13: return "NT_PROCSTAT_RLIMIT"; case 14: return "NT_PROCSTAT_OSREL"; case 15: return "NT_PROCSTAT_PSSTRINGS"; case 16: return "NT_PROCSTAT_AUXV"; case 17: return "NT_PTLWPINFO"; case 0x100: return "NT_PPC_VMX (ppc Altivec registers)"; case 0x102: return "NT_PPC_VSX (ppc VSX registers)"; case 0x202: return "NT_X86_XSTATE (x86 XSAVE extended state)"; case 0x400: return "NT_ARM_VFP (arm VFP registers)"; default: return (note_type_unknown(nt)); } } static const char * note_type_linux_core(unsigned int nt) { switch (nt) { case 1: return "NT_PRSTATUS (Process status)"; case 2: return "NT_FPREGSET (Floating point information)"; case 3: return "NT_PRPSINFO (Process information)"; case 4: return "NT_TASKSTRUCT (Task structure)"; case 6: return "NT_AUXV (Auxiliary vector)"; case 10: return "NT_PSTATUS (Linux process status)"; case 12: return "NT_FPREGS (Linux floating point regset)"; case 13: return "NT_PSINFO (Linux process information)"; case 16: return "NT_LWPSTATUS (Linux lwpstatus_t type)"; case 17: return "NT_LWPSINFO (Linux lwpinfo_t type)"; case 18: return "NT_WIN32PSTATUS (win32_pstatus structure)"; case 0x100: return "NT_PPC_VMX (ppc Altivec registers)"; case 0x102: return "NT_PPC_VSX (ppc VSX registers)"; case 0x202: return "NT_X86_XSTATE (x86 XSAVE extended state)"; case 0x300: return "NT_S390_HIGH_GPRS (s390 upper register halves)"; case 0x301: return "NT_S390_TIMER (s390 timer register)"; case 0x302: return "NT_S390_TODCMP (s390 TOD comparator register)"; case 0x303: return "NT_S390_TODPREG (s390 TOD programmable register)"; case 0x304: return "NT_S390_CTRS (s390 control registers)"; case 0x305: return "NT_S390_PREFIX (s390 prefix register)"; case 0x400: return "NT_ARM_VFP (arm VFP registers)"; case 0x46494c45UL: return "NT_FILE (mapped files)"; case 0x46E62B7FUL: return "NT_PRXFPREG (Linux user_xfpregs structure)"; case 0x53494749UL: return "NT_SIGINFO (siginfo_t data)"; default: return (note_type_unknown(nt)); } } static const char * note_type_gnu(unsigned int nt) { switch (nt) { case 1: return "NT_GNU_ABI_TAG"; case 2: return "NT_GNU_HWCAP (Hardware capabilities)"; case 3: return "NT_GNU_BUILD_ID (Build id set by ld(1))"; case 4: return "NT_GNU_GOLD_VERSION (GNU gold version)"; case 5: return "NT_GNU_PROPERTY_TYPE_0"; default: return (note_type_unknown(nt)); } } static const char * note_type_netbsd(unsigned int nt) { switch (nt) { case 1: return "NT_NETBSD_IDENT"; default: return (note_type_unknown(nt)); } } static const char * note_type_openbsd(unsigned int nt) { switch (nt) { case 1: return "NT_OPENBSD_IDENT"; default: return (note_type_unknown(nt)); } } static const char * note_type_unknown(unsigned int nt) { static char s_nt[32]; snprintf(s_nt, sizeof(s_nt), nt >= 0x100 ? "" : "", nt); return (s_nt); } static const char * note_type_xen(unsigned int nt) { switch (nt) { case 0: return "XEN_ELFNOTE_INFO"; case 1: return "XEN_ELFNOTE_ENTRY"; case 2: return "XEN_ELFNOTE_HYPERCALL_PAGE"; case 3: return "XEN_ELFNOTE_VIRT_BASE"; case 4: return "XEN_ELFNOTE_PADDR_OFFSET"; case 5: return "XEN_ELFNOTE_XEN_VERSION"; case 6: return "XEN_ELFNOTE_GUEST_OS"; case 7: return "XEN_ELFNOTE_GUEST_VERSION"; case 8: return "XEN_ELFNOTE_LOADER"; case 9: return "XEN_ELFNOTE_PAE_MODE"; case 10: return "XEN_ELFNOTE_FEATURES"; case 11: return "XEN_ELFNOTE_BSD_SYMTAB"; case 12: return "XEN_ELFNOTE_HV_START_LOW"; case 13: return "XEN_ELFNOTE_L1_MFN_VALID"; case 14: return "XEN_ELFNOTE_SUSPEND_CANCEL"; case 15: return "XEN_ELFNOTE_INIT_P2M"; case 16: return "XEN_ELFNOTE_MOD_START_PFN"; case 17: return "XEN_ELFNOTE_SUPPORTED_FEATURES"; case 18: return "XEN_ELFNOTE_PHYS32_ENTRY"; default: return (note_type_unknown(nt)); } } static struct { const char *name; int value; } l_flag[] = { {"EXACT_MATCH", LL_EXACT_MATCH}, {"IGNORE_INT_VER", LL_IGNORE_INT_VER}, {"REQUIRE_MINOR", LL_REQUIRE_MINOR}, {"EXPORTS", LL_EXPORTS}, {"DELAY_LOAD", LL_DELAY_LOAD}, {"DELTA", LL_DELTA}, {NULL, 0} }; static struct mips_option mips_exceptions_option[] = { {OEX_PAGE0, "PAGE0"}, {OEX_SMM, "SMM"}, {OEX_PRECISEFP, "PRECISEFP"}, {OEX_DISMISS, "DISMISS"}, {0, NULL} }; static struct mips_option mips_pad_option[] = { {OPAD_PREFIX, "PREFIX"}, {OPAD_POSTFIX, "POSTFIX"}, {OPAD_SYMBOL, "SYMBOL"}, {0, NULL} }; static struct mips_option mips_hwpatch_option[] = { {OHW_R4KEOP, "R4KEOP"}, {OHW_R8KPFETCH, "R8KPFETCH"}, {OHW_R5KEOP, "R5KEOP"}, {OHW_R5KCVTL, "R5KCVTL"}, {0, NULL} }; static struct mips_option mips_hwa_option[] = { {OHWA0_R4KEOP_CHECKED, "R4KEOP_CHECKED"}, {OHWA0_R4KEOP_CLEAN, "R4KEOP_CLEAN"}, {0, NULL} }; static struct mips_option mips_hwo_option[] = { {OHWO0_FIXADE, "FIXADE"}, {0, NULL} }; static const char * option_kind(uint8_t kind) { static char s_kind[32]; switch (kind) { case ODK_NULL: return "NULL"; case ODK_REGINFO: return "REGINFO"; case ODK_EXCEPTIONS: return "EXCEPTIONS"; case ODK_PAD: return "PAD"; case ODK_HWPATCH: return "HWPATCH"; case ODK_FILL: return "FILL"; case ODK_TAGS: return "TAGS"; case ODK_HWAND: return "HWAND"; case ODK_HWOR: return "HWOR"; case ODK_GP_GROUP: return "GP_GROUP"; case ODK_IDENT: return "IDENT"; default: snprintf(s_kind, sizeof(s_kind), "", kind); return (s_kind); } } static const char * top_tag(unsigned int tag) { static char s_top_tag[32]; switch (tag) { case 1: return "File Attributes"; case 2: return "Section Attributes"; case 3: return "Symbol Attributes"; default: snprintf(s_top_tag, sizeof(s_top_tag), "Unknown tag: %u", tag); return (s_top_tag); } } static const char * aeabi_cpu_arch(uint64_t arch) { static char s_cpu_arch[32]; switch (arch) { case 0: return "Pre-V4"; case 1: return "ARM v4"; case 2: return "ARM v4T"; case 3: return "ARM v5T"; case 4: return "ARM v5TE"; case 5: return "ARM v5TEJ"; case 6: return "ARM v6"; case 7: return "ARM v6KZ"; case 8: return "ARM v6T2"; case 9: return "ARM v6K"; case 10: return "ARM v7"; case 11: return "ARM v6-M"; case 12: return "ARM v6S-M"; case 13: return "ARM v7E-M"; default: snprintf(s_cpu_arch, sizeof(s_cpu_arch), "Unknown (%ju)", (uintmax_t) arch); return (s_cpu_arch); } } static const char * aeabi_cpu_arch_profile(uint64_t pf) { static char s_arch_profile[32]; switch (pf) { case 0: return "Not applicable"; case 0x41: /* 'A' */ return "Application Profile"; case 0x52: /* 'R' */ return "Real-Time Profile"; case 0x4D: /* 'M' */ return "Microcontroller Profile"; case 0x53: /* 'S' */ return "Application or Real-Time Profile"; default: snprintf(s_arch_profile, sizeof(s_arch_profile), "Unknown (%ju)\n", (uintmax_t) pf); return (s_arch_profile); } } static const char * aeabi_arm_isa(uint64_t ai) { static char s_ai[32]; switch (ai) { case 0: return "No"; case 1: return "Yes"; default: snprintf(s_ai, sizeof(s_ai), "Unknown (%ju)\n", (uintmax_t) ai); return (s_ai); } } static const char * aeabi_thumb_isa(uint64_t ti) { static char s_ti[32]; switch (ti) { case 0: return "No"; case 1: return "16-bit Thumb"; case 2: return "32-bit Thumb"; default: snprintf(s_ti, sizeof(s_ti), "Unknown (%ju)\n", (uintmax_t) ti); return (s_ti); } } static const char * aeabi_fp_arch(uint64_t fp) { static char s_fp_arch[32]; switch (fp) { case 0: return "No"; case 1: return "VFPv1"; case 2: return "VFPv2"; case 3: return "VFPv3"; case 4: return "VFPv3-D16"; case 5: return "VFPv4"; case 6: return "VFPv4-D16"; default: snprintf(s_fp_arch, sizeof(s_fp_arch), "Unknown (%ju)", (uintmax_t) fp); return (s_fp_arch); } } static const char * aeabi_wmmx_arch(uint64_t wmmx) { static char s_wmmx[32]; switch (wmmx) { case 0: return "No"; case 1: return "WMMXv1"; case 2: return "WMMXv2"; default: snprintf(s_wmmx, sizeof(s_wmmx), "Unknown (%ju)", (uintmax_t) wmmx); return (s_wmmx); } } static const char * aeabi_adv_simd_arch(uint64_t simd) { static char s_simd[32]; switch (simd) { case 0: return "No"; case 1: return "NEONv1"; case 2: return "NEONv2"; default: snprintf(s_simd, sizeof(s_simd), "Unknown (%ju)", (uintmax_t) simd); return (s_simd); } } static const char * aeabi_pcs_config(uint64_t pcs) { static char s_pcs[32]; switch (pcs) { case 0: return "None"; case 1: return "Bare platform"; case 2: return "Linux"; case 3: return "Linux DSO"; case 4: return "Palm OS 2004"; case 5: return "Palm OS (future)"; case 6: return "Symbian OS 2004"; case 7: return "Symbian OS (future)"; default: snprintf(s_pcs, sizeof(s_pcs), "Unknown (%ju)", (uintmax_t) pcs); return (s_pcs); } } static const char * aeabi_pcs_r9(uint64_t r9) { static char s_r9[32]; switch (r9) { case 0: return "V6"; case 1: return "SB"; case 2: return "TLS pointer"; case 3: return "Unused"; default: snprintf(s_r9, sizeof(s_r9), "Unknown (%ju)", (uintmax_t) r9); return (s_r9); } } static const char * aeabi_pcs_rw(uint64_t rw) { static char s_rw[32]; switch (rw) { case 0: return "Absolute"; case 1: return "PC-relative"; case 2: return "SB-relative"; case 3: return "None"; default: snprintf(s_rw, sizeof(s_rw), "Unknown (%ju)", (uintmax_t) rw); return (s_rw); } } static const char * aeabi_pcs_ro(uint64_t ro) { static char s_ro[32]; switch (ro) { case 0: return "Absolute"; case 1: return "PC-relative"; case 2: return "None"; default: snprintf(s_ro, sizeof(s_ro), "Unknown (%ju)", (uintmax_t) ro); return (s_ro); } } static const char * aeabi_pcs_got(uint64_t got) { static char s_got[32]; switch (got) { case 0: return "None"; case 1: return "direct"; case 2: return "indirect via GOT"; default: snprintf(s_got, sizeof(s_got), "Unknown (%ju)", (uintmax_t) got); return (s_got); } } static const char * aeabi_pcs_wchar_t(uint64_t wt) { static char s_wt[32]; switch (wt) { case 0: return "None"; case 2: return "wchar_t size 2"; case 4: return "wchar_t size 4"; default: snprintf(s_wt, sizeof(s_wt), "Unknown (%ju)", (uintmax_t) wt); return (s_wt); } } static const char * aeabi_enum_size(uint64_t es) { static char s_es[32]; switch (es) { case 0: return "None"; case 1: return "smallest"; case 2: return "32-bit"; case 3: return "visible 32-bit"; default: snprintf(s_es, sizeof(s_es), "Unknown (%ju)", (uintmax_t) es); return (s_es); } } static const char * aeabi_align_needed(uint64_t an) { static char s_align_n[64]; switch (an) { case 0: return "No"; case 1: return "8-byte align"; case 2: return "4-byte align"; case 3: return "Reserved"; default: if (an >= 4 && an <= 12) snprintf(s_align_n, sizeof(s_align_n), "8-byte align" " and up to 2^%ju-byte extended align", (uintmax_t) an); else snprintf(s_align_n, sizeof(s_align_n), "Unknown (%ju)", (uintmax_t) an); return (s_align_n); } } static const char * aeabi_align_preserved(uint64_t ap) { static char s_align_p[128]; switch (ap) { case 0: return "No"; case 1: return "8-byte align"; case 2: return "8-byte align and SP % 8 == 0"; case 3: return "Reserved"; default: if (ap >= 4 && ap <= 12) snprintf(s_align_p, sizeof(s_align_p), "8-byte align" " and SP %% 8 == 0 and up to 2^%ju-byte extended" " align", (uintmax_t) ap); else snprintf(s_align_p, sizeof(s_align_p), "Unknown (%ju)", (uintmax_t) ap); return (s_align_p); } } static const char * aeabi_fp_rounding(uint64_t fr) { static char s_fp_r[32]; switch (fr) { case 0: return "Unused"; case 1: return "Needed"; default: snprintf(s_fp_r, sizeof(s_fp_r), "Unknown (%ju)", (uintmax_t) fr); return (s_fp_r); } } static const char * aeabi_fp_denormal(uint64_t fd) { static char s_fp_d[32]; switch (fd) { case 0: return "Unused"; case 1: return "Needed"; case 2: return "Sign Only"; default: snprintf(s_fp_d, sizeof(s_fp_d), "Unknown (%ju)", (uintmax_t) fd); return (s_fp_d); } } static const char * aeabi_fp_exceptions(uint64_t fe) { static char s_fp_e[32]; switch (fe) { case 0: return "Unused"; case 1: return "Needed"; default: snprintf(s_fp_e, sizeof(s_fp_e), "Unknown (%ju)", (uintmax_t) fe); return (s_fp_e); } } static const char * aeabi_fp_user_exceptions(uint64_t fu) { static char s_fp_u[32]; switch (fu) { case 0: return "Unused"; case 1: return "Needed"; default: snprintf(s_fp_u, sizeof(s_fp_u), "Unknown (%ju)", (uintmax_t) fu); return (s_fp_u); } } static const char * aeabi_fp_number_model(uint64_t fn) { static char s_fp_n[32]; switch (fn) { case 0: return "Unused"; case 1: return "IEEE 754 normal"; case 2: return "RTABI"; case 3: return "IEEE 754"; default: snprintf(s_fp_n, sizeof(s_fp_n), "Unknown (%ju)", (uintmax_t) fn); return (s_fp_n); } } static const char * aeabi_fp_16bit_format(uint64_t fp16) { static char s_fp_16[64]; switch (fp16) { case 0: return "None"; case 1: return "IEEE 754"; case 2: return "VFPv3/Advanced SIMD (alternative format)"; default: snprintf(s_fp_16, sizeof(s_fp_16), "Unknown (%ju)", (uintmax_t) fp16); return (s_fp_16); } } static const char * aeabi_mpext(uint64_t mp) { static char s_mp[32]; switch (mp) { case 0: return "Not allowed"; case 1: return "Allowed"; default: snprintf(s_mp, sizeof(s_mp), "Unknown (%ju)", (uintmax_t) mp); return (s_mp); } } static const char * aeabi_div(uint64_t du) { static char s_du[32]; switch (du) { case 0: return "Yes (V7-R/V7-M)"; case 1: return "No"; case 2: return "Yes (V7-A)"; default: snprintf(s_du, sizeof(s_du), "Unknown (%ju)", (uintmax_t) du); return (s_du); } } static const char * aeabi_t2ee(uint64_t t2ee) { static char s_t2ee[32]; switch (t2ee) { case 0: return "Not allowed"; case 1: return "Allowed"; default: snprintf(s_t2ee, sizeof(s_t2ee), "Unknown(%ju)", (uintmax_t) t2ee); return (s_t2ee); } } static const char * aeabi_hardfp(uint64_t hfp) { static char s_hfp[32]; switch (hfp) { case 0: return "Tag_FP_arch"; case 1: return "only SP"; case 2: return "only DP"; case 3: return "both SP and DP"; default: snprintf(s_hfp, sizeof(s_hfp), "Unknown (%ju)", (uintmax_t) hfp); return (s_hfp); } } static const char * aeabi_vfp_args(uint64_t va) { static char s_va[32]; switch (va) { case 0: return "AAPCS (base variant)"; case 1: return "AAPCS (VFP variant)"; case 2: return "toolchain-specific"; default: snprintf(s_va, sizeof(s_va), "Unknown (%ju)", (uintmax_t) va); return (s_va); } } static const char * aeabi_wmmx_args(uint64_t wa) { static char s_wa[32]; switch (wa) { case 0: return "AAPCS (base variant)"; case 1: return "Intel WMMX"; case 2: return "toolchain-specific"; default: snprintf(s_wa, sizeof(s_wa), "Unknown(%ju)", (uintmax_t) wa); return (s_wa); } } static const char * aeabi_unaligned_access(uint64_t ua) { static char s_ua[32]; switch (ua) { case 0: return "Not allowed"; case 1: return "Allowed"; default: snprintf(s_ua, sizeof(s_ua), "Unknown(%ju)", (uintmax_t) ua); return (s_ua); } } static const char * aeabi_fp_hpext(uint64_t fh) { static char s_fh[32]; switch (fh) { case 0: return "Not allowed"; case 1: return "Allowed"; default: snprintf(s_fh, sizeof(s_fh), "Unknown(%ju)", (uintmax_t) fh); return (s_fh); } } static const char * aeabi_optm_goal(uint64_t og) { static char s_og[32]; switch (og) { case 0: return "None"; case 1: return "Speed"; case 2: return "Speed aggressive"; case 3: return "Space"; case 4: return "Space aggressive"; case 5: return "Debugging"; case 6: return "Best Debugging"; default: snprintf(s_og, sizeof(s_og), "Unknown(%ju)", (uintmax_t) og); return (s_og); } } static const char * aeabi_fp_optm_goal(uint64_t fog) { static char s_fog[32]; switch (fog) { case 0: return "None"; case 1: return "Speed"; case 2: return "Speed aggressive"; case 3: return "Space"; case 4: return "Space aggressive"; case 5: return "Accurary"; case 6: return "Best Accurary"; default: snprintf(s_fog, sizeof(s_fog), "Unknown(%ju)", (uintmax_t) fog); return (s_fog); } } static const char * aeabi_virtual(uint64_t vt) { static char s_virtual[64]; switch (vt) { case 0: return "No"; case 1: return "TrustZone"; case 2: return "Virtualization extension"; case 3: return "TrustZone and virtualization extension"; default: snprintf(s_virtual, sizeof(s_virtual), "Unknown(%ju)", (uintmax_t) vt); return (s_virtual); } } static struct { uint64_t tag; const char *s_tag; const char *(*get_desc)(uint64_t val); } aeabi_tags[] = { {4, "Tag_CPU_raw_name", NULL}, {5, "Tag_CPU_name", NULL}, {6, "Tag_CPU_arch", aeabi_cpu_arch}, {7, "Tag_CPU_arch_profile", aeabi_cpu_arch_profile}, {8, "Tag_ARM_ISA_use", aeabi_arm_isa}, {9, "Tag_THUMB_ISA_use", aeabi_thumb_isa}, {10, "Tag_FP_arch", aeabi_fp_arch}, {11, "Tag_WMMX_arch", aeabi_wmmx_arch}, {12, "Tag_Advanced_SIMD_arch", aeabi_adv_simd_arch}, {13, "Tag_PCS_config", aeabi_pcs_config}, {14, "Tag_ABI_PCS_R9_use", aeabi_pcs_r9}, {15, "Tag_ABI_PCS_RW_data", aeabi_pcs_rw}, {16, "Tag_ABI_PCS_RO_data", aeabi_pcs_ro}, {17, "Tag_ABI_PCS_GOT_use", aeabi_pcs_got}, {18, "Tag_ABI_PCS_wchar_t", aeabi_pcs_wchar_t}, {19, "Tag_ABI_FP_rounding", aeabi_fp_rounding}, {20, "Tag_ABI_FP_denormal", aeabi_fp_denormal}, {21, "Tag_ABI_FP_exceptions", aeabi_fp_exceptions}, {22, "Tag_ABI_FP_user_exceptions", aeabi_fp_user_exceptions}, {23, "Tag_ABI_FP_number_model", aeabi_fp_number_model}, {24, "Tag_ABI_align_needed", aeabi_align_needed}, {25, "Tag_ABI_align_preserved", aeabi_align_preserved}, {26, "Tag_ABI_enum_size", aeabi_enum_size}, {27, "Tag_ABI_HardFP_use", aeabi_hardfp}, {28, "Tag_ABI_VFP_args", aeabi_vfp_args}, {29, "Tag_ABI_WMMX_args", aeabi_wmmx_args}, {30, "Tag_ABI_optimization_goals", aeabi_optm_goal}, {31, "Tag_ABI_FP_optimization_goals", aeabi_fp_optm_goal}, {32, "Tag_compatibility", NULL}, {34, "Tag_CPU_unaligned_access", aeabi_unaligned_access}, {36, "Tag_FP_HP_extension", aeabi_fp_hpext}, {38, "Tag_ABI_FP_16bit_format", aeabi_fp_16bit_format}, {42, "Tag_MPextension_use", aeabi_mpext}, {44, "Tag_DIV_use", aeabi_div}, {64, "Tag_nodefaults", NULL}, {65, "Tag_also_compatible_with", NULL}, {66, "Tag_T2EE_use", aeabi_t2ee}, {67, "Tag_conformance", NULL}, {68, "Tag_Virtualization_use", aeabi_virtual}, {70, "Tag_MPextension_use", aeabi_mpext}, }; static const char * mips_abi_fp(uint64_t fp) { static char s_mips_abi_fp[64]; switch (fp) { case 0: return "N/A"; case 1: return "Hard float (double precision)"; case 2: return "Hard float (single precision)"; case 3: return "Soft float"; case 4: return "64-bit float (-mips32r2 -mfp64)"; default: snprintf(s_mips_abi_fp, sizeof(s_mips_abi_fp), "Unknown(%ju)", (uintmax_t) fp); return (s_mips_abi_fp); } } static const char * ppc_abi_fp(uint64_t fp) { static char s_ppc_abi_fp[64]; switch (fp) { case 0: return "N/A"; case 1: return "Hard float (double precision)"; case 2: return "Soft float"; case 3: return "Hard float (single precision)"; default: snprintf(s_ppc_abi_fp, sizeof(s_ppc_abi_fp), "Unknown(%ju)", (uintmax_t) fp); return (s_ppc_abi_fp); } } static const char * ppc_abi_vector(uint64_t vec) { static char s_vec[64]; switch (vec) { case 0: return "N/A"; case 1: return "Generic purpose registers"; case 2: return "AltiVec registers"; case 3: return "SPE registers"; default: snprintf(s_vec, sizeof(s_vec), "Unknown(%ju)", (uintmax_t) vec); return (s_vec); } } static const char * dwarf_reg(unsigned int mach, unsigned int reg) { switch (mach) { case EM_386: case EM_IAMCU: switch (reg) { case 0: return "eax"; case 1: return "ecx"; case 2: return "edx"; case 3: return "ebx"; case 4: return "esp"; case 5: return "ebp"; case 6: return "esi"; case 7: return "edi"; case 8: return "eip"; case 9: return "eflags"; case 11: return "st0"; case 12: return "st1"; case 13: return "st2"; case 14: return "st3"; case 15: return "st4"; case 16: return "st5"; case 17: return "st6"; case 18: return "st7"; case 21: return "xmm0"; case 22: return "xmm1"; case 23: return "xmm2"; case 24: return "xmm3"; case 25: return "xmm4"; case 26: return "xmm5"; case 27: return "xmm6"; case 28: return "xmm7"; case 29: return "mm0"; case 30: return "mm1"; case 31: return "mm2"; case 32: return "mm3"; case 33: return "mm4"; case 34: return "mm5"; case 35: return "mm6"; case 36: return "mm7"; case 37: return "fcw"; case 38: return "fsw"; case 39: return "mxcsr"; case 40: return "es"; case 41: return "cs"; case 42: return "ss"; case 43: return "ds"; case 44: return "fs"; case 45: return "gs"; case 48: return "tr"; case 49: return "ldtr"; default: return (NULL); } case EM_RISCV: switch (reg) { case 0: return "zero"; case 1: return "ra"; case 2: return "sp"; case 3: return "gp"; case 4: return "tp"; case 5: return "t0"; case 6: return "t1"; case 7: return "t2"; case 8: return "s0"; case 9: return "s1"; case 10: return "a0"; case 11: return "a1"; case 12: return "a2"; case 13: return "a3"; case 14: return "a4"; case 15: return "a5"; case 16: return "a6"; case 17: return "a7"; case 18: return "s2"; case 19: return "s3"; case 20: return "s4"; case 21: return "s5"; case 22: return "s6"; case 23: return "s7"; case 24: return "s8"; case 25: return "s9"; case 26: return "s10"; case 27: return "s11"; case 28: return "t3"; case 29: return "t4"; case 30: return "t5"; case 31: return "t6"; case 32: return "ft0"; case 33: return "ft1"; case 34: return "ft2"; case 35: return "ft3"; case 36: return "ft4"; case 37: return "ft5"; case 38: return "ft6"; case 39: return "ft7"; case 40: return "fs0"; case 41: return "fs1"; case 42: return "fa0"; case 43: return "fa1"; case 44: return "fa2"; case 45: return "fa3"; case 46: return "fa4"; case 47: return "fa5"; case 48: return "fa6"; case 49: return "fa7"; case 50: return "fs2"; case 51: return "fs3"; case 52: return "fs4"; case 53: return "fs5"; case 54: return "fs6"; case 55: return "fs7"; case 56: return "fs8"; case 57: return "fs9"; case 58: return "fs10"; case 59: return "fs11"; case 60: return "ft8"; case 61: return "ft9"; case 62: return "ft10"; case 63: return "ft11"; default: return (NULL); } case EM_X86_64: switch (reg) { case 0: return "rax"; case 1: return "rdx"; case 2: return "rcx"; case 3: return "rbx"; case 4: return "rsi"; case 5: return "rdi"; case 6: return "rbp"; case 7: return "rsp"; case 16: return "rip"; case 17: return "xmm0"; case 18: return "xmm1"; case 19: return "xmm2"; case 20: return "xmm3"; case 21: return "xmm4"; case 22: return "xmm5"; case 23: return "xmm6"; case 24: return "xmm7"; case 25: return "xmm8"; case 26: return "xmm9"; case 27: return "xmm10"; case 28: return "xmm11"; case 29: return "xmm12"; case 30: return "xmm13"; case 31: return "xmm14"; case 32: return "xmm15"; case 33: return "st0"; case 34: return "st1"; case 35: return "st2"; case 36: return "st3"; case 37: return "st4"; case 38: return "st5"; case 39: return "st6"; case 40: return "st7"; case 41: return "mm0"; case 42: return "mm1"; case 43: return "mm2"; case 44: return "mm3"; case 45: return "mm4"; case 46: return "mm5"; case 47: return "mm6"; case 48: return "mm7"; case 49: return "rflags"; case 50: return "es"; case 51: return "cs"; case 52: return "ss"; case 53: return "ds"; case 54: return "fs"; case 55: return "gs"; case 58: return "fs.base"; case 59: return "gs.base"; case 62: return "tr"; case 63: return "ldtr"; case 64: return "mxcsr"; case 65: return "fcw"; case 66: return "fsw"; default: return (NULL); } default: return (NULL); } } static void dump_ehdr(struct readelf *re) { size_t phnum, shnum, shstrndx; int i; printf("ELF Header:\n"); /* e_ident[]. */ printf(" Magic: "); for (i = 0; i < EI_NIDENT; i++) printf("%.2x ", re->ehdr.e_ident[i]); putchar('\n'); /* EI_CLASS. */ printf("%-37s%s\n", " Class:", elf_class(re->ehdr.e_ident[EI_CLASS])); /* EI_DATA. */ printf("%-37s%s\n", " Data:", elf_endian(re->ehdr.e_ident[EI_DATA])); /* EI_VERSION. */ printf("%-37s%d %s\n", " Version:", re->ehdr.e_ident[EI_VERSION], elf_ver(re->ehdr.e_ident[EI_VERSION])); /* EI_OSABI. */ printf("%-37s%s\n", " OS/ABI:", elf_osabi(re->ehdr.e_ident[EI_OSABI])); /* EI_ABIVERSION. */ printf("%-37s%d\n", " ABI Version:", re->ehdr.e_ident[EI_ABIVERSION]); /* e_type. */ printf("%-37s%s\n", " Type:", elf_type(re->ehdr.e_type)); /* e_machine. */ printf("%-37s%s\n", " Machine:", elf_machine(re->ehdr.e_machine)); /* e_version. */ printf("%-37s%#x\n", " Version:", re->ehdr.e_version); /* e_entry. */ printf("%-37s%#jx\n", " Entry point address:", (uintmax_t)re->ehdr.e_entry); /* e_phoff. */ printf("%-37s%ju (bytes into file)\n", " Start of program headers:", (uintmax_t)re->ehdr.e_phoff); /* e_shoff. */ printf("%-37s%ju (bytes into file)\n", " Start of section headers:", (uintmax_t)re->ehdr.e_shoff); /* e_flags. */ printf("%-37s%#x", " Flags:", re->ehdr.e_flags); dump_eflags(re, re->ehdr.e_flags); putchar('\n'); /* e_ehsize. */ printf("%-37s%u (bytes)\n", " Size of this header:", re->ehdr.e_ehsize); /* e_phentsize. */ printf("%-37s%u (bytes)\n", " Size of program headers:", re->ehdr.e_phentsize); /* e_phnum. */ printf("%-37s%u", " Number of program headers:", re->ehdr.e_phnum); if (re->ehdr.e_phnum == PN_XNUM) { /* Extended program header numbering is in use. */ if (elf_getphnum(re->elf, &phnum)) printf(" (%zu)", phnum); } putchar('\n'); /* e_shentsize. */ printf("%-37s%u (bytes)\n", " Size of section headers:", re->ehdr.e_shentsize); /* e_shnum. */ printf("%-37s%u", " Number of section headers:", re->ehdr.e_shnum); if (re->ehdr.e_shnum == SHN_UNDEF) { /* Extended section numbering is in use. */ if (elf_getshnum(re->elf, &shnum)) printf(" (%ju)", (uintmax_t)shnum); } putchar('\n'); /* e_shstrndx. */ printf("%-37s%u", " Section header string table index:", re->ehdr.e_shstrndx); if (re->ehdr.e_shstrndx == SHN_XINDEX) { /* Extended section numbering is in use. */ if (elf_getshstrndx(re->elf, &shstrndx)) printf(" (%ju)", (uintmax_t)shstrndx); } putchar('\n'); } static void dump_eflags(struct readelf *re, uint64_t e_flags) { struct eflags_desc *edesc; int arm_eabi; edesc = NULL; switch (re->ehdr.e_machine) { case EM_ARM: arm_eabi = (e_flags & EF_ARM_EABIMASK) >> 24; if (arm_eabi == 0) printf(", GNU EABI"); else if (arm_eabi <= 5) printf(", Version%d EABI", arm_eabi); edesc = arm_eflags_desc; break; case EM_MIPS: case EM_MIPS_RS3_LE: switch ((e_flags & EF_MIPS_ARCH) >> 28) { case 0: printf(", mips1"); break; case 1: printf(", mips2"); break; case 2: printf(", mips3"); break; case 3: printf(", mips4"); break; case 4: printf(", mips5"); break; case 5: printf(", mips32"); break; case 6: printf(", mips64"); break; case 7: printf(", mips32r2"); break; case 8: printf(", mips64r2"); break; default: break; } switch ((e_flags & 0x00FF0000) >> 16) { case 0x81: printf(", 3900"); break; case 0x82: printf(", 4010"); break; case 0x83: printf(", 4100"); break; case 0x85: printf(", 4650"); break; case 0x87: printf(", 4120"); break; case 0x88: printf(", 4111"); break; case 0x8a: printf(", sb1"); break; case 0x8b: printf(", octeon"); break; case 0x8c: printf(", xlr"); break; case 0x91: printf(", 5400"); break; case 0x98: printf(", 5500"); break; case 0x99: printf(", 9000"); break; case 0xa0: printf(", loongson-2e"); break; case 0xa1: printf(", loongson-2f"); break; default: break; } switch ((e_flags & 0x0000F000) >> 12) { case 1: printf(", o32"); break; case 2: printf(", o64"); break; case 3: printf(", eabi32"); break; case 4: printf(", eabi64"); break; default: break; } edesc = mips_eflags_desc; break; case EM_PPC64: switch (e_flags) { case 0: printf(", Unspecified or Power ELF V1 ABI"); break; case 1: printf(", Power ELF V1 ABI"); break; case 2: printf(", OpenPOWER ELF V2 ABI"); break; default: break; } /* FALLTHROUGH */ case EM_PPC: edesc = powerpc_eflags_desc; break; case EM_RISCV: switch (e_flags & EF_RISCV_FLOAT_ABI_MASK) { case EF_RISCV_FLOAT_ABI_SOFT: printf(", soft-float ABI"); break; case EF_RISCV_FLOAT_ABI_SINGLE: printf(", single-float ABI"); break; case EF_RISCV_FLOAT_ABI_DOUBLE: printf(", double-float ABI"); break; case EF_RISCV_FLOAT_ABI_QUAD: printf(", quad-float ABI"); break; } edesc = riscv_eflags_desc; break; case EM_SPARC: case EM_SPARC32PLUS: case EM_SPARCV9: switch ((e_flags & EF_SPARCV9_MM)) { case EF_SPARCV9_TSO: printf(", tso"); break; case EF_SPARCV9_PSO: printf(", pso"); break; case EF_SPARCV9_MM: printf(", rmo"); break; default: break; } edesc = sparc_eflags_desc; break; default: break; } if (edesc != NULL) { while (edesc->desc != NULL) { if (e_flags & edesc->flag) printf(", %s", edesc->desc); edesc++; } } } static void dump_phdr(struct readelf *re) { const char *rawfile; GElf_Phdr phdr; size_t phnum, size; int i, j; #define PH_HDR "Type", "Offset", "VirtAddr", "PhysAddr", "FileSiz", \ "MemSiz", "Flg", "Align" #define PH_CT phdr_type(re->ehdr.e_machine, phdr.p_type), \ (uintmax_t)phdr.p_offset, (uintmax_t)phdr.p_vaddr, \ (uintmax_t)phdr.p_paddr, (uintmax_t)phdr.p_filesz, \ (uintmax_t)phdr.p_memsz, \ phdr.p_flags & PF_R ? 'R' : ' ', \ phdr.p_flags & PF_W ? 'W' : ' ', \ phdr.p_flags & PF_X ? 'E' : ' ', \ (uintmax_t)phdr.p_align if (elf_getphnum(re->elf, &phnum) == 0) { warnx("elf_getphnum failed: %s", elf_errmsg(-1)); return; } if (phnum == 0) { printf("\nThere are no program headers in this file.\n"); return; } printf("\nElf file type is %s", elf_type(re->ehdr.e_type)); printf("\nEntry point 0x%jx\n", (uintmax_t)re->ehdr.e_entry); printf("There are %ju program headers, starting at offset %ju\n", (uintmax_t)phnum, (uintmax_t)re->ehdr.e_phoff); /* Dump program headers. */ printf("\nProgram Headers:\n"); if (re->ec == ELFCLASS32) printf(" %-15s%-9s%-11s%-11s%-8s%-8s%-4s%s\n", PH_HDR); else if (re->options & RE_WW) printf(" %-15s%-9s%-19s%-19s%-9s%-9s%-4s%s\n", PH_HDR); else printf(" %-15s%-19s%-19s%s\n %-19s%-20s" "%-7s%s\n", PH_HDR); for (i = 0; (size_t) i < phnum; i++) { if (gelf_getphdr(re->elf, i, &phdr) != &phdr) { warnx("gelf_getphdr failed: %s", elf_errmsg(-1)); continue; } /* TODO: Add arch-specific segment type dump. */ if (re->ec == ELFCLASS32) printf(" %-14.14s 0x%6.6jx 0x%8.8jx 0x%8.8jx " "0x%5.5jx 0x%5.5jx %c%c%c %#jx\n", PH_CT); else if (re->options & RE_WW) printf(" %-14.14s 0x%6.6jx 0x%16.16jx 0x%16.16jx " "0x%6.6jx 0x%6.6jx %c%c%c %#jx\n", PH_CT); else printf(" %-14.14s 0x%16.16jx 0x%16.16jx 0x%16.16jx\n" " 0x%16.16jx 0x%16.16jx %c%c%c" " %#jx\n", PH_CT); if (phdr.p_type == PT_INTERP) { if ((rawfile = elf_rawfile(re->elf, &size)) == NULL) { warnx("elf_rawfile failed: %s", elf_errmsg(-1)); continue; } if (phdr.p_offset >= size) { warnx("invalid program header offset"); continue; } printf(" [Requesting program interpreter: %s]\n", rawfile + phdr.p_offset); } } /* Dump section to segment mapping. */ if (re->shnum == 0) return; printf("\n Section to Segment mapping:\n"); printf(" Segment Sections...\n"); for (i = 0; (size_t)i < phnum; i++) { if (gelf_getphdr(re->elf, i, &phdr) != &phdr) { warnx("gelf_getphdr failed: %s", elf_errmsg(-1)); continue; } printf(" %2.2d ", i); /* skip NULL section. */ for (j = 1; (size_t)j < re->shnum; j++) { if (re->sl[j].off < phdr.p_offset) continue; if (re->sl[j].off + re->sl[j].sz > phdr.p_offset + phdr.p_filesz && re->sl[j].type != SHT_NOBITS) continue; if (re->sl[j].addr < phdr.p_vaddr || re->sl[j].addr + re->sl[j].sz > phdr.p_vaddr + phdr.p_memsz) continue; if (phdr.p_type == PT_TLS && (re->sl[j].flags & SHF_TLS) == 0) continue; printf("%s ", re->sl[j].name); } printf("\n"); } #undef PH_HDR #undef PH_CT } static char * section_flags(struct readelf *re, struct section *s) { #define BUF_SZ 256 static char buf[BUF_SZ]; int i, p, nb; p = 0; nb = re->ec == ELFCLASS32 ? 8 : 16; if (re->options & RE_T) { snprintf(buf, BUF_SZ, "[%*.*jx]: ", nb, nb, (uintmax_t)s->flags); p += nb + 4; } for (i = 0; section_flag[i].ln != NULL; i++) { if ((s->flags & section_flag[i].value) == 0) continue; if (re->options & RE_T) { snprintf(&buf[p], BUF_SZ - p, "%s, ", section_flag[i].ln); p += strlen(section_flag[i].ln) + 2; } else buf[p++] = section_flag[i].sn; } if (re->options & RE_T && p > nb + 4) p -= 2; buf[p] = '\0'; return (buf); } static void dump_shdr(struct readelf *re) { struct section *s; int i; #define S_HDR "[Nr] Name", "Type", "Addr", "Off", "Size", "ES", \ "Flg", "Lk", "Inf", "Al" #define S_HDRL "[Nr] Name", "Type", "Address", "Offset", "Size", \ "EntSize", "Flags", "Link", "Info", "Align" #define ST_HDR "[Nr] Name", "Type", "Addr", "Off", "Size", "ES", \ "Lk", "Inf", "Al", "Flags" #define ST_HDRL "[Nr] Name", "Type", "Address", "Offset", "Link", \ "Size", "EntSize", "Info", "Align", "Flags" #define S_CT i, s->name, section_type(re->ehdr.e_machine, s->type), \ (uintmax_t)s->addr, (uintmax_t)s->off, (uintmax_t)s->sz,\ (uintmax_t)s->entsize, section_flags(re, s), \ s->link, s->info, (uintmax_t)s->align #define ST_CT i, s->name, section_type(re->ehdr.e_machine, s->type), \ (uintmax_t)s->addr, (uintmax_t)s->off, (uintmax_t)s->sz,\ (uintmax_t)s->entsize, s->link, s->info, \ (uintmax_t)s->align, section_flags(re, s) #define ST_CTL i, s->name, section_type(re->ehdr.e_machine, s->type), \ (uintmax_t)s->addr, (uintmax_t)s->off, s->link, \ (uintmax_t)s->sz, (uintmax_t)s->entsize, s->info, \ (uintmax_t)s->align, section_flags(re, s) if (re->shnum == 0) { printf("\nThere are no sections in this file.\n"); return; } printf("There are %ju section headers, starting at offset 0x%jx:\n", (uintmax_t)re->shnum, (uintmax_t)re->ehdr.e_shoff); printf("\nSection Headers:\n"); if (re->ec == ELFCLASS32) { if (re->options & RE_T) printf(" %s\n %-16s%-9s%-7s%-7s%-5s%-3s%-4s%s\n" "%12s\n", ST_HDR); else printf(" %-23s%-16s%-9s%-7s%-7s%-3s%-4s%-3s%-4s%s\n", S_HDR); } else if (re->options & RE_WW) { if (re->options & RE_T) printf(" %s\n %-16s%-17s%-7s%-7s%-5s%-3s%-4s%s\n" "%12s\n", ST_HDR); else printf(" %-23s%-16s%-17s%-7s%-7s%-3s%-4s%-3s%-4s%s\n", S_HDR); } else { if (re->options & RE_T) printf(" %s\n %-18s%-17s%-18s%s\n %-18s" "%-17s%-18s%s\n%12s\n", ST_HDRL); else printf(" %-23s%-17s%-18s%s\n %-18s%-17s%-7s%" "-6s%-6s%s\n", S_HDRL); } for (i = 0; (size_t)i < re->shnum; i++) { s = &re->sl[i]; if (re->ec == ELFCLASS32) { if (re->options & RE_T) printf(" [%2d] %s\n %-15.15s %8.8jx" " %6.6jx %6.6jx %2.2jx %2u %3u %2ju\n" " %s\n", ST_CT); else printf(" [%2d] %-17.17s %-15.15s %8.8jx" " %6.6jx %6.6jx %2.2jx %3s %2u %3u %2ju\n", S_CT); } else if (re->options & RE_WW) { if (re->options & RE_T) printf(" [%2d] %s\n %-15.15s %16.16jx" " %6.6jx %6.6jx %2.2jx %2u %3u %2ju\n" " %s\n", ST_CT); else printf(" [%2d] %-17.17s %-15.15s %16.16jx" " %6.6jx %6.6jx %2.2jx %3s %2u %3u %2ju\n", S_CT); } else { if (re->options & RE_T) printf(" [%2d] %s\n %-15.15s %16.16jx" " %16.16jx %u\n %16.16jx %16.16jx" " %-16u %ju\n %s\n", ST_CTL); else printf(" [%2d] %-17.17s %-15.15s %16.16jx" " %8.8jx\n %16.16jx %16.16jx " "%3s %2u %3u %ju\n", S_CT); } } if ((re->options & RE_T) == 0) printf("Key to Flags:\n W (write), A (alloc)," " X (execute), M (merge), S (strings)\n" " I (info), L (link order), G (group), x (unknown)\n" " O (extra OS processing required)" " o (OS specific), p (processor specific)\n"); #undef S_HDR #undef S_HDRL #undef ST_HDR #undef ST_HDRL #undef S_CT #undef ST_CT #undef ST_CTL } /* * Return number of entries in the given section. We'd prefer ent_count be a * size_t *, but libelf APIs already use int for section indices. */ static int get_ent_count(struct section *s, int *ent_count) { if (s->entsize == 0) { warnx("section %s has entry size 0", s->name); return (0); } else if (s->sz / s->entsize > INT_MAX) { warnx("section %s has invalid section count", s->name); return (0); } *ent_count = (int)(s->sz / s->entsize); return (1); } static void dump_dynamic(struct readelf *re) { GElf_Dyn dyn; Elf_Data *d; struct section *s; int elferr, i, is_dynamic, j, jmax, nentries; is_dynamic = 0; for (i = 0; (size_t)i < re->shnum; i++) { s = &re->sl[i]; if (s->type != SHT_DYNAMIC) continue; (void) elf_errno(); if ((d = elf_getdata(s->scn, NULL)) == NULL) { elferr = elf_errno(); if (elferr != 0) warnx("elf_getdata failed: %s", elf_errmsg(-1)); continue; } if (d->d_size <= 0) continue; is_dynamic = 1; /* Determine the actual number of table entries. */ nentries = 0; if (!get_ent_count(s, &jmax)) continue; for (j = 0; j < jmax; j++) { if (gelf_getdyn(d, j, &dyn) != &dyn) { warnx("gelf_getdyn failed: %s", elf_errmsg(-1)); continue; } nentries ++; if (dyn.d_tag == DT_NULL) break; } printf("\nDynamic section at offset 0x%jx", (uintmax_t)s->off); printf(" contains %u entries:\n", nentries); if (re->ec == ELFCLASS32) printf("%5s%12s%28s\n", "Tag", "Type", "Name/Value"); else printf("%5s%20s%28s\n", "Tag", "Type", "Name/Value"); for (j = 0; j < nentries; j++) { if (gelf_getdyn(d, j, &dyn) != &dyn) continue; /* Dump dynamic entry type. */ if (re->ec == ELFCLASS32) printf(" 0x%8.8jx", (uintmax_t)dyn.d_tag); else printf(" 0x%16.16jx", (uintmax_t)dyn.d_tag); printf(" %-20s", dt_type(re->ehdr.e_machine, dyn.d_tag)); /* Dump dynamic entry value. */ dump_dyn_val(re, &dyn, s->link); } } if (!is_dynamic) printf("\nThere is no dynamic section in this file.\n"); } static char * timestamp(time_t ti) { static char ts[32]; struct tm *t; t = gmtime(&ti); snprintf(ts, sizeof(ts), "%04d-%02d-%02dT%02d:%02d:%02d", t->tm_year + 1900, t->tm_mon + 1, t->tm_mday, t->tm_hour, t->tm_min, t->tm_sec); return (ts); } static const char * dyn_str(struct readelf *re, uint32_t stab, uint64_t d_val) { const char *name; if (stab == SHN_UNDEF) name = "ERROR"; else if ((name = elf_strptr(re->elf, stab, d_val)) == NULL) { (void) elf_errno(); /* clear error */ name = "ERROR"; } return (name); } static void dump_arch_dyn_val(struct readelf *re, GElf_Dyn *dyn) { switch (re->ehdr.e_machine) { case EM_MIPS: case EM_MIPS_RS3_LE: switch (dyn->d_tag) { case DT_MIPS_RLD_VERSION: case DT_MIPS_LOCAL_GOTNO: case DT_MIPS_CONFLICTNO: case DT_MIPS_LIBLISTNO: case DT_MIPS_SYMTABNO: case DT_MIPS_UNREFEXTNO: case DT_MIPS_GOTSYM: case DT_MIPS_HIPAGENO: case DT_MIPS_DELTA_CLASS_NO: case DT_MIPS_DELTA_INSTANCE_NO: case DT_MIPS_DELTA_RELOC_NO: case DT_MIPS_DELTA_SYM_NO: case DT_MIPS_DELTA_CLASSSYM_NO: case DT_MIPS_LOCALPAGE_GOTIDX: case DT_MIPS_LOCAL_GOTIDX: case DT_MIPS_HIDDEN_GOTIDX: case DT_MIPS_PROTECTED_GOTIDX: printf(" %ju\n", (uintmax_t) dyn->d_un.d_val); break; case DT_MIPS_ICHECKSUM: case DT_MIPS_FLAGS: case DT_MIPS_BASE_ADDRESS: case DT_MIPS_CONFLICT: case DT_MIPS_LIBLIST: case DT_MIPS_RLD_MAP: case DT_MIPS_DELTA_CLASS: case DT_MIPS_DELTA_INSTANCE: case DT_MIPS_DELTA_RELOC: case DT_MIPS_DELTA_SYM: case DT_MIPS_DELTA_CLASSSYM: case DT_MIPS_CXX_FLAGS: case DT_MIPS_PIXIE_INIT: case DT_MIPS_SYMBOL_LIB: case DT_MIPS_OPTIONS: case DT_MIPS_INTERFACE: case DT_MIPS_DYNSTR_ALIGN: case DT_MIPS_INTERFACE_SIZE: case DT_MIPS_RLD_TEXT_RESOLVE_ADDR: case DT_MIPS_COMPACT_SIZE: case DT_MIPS_GP_VALUE: case DT_MIPS_AUX_DYNAMIC: case DT_MIPS_PLTGOT: case DT_MIPS_RLD_OBJ_UPDATE: case DT_MIPS_RWPLT: printf(" 0x%jx\n", (uintmax_t) dyn->d_un.d_val); break; case DT_MIPS_IVERSION: case DT_MIPS_PERF_SUFFIX: case DT_MIPS_TIME_STAMP: printf(" %s\n", timestamp(dyn->d_un.d_val)); break; default: printf("\n"); break; } break; default: printf("\n"); break; } } static void dump_flags(struct flag_desc *desc, uint64_t val) { struct flag_desc *fd; for (fd = desc; fd->flag != 0; fd++) { if (val & fd->flag) { val &= ~fd->flag; printf(" %s", fd->desc); } } if (val != 0) printf(" unknown (0x%jx)", (uintmax_t)val); printf("\n"); } static struct flag_desc dt_flags[] = { { DF_ORIGIN, "ORIGIN" }, { DF_SYMBOLIC, "SYMBOLIC" }, { DF_TEXTREL, "TEXTREL" }, { DF_BIND_NOW, "BIND_NOW" }, { DF_STATIC_TLS, "STATIC_TLS" }, { 0, NULL } }; static struct flag_desc dt_flags_1[] = { { DF_1_BIND_NOW, "NOW" }, { DF_1_GLOBAL, "GLOBAL" }, { 0x4, "GROUP" }, { DF_1_NODELETE, "NODELETE" }, { DF_1_LOADFLTR, "LOADFLTR" }, { 0x20, "INITFIRST" }, { DF_1_NOOPEN, "NOOPEN" }, { DF_1_ORIGIN, "ORIGIN" }, { 0x100, "DIRECT" }, { DF_1_INTERPOSE, "INTERPOSE" }, { DF_1_NODEFLIB, "NODEFLIB" }, { 0x1000, "NODUMP" }, { 0x2000, "CONFALT" }, { 0x4000, "ENDFILTEE" }, { 0x8000, "DISPRELDNE" }, { 0x10000, "DISPRELPND" }, { 0x20000, "NODIRECT" }, { 0x40000, "IGNMULDEF" }, { 0x80000, "NOKSYMS" }, { 0x100000, "NOHDR" }, { 0x200000, "EDITED" }, { 0x400000, "NORELOC" }, { 0x800000, "SYMINTPOSE" }, { 0x1000000, "GLOBAUDIT" }, { 0, NULL } }; static void dump_dyn_val(struct readelf *re, GElf_Dyn *dyn, uint32_t stab) { const char *name; if (dyn->d_tag >= DT_LOPROC && dyn->d_tag <= DT_HIPROC && dyn->d_tag != DT_AUXILIARY && dyn->d_tag != DT_FILTER) { dump_arch_dyn_val(re, dyn); return; } /* These entry values are index into the string table. */ name = NULL; if (dyn->d_tag == DT_AUXILIARY || dyn->d_tag == DT_FILTER || dyn->d_tag == DT_NEEDED || dyn->d_tag == DT_SONAME || dyn->d_tag == DT_RPATH || dyn->d_tag == DT_RUNPATH) name = dyn_str(re, stab, dyn->d_un.d_val); switch(dyn->d_tag) { case DT_NULL: case DT_PLTGOT: case DT_HASH: case DT_STRTAB: case DT_SYMTAB: case DT_RELA: case DT_INIT: case DT_SYMBOLIC: case DT_REL: case DT_DEBUG: case DT_TEXTREL: case DT_JMPREL: case DT_FINI: case DT_VERDEF: case DT_VERNEED: case DT_VERSYM: case DT_GNU_HASH: case DT_GNU_LIBLIST: case DT_GNU_CONFLICT: printf(" 0x%jx\n", (uintmax_t) dyn->d_un.d_val); break; case DT_PLTRELSZ: case DT_RELASZ: case DT_RELAENT: case DT_STRSZ: case DT_SYMENT: case DT_RELSZ: case DT_RELENT: case DT_PREINIT_ARRAYSZ: case DT_INIT_ARRAYSZ: case DT_FINI_ARRAYSZ: case DT_GNU_CONFLICTSZ: case DT_GNU_LIBLISTSZ: printf(" %ju (bytes)\n", (uintmax_t) dyn->d_un.d_val); break; case DT_RELACOUNT: case DT_RELCOUNT: case DT_VERDEFNUM: case DT_VERNEEDNUM: printf(" %ju\n", (uintmax_t) dyn->d_un.d_val); break; case DT_AUXILIARY: printf(" Auxiliary library: [%s]\n", name); break; case DT_FILTER: printf(" Filter library: [%s]\n", name); break; case DT_NEEDED: printf(" Shared library: [%s]\n", name); break; case DT_SONAME: printf(" Library soname: [%s]\n", name); break; case DT_RPATH: printf(" Library rpath: [%s]\n", name); break; case DT_RUNPATH: printf(" Library runpath: [%s]\n", name); break; case DT_PLTREL: printf(" %s\n", dt_type(re->ehdr.e_machine, dyn->d_un.d_val)); break; case DT_GNU_PRELINKED: printf(" %s\n", timestamp(dyn->d_un.d_val)); break; case DT_FLAGS: dump_flags(dt_flags, dyn->d_un.d_val); break; case DT_FLAGS_1: dump_flags(dt_flags_1, dyn->d_un.d_val); break; default: printf("\n"); } } static void dump_rel(struct readelf *re, struct section *s, Elf_Data *d) { GElf_Rel r; const char *symname; uint64_t symval; int i, len; uint32_t type; uint8_t type2, type3; if (s->link >= re->shnum) return; #define REL_HDR "r_offset", "r_info", "r_type", "st_value", "st_name" #define REL_CT32 (uintmax_t)r.r_offset, (uintmax_t)r.r_info, \ elftc_reloc_type_str(re->ehdr.e_machine, \ ELF32_R_TYPE(r.r_info)), (uintmax_t)symval, symname #define REL_CT64 (uintmax_t)r.r_offset, (uintmax_t)r.r_info, \ elftc_reloc_type_str(re->ehdr.e_machine, type), \ (uintmax_t)symval, symname printf("\nRelocation section (%s):\n", s->name); if (re->ec == ELFCLASS32) printf("%-8s %-8s %-19s %-8s %s\n", REL_HDR); else { if (re->options & RE_WW) printf("%-16s %-16s %-24s %-16s %s\n", REL_HDR); else printf("%-12s %-12s %-19s %-16s %s\n", REL_HDR); } assert(d->d_size == s->sz); if (!get_ent_count(s, &len)) return; for (i = 0; i < len; i++) { if (gelf_getrel(d, i, &r) != &r) { warnx("gelf_getrel failed: %s", elf_errmsg(-1)); continue; } symname = get_symbol_name(re, s->link, GELF_R_SYM(r.r_info)); symval = get_symbol_value(re, s->link, GELF_R_SYM(r.r_info)); if (re->ec == ELFCLASS32) { r.r_info = ELF32_R_INFO(ELF64_R_SYM(r.r_info), ELF64_R_TYPE(r.r_info)); printf("%8.8jx %8.8jx %-19.19s %8.8jx %s\n", REL_CT32); } else { type = ELF64_R_TYPE(r.r_info); if (re->ehdr.e_machine == EM_MIPS) { type2 = (type >> 8) & 0xFF; type3 = (type >> 16) & 0xFF; type = type & 0xFF; } else { type2 = type3 = 0; } if (re->options & RE_WW) printf("%16.16jx %16.16jx %-24.24s" " %16.16jx %s\n", REL_CT64); else printf("%12.12jx %12.12jx %-19.19s" " %16.16jx %s\n", REL_CT64); if (re->ehdr.e_machine == EM_MIPS) { if (re->options & RE_WW) { printf("%32s: %s\n", "Type2", elftc_reloc_type_str(EM_MIPS, type2)); printf("%32s: %s\n", "Type3", elftc_reloc_type_str(EM_MIPS, type3)); } else { printf("%24s: %s\n", "Type2", elftc_reloc_type_str(EM_MIPS, type2)); printf("%24s: %s\n", "Type3", elftc_reloc_type_str(EM_MIPS, type3)); } } } } #undef REL_HDR #undef REL_CT } static void dump_rela(struct readelf *re, struct section *s, Elf_Data *d) { GElf_Rela r; const char *symname; uint64_t symval; int i, len; uint32_t type; uint8_t type2, type3; if (s->link >= re->shnum) return; #define RELA_HDR "r_offset", "r_info", "r_type", "st_value", \ "st_name + r_addend" #define RELA_CT32 (uintmax_t)r.r_offset, (uintmax_t)r.r_info, \ elftc_reloc_type_str(re->ehdr.e_machine, \ ELF32_R_TYPE(r.r_info)), (uintmax_t)symval, symname #define RELA_CT64 (uintmax_t)r.r_offset, (uintmax_t)r.r_info, \ elftc_reloc_type_str(re->ehdr.e_machine, type), \ (uintmax_t)symval, symname printf("\nRelocation section with addend (%s):\n", s->name); if (re->ec == ELFCLASS32) printf("%-8s %-8s %-19s %-8s %s\n", RELA_HDR); else { if (re->options & RE_WW) printf("%-16s %-16s %-24s %-16s %s\n", RELA_HDR); else printf("%-12s %-12s %-19s %-16s %s\n", RELA_HDR); } assert(d->d_size == s->sz); if (!get_ent_count(s, &len)) return; for (i = 0; i < len; i++) { if (gelf_getrela(d, i, &r) != &r) { warnx("gelf_getrel failed: %s", elf_errmsg(-1)); continue; } symname = get_symbol_name(re, s->link, GELF_R_SYM(r.r_info)); symval = get_symbol_value(re, s->link, GELF_R_SYM(r.r_info)); if (re->ec == ELFCLASS32) { r.r_info = ELF32_R_INFO(ELF64_R_SYM(r.r_info), ELF64_R_TYPE(r.r_info)); printf("%8.8jx %8.8jx %-19.19s %8.8jx %s", RELA_CT32); printf(" + %x\n", (uint32_t) r.r_addend); } else { type = ELF64_R_TYPE(r.r_info); if (re->ehdr.e_machine == EM_MIPS) { type2 = (type >> 8) & 0xFF; type3 = (type >> 16) & 0xFF; type = type & 0xFF; } else { type2 = type3 = 0; } if (re->options & RE_WW) printf("%16.16jx %16.16jx %-24.24s" " %16.16jx %s", RELA_CT64); else printf("%12.12jx %12.12jx %-19.19s" " %16.16jx %s", RELA_CT64); printf(" + %jx\n", (uintmax_t) r.r_addend); if (re->ehdr.e_machine == EM_MIPS) { if (re->options & RE_WW) { printf("%32s: %s\n", "Type2", elftc_reloc_type_str(EM_MIPS, type2)); printf("%32s: %s\n", "Type3", elftc_reloc_type_str(EM_MIPS, type3)); } else { printf("%24s: %s\n", "Type2", elftc_reloc_type_str(EM_MIPS, type2)); printf("%24s: %s\n", "Type3", elftc_reloc_type_str(EM_MIPS, type3)); } } } } #undef RELA_HDR #undef RELA_CT } static void dump_reloc(struct readelf *re) { struct section *s; Elf_Data *d; int i, elferr; for (i = 0; (size_t)i < re->shnum; i++) { s = &re->sl[i]; if (s->type == SHT_REL || s->type == SHT_RELA) { (void) elf_errno(); if ((d = elf_getdata(s->scn, NULL)) == NULL) { elferr = elf_errno(); if (elferr != 0) warnx("elf_getdata failed: %s", elf_errmsg(elferr)); continue; } if (s->type == SHT_REL) dump_rel(re, s, d); else dump_rela(re, s, d); } } } static void dump_symtab(struct readelf *re, int i) { struct section *s; Elf_Data *d; GElf_Sym sym; const char *name; uint32_t stab; int elferr, j, len; uint16_t vs; s = &re->sl[i]; if (s->link >= re->shnum) return; stab = s->link; (void) elf_errno(); if ((d = elf_getdata(s->scn, NULL)) == NULL) { elferr = elf_errno(); if (elferr != 0) warnx("elf_getdata failed: %s", elf_errmsg(elferr)); return; } if (d->d_size <= 0) return; if (!get_ent_count(s, &len)) return; printf("Symbol table (%s)", s->name); printf(" contains %d entries:\n", len); printf("%7s%9s%14s%5s%8s%6s%9s%5s\n", "Num:", "Value", "Size", "Type", "Bind", "Vis", "Ndx", "Name"); for (j = 0; j < len; j++) { if (gelf_getsym(d, j, &sym) != &sym) { warnx("gelf_getsym failed: %s", elf_errmsg(-1)); continue; } printf("%6d:", j); printf(" %16.16jx", (uintmax_t) sym.st_value); printf(" %5ju", (uintmax_t) sym.st_size); printf(" %-7s", st_type(re->ehdr.e_machine, re->ehdr.e_ident[EI_OSABI], GELF_ST_TYPE(sym.st_info))); printf(" %-6s", st_bind(GELF_ST_BIND(sym.st_info))); printf(" %-8s", st_vis(GELF_ST_VISIBILITY(sym.st_other))); printf(" %3s", st_shndx(sym.st_shndx)); if ((name = elf_strptr(re->elf, stab, sym.st_name)) != NULL) printf(" %s", name); /* Append symbol version string for SHT_DYNSYM symbol table. */ if (s->type == SHT_DYNSYM && re->ver != NULL && re->vs != NULL && re->vs[j] > 1) { vs = re->vs[j] & VERSYM_VERSION; if (vs >= re->ver_sz || re->ver[vs].name == NULL) { warnx("invalid versym version index %u", vs); break; } if (re->vs[j] & VERSYM_HIDDEN || re->ver[vs].type == 0) printf("@%s (%d)", re->ver[vs].name, vs); else printf("@@%s (%d)", re->ver[vs].name, vs); } putchar('\n'); } } static void dump_symtabs(struct readelf *re) { GElf_Dyn dyn; Elf_Data *d; struct section *s; uint64_t dyn_off; int elferr, i, len; /* * If -D is specified, only dump the symbol table specified by * the DT_SYMTAB entry in the .dynamic section. */ dyn_off = 0; if (re->options & RE_DD) { s = NULL; for (i = 0; (size_t)i < re->shnum; i++) if (re->sl[i].type == SHT_DYNAMIC) { s = &re->sl[i]; break; } if (s == NULL) return; (void) elf_errno(); if ((d = elf_getdata(s->scn, NULL)) == NULL) { elferr = elf_errno(); if (elferr != 0) warnx("elf_getdata failed: %s", elf_errmsg(-1)); return; } if (d->d_size <= 0) return; if (!get_ent_count(s, &len)) return; for (i = 0; i < len; i++) { if (gelf_getdyn(d, i, &dyn) != &dyn) { warnx("gelf_getdyn failed: %s", elf_errmsg(-1)); continue; } if (dyn.d_tag == DT_SYMTAB) { dyn_off = dyn.d_un.d_val; break; } } } /* Find and dump symbol tables. */ for (i = 0; (size_t)i < re->shnum; i++) { s = &re->sl[i]; if (s->type == SHT_SYMTAB || s->type == SHT_DYNSYM) { if (re->options & RE_DD) { if (dyn_off == s->addr) { dump_symtab(re, i); break; } } else dump_symtab(re, i); } } } static void dump_svr4_hash(struct section *s) { Elf_Data *d; uint32_t *buf; uint32_t nbucket, nchain; uint32_t *bucket, *chain; uint32_t *bl, *c, maxl, total; int elferr, i, j; /* Read and parse the content of .hash section. */ (void) elf_errno(); if ((d = elf_getdata(s->scn, NULL)) == NULL) { elferr = elf_errno(); if (elferr != 0) warnx("elf_getdata failed: %s", elf_errmsg(elferr)); return; } if (d->d_size < 2 * sizeof(uint32_t)) { warnx(".hash section too small"); return; } buf = d->d_buf; nbucket = buf[0]; nchain = buf[1]; if (nbucket <= 0 || nchain <= 0) { warnx("Malformed .hash section"); return; } if (d->d_size != (nbucket + nchain + 2) * sizeof(uint32_t)) { warnx("Malformed .hash section"); return; } bucket = &buf[2]; chain = &buf[2 + nbucket]; maxl = 0; if ((bl = calloc(nbucket, sizeof(*bl))) == NULL) errx(EXIT_FAILURE, "calloc failed"); for (i = 0; (uint32_t)i < nbucket; i++) for (j = bucket[i]; j > 0 && (uint32_t)j < nchain; j = chain[j]) if (++bl[i] > maxl) maxl = bl[i]; if ((c = calloc(maxl + 1, sizeof(*c))) == NULL) errx(EXIT_FAILURE, "calloc failed"); for (i = 0; (uint32_t)i < nbucket; i++) c[bl[i]]++; printf("\nHistogram for bucket list length (total of %u buckets):\n", nbucket); printf(" Length\tNumber\t\t%% of total\tCoverage\n"); total = 0; for (i = 0; (uint32_t)i <= maxl; i++) { total += c[i] * i; printf("%7u\t%-10u\t(%5.1f%%)\t%5.1f%%\n", i, c[i], c[i] * 100.0 / nbucket, total * 100.0 / (nchain - 1)); } free(c); free(bl); } static void dump_svr4_hash64(struct readelf *re, struct section *s) { Elf_Data *d, dst; uint64_t *buf; uint64_t nbucket, nchain; uint64_t *bucket, *chain; uint64_t *bl, *c, maxl, total; int elferr, i, j; /* * ALPHA uses 64-bit hash entries. Since libelf assumes that * .hash section contains only 32-bit entry, an explicit * gelf_xlatetom is needed here. */ (void) elf_errno(); if ((d = elf_rawdata(s->scn, NULL)) == NULL) { elferr = elf_errno(); if (elferr != 0) warnx("elf_rawdata failed: %s", elf_errmsg(elferr)); return; } d->d_type = ELF_T_XWORD; memcpy(&dst, d, sizeof(Elf_Data)); if (gelf_xlatetom(re->elf, &dst, d, re->ehdr.e_ident[EI_DATA]) != &dst) { warnx("gelf_xlatetom failed: %s", elf_errmsg(-1)); return; } if (dst.d_size < 2 * sizeof(uint64_t)) { warnx(".hash section too small"); return; } buf = dst.d_buf; nbucket = buf[0]; nchain = buf[1]; if (nbucket <= 0 || nchain <= 0) { warnx("Malformed .hash section"); return; } if (d->d_size != (nbucket + nchain + 2) * sizeof(uint32_t)) { warnx("Malformed .hash section"); return; } bucket = &buf[2]; chain = &buf[2 + nbucket]; maxl = 0; if ((bl = calloc(nbucket, sizeof(*bl))) == NULL) errx(EXIT_FAILURE, "calloc failed"); for (i = 0; (uint32_t)i < nbucket; i++) for (j = bucket[i]; j > 0 && (uint32_t)j < nchain; j = chain[j]) if (++bl[i] > maxl) maxl = bl[i]; if ((c = calloc(maxl + 1, sizeof(*c))) == NULL) errx(EXIT_FAILURE, "calloc failed"); for (i = 0; (uint64_t)i < nbucket; i++) c[bl[i]]++; printf("Histogram for bucket list length (total of %ju buckets):\n", (uintmax_t)nbucket); printf(" Length\tNumber\t\t%% of total\tCoverage\n"); total = 0; for (i = 0; (uint64_t)i <= maxl; i++) { total += c[i] * i; printf("%7u\t%-10ju\t(%5.1f%%)\t%5.1f%%\n", i, (uintmax_t)c[i], c[i] * 100.0 / nbucket, total * 100.0 / (nchain - 1)); } free(c); free(bl); } static void dump_gnu_hash(struct readelf *re, struct section *s) { struct section *ds; Elf_Data *d; uint32_t *buf; uint32_t *bucket, *chain; uint32_t nbucket, nchain, symndx, maskwords; uint32_t *bl, *c, maxl, total; int elferr, dynsymcount, i, j; (void) elf_errno(); if ((d = elf_getdata(s->scn, NULL)) == NULL) { elferr = elf_errno(); if (elferr != 0) warnx("elf_getdata failed: %s", elf_errmsg(elferr)); return; } if (d->d_size < 4 * sizeof(uint32_t)) { warnx(".gnu.hash section too small"); return; } buf = d->d_buf; nbucket = buf[0]; symndx = buf[1]; maskwords = buf[2]; buf += 4; if (s->link >= re->shnum) return; ds = &re->sl[s->link]; if (!get_ent_count(ds, &dynsymcount)) return; if (symndx >= (uint32_t)dynsymcount) { warnx("Malformed .gnu.hash section (symndx out of range)"); return; } nchain = dynsymcount - symndx; if (d->d_size != 4 * sizeof(uint32_t) + maskwords * (re->ec == ELFCLASS32 ? sizeof(uint32_t) : sizeof(uint64_t)) + (nbucket + nchain) * sizeof(uint32_t)) { warnx("Malformed .gnu.hash section"); return; } bucket = buf + (re->ec == ELFCLASS32 ? maskwords : maskwords * 2); chain = bucket + nbucket; maxl = 0; if ((bl = calloc(nbucket, sizeof(*bl))) == NULL) errx(EXIT_FAILURE, "calloc failed"); for (i = 0; (uint32_t)i < nbucket; i++) for (j = bucket[i]; j > 0 && (uint32_t)j - symndx < nchain; j++) { if (++bl[i] > maxl) maxl = bl[i]; if (chain[j - symndx] & 1) break; } if ((c = calloc(maxl + 1, sizeof(*c))) == NULL) errx(EXIT_FAILURE, "calloc failed"); for (i = 0; (uint32_t)i < nbucket; i++) c[bl[i]]++; printf("Histogram for bucket list length (total of %u buckets):\n", nbucket); printf(" Length\tNumber\t\t%% of total\tCoverage\n"); total = 0; for (i = 0; (uint32_t)i <= maxl; i++) { total += c[i] * i; printf("%7u\t%-10u\t(%5.1f%%)\t%5.1f%%\n", i, c[i], c[i] * 100.0 / nbucket, total * 100.0 / (nchain - 1)); } free(c); free(bl); } static struct flag_desc gnu_property_x86_feature_1_and_bits[] = { { GNU_PROPERTY_X86_FEATURE_1_IBT, "IBT" }, { GNU_PROPERTY_X86_FEATURE_1_SHSTK, "SHSTK" }, { 0, NULL } }; static void dump_gnu_property_type_0(struct readelf *re, const char *buf, size_t sz) { size_t i; uint32_t type, prop_sz; printf(" Properties: "); while (sz > 0) { if (sz < 8) goto bad; type = *(const uint32_t *)(const void *)buf; prop_sz = *(const uint32_t *)(const void *)(buf + 4); buf += 8; sz -= 8; if (prop_sz > sz) goto bad; if (type >= GNU_PROPERTY_LOPROC && type <= GNU_PROPERTY_HIPROC) { if (re->ehdr.e_machine != EM_X86_64) { printf("machine type %x unknown\n", re->ehdr.e_machine); goto unknown; } switch (type) { case GNU_PROPERTY_X86_FEATURE_1_AND: printf("x86 features:"); if (prop_sz != 4) goto bad; dump_flags(gnu_property_x86_feature_1_and_bits, *(const uint32_t *)(const void *)buf); break; } } buf += roundup2(prop_sz, 8); sz -= roundup2(prop_sz, 8); } return; bad: printf("corrupt GNU property\n"); unknown: printf("remaining description data:"); for (i = 0; i < sz; i++) printf(" %02x", (unsigned char)buf[i]); printf("\n"); } static void dump_hash(struct readelf *re) { struct section *s; int i; for (i = 0; (size_t) i < re->shnum; i++) { s = &re->sl[i]; if (s->type == SHT_HASH || s->type == SHT_GNU_HASH) { if (s->type == SHT_GNU_HASH) dump_gnu_hash(re, s); else if (re->ehdr.e_machine == EM_ALPHA && s->entsize == 8) dump_svr4_hash64(re, s); else dump_svr4_hash(s); } } } static void dump_notes(struct readelf *re) { struct section *s; const char *rawfile; GElf_Phdr phdr; Elf_Data *d; size_t filesize, phnum; int i, elferr; if (re->ehdr.e_type == ET_CORE) { /* * Search program headers in the core file for * PT_NOTE entry. */ if (elf_getphnum(re->elf, &phnum) == 0) { warnx("elf_getphnum failed: %s", elf_errmsg(-1)); return; } if (phnum == 0) return; if ((rawfile = elf_rawfile(re->elf, &filesize)) == NULL) { warnx("elf_rawfile failed: %s", elf_errmsg(-1)); return; } for (i = 0; (size_t) i < phnum; i++) { if (gelf_getphdr(re->elf, i, &phdr) != &phdr) { warnx("gelf_getphdr failed: %s", elf_errmsg(-1)); continue; } if (phdr.p_type == PT_NOTE) { if (phdr.p_offset >= filesize || phdr.p_filesz > filesize - phdr.p_offset) { warnx("invalid PHDR offset"); continue; } dump_notes_content(re, rawfile + phdr.p_offset, phdr.p_filesz, phdr.p_offset); } } } else { /* * For objects other than core files, Search for * SHT_NOTE sections. */ for (i = 0; (size_t) i < re->shnum; i++) { s = &re->sl[i]; if (s->type == SHT_NOTE) { (void) elf_errno(); if ((d = elf_getdata(s->scn, NULL)) == NULL) { elferr = elf_errno(); if (elferr != 0) warnx("elf_getdata failed: %s", elf_errmsg(elferr)); continue; } dump_notes_content(re, d->d_buf, d->d_size, s->off); } } } } static struct flag_desc note_feature_ctl_flags[] = { { NT_FREEBSD_FCTL_ASLR_DISABLE, "ASLR_DISABLE" }, { NT_FREEBSD_FCTL_PROTMAX_DISABLE, "PROTMAX_DISABLE" }, { NT_FREEBSD_FCTL_STKGAP_DISABLE, "STKGAP_DISABLE" }, { NT_FREEBSD_FCTL_WXNEEDED, "WXNEEDED" }, { 0, NULL } }; static void dump_note_string(const char *description, const char *s, size_t len) { size_t i; int printable = 1; if (len == 0 || s[--len] != '\0') { printable = 0; } else { for (i = 0; i < len; i++) { if (!isprint(s[i])) { printable = 0; break; } } } if (printable) { printf(" %s: %s\n", description, s); } else { printf(" description data:"); for (i = 0; i < len; i++) printf(" %02x", (unsigned char)s[i]); printf("\n"); } } static void dump_notes_data(struct readelf *re, const char *name, uint32_t type, const char *buf, size_t sz) { size_t i; const uint32_t *ubuf; /* Note data is at least 4-byte aligned. */ if (((uintptr_t)buf & 3) != 0) { warnx("bad note data alignment"); goto unknown; } ubuf = (const uint32_t *)(const void *)buf; if (strcmp(name, "FreeBSD") == 0) { switch (type) { case NT_FREEBSD_ABI_TAG: if (sz != 4) goto unknown; printf(" ABI tag: %u\n", ubuf[0]); return; /* NT_FREEBSD_NOINIT_TAG carries no data, treat as unknown. */ case NT_FREEBSD_ARCH_TAG: if (sz != 4) goto unknown; printf(" Arch tag: %x\n", ubuf[0]); return; case NT_FREEBSD_FEATURE_CTL: if (sz != 4) goto unknown; printf(" Features:"); dump_flags(note_feature_ctl_flags, ubuf[0]); return; } } else if (strcmp(name, "GNU") == 0) { switch (type) { case NT_GNU_PROPERTY_TYPE_0: dump_gnu_property_type_0(re, buf, sz); return; + case NT_GNU_BUILD_ID: + printf(" Build ID: "); + for (i = 0; i < sz; i++) + printf("%02x", (unsigned char)buf[i]); + printf("\n"); + return; } } else if (strcmp(name, "Xen") == 0) { switch (type) { case 5: dump_note_string("Xen version", buf, sz); return; case 6: dump_note_string("Guest OS", buf, sz); return; case 7: dump_note_string("Guest version", buf, sz); return; case 8: dump_note_string("Loader", buf, sz); return; case 9: dump_note_string("PAE mode", buf, sz); return; case 10: dump_note_string("Features", buf, sz); return; case 11: dump_note_string("BSD symtab", buf, sz); return; } } unknown: printf(" description data:"); for (i = 0; i < sz; i++) printf(" %02x", (unsigned char)buf[i]); printf("\n"); } static void dump_notes_content(struct readelf *re, const char *buf, size_t sz, off_t off) { Elf_Note *note; const char *end, *name; uint32_t namesz, descsz; printf("\nNotes at offset %#010jx with length %#010jx:\n", (uintmax_t) off, (uintmax_t) sz); printf(" %-13s %-15s %s\n", "Owner", "Data size", "Description"); end = buf + sz; while (buf < end) { if (buf + sizeof(*note) > end) { warnx("invalid note header"); return; } note = (Elf_Note *)(uintptr_t) buf; namesz = roundup2(note->n_namesz, 4); descsz = roundup2(note->n_descsz, 4); if (namesz < note->n_namesz || descsz < note->n_descsz || buf + namesz + descsz > end) { warnx("invalid note header"); return; } buf += sizeof(Elf_Note); name = buf; buf += namesz; /* * The name field is required to be nul-terminated, and * n_namesz includes the terminating nul in observed * implementations (contrary to the ELF-64 spec). A special * case is needed for cores generated by some older Linux * versions, which write a note named "CORE" without a nul * terminator and n_namesz = 4. */ if (note->n_namesz == 0) name = ""; else if (note->n_namesz == 4 && strncmp(name, "CORE", 4) == 0) name = "CORE"; else if (strnlen(name, note->n_namesz) >= note->n_namesz) name = ""; printf(" %-13s %#010jx", name, (uintmax_t) note->n_descsz); printf(" %s\n", note_type(name, re->ehdr.e_type, note->n_type)); dump_notes_data(re, name, note->n_type, buf, note->n_descsz); buf += descsz; } } /* * Symbol versioning sections are the same for 32bit and 64bit * ELF objects. */ #define Elf_Verdef Elf32_Verdef #define Elf_Verdaux Elf32_Verdaux #define Elf_Verneed Elf32_Verneed #define Elf_Vernaux Elf32_Vernaux #define SAVE_VERSION_NAME(x, n, t) \ do { \ while (x >= re->ver_sz) { \ nv = realloc(re->ver, \ sizeof(*re->ver) * re->ver_sz * 2); \ if (nv == NULL) { \ warn("realloc failed"); \ free(re->ver); \ return; \ } \ re->ver = nv; \ for (i = re->ver_sz; i < re->ver_sz * 2; i++) { \ re->ver[i].name = NULL; \ re->ver[i].type = 0; \ } \ re->ver_sz *= 2; \ } \ if (x > 1) { \ re->ver[x].name = n; \ re->ver[x].type = t; \ } \ } while (0) static void dump_verdef(struct readelf *re, int dump) { struct section *s; struct symver *nv; Elf_Data *d; Elf_Verdef *vd; Elf_Verdaux *vda; uint8_t *buf, *end, *buf2; const char *name; int elferr, i, j; if ((s = re->vd_s) == NULL) return; if (s->link >= re->shnum) return; if (re->ver == NULL) { re->ver_sz = 16; if ((re->ver = calloc(re->ver_sz, sizeof(*re->ver))) == NULL) { warn("calloc failed"); return; } re->ver[0].name = "*local*"; re->ver[1].name = "*global*"; } if (dump) printf("\nVersion definition section (%s):\n", s->name); (void) elf_errno(); if ((d = elf_getdata(s->scn, NULL)) == NULL) { elferr = elf_errno(); if (elferr != 0) warnx("elf_getdata failed: %s", elf_errmsg(elferr)); return; } if (d->d_size == 0) return; buf = d->d_buf; end = buf + d->d_size; while (buf + sizeof(Elf_Verdef) <= end) { vd = (Elf_Verdef *) (uintptr_t) buf; if (dump) { printf(" 0x%4.4lx", (unsigned long) (buf - (uint8_t *)d->d_buf)); printf(" vd_version: %u vd_flags: %d" " vd_ndx: %u vd_cnt: %u", vd->vd_version, vd->vd_flags, vd->vd_ndx, vd->vd_cnt); } buf2 = buf + vd->vd_aux; j = 0; while (buf2 + sizeof(Elf_Verdaux) <= end && j < vd->vd_cnt) { vda = (Elf_Verdaux *) (uintptr_t) buf2; name = get_string(re, s->link, vda->vda_name); if (j == 0) { if (dump) printf(" vda_name: %s\n", name); SAVE_VERSION_NAME((int)vd->vd_ndx, name, 1); } else if (dump) printf(" 0x%4.4lx parent: %s\n", (unsigned long) (buf2 - (uint8_t *)d->d_buf), name); if (vda->vda_next == 0) break; buf2 += vda->vda_next; j++; } if (vd->vd_next == 0) break; buf += vd->vd_next; } } static void dump_verneed(struct readelf *re, int dump) { struct section *s; struct symver *nv; Elf_Data *d; Elf_Verneed *vn; Elf_Vernaux *vna; uint8_t *buf, *end, *buf2; const char *name; int elferr, i, j; if ((s = re->vn_s) == NULL) return; if (s->link >= re->shnum) return; if (re->ver == NULL) { re->ver_sz = 16; if ((re->ver = calloc(re->ver_sz, sizeof(*re->ver))) == NULL) { warn("calloc failed"); return; } re->ver[0].name = "*local*"; re->ver[1].name = "*global*"; } if (dump) printf("\nVersion needed section (%s):\n", s->name); (void) elf_errno(); if ((d = elf_getdata(s->scn, NULL)) == NULL) { elferr = elf_errno(); if (elferr != 0) warnx("elf_getdata failed: %s", elf_errmsg(elferr)); return; } if (d->d_size == 0) return; buf = d->d_buf; end = buf + d->d_size; while (buf + sizeof(Elf_Verneed) <= end) { vn = (Elf_Verneed *) (uintptr_t) buf; if (dump) { printf(" 0x%4.4lx", (unsigned long) (buf - (uint8_t *)d->d_buf)); printf(" vn_version: %u vn_file: %s vn_cnt: %u\n", vn->vn_version, get_string(re, s->link, vn->vn_file), vn->vn_cnt); } buf2 = buf + vn->vn_aux; j = 0; while (buf2 + sizeof(Elf_Vernaux) <= end && j < vn->vn_cnt) { vna = (Elf32_Vernaux *) (uintptr_t) buf2; if (dump) printf(" 0x%4.4lx", (unsigned long) (buf2 - (uint8_t *)d->d_buf)); name = get_string(re, s->link, vna->vna_name); if (dump) printf(" vna_name: %s vna_flags: %u" " vna_other: %u\n", name, vna->vna_flags, vna->vna_other); SAVE_VERSION_NAME((int)vna->vna_other, name, 0); if (vna->vna_next == 0) break; buf2 += vna->vna_next; j++; } if (vn->vn_next == 0) break; buf += vn->vn_next; } } static void dump_versym(struct readelf *re) { int i; uint16_t vs; if (re->vs_s == NULL || re->ver == NULL || re->vs == NULL) return; printf("\nVersion symbol section (%s):\n", re->vs_s->name); for (i = 0; i < re->vs_sz; i++) { if ((i & 3) == 0) { if (i > 0) putchar('\n'); printf(" %03x:", i); } vs = re->vs[i] & VERSYM_VERSION; if (vs >= re->ver_sz || re->ver[vs].name == NULL) { warnx("invalid versym version index %u", re->vs[i]); break; } if (re->vs[i] & VERSYM_HIDDEN) printf(" %3xh %-12s ", vs, re->ver[re->vs[i] & VERSYM_VERSION].name); else printf(" %3x %-12s ", vs, re->ver[re->vs[i]].name); } putchar('\n'); } static void dump_ver(struct readelf *re) { if (re->vs_s && re->ver && re->vs) dump_versym(re); if (re->vd_s) dump_verdef(re, 1); if (re->vn_s) dump_verneed(re, 1); } static void search_ver(struct readelf *re) { struct section *s; Elf_Data *d; int elferr, i; for (i = 0; (size_t) i < re->shnum; i++) { s = &re->sl[i]; if (s->type == SHT_SUNW_versym) re->vs_s = s; if (s->type == SHT_SUNW_verneed) re->vn_s = s; if (s->type == SHT_SUNW_verdef) re->vd_s = s; } if (re->vd_s) dump_verdef(re, 0); if (re->vn_s) dump_verneed(re, 0); if (re->vs_s && re->ver != NULL) { (void) elf_errno(); if ((d = elf_getdata(re->vs_s->scn, NULL)) == NULL) { elferr = elf_errno(); if (elferr != 0) warnx("elf_getdata failed: %s", elf_errmsg(elferr)); return; } if (d->d_size == 0) return; re->vs = d->d_buf; re->vs_sz = d->d_size / sizeof(Elf32_Half); } } #undef Elf_Verdef #undef Elf_Verdaux #undef Elf_Verneed #undef Elf_Vernaux #undef SAVE_VERSION_NAME /* * Elf32_Lib and Elf64_Lib are identical. */ #define Elf_Lib Elf32_Lib static void dump_liblist(struct readelf *re) { struct section *s; struct tm *t; time_t ti; char tbuf[20]; Elf_Data *d; Elf_Lib *lib; int i, j, k, elferr, first, len; for (i = 0; (size_t) i < re->shnum; i++) { s = &re->sl[i]; if (s->type != SHT_GNU_LIBLIST) continue; if (s->link >= re->shnum) continue; (void) elf_errno(); if ((d = elf_getdata(s->scn, NULL)) == NULL) { elferr = elf_errno(); if (elferr != 0) warnx("elf_getdata failed: %s", elf_errmsg(elferr)); continue; } if (d->d_size <= 0) continue; lib = d->d_buf; if (!get_ent_count(s, &len)) continue; printf("\nLibrary list section '%s' ", s->name); printf("contains %d entries:\n", len); printf("%12s%24s%18s%10s%6s\n", "Library", "Time Stamp", "Checksum", "Version", "Flags"); for (j = 0; (uint64_t) j < s->sz / s->entsize; j++) { printf("%3d: ", j); printf("%-20.20s ", get_string(re, s->link, lib->l_name)); ti = lib->l_time_stamp; t = gmtime(&ti); snprintf(tbuf, sizeof(tbuf), "%04d-%02d-%02dT%02d:%02d" ":%2d", t->tm_year + 1900, t->tm_mon + 1, t->tm_mday, t->tm_hour, t->tm_min, t->tm_sec); printf("%-19.19s ", tbuf); printf("0x%08x ", lib->l_checksum); printf("%-7d %#x", lib->l_version, lib->l_flags); if (lib->l_flags != 0) { first = 1; putchar('('); for (k = 0; l_flag[k].name != NULL; k++) { if ((l_flag[k].value & lib->l_flags) == 0) continue; if (!first) putchar(','); else first = 0; printf("%s", l_flag[k].name); } putchar(')'); } putchar('\n'); lib++; } } } #undef Elf_Lib static void dump_section_groups(struct readelf *re) { struct section *s; const char *symname; Elf_Data *d; uint32_t *w; int i, j, elferr; size_t n; for (i = 0; (size_t) i < re->shnum; i++) { s = &re->sl[i]; if (s->type != SHT_GROUP) continue; if (s->link >= re->shnum) continue; (void) elf_errno(); if ((d = elf_getdata(s->scn, NULL)) == NULL) { elferr = elf_errno(); if (elferr != 0) warnx("elf_getdata failed: %s", elf_errmsg(elferr)); continue; } if (d->d_size <= 0) continue; w = d->d_buf; /* We only support COMDAT section. */ #ifndef GRP_COMDAT #define GRP_COMDAT 0x1 #endif if ((*w++ & GRP_COMDAT) == 0) return; if (s->entsize == 0) s->entsize = 4; symname = get_symbol_name(re, s->link, s->info); n = s->sz / s->entsize; if (n-- < 1) return; printf("\nCOMDAT group section [%5d] `%s' [%s] contains %ju" " sections:\n", i, s->name, symname, (uintmax_t)n); printf(" %-10.10s %s\n", "[Index]", "Name"); for (j = 0; (size_t) j < n; j++, w++) { if (*w >= re->shnum) { warnx("invalid section index: %u", *w); continue; } printf(" [%5u] %s\n", *w, re->sl[*w].name); } } } static uint8_t * dump_unknown_tag(uint64_t tag, uint8_t *p, uint8_t *pe) { uint64_t val; /* * According to ARM EABI: For tags > 32, even numbered tags have * a ULEB128 param and odd numbered ones have NUL-terminated * string param. This rule probably also applies for tags <= 32 * if the object arch is not ARM. */ printf(" Tag_unknown_%ju: ", (uintmax_t) tag); if (tag & 1) { printf("%s\n", (char *) p); p += strlen((char *) p) + 1; } else { val = _decode_uleb128(&p, pe); printf("%ju\n", (uintmax_t) val); } return (p); } static uint8_t * dump_compatibility_tag(uint8_t *p, uint8_t *pe) { uint64_t val; val = _decode_uleb128(&p, pe); printf("flag = %ju, vendor = %s\n", (uintmax_t) val, p); p += strlen((char *) p) + 1; return (p); } static void dump_arm_attributes(struct readelf *re, uint8_t *p, uint8_t *pe) { uint64_t tag, val; size_t i; int found, desc; (void) re; while (p < pe) { tag = _decode_uleb128(&p, pe); found = desc = 0; for (i = 0; i < sizeof(aeabi_tags) / sizeof(aeabi_tags[0]); i++) { if (tag == aeabi_tags[i].tag) { found = 1; printf(" %s: ", aeabi_tags[i].s_tag); if (aeabi_tags[i].get_desc) { desc = 1; val = _decode_uleb128(&p, pe); printf("%s\n", aeabi_tags[i].get_desc(val)); } break; } if (tag < aeabi_tags[i].tag) break; } if (!found) { p = dump_unknown_tag(tag, p, pe); continue; } if (desc) continue; switch (tag) { case 4: /* Tag_CPU_raw_name */ case 5: /* Tag_CPU_name */ case 67: /* Tag_conformance */ printf("%s\n", (char *) p); p += strlen((char *) p) + 1; break; case 32: /* Tag_compatibility */ p = dump_compatibility_tag(p, pe); break; case 64: /* Tag_nodefaults */ /* ignored, written as 0. */ (void) _decode_uleb128(&p, pe); printf("True\n"); break; case 65: /* Tag_also_compatible_with */ val = _decode_uleb128(&p, pe); /* Must be Tag_CPU_arch */ if (val != 6) { printf("unknown\n"); break; } val = _decode_uleb128(&p, pe); printf("%s\n", aeabi_cpu_arch(val)); /* Skip NUL terminator. */ p++; break; default: putchar('\n'); break; } } } #ifndef Tag_GNU_MIPS_ABI_FP #define Tag_GNU_MIPS_ABI_FP 4 #endif static void dump_mips_attributes(struct readelf *re, uint8_t *p, uint8_t *pe) { uint64_t tag, val; (void) re; while (p < pe) { tag = _decode_uleb128(&p, pe); switch (tag) { case Tag_GNU_MIPS_ABI_FP: val = _decode_uleb128(&p, pe); printf(" Tag_GNU_MIPS_ABI_FP: %s\n", mips_abi_fp(val)); break; case 32: /* Tag_compatibility */ p = dump_compatibility_tag(p, pe); break; default: p = dump_unknown_tag(tag, p, pe); break; } } } #ifndef Tag_GNU_Power_ABI_FP #define Tag_GNU_Power_ABI_FP 4 #endif #ifndef Tag_GNU_Power_ABI_Vector #define Tag_GNU_Power_ABI_Vector 8 #endif static void dump_ppc_attributes(uint8_t *p, uint8_t *pe) { uint64_t tag, val; while (p < pe) { tag = _decode_uleb128(&p, pe); switch (tag) { case Tag_GNU_Power_ABI_FP: val = _decode_uleb128(&p, pe); printf(" Tag_GNU_Power_ABI_FP: %s\n", ppc_abi_fp(val)); break; case Tag_GNU_Power_ABI_Vector: val = _decode_uleb128(&p, pe); printf(" Tag_GNU_Power_ABI_Vector: %s\n", ppc_abi_vector(val)); break; case 32: /* Tag_compatibility */ p = dump_compatibility_tag(p, pe); break; default: p = dump_unknown_tag(tag, p, pe); break; } } } static void dump_attributes(struct readelf *re) { struct section *s; Elf_Data *d; uint8_t *p, *pe, *sp; size_t len, seclen, nlen, sublen; uint64_t val; int tag, i, elferr; for (i = 0; (size_t) i < re->shnum; i++) { s = &re->sl[i]; if (s->type != SHT_GNU_ATTRIBUTES && (re->ehdr.e_machine != EM_ARM || s->type != SHT_LOPROC + 3)) continue; (void) elf_errno(); if ((d = elf_rawdata(s->scn, NULL)) == NULL) { elferr = elf_errno(); if (elferr != 0) warnx("elf_rawdata failed: %s", elf_errmsg(elferr)); continue; } if (d->d_size <= 0) continue; p = d->d_buf; pe = p + d->d_size; if (*p != 'A') { printf("Unknown Attribute Section Format: %c\n", (char) *p); continue; } len = d->d_size - 1; p++; while (len > 0) { if (len < 4) { warnx("truncated attribute section length"); return; } seclen = re->dw_decode(&p, 4); if (seclen > len) { warnx("invalid attribute section length"); return; } len -= seclen; nlen = strlen((char *) p) + 1; if (nlen + 4 > seclen) { warnx("invalid attribute section name"); return; } printf("Attribute Section: %s\n", (char *) p); p += nlen; seclen -= nlen + 4; while (seclen > 0) { sp = p; tag = *p++; sublen = re->dw_decode(&p, 4); if (sublen > seclen) { warnx("invalid attribute sub-section" " length"); return; } seclen -= sublen; printf("%s", top_tag(tag)); if (tag == 2 || tag == 3) { putchar(':'); for (;;) { val = _decode_uleb128(&p, pe); if (val == 0) break; printf(" %ju", (uintmax_t) val); } } putchar('\n'); if (re->ehdr.e_machine == EM_ARM && s->type == SHT_LOPROC + 3) dump_arm_attributes(re, p, sp + sublen); else if (re->ehdr.e_machine == EM_MIPS || re->ehdr.e_machine == EM_MIPS_RS3_LE) dump_mips_attributes(re, p, sp + sublen); else if (re->ehdr.e_machine == EM_PPC) dump_ppc_attributes(p, sp + sublen); p = sp + sublen; } } } } static void dump_mips_specific_info(struct readelf *re) { struct section *s; int i; s = NULL; for (i = 0; (size_t) i < re->shnum; i++) { s = &re->sl[i]; if (s->name != NULL && (!strcmp(s->name, ".MIPS.options") || (s->type == SHT_MIPS_OPTIONS))) { dump_mips_options(re, s); } } if (s->name != NULL && (!strcmp(s->name, ".MIPS.abiflags") || (s->type == SHT_MIPS_ABIFLAGS))) dump_mips_abiflags(re, s); /* * Dump .reginfo if present (although it will be ignored by an OS if a * .MIPS.options section is present, according to SGI mips64 spec). */ for (i = 0; (size_t) i < re->shnum; i++) { s = &re->sl[i]; if (s->name != NULL && (!strcmp(s->name, ".reginfo") || (s->type == SHT_MIPS_REGINFO))) dump_mips_reginfo(re, s); } } static void dump_mips_abiflags(struct readelf *re, struct section *s) { Elf_Data *d; uint8_t *p; int elferr; uint32_t isa_ext, ases, flags1, flags2; uint16_t version; uint8_t isa_level, isa_rev, gpr_size, cpr1_size, cpr2_size, fp_abi; if ((d = elf_rawdata(s->scn, NULL)) == NULL) { elferr = elf_errno(); if (elferr != 0) warnx("elf_rawdata failed: %s", elf_errmsg(elferr)); return; } if (d->d_size != 24) { warnx("invalid MIPS abiflags section size"); return; } p = d->d_buf; version = re->dw_decode(&p, 2); printf("MIPS ABI Flags Version: %u", version); if (version != 0) { printf(" (unknown)\n\n"); return; } printf("\n\n"); isa_level = re->dw_decode(&p, 1); isa_rev = re->dw_decode(&p, 1); gpr_size = re->dw_decode(&p, 1); cpr1_size = re->dw_decode(&p, 1); cpr2_size = re->dw_decode(&p, 1); fp_abi = re->dw_decode(&p, 1); isa_ext = re->dw_decode(&p, 4); ases = re->dw_decode(&p, 4); flags1 = re->dw_decode(&p, 4); flags2 = re->dw_decode(&p, 4); printf("ISA: "); if (isa_rev <= 1) printf("MIPS%u\n", isa_level); else printf("MIPS%ur%u\n", isa_level, isa_rev); printf("GPR size: %d\n", get_mips_register_size(gpr_size)); printf("CPR1 size: %d\n", get_mips_register_size(cpr1_size)); printf("CPR2 size: %d\n", get_mips_register_size(cpr2_size)); printf("FP ABI: "); switch (fp_abi) { case 3: printf("Soft float"); break; default: printf("%u", fp_abi); break; } printf("\nISA Extension: %u\n", isa_ext); printf("ASEs: %u\n", ases); printf("FLAGS 1: %08x\n", flags1); printf("FLAGS 2: %08x\n", flags2); } static int get_mips_register_size(uint8_t flag) { switch (flag) { case 0: return 0; case 1: return 32; case 2: return 64; case 3: return 128; default: return -1; } } static void dump_mips_reginfo(struct readelf *re, struct section *s) { Elf_Data *d; int elferr, len; (void) elf_errno(); if ((d = elf_rawdata(s->scn, NULL)) == NULL) { elferr = elf_errno(); if (elferr != 0) warnx("elf_rawdata failed: %s", elf_errmsg(elferr)); return; } if (d->d_size <= 0) return; if (!get_ent_count(s, &len)) return; printf("\nSection '%s' contains %d entries:\n", s->name, len); dump_mips_odk_reginfo(re, d->d_buf, d->d_size); } static void dump_mips_options(struct readelf *re, struct section *s) { Elf_Data *d; uint32_t info; uint16_t sndx; uint8_t *p, *pe; uint8_t kind, size; int elferr; (void) elf_errno(); if ((d = elf_rawdata(s->scn, NULL)) == NULL) { elferr = elf_errno(); if (elferr != 0) warnx("elf_rawdata failed: %s", elf_errmsg(elferr)); return; } if (d->d_size == 0) return; printf("\nSection %s contains:\n", s->name); p = d->d_buf; pe = p + d->d_size; while (p < pe) { if (pe - p < 8) { warnx("Truncated MIPS option header"); return; } kind = re->dw_decode(&p, 1); size = re->dw_decode(&p, 1); sndx = re->dw_decode(&p, 2); info = re->dw_decode(&p, 4); if (size < 8 || size - 8 > pe - p) { warnx("Malformed MIPS option header"); return; } size -= 8; switch (kind) { case ODK_REGINFO: dump_mips_odk_reginfo(re, p, size); break; case ODK_EXCEPTIONS: printf(" EXCEPTIONS FPU_MIN: %#x\n", info & OEX_FPU_MIN); printf("%11.11s FPU_MAX: %#x\n", "", info & OEX_FPU_MAX); dump_mips_option_flags("", mips_exceptions_option, info); break; case ODK_PAD: printf(" %-10.10s section: %ju\n", "OPAD", (uintmax_t) sndx); dump_mips_option_flags("", mips_pad_option, info); break; case ODK_HWPATCH: dump_mips_option_flags("HWPATCH", mips_hwpatch_option, info); break; case ODK_HWAND: dump_mips_option_flags("HWAND", mips_hwa_option, info); break; case ODK_HWOR: dump_mips_option_flags("HWOR", mips_hwo_option, info); break; case ODK_FILL: printf(" %-10.10s %#jx\n", "FILL", (uintmax_t) info); break; case ODK_TAGS: printf(" %-10.10s\n", "TAGS"); break; case ODK_GP_GROUP: printf(" %-10.10s GP group number: %#x\n", "GP_GROUP", info & 0xFFFF); if (info & 0x10000) printf(" %-10.10s GP group is " "self-contained\n", ""); break; case ODK_IDENT: printf(" %-10.10s default GP group number: %#x\n", "IDENT", info & 0xFFFF); if (info & 0x10000) printf(" %-10.10s default GP group is " "self-contained\n", ""); break; case ODK_PAGESIZE: printf(" %-10.10s\n", "PAGESIZE"); break; default: break; } p += size; } } static void dump_mips_option_flags(const char *name, struct mips_option *opt, uint64_t info) { int first; first = 1; for (; opt->desc != NULL; opt++) { if (info & opt->flag) { printf(" %-10.10s %s\n", first ? name : "", opt->desc); first = 0; } } } static void dump_mips_odk_reginfo(struct readelf *re, uint8_t *p, size_t sz) { uint32_t ri_gprmask; uint32_t ri_cprmask[4]; uint64_t ri_gp_value; uint8_t *pe; int i; pe = p + sz; while (p < pe) { ri_gprmask = re->dw_decode(&p, 4); /* Skip ri_pad padding field for mips64. */ if (re->ec == ELFCLASS64) re->dw_decode(&p, 4); for (i = 0; i < 4; i++) ri_cprmask[i] = re->dw_decode(&p, 4); if (re->ec == ELFCLASS32) ri_gp_value = re->dw_decode(&p, 4); else ri_gp_value = re->dw_decode(&p, 8); printf(" %s ", option_kind(ODK_REGINFO)); printf("ri_gprmask: 0x%08jx\n", (uintmax_t) ri_gprmask); for (i = 0; i < 4; i++) printf("%11.11s ri_cprmask[%d]: 0x%08jx\n", "", i, (uintmax_t) ri_cprmask[i]); printf("%12.12s", ""); printf("ri_gp_value: %#jx\n", (uintmax_t) ri_gp_value); } } static void dump_arch_specific_info(struct readelf *re) { dump_liblist(re); dump_attributes(re); switch (re->ehdr.e_machine) { case EM_MIPS: case EM_MIPS_RS3_LE: dump_mips_specific_info(re); default: break; } } static const char * dwarf_regname(struct readelf *re, unsigned int num) { static char rx[32]; const char *rn; if ((rn = dwarf_reg(re->ehdr.e_machine, num)) != NULL) return (rn); snprintf(rx, sizeof(rx), "r%u", num); return (rx); } static void dump_dwarf_line(struct readelf *re) { struct section *s; Dwarf_Die die; Dwarf_Error de; Dwarf_Half tag, version, pointer_size; Dwarf_Unsigned offset, endoff, length, hdrlen, dirndx, mtime, fsize; Dwarf_Small minlen, defstmt, lrange, opbase, oplen; Elf_Data *d; char *pn; uint64_t address, file, line, column, isa, opsize, udelta; int64_t sdelta; uint8_t *p, *pe; int8_t lbase; int i, is_stmt, dwarf_size, elferr, ret; printf("\nDump of debug contents of section .debug_line:\n"); s = NULL; for (i = 0; (size_t) i < re->shnum; i++) { s = &re->sl[i]; if (s->name != NULL && !strcmp(s->name, ".debug_line")) break; } if ((size_t) i >= re->shnum) return; (void) elf_errno(); if ((d = elf_getdata(s->scn, NULL)) == NULL) { elferr = elf_errno(); if (elferr != 0) warnx("elf_getdata failed: %s", elf_errmsg(-1)); return; } if (d->d_size <= 0) return; while ((ret = dwarf_next_cu_header(re->dbg, NULL, NULL, NULL, NULL, NULL, &de)) == DW_DLV_OK) { die = NULL; while (dwarf_siblingof(re->dbg, die, &die, &de) == DW_DLV_OK) { if (dwarf_tag(die, &tag, &de) != DW_DLV_OK) { warnx("dwarf_tag failed: %s", dwarf_errmsg(de)); return; } /* XXX: What about DW_TAG_partial_unit? */ if (tag == DW_TAG_compile_unit) break; } if (die == NULL) { warnx("could not find DW_TAG_compile_unit die"); return; } if (dwarf_attrval_unsigned(die, DW_AT_stmt_list, &offset, &de) != DW_DLV_OK) continue; length = re->dw_read(d, &offset, 4); if (length == 0xffffffff) { dwarf_size = 8; length = re->dw_read(d, &offset, 8); } else dwarf_size = 4; if (length > d->d_size - offset) { warnx("invalid .dwarf_line section"); continue; } endoff = offset + length; pe = (uint8_t *) d->d_buf + endoff; version = re->dw_read(d, &offset, 2); hdrlen = re->dw_read(d, &offset, dwarf_size); minlen = re->dw_read(d, &offset, 1); defstmt = re->dw_read(d, &offset, 1); lbase = re->dw_read(d, &offset, 1); lrange = re->dw_read(d, &offset, 1); opbase = re->dw_read(d, &offset, 1); printf("\n"); printf(" Length:\t\t\t%ju\n", (uintmax_t) length); printf(" DWARF version:\t\t%u\n", version); printf(" Prologue Length:\t\t%ju\n", (uintmax_t) hdrlen); printf(" Minimum Instruction Length:\t%u\n", minlen); printf(" Initial value of 'is_stmt':\t%u\n", defstmt); printf(" Line Base:\t\t\t%d\n", lbase); printf(" Line Range:\t\t\t%u\n", lrange); printf(" Opcode Base:\t\t\t%u\n", opbase); (void) dwarf_get_address_size(re->dbg, &pointer_size, &de); printf(" (Pointer size:\t\t%u)\n", pointer_size); printf("\n"); printf(" Opcodes:\n"); for (i = 1; i < opbase; i++) { oplen = re->dw_read(d, &offset, 1); printf(" Opcode %d has %u args\n", i, oplen); } printf("\n"); printf(" The Directory Table:\n"); p = (uint8_t *) d->d_buf + offset; while (*p != '\0') { printf(" %s\n", (char *) p); p += strlen((char *) p) + 1; } p++; printf("\n"); printf(" The File Name Table:\n"); printf(" Entry\tDir\tTime\tSize\tName\n"); i = 0; while (*p != '\0') { i++; pn = (char *) p; p += strlen(pn) + 1; dirndx = _decode_uleb128(&p, pe); mtime = _decode_uleb128(&p, pe); fsize = _decode_uleb128(&p, pe); printf(" %d\t%ju\t%ju\t%ju\t%s\n", i, (uintmax_t) dirndx, (uintmax_t) mtime, (uintmax_t) fsize, pn); } #define RESET_REGISTERS \ do { \ address = 0; \ file = 1; \ line = 1; \ column = 0; \ is_stmt = defstmt; \ } while(0) #define LINE(x) (lbase + (((x) - opbase) % lrange)) #define ADDRESS(x) ((((x) - opbase) / lrange) * minlen) p++; printf("\n"); printf(" Line Number Statements:\n"); RESET_REGISTERS; while (p < pe) { if (*p == 0) { /* * Extended Opcodes. */ p++; opsize = _decode_uleb128(&p, pe); printf(" Extended opcode %u: ", *p); switch (*p) { case DW_LNE_end_sequence: p++; RESET_REGISTERS; printf("End of Sequence\n"); break; case DW_LNE_set_address: p++; address = re->dw_decode(&p, pointer_size); printf("set Address to %#jx\n", (uintmax_t) address); break; case DW_LNE_define_file: p++; pn = (char *) p; p += strlen(pn) + 1; dirndx = _decode_uleb128(&p, pe); mtime = _decode_uleb128(&p, pe); fsize = _decode_uleb128(&p, pe); printf("define new file: %s\n", pn); break; default: /* Unrecognized extened opcodes. */ p += opsize; printf("unknown opcode\n"); } } else if (*p > 0 && *p < opbase) { /* * Standard Opcodes. */ switch(*p++) { case DW_LNS_copy: printf(" Copy\n"); break; case DW_LNS_advance_pc: udelta = _decode_uleb128(&p, pe) * minlen; address += udelta; printf(" Advance PC by %ju to %#jx\n", (uintmax_t) udelta, (uintmax_t) address); break; case DW_LNS_advance_line: sdelta = _decode_sleb128(&p, pe); line += sdelta; printf(" Advance Line by %jd to %ju\n", (intmax_t) sdelta, (uintmax_t) line); break; case DW_LNS_set_file: file = _decode_uleb128(&p, pe); printf(" Set File to %ju\n", (uintmax_t) file); break; case DW_LNS_set_column: column = _decode_uleb128(&p, pe); printf(" Set Column to %ju\n", (uintmax_t) column); break; case DW_LNS_negate_stmt: is_stmt = !is_stmt; printf(" Set is_stmt to %d\n", is_stmt); break; case DW_LNS_set_basic_block: printf(" Set basic block flag\n"); break; case DW_LNS_const_add_pc: address += ADDRESS(255); printf(" Advance PC by constant %ju" " to %#jx\n", (uintmax_t) ADDRESS(255), (uintmax_t) address); break; case DW_LNS_fixed_advance_pc: udelta = re->dw_decode(&p, 2); address += udelta; printf(" Advance PC by fixed value " "%ju to %#jx\n", (uintmax_t) udelta, (uintmax_t) address); break; case DW_LNS_set_prologue_end: printf(" Set prologue end flag\n"); break; case DW_LNS_set_epilogue_begin: printf(" Set epilogue begin flag\n"); break; case DW_LNS_set_isa: isa = _decode_uleb128(&p, pe); printf(" Set isa to %ju\n", (uintmax_t) isa); break; default: /* Unrecognized extended opcodes. */ printf(" Unknown extended opcode %u\n", *(p - 1)); break; } } else { /* * Special Opcodes. */ line += LINE(*p); address += ADDRESS(*p); printf(" Special opcode %u: advance Address " "by %ju to %#jx and Line by %jd to %ju\n", *p - opbase, (uintmax_t) ADDRESS(*p), (uintmax_t) address, (intmax_t) LINE(*p), (uintmax_t) line); p++; } } } if (ret == DW_DLV_ERROR) warnx("dwarf_next_cu_header: %s", dwarf_errmsg(de)); #undef RESET_REGISTERS #undef LINE #undef ADDRESS } static void dump_dwarf_line_decoded(struct readelf *re) { Dwarf_Die die; Dwarf_Line *linebuf, ln; Dwarf_Addr lineaddr; Dwarf_Signed linecount, srccount; Dwarf_Unsigned lineno, fn; Dwarf_Error de; const char *dir, *file; char **srcfiles; int i, ret; printf("Decoded dump of debug contents of section .debug_line:\n\n"); while ((ret = dwarf_next_cu_header(re->dbg, NULL, NULL, NULL, NULL, NULL, &de)) == DW_DLV_OK) { if (dwarf_siblingof(re->dbg, NULL, &die, &de) != DW_DLV_OK) continue; if (dwarf_attrval_string(die, DW_AT_name, &file, &de) != DW_DLV_OK) file = NULL; if (dwarf_attrval_string(die, DW_AT_comp_dir, &dir, &de) != DW_DLV_OK) dir = NULL; printf("CU: "); if (dir && file && file[0] != '/') printf("%s/", dir); if (file) printf("%s", file); putchar('\n'); printf("%-37s %11s %s\n", "Filename", "Line Number", "Starting Address"); if (dwarf_srclines(die, &linebuf, &linecount, &de) != DW_DLV_OK) continue; if (dwarf_srcfiles(die, &srcfiles, &srccount, &de) != DW_DLV_OK) continue; for (i = 0; i < linecount; i++) { ln = linebuf[i]; if (dwarf_line_srcfileno(ln, &fn, &de) != DW_DLV_OK) continue; if (dwarf_lineno(ln, &lineno, &de) != DW_DLV_OK) continue; if (dwarf_lineaddr(ln, &lineaddr, &de) != DW_DLV_OK) continue; printf("%-37s %11ju %#18jx\n", basename(srcfiles[fn - 1]), (uintmax_t) lineno, (uintmax_t) lineaddr); } putchar('\n'); } } static void dump_dwarf_die(struct readelf *re, Dwarf_Die die, int level) { Dwarf_Attribute *attr_list; Dwarf_Die ret_die; Dwarf_Off dieoff, cuoff, culen, attroff; Dwarf_Unsigned ate, lang, v_udata, v_sig; Dwarf_Signed attr_count, v_sdata; Dwarf_Off v_off; Dwarf_Addr v_addr; Dwarf_Half tag, attr, form; Dwarf_Block *v_block; Dwarf_Bool v_bool, is_info; Dwarf_Sig8 v_sig8; Dwarf_Error de; Dwarf_Ptr v_expr; const char *tag_str, *attr_str, *ate_str, *lang_str; char unk_tag[32], unk_attr[32]; char *v_str; uint8_t *b, *p; int i, j, abc, ret; if (dwarf_dieoffset(die, &dieoff, &de) != DW_DLV_OK) { warnx("dwarf_dieoffset failed: %s", dwarf_errmsg(de)); goto cont_search; } printf(" <%d><%jx>: ", level, (uintmax_t) dieoff); if (dwarf_die_CU_offset_range(die, &cuoff, &culen, &de) != DW_DLV_OK) { warnx("dwarf_die_CU_offset_range failed: %s", dwarf_errmsg(de)); cuoff = 0; } abc = dwarf_die_abbrev_code(die); if (dwarf_tag(die, &tag, &de) != DW_DLV_OK) { warnx("dwarf_tag failed: %s", dwarf_errmsg(de)); goto cont_search; } if (dwarf_get_TAG_name(tag, &tag_str) != DW_DLV_OK) { snprintf(unk_tag, sizeof(unk_tag), "[Unknown Tag: %#x]", tag); tag_str = unk_tag; } printf("Abbrev Number: %d (%s)\n", abc, tag_str); if ((ret = dwarf_attrlist(die, &attr_list, &attr_count, &de)) != DW_DLV_OK) { if (ret == DW_DLV_ERROR) warnx("dwarf_attrlist failed: %s", dwarf_errmsg(de)); goto cont_search; } for (i = 0; i < attr_count; i++) { if (dwarf_whatform(attr_list[i], &form, &de) != DW_DLV_OK) { warnx("dwarf_whatform failed: %s", dwarf_errmsg(de)); continue; } if (dwarf_whatattr(attr_list[i], &attr, &de) != DW_DLV_OK) { warnx("dwarf_whatattr failed: %s", dwarf_errmsg(de)); continue; } if (dwarf_get_AT_name(attr, &attr_str) != DW_DLV_OK) { snprintf(unk_attr, sizeof(unk_attr), "[Unknown AT: %#x]", attr); attr_str = unk_attr; } if (dwarf_attroffset(attr_list[i], &attroff, &de) != DW_DLV_OK) { warnx("dwarf_attroffset failed: %s", dwarf_errmsg(de)); attroff = 0; } printf(" <%jx> %-18s: ", (uintmax_t) attroff, attr_str); switch (form) { case DW_FORM_ref_addr: case DW_FORM_sec_offset: if (dwarf_global_formref(attr_list[i], &v_off, &de) != DW_DLV_OK) { warnx("dwarf_global_formref failed: %s", dwarf_errmsg(de)); continue; } if (form == DW_FORM_ref_addr) printf("<0x%jx>", (uintmax_t) v_off); else printf("0x%jx", (uintmax_t) v_off); break; case DW_FORM_ref1: case DW_FORM_ref2: case DW_FORM_ref4: case DW_FORM_ref8: case DW_FORM_ref_udata: if (dwarf_formref(attr_list[i], &v_off, &de) != DW_DLV_OK) { warnx("dwarf_formref failed: %s", dwarf_errmsg(de)); continue; } v_off += cuoff; printf("<0x%jx>", (uintmax_t) v_off); break; case DW_FORM_addr: if (dwarf_formaddr(attr_list[i], &v_addr, &de) != DW_DLV_OK) { warnx("dwarf_formaddr failed: %s", dwarf_errmsg(de)); continue; } printf("%#jx", (uintmax_t) v_addr); break; case DW_FORM_data1: case DW_FORM_data2: case DW_FORM_data4: case DW_FORM_data8: case DW_FORM_udata: if (dwarf_formudata(attr_list[i], &v_udata, &de) != DW_DLV_OK) { warnx("dwarf_formudata failed: %s", dwarf_errmsg(de)); continue; } if (attr == DW_AT_high_pc) printf("0x%jx", (uintmax_t) v_udata); else printf("%ju", (uintmax_t) v_udata); break; case DW_FORM_sdata: if (dwarf_formsdata(attr_list[i], &v_sdata, &de) != DW_DLV_OK) { warnx("dwarf_formudata failed: %s", dwarf_errmsg(de)); continue; } printf("%jd", (intmax_t) v_sdata); break; case DW_FORM_flag: if (dwarf_formflag(attr_list[i], &v_bool, &de) != DW_DLV_OK) { warnx("dwarf_formflag failed: %s", dwarf_errmsg(de)); continue; } printf("%jd", (intmax_t) v_bool); break; case DW_FORM_flag_present: putchar('1'); break; case DW_FORM_string: case DW_FORM_strp: if (dwarf_formstring(attr_list[i], &v_str, &de) != DW_DLV_OK) { warnx("dwarf_formstring failed: %s", dwarf_errmsg(de)); continue; } if (form == DW_FORM_string) printf("%s", v_str); else printf("(indirect string) %s", v_str); break; case DW_FORM_block: case DW_FORM_block1: case DW_FORM_block2: case DW_FORM_block4: if (dwarf_formblock(attr_list[i], &v_block, &de) != DW_DLV_OK) { warnx("dwarf_formblock failed: %s", dwarf_errmsg(de)); continue; } printf("%ju byte block:", (uintmax_t) v_block->bl_len); b = v_block->bl_data; for (j = 0; (Dwarf_Unsigned) j < v_block->bl_len; j++) printf(" %x", b[j]); printf("\t("); dump_dwarf_block(re, v_block->bl_data, v_block->bl_len); putchar(')'); break; case DW_FORM_exprloc: if (dwarf_formexprloc(attr_list[i], &v_udata, &v_expr, &de) != DW_DLV_OK) { warnx("dwarf_formexprloc failed: %s", dwarf_errmsg(de)); continue; } printf("%ju byte block:", (uintmax_t) v_udata); b = v_expr; for (j = 0; (Dwarf_Unsigned) j < v_udata; j++) printf(" %x", b[j]); printf("\t("); dump_dwarf_block(re, v_expr, v_udata); putchar(')'); break; case DW_FORM_ref_sig8: if (dwarf_formsig8(attr_list[i], &v_sig8, &de) != DW_DLV_OK) { warnx("dwarf_formsig8 failed: %s", dwarf_errmsg(de)); continue; } p = (uint8_t *)(uintptr_t) &v_sig8.signature[0]; v_sig = re->dw_decode(&p, 8); printf("signature: 0x%jx", (uintmax_t) v_sig); } switch (attr) { case DW_AT_encoding: if (dwarf_attrval_unsigned(die, attr, &ate, &de) != DW_DLV_OK) break; if (dwarf_get_ATE_name(ate, &ate_str) != DW_DLV_OK) ate_str = "DW_ATE_UNKNOWN"; printf("\t(%s)", &ate_str[strlen("DW_ATE_")]); break; case DW_AT_language: if (dwarf_attrval_unsigned(die, attr, &lang, &de) != DW_DLV_OK) break; if (dwarf_get_LANG_name(lang, &lang_str) != DW_DLV_OK) break; printf("\t(%s)", &lang_str[strlen("DW_LANG_")]); break; case DW_AT_location: case DW_AT_string_length: case DW_AT_return_addr: case DW_AT_data_member_location: case DW_AT_frame_base: case DW_AT_segment: case DW_AT_static_link: case DW_AT_use_location: case DW_AT_vtable_elem_location: switch (form) { case DW_FORM_data4: case DW_FORM_data8: case DW_FORM_sec_offset: printf("\t(location list)"); break; default: break; } default: break; } putchar('\n'); } cont_search: /* Search children. */ ret = dwarf_child(die, &ret_die, &de); if (ret == DW_DLV_ERROR) warnx("dwarf_child: %s", dwarf_errmsg(de)); else if (ret == DW_DLV_OK) dump_dwarf_die(re, ret_die, level + 1); /* Search sibling. */ is_info = dwarf_get_die_infotypes_flag(die); ret = dwarf_siblingof_b(re->dbg, die, &ret_die, is_info, &de); if (ret == DW_DLV_ERROR) warnx("dwarf_siblingof: %s", dwarf_errmsg(de)); else if (ret == DW_DLV_OK) dump_dwarf_die(re, ret_die, level); dwarf_dealloc(re->dbg, die, DW_DLA_DIE); } static void set_cu_context(struct readelf *re, Dwarf_Half psize, Dwarf_Half osize, Dwarf_Half ver) { re->cu_psize = psize; re->cu_osize = osize; re->cu_ver = ver; } static void dump_dwarf_info(struct readelf *re, Dwarf_Bool is_info) { struct section *s; Dwarf_Die die; Dwarf_Error de; Dwarf_Half tag, version, pointer_size, off_size; Dwarf_Off cu_offset, cu_length; Dwarf_Off aboff; Dwarf_Unsigned typeoff; Dwarf_Sig8 sig8; Dwarf_Unsigned sig; uint8_t *p; const char *sn; int i, ret; sn = is_info ? ".debug_info" : ".debug_types"; s = NULL; for (i = 0; (size_t) i < re->shnum; i++) { s = &re->sl[i]; if (s->name != NULL && !strcmp(s->name, sn)) break; } if ((size_t) i >= re->shnum) return; do { printf("\nDump of debug contents of section %s:\n", sn); while ((ret = dwarf_next_cu_header_c(re->dbg, is_info, NULL, &version, &aboff, &pointer_size, &off_size, NULL, &sig8, &typeoff, NULL, &de)) == DW_DLV_OK) { set_cu_context(re, pointer_size, off_size, version); die = NULL; while (dwarf_siblingof_b(re->dbg, die, &die, is_info, &de) == DW_DLV_OK) { if (dwarf_tag(die, &tag, &de) != DW_DLV_OK) { warnx("dwarf_tag failed: %s", dwarf_errmsg(de)); continue; } /* XXX: What about DW_TAG_partial_unit? */ if ((is_info && tag == DW_TAG_compile_unit) || (!is_info && tag == DW_TAG_type_unit)) break; } if (die == NULL && is_info) { warnx("could not find DW_TAG_compile_unit " "die"); continue; } else if (die == NULL && !is_info) { warnx("could not find DW_TAG_type_unit die"); continue; } if (dwarf_die_CU_offset_range(die, &cu_offset, &cu_length, &de) != DW_DLV_OK) { warnx("dwarf_die_CU_offset failed: %s", dwarf_errmsg(de)); continue; } cu_length -= off_size == 4 ? 4 : 12; sig = 0; if (!is_info) { p = (uint8_t *)(uintptr_t) &sig8.signature[0]; sig = re->dw_decode(&p, 8); } printf("\n Type Unit @ offset 0x%jx:\n", (uintmax_t) cu_offset); printf(" Length:\t\t%#jx (%d-bit)\n", (uintmax_t) cu_length, off_size == 4 ? 32 : 64); printf(" Version:\t\t%u\n", version); printf(" Abbrev Offset:\t0x%jx\n", (uintmax_t) aboff); printf(" Pointer Size:\t%u\n", pointer_size); if (!is_info) { printf(" Signature:\t\t0x%016jx\n", (uintmax_t) sig); printf(" Type Offset:\t0x%jx\n", (uintmax_t) typeoff); } dump_dwarf_die(re, die, 0); } if (ret == DW_DLV_ERROR) warnx("dwarf_next_cu_header: %s", dwarf_errmsg(de)); if (is_info) break; } while (dwarf_next_types_section(re->dbg, &de) == DW_DLV_OK); } static void dump_dwarf_abbrev(struct readelf *re) { Dwarf_Abbrev ab; Dwarf_Off aboff, atoff; Dwarf_Unsigned length, attr_count; Dwarf_Signed flag, form; Dwarf_Half tag, attr; Dwarf_Error de; const char *tag_str, *attr_str, *form_str; char unk_tag[32], unk_attr[32], unk_form[32]; int i, j, ret; printf("\nContents of section .debug_abbrev:\n\n"); while ((ret = dwarf_next_cu_header(re->dbg, NULL, NULL, &aboff, NULL, NULL, &de)) == DW_DLV_OK) { printf(" Number TAG\n"); i = 0; while ((ret = dwarf_get_abbrev(re->dbg, aboff, &ab, &length, &attr_count, &de)) == DW_DLV_OK) { if (length == 1) { dwarf_dealloc(re->dbg, ab, DW_DLA_ABBREV); break; } aboff += length; printf("%4d", ++i); if (dwarf_get_abbrev_tag(ab, &tag, &de) != DW_DLV_OK) { warnx("dwarf_get_abbrev_tag failed: %s", dwarf_errmsg(de)); goto next_abbrev; } if (dwarf_get_TAG_name(tag, &tag_str) != DW_DLV_OK) { snprintf(unk_tag, sizeof(unk_tag), "[Unknown Tag: %#x]", tag); tag_str = unk_tag; } if (dwarf_get_abbrev_children_flag(ab, &flag, &de) != DW_DLV_OK) { warnx("dwarf_get_abbrev_children_flag failed:" " %s", dwarf_errmsg(de)); goto next_abbrev; } printf(" %s %s\n", tag_str, flag ? "[has children]" : "[no children]"); for (j = 0; (Dwarf_Unsigned) j < attr_count; j++) { if (dwarf_get_abbrev_entry(ab, (Dwarf_Signed) j, &attr, &form, &atoff, &de) != DW_DLV_OK) { warnx("dwarf_get_abbrev_entry failed:" " %s", dwarf_errmsg(de)); continue; } if (dwarf_get_AT_name(attr, &attr_str) != DW_DLV_OK) { snprintf(unk_attr, sizeof(unk_attr), "[Unknown AT: %#x]", attr); attr_str = unk_attr; } if (dwarf_get_FORM_name(form, &form_str) != DW_DLV_OK) { snprintf(unk_form, sizeof(unk_form), "[Unknown Form: %#x]", (Dwarf_Half) form); form_str = unk_form; } printf(" %-18s %s\n", attr_str, form_str); } next_abbrev: dwarf_dealloc(re->dbg, ab, DW_DLA_ABBREV); } if (ret != DW_DLV_OK) warnx("dwarf_get_abbrev: %s", dwarf_errmsg(de)); } if (ret == DW_DLV_ERROR) warnx("dwarf_next_cu_header: %s", dwarf_errmsg(de)); } static void dump_dwarf_pubnames(struct readelf *re) { struct section *s; Dwarf_Off die_off; Dwarf_Unsigned offset, length, nt_cu_offset, nt_cu_length; Dwarf_Signed cnt; Dwarf_Global *globs; Dwarf_Half nt_version; Dwarf_Error de; Elf_Data *d; char *glob_name; int i, dwarf_size, elferr; printf("\nContents of the .debug_pubnames section:\n"); s = NULL; for (i = 0; (size_t) i < re->shnum; i++) { s = &re->sl[i]; if (s->name != NULL && !strcmp(s->name, ".debug_pubnames")) break; } if ((size_t) i >= re->shnum) return; (void) elf_errno(); if ((d = elf_getdata(s->scn, NULL)) == NULL) { elferr = elf_errno(); if (elferr != 0) warnx("elf_getdata failed: %s", elf_errmsg(-1)); return; } if (d->d_size <= 0) return; /* Read in .debug_pubnames section table header. */ offset = 0; length = re->dw_read(d, &offset, 4); if (length == 0xffffffff) { dwarf_size = 8; length = re->dw_read(d, &offset, 8); } else dwarf_size = 4; if (length > d->d_size - offset) { warnx("invalid .dwarf_pubnames section"); return; } nt_version = re->dw_read(d, &offset, 2); nt_cu_offset = re->dw_read(d, &offset, dwarf_size); nt_cu_length = re->dw_read(d, &offset, dwarf_size); printf(" Length:\t\t\t\t%ju\n", (uintmax_t) length); printf(" Version:\t\t\t\t%u\n", nt_version); printf(" Offset into .debug_info section:\t%ju\n", (uintmax_t) nt_cu_offset); printf(" Size of area in .debug_info section:\t%ju\n", (uintmax_t) nt_cu_length); if (dwarf_get_globals(re->dbg, &globs, &cnt, &de) != DW_DLV_OK) { warnx("dwarf_get_globals failed: %s", dwarf_errmsg(de)); return; } printf("\n Offset Name\n"); for (i = 0; i < cnt; i++) { if (dwarf_globname(globs[i], &glob_name, &de) != DW_DLV_OK) { warnx("dwarf_globname failed: %s", dwarf_errmsg(de)); continue; } if (dwarf_global_die_offset(globs[i], &die_off, &de) != DW_DLV_OK) { warnx("dwarf_global_die_offset failed: %s", dwarf_errmsg(de)); continue; } printf(" %-11ju %s\n", (uintmax_t) die_off, glob_name); } } static void dump_dwarf_aranges(struct readelf *re) { struct section *s; Dwarf_Arange *aranges; Dwarf_Addr start; Dwarf_Unsigned offset, length, as_cu_offset; Dwarf_Off die_off; Dwarf_Signed cnt; Dwarf_Half as_version, as_addrsz, as_segsz; Dwarf_Error de; Elf_Data *d; int i, dwarf_size, elferr; printf("\nContents of section .debug_aranges:\n"); s = NULL; for (i = 0; (size_t) i < re->shnum; i++) { s = &re->sl[i]; if (s->name != NULL && !strcmp(s->name, ".debug_aranges")) break; } if ((size_t) i >= re->shnum) return; (void) elf_errno(); if ((d = elf_getdata(s->scn, NULL)) == NULL) { elferr = elf_errno(); if (elferr != 0) warnx("elf_getdata failed: %s", elf_errmsg(-1)); return; } if (d->d_size <= 0) return; /* Read in the .debug_aranges section table header. */ offset = 0; length = re->dw_read(d, &offset, 4); if (length == 0xffffffff) { dwarf_size = 8; length = re->dw_read(d, &offset, 8); } else dwarf_size = 4; if (length > d->d_size - offset) { warnx("invalid .dwarf_aranges section"); return; } as_version = re->dw_read(d, &offset, 2); as_cu_offset = re->dw_read(d, &offset, dwarf_size); as_addrsz = re->dw_read(d, &offset, 1); as_segsz = re->dw_read(d, &offset, 1); printf(" Length:\t\t\t%ju\n", (uintmax_t) length); printf(" Version:\t\t\t%u\n", as_version); printf(" Offset into .debug_info:\t%ju\n", (uintmax_t) as_cu_offset); printf(" Pointer Size:\t\t\t%u\n", as_addrsz); printf(" Segment Size:\t\t\t%u\n", as_segsz); if (dwarf_get_aranges(re->dbg, &aranges, &cnt, &de) != DW_DLV_OK) { warnx("dwarf_get_aranges failed: %s", dwarf_errmsg(de)); return; } printf("\n Address Length\n"); for (i = 0; i < cnt; i++) { if (dwarf_get_arange_info(aranges[i], &start, &length, &die_off, &de) != DW_DLV_OK) { warnx("dwarf_get_arange_info failed: %s", dwarf_errmsg(de)); continue; } printf(" %08jx %ju\n", (uintmax_t) start, (uintmax_t) length); } } static void dump_dwarf_ranges_foreach(struct readelf *re, Dwarf_Die die, Dwarf_Addr base) { Dwarf_Attribute *attr_list; Dwarf_Ranges *ranges; Dwarf_Die ret_die; Dwarf_Error de; Dwarf_Addr base0; Dwarf_Half attr; Dwarf_Signed attr_count, cnt; Dwarf_Unsigned off, bytecnt; int i, j, ret; if ((ret = dwarf_attrlist(die, &attr_list, &attr_count, &de)) != DW_DLV_OK) { if (ret == DW_DLV_ERROR) warnx("dwarf_attrlist failed: %s", dwarf_errmsg(de)); goto cont_search; } for (i = 0; i < attr_count; i++) { if (dwarf_whatattr(attr_list[i], &attr, &de) != DW_DLV_OK) { warnx("dwarf_whatattr failed: %s", dwarf_errmsg(de)); continue; } if (attr != DW_AT_ranges) continue; if (dwarf_formudata(attr_list[i], &off, &de) != DW_DLV_OK) { warnx("dwarf_formudata failed: %s", dwarf_errmsg(de)); continue; } if (dwarf_get_ranges(re->dbg, (Dwarf_Off) off, &ranges, &cnt, &bytecnt, &de) != DW_DLV_OK) continue; base0 = base; for (j = 0; j < cnt; j++) { printf(" %08jx ", (uintmax_t) off); if (ranges[j].dwr_type == DW_RANGES_END) { printf("%s\n", ""); continue; } else if (ranges[j].dwr_type == DW_RANGES_ADDRESS_SELECTION) { base0 = ranges[j].dwr_addr2; continue; } if (re->ec == ELFCLASS32) printf("%08jx %08jx\n", (uintmax_t) (ranges[j].dwr_addr1 + base0), (uintmax_t) (ranges[j].dwr_addr2 + base0)); else printf("%016jx %016jx\n", (uintmax_t) (ranges[j].dwr_addr1 + base0), (uintmax_t) (ranges[j].dwr_addr2 + base0)); } } cont_search: /* Search children. */ ret = dwarf_child(die, &ret_die, &de); if (ret == DW_DLV_ERROR) warnx("dwarf_child: %s", dwarf_errmsg(de)); else if (ret == DW_DLV_OK) dump_dwarf_ranges_foreach(re, ret_die, base); /* Search sibling. */ ret = dwarf_siblingof(re->dbg, die, &ret_die, &de); if (ret == DW_DLV_ERROR) warnx("dwarf_siblingof: %s", dwarf_errmsg(de)); else if (ret == DW_DLV_OK) dump_dwarf_ranges_foreach(re, ret_die, base); } static void dump_dwarf_ranges(struct readelf *re) { Dwarf_Ranges *ranges; Dwarf_Die die; Dwarf_Signed cnt; Dwarf_Unsigned bytecnt; Dwarf_Half tag; Dwarf_Error de; Dwarf_Unsigned lowpc; int ret; if (dwarf_get_ranges(re->dbg, 0, &ranges, &cnt, &bytecnt, &de) != DW_DLV_OK) return; printf("Contents of the .debug_ranges section:\n\n"); if (re->ec == ELFCLASS32) printf(" %-8s %-8s %s\n", "Offset", "Begin", "End"); else printf(" %-8s %-16s %s\n", "Offset", "Begin", "End"); while ((ret = dwarf_next_cu_header(re->dbg, NULL, NULL, NULL, NULL, NULL, &de)) == DW_DLV_OK) { die = NULL; if (dwarf_siblingof(re->dbg, die, &die, &de) != DW_DLV_OK) continue; if (dwarf_tag(die, &tag, &de) != DW_DLV_OK) { warnx("dwarf_tag failed: %s", dwarf_errmsg(de)); continue; } /* XXX: What about DW_TAG_partial_unit? */ lowpc = 0; if (tag == DW_TAG_compile_unit) { if (dwarf_attrval_unsigned(die, DW_AT_low_pc, &lowpc, &de) != DW_DLV_OK) lowpc = 0; } dump_dwarf_ranges_foreach(re, die, (Dwarf_Addr) lowpc); } putchar('\n'); } static void dump_dwarf_macinfo(struct readelf *re) { Dwarf_Unsigned offset; Dwarf_Signed cnt; Dwarf_Macro_Details *md; Dwarf_Error de; const char *mi_str; char unk_mi[32]; int i; #define _MAX_MACINFO_ENTRY 65535 printf("\nContents of section .debug_macinfo:\n\n"); offset = 0; while (dwarf_get_macro_details(re->dbg, offset, _MAX_MACINFO_ENTRY, &cnt, &md, &de) == DW_DLV_OK) { for (i = 0; i < cnt; i++) { offset = md[i].dmd_offset + 1; if (md[i].dmd_type == 0) break; if (dwarf_get_MACINFO_name(md[i].dmd_type, &mi_str) != DW_DLV_OK) { snprintf(unk_mi, sizeof(unk_mi), "[Unknown MACINFO: %#x]", md[i].dmd_type); mi_str = unk_mi; } printf(" %s", mi_str); switch (md[i].dmd_type) { case DW_MACINFO_define: case DW_MACINFO_undef: printf(" - lineno : %jd macro : %s\n", (intmax_t) md[i].dmd_lineno, md[i].dmd_macro); break; case DW_MACINFO_start_file: printf(" - lineno : %jd filenum : %jd\n", (intmax_t) md[i].dmd_lineno, (intmax_t) md[i].dmd_fileindex); break; default: putchar('\n'); break; } } } #undef _MAX_MACINFO_ENTRY } static void dump_dwarf_frame_inst(struct readelf *re, Dwarf_Cie cie, uint8_t *insts, Dwarf_Unsigned len, Dwarf_Unsigned caf, Dwarf_Signed daf, Dwarf_Addr pc, Dwarf_Debug dbg) { Dwarf_Frame_Op *oplist; Dwarf_Signed opcnt, delta; Dwarf_Small op; Dwarf_Error de; const char *op_str; char unk_op[32]; int i; if (dwarf_expand_frame_instructions(cie, insts, len, &oplist, &opcnt, &de) != DW_DLV_OK) { warnx("dwarf_expand_frame_instructions failed: %s", dwarf_errmsg(de)); return; } for (i = 0; i < opcnt; i++) { if (oplist[i].fp_base_op != 0) op = oplist[i].fp_base_op << 6; else op = oplist[i].fp_extended_op; if (dwarf_get_CFA_name(op, &op_str) != DW_DLV_OK) { snprintf(unk_op, sizeof(unk_op), "[Unknown CFA: %#x]", op); op_str = unk_op; } printf(" %s", op_str); switch (op) { case DW_CFA_advance_loc: delta = oplist[i].fp_offset * caf; pc += delta; printf(": %ju to %08jx", (uintmax_t) delta, (uintmax_t) pc); break; case DW_CFA_offset: case DW_CFA_offset_extended: case DW_CFA_offset_extended_sf: delta = oplist[i].fp_offset * daf; printf(": r%u (%s) at cfa%+jd", oplist[i].fp_register, dwarf_regname(re, oplist[i].fp_register), (intmax_t) delta); break; case DW_CFA_restore: printf(": r%u (%s)", oplist[i].fp_register, dwarf_regname(re, oplist[i].fp_register)); break; case DW_CFA_set_loc: pc = oplist[i].fp_offset; printf(": to %08jx", (uintmax_t) pc); break; case DW_CFA_advance_loc1: case DW_CFA_advance_loc2: case DW_CFA_advance_loc4: pc += oplist[i].fp_offset; printf(": %jd to %08jx", (intmax_t) oplist[i].fp_offset, (uintmax_t) pc); break; case DW_CFA_def_cfa: printf(": r%u (%s) ofs %ju", oplist[i].fp_register, dwarf_regname(re, oplist[i].fp_register), (uintmax_t) oplist[i].fp_offset); break; case DW_CFA_def_cfa_sf: printf(": r%u (%s) ofs %jd", oplist[i].fp_register, dwarf_regname(re, oplist[i].fp_register), (intmax_t) (oplist[i].fp_offset * daf)); break; case DW_CFA_def_cfa_register: printf(": r%u (%s)", oplist[i].fp_register, dwarf_regname(re, oplist[i].fp_register)); break; case DW_CFA_def_cfa_offset: printf(": %ju", (uintmax_t) oplist[i].fp_offset); break; case DW_CFA_def_cfa_offset_sf: printf(": %jd", (intmax_t) (oplist[i].fp_offset * daf)); break; default: break; } putchar('\n'); } dwarf_dealloc(dbg, oplist, DW_DLA_FRAME_BLOCK); } static char * get_regoff_str(struct readelf *re, Dwarf_Half reg, Dwarf_Addr off) { static char rs[16]; if (reg == DW_FRAME_UNDEFINED_VAL || reg == DW_FRAME_REG_INITIAL_VALUE) snprintf(rs, sizeof(rs), "%c", 'u'); else if (reg == DW_FRAME_CFA_COL) snprintf(rs, sizeof(rs), "c%+jd", (intmax_t) off); else snprintf(rs, sizeof(rs), "%s%+jd", dwarf_regname(re, reg), (intmax_t) off); return (rs); } static int dump_dwarf_frame_regtable(struct readelf *re, Dwarf_Fde fde, Dwarf_Addr pc, Dwarf_Unsigned func_len, Dwarf_Half cie_ra) { Dwarf_Regtable rt; Dwarf_Addr row_pc, end_pc, pre_pc, cur_pc; Dwarf_Error de; char *vec; int i; #define BIT_SET(v, n) (v[(n)>>3] |= 1U << ((n) & 7)) #define BIT_CLR(v, n) (v[(n)>>3] &= ~(1U << ((n) & 7))) #define BIT_ISSET(v, n) (v[(n)>>3] & (1U << ((n) & 7))) #define RT(x) rt.rules[(x)] vec = calloc((DW_REG_TABLE_SIZE + 7) / 8, 1); if (vec == NULL) err(EXIT_FAILURE, "calloc failed"); pre_pc = ~((Dwarf_Addr) 0); cur_pc = pc; end_pc = pc + func_len; for (; cur_pc < end_pc; cur_pc++) { if (dwarf_get_fde_info_for_all_regs(fde, cur_pc, &rt, &row_pc, &de) != DW_DLV_OK) { free(vec); warnx("dwarf_get_fde_info_for_all_regs failed: %s\n", dwarf_errmsg(de)); return (-1); } if (row_pc == pre_pc) continue; pre_pc = row_pc; for (i = 1; i < DW_REG_TABLE_SIZE; i++) { if (rt.rules[i].dw_regnum != DW_FRAME_REG_INITIAL_VALUE) BIT_SET(vec, i); } } printf(" LOC CFA "); for (i = 1; i < DW_REG_TABLE_SIZE; i++) { if (BIT_ISSET(vec, i)) { if ((Dwarf_Half) i == cie_ra) printf("ra "); else printf("%-5s", dwarf_regname(re, (unsigned int) i)); } } putchar('\n'); pre_pc = ~((Dwarf_Addr) 0); cur_pc = pc; end_pc = pc + func_len; for (; cur_pc < end_pc; cur_pc++) { if (dwarf_get_fde_info_for_all_regs(fde, cur_pc, &rt, &row_pc, &de) != DW_DLV_OK) { warnx("dwarf_get_fde_info_for_all_regs failed: %s\n", dwarf_errmsg(de)); return (-1); } if (row_pc == pre_pc) continue; pre_pc = row_pc; printf("%08jx ", (uintmax_t) row_pc); printf("%-8s ", get_regoff_str(re, RT(0).dw_regnum, RT(0).dw_offset)); for (i = 1; i < DW_REG_TABLE_SIZE; i++) { if (BIT_ISSET(vec, i)) { printf("%-5s", get_regoff_str(re, RT(i).dw_regnum, RT(i).dw_offset)); } } putchar('\n'); } free(vec); return (0); #undef BIT_SET #undef BIT_CLR #undef BIT_ISSET #undef RT } static void dump_dwarf_frame_section(struct readelf *re, struct section *s, int alt) { Dwarf_Cie *cie_list, cie, pre_cie; Dwarf_Fde *fde_list, fde; Dwarf_Off cie_offset, fde_offset; Dwarf_Unsigned cie_length, fde_instlen; Dwarf_Unsigned cie_caf, cie_daf, cie_instlen, func_len, fde_length; Dwarf_Signed cie_count, fde_count, cie_index; Dwarf_Addr low_pc; Dwarf_Half cie_ra; Dwarf_Small cie_version; Dwarf_Ptr fde_addr, fde_inst, cie_inst; char *cie_aug, c; int i, eh_frame; Dwarf_Error de; printf("\nThe section %s contains:\n\n", s->name); if (!strcmp(s->name, ".debug_frame")) { eh_frame = 0; if (dwarf_get_fde_list(re->dbg, &cie_list, &cie_count, &fde_list, &fde_count, &de) != DW_DLV_OK) { warnx("dwarf_get_fde_list failed: %s", dwarf_errmsg(de)); return; } } else if (!strcmp(s->name, ".eh_frame")) { eh_frame = 1; if (dwarf_get_fde_list_eh(re->dbg, &cie_list, &cie_count, &fde_list, &fde_count, &de) != DW_DLV_OK) { warnx("dwarf_get_fde_list_eh failed: %s", dwarf_errmsg(de)); return; } } else return; pre_cie = NULL; for (i = 0; i < fde_count; i++) { if (dwarf_get_fde_n(fde_list, i, &fde, &de) != DW_DLV_OK) { warnx("dwarf_get_fde_n failed: %s", dwarf_errmsg(de)); continue; } if (dwarf_get_cie_of_fde(fde, &cie, &de) != DW_DLV_OK) { warnx("dwarf_get_fde_n failed: %s", dwarf_errmsg(de)); continue; } if (dwarf_get_fde_range(fde, &low_pc, &func_len, &fde_addr, &fde_length, &cie_offset, &cie_index, &fde_offset, &de) != DW_DLV_OK) { warnx("dwarf_get_fde_range failed: %s", dwarf_errmsg(de)); continue; } if (dwarf_get_fde_instr_bytes(fde, &fde_inst, &fde_instlen, &de) != DW_DLV_OK) { warnx("dwarf_get_fde_instr_bytes failed: %s", dwarf_errmsg(de)); continue; } if (pre_cie == NULL || cie != pre_cie) { pre_cie = cie; if (dwarf_get_cie_info(cie, &cie_length, &cie_version, &cie_aug, &cie_caf, &cie_daf, &cie_ra, &cie_inst, &cie_instlen, &de) != DW_DLV_OK) { warnx("dwarf_get_cie_info failed: %s", dwarf_errmsg(de)); continue; } printf("%08jx %08jx %8.8jx CIE", (uintmax_t) cie_offset, (uintmax_t) cie_length, (uintmax_t) (eh_frame ? 0 : ~0U)); if (!alt) { putchar('\n'); printf(" Version:\t\t\t%u\n", cie_version); printf(" Augmentation:\t\t\t\""); while ((c = *cie_aug++) != '\0') putchar(c); printf("\"\n"); printf(" Code alignment factor:\t%ju\n", (uintmax_t) cie_caf); printf(" Data alignment factor:\t%jd\n", (intmax_t) cie_daf); printf(" Return address column:\t%ju\n", (uintmax_t) cie_ra); putchar('\n'); dump_dwarf_frame_inst(re, cie, cie_inst, cie_instlen, cie_caf, cie_daf, 0, re->dbg); putchar('\n'); } else { printf(" \""); while ((c = *cie_aug++) != '\0') putchar(c); putchar('"'); printf(" cf=%ju df=%jd ra=%ju\n", (uintmax_t) cie_caf, (uintmax_t) cie_daf, (uintmax_t) cie_ra); dump_dwarf_frame_regtable(re, fde, low_pc, 1, cie_ra); putchar('\n'); } } printf("%08jx %08jx %08jx FDE cie=%08jx pc=%08jx..%08jx\n", (uintmax_t) fde_offset, (uintmax_t) fde_length, (uintmax_t) cie_offset, (uintmax_t) (eh_frame ? fde_offset + 4 - cie_offset : cie_offset), (uintmax_t) low_pc, (uintmax_t) (low_pc + func_len)); if (!alt) dump_dwarf_frame_inst(re, cie, fde_inst, fde_instlen, cie_caf, cie_daf, low_pc, re->dbg); else dump_dwarf_frame_regtable(re, fde, low_pc, func_len, cie_ra); putchar('\n'); } } static void dump_dwarf_frame(struct readelf *re, int alt) { struct section *s; int i; (void) dwarf_set_frame_cfa_value(re->dbg, DW_FRAME_CFA_COL); for (i = 0; (size_t) i < re->shnum; i++) { s = &re->sl[i]; if (s->name != NULL && (!strcmp(s->name, ".debug_frame") || !strcmp(s->name, ".eh_frame"))) dump_dwarf_frame_section(re, s, alt); } } static void dump_dwarf_str(struct readelf *re) { struct section *s; Elf_Data *d; unsigned char *p; int elferr, end, i, j; printf("\nContents of section .debug_str:\n"); s = NULL; for (i = 0; (size_t) i < re->shnum; i++) { s = &re->sl[i]; if (s->name != NULL && !strcmp(s->name, ".debug_str")) break; } if ((size_t) i >= re->shnum) return; (void) elf_errno(); if ((d = elf_getdata(s->scn, NULL)) == NULL) { elferr = elf_errno(); if (elferr != 0) warnx("elf_getdata failed: %s", elf_errmsg(-1)); return; } if (d->d_size <= 0) return; for (i = 0, p = d->d_buf; (size_t) i < d->d_size; i += 16) { printf(" 0x%08x", (unsigned int) i); if ((size_t) i + 16 > d->d_size) end = d->d_size; else end = i + 16; for (j = i; j < i + 16; j++) { if ((j - i) % 4 == 0) putchar(' '); if (j >= end) { printf(" "); continue; } printf("%02x", (uint8_t) p[j]); } putchar(' '); for (j = i; j < end; j++) { if (isprint(p[j])) putchar(p[j]); else if (p[j] == 0) putchar('.'); else putchar(' '); } putchar('\n'); } } static int loc_at_comparator(const void *la1, const void *la2) { const struct loc_at *left, *right; left = (const struct loc_at *)la1; right = (const struct loc_at *)la2; if (left->la_off > right->la_off) return (1); else if (left->la_off < right->la_off) return (-1); else return (0); } static void search_loclist_at(struct readelf *re, Dwarf_Die die, Dwarf_Unsigned lowpc, struct loc_at **la_list, size_t *la_list_len, size_t *la_list_cap) { struct loc_at *la; Dwarf_Attribute *attr_list; Dwarf_Die ret_die; Dwarf_Unsigned off; Dwarf_Off ref; Dwarf_Signed attr_count; Dwarf_Half attr, form; Dwarf_Bool is_info; Dwarf_Error de; int i, ret; is_info = dwarf_get_die_infotypes_flag(die); if ((ret = dwarf_attrlist(die, &attr_list, &attr_count, &de)) != DW_DLV_OK) { if (ret == DW_DLV_ERROR) warnx("dwarf_attrlist failed: %s", dwarf_errmsg(de)); goto cont_search; } for (i = 0; i < attr_count; i++) { if (dwarf_whatattr(attr_list[i], &attr, &de) != DW_DLV_OK) { warnx("dwarf_whatattr failed: %s", dwarf_errmsg(de)); continue; } if (attr != DW_AT_location && attr != DW_AT_string_length && attr != DW_AT_return_addr && attr != DW_AT_data_member_location && attr != DW_AT_frame_base && attr != DW_AT_segment && attr != DW_AT_static_link && attr != DW_AT_use_location && attr != DW_AT_vtable_elem_location) continue; if (dwarf_whatform(attr_list[i], &form, &de) != DW_DLV_OK) { warnx("dwarf_whatform failed: %s", dwarf_errmsg(de)); continue; } if (form == DW_FORM_data4 || form == DW_FORM_data8) { if (dwarf_formudata(attr_list[i], &off, &de) != DW_DLV_OK) { warnx("dwarf_formudata failed: %s", dwarf_errmsg(de)); continue; } } else if (form == DW_FORM_sec_offset) { if (dwarf_global_formref(attr_list[i], &ref, &de) != DW_DLV_OK) { warnx("dwarf_global_formref failed: %s", dwarf_errmsg(de)); continue; } off = ref; } else continue; if (*la_list_cap == *la_list_len) { *la_list = realloc(*la_list, *la_list_cap * 2 * sizeof(**la_list)); if (*la_list == NULL) err(EXIT_FAILURE, "realloc failed"); *la_list_cap *= 2; } la = &((*la_list)[*la_list_len]); la->la_at = attr_list[i]; la->la_off = off; la->la_lowpc = lowpc; la->la_cu_psize = re->cu_psize; la->la_cu_osize = re->cu_osize; la->la_cu_ver = re->cu_ver; (*la_list_len)++; } cont_search: /* Search children. */ ret = dwarf_child(die, &ret_die, &de); if (ret == DW_DLV_ERROR) warnx("dwarf_child: %s", dwarf_errmsg(de)); else if (ret == DW_DLV_OK) search_loclist_at(re, ret_die, lowpc, la_list, la_list_len, la_list_cap); /* Search sibling. */ ret = dwarf_siblingof_b(re->dbg, die, &ret_die, is_info, &de); if (ret == DW_DLV_ERROR) warnx("dwarf_siblingof: %s", dwarf_errmsg(de)); else if (ret == DW_DLV_OK) search_loclist_at(re, ret_die, lowpc, la_list, la_list_len, la_list_cap); } static void dump_dwarf_loc(struct readelf *re, Dwarf_Loc *lr) { const char *op_str; char unk_op[32]; uint8_t *b, n; int i; if (dwarf_get_OP_name(lr->lr_atom, &op_str) != DW_DLV_OK) { snprintf(unk_op, sizeof(unk_op), "[Unknown OP: %#x]", lr->lr_atom); op_str = unk_op; } printf("%s", op_str); switch (lr->lr_atom) { case DW_OP_reg0: case DW_OP_reg1: case DW_OP_reg2: case DW_OP_reg3: case DW_OP_reg4: case DW_OP_reg5: case DW_OP_reg6: case DW_OP_reg7: case DW_OP_reg8: case DW_OP_reg9: case DW_OP_reg10: case DW_OP_reg11: case DW_OP_reg12: case DW_OP_reg13: case DW_OP_reg14: case DW_OP_reg15: case DW_OP_reg16: case DW_OP_reg17: case DW_OP_reg18: case DW_OP_reg19: case DW_OP_reg20: case DW_OP_reg21: case DW_OP_reg22: case DW_OP_reg23: case DW_OP_reg24: case DW_OP_reg25: case DW_OP_reg26: case DW_OP_reg27: case DW_OP_reg28: case DW_OP_reg29: case DW_OP_reg30: case DW_OP_reg31: printf(" (%s)", dwarf_regname(re, lr->lr_atom - DW_OP_reg0)); break; case DW_OP_deref: case DW_OP_lit0: case DW_OP_lit1: case DW_OP_lit2: case DW_OP_lit3: case DW_OP_lit4: case DW_OP_lit5: case DW_OP_lit6: case DW_OP_lit7: case DW_OP_lit8: case DW_OP_lit9: case DW_OP_lit10: case DW_OP_lit11: case DW_OP_lit12: case DW_OP_lit13: case DW_OP_lit14: case DW_OP_lit15: case DW_OP_lit16: case DW_OP_lit17: case DW_OP_lit18: case DW_OP_lit19: case DW_OP_lit20: case DW_OP_lit21: case DW_OP_lit22: case DW_OP_lit23: case DW_OP_lit24: case DW_OP_lit25: case DW_OP_lit26: case DW_OP_lit27: case DW_OP_lit28: case DW_OP_lit29: case DW_OP_lit30: case DW_OP_lit31: case DW_OP_dup: case DW_OP_drop: case DW_OP_over: case DW_OP_swap: case DW_OP_rot: case DW_OP_xderef: case DW_OP_abs: case DW_OP_and: case DW_OP_div: case DW_OP_minus: case DW_OP_mod: case DW_OP_mul: case DW_OP_neg: case DW_OP_not: case DW_OP_or: case DW_OP_plus: case DW_OP_shl: case DW_OP_shr: case DW_OP_shra: case DW_OP_xor: case DW_OP_eq: case DW_OP_ge: case DW_OP_gt: case DW_OP_le: case DW_OP_lt: case DW_OP_ne: case DW_OP_nop: case DW_OP_push_object_address: case DW_OP_form_tls_address: case DW_OP_call_frame_cfa: case DW_OP_stack_value: case DW_OP_GNU_push_tls_address: case DW_OP_GNU_uninit: break; case DW_OP_const1u: case DW_OP_pick: case DW_OP_deref_size: case DW_OP_xderef_size: case DW_OP_const2u: case DW_OP_bra: case DW_OP_skip: case DW_OP_const4u: case DW_OP_const8u: case DW_OP_constu: case DW_OP_plus_uconst: case DW_OP_regx: case DW_OP_piece: printf(": %ju", (uintmax_t) lr->lr_number); break; case DW_OP_const1s: case DW_OP_const2s: case DW_OP_const4s: case DW_OP_const8s: case DW_OP_consts: printf(": %jd", (intmax_t) lr->lr_number); break; case DW_OP_breg0: case DW_OP_breg1: case DW_OP_breg2: case DW_OP_breg3: case DW_OP_breg4: case DW_OP_breg5: case DW_OP_breg6: case DW_OP_breg7: case DW_OP_breg8: case DW_OP_breg9: case DW_OP_breg10: case DW_OP_breg11: case DW_OP_breg12: case DW_OP_breg13: case DW_OP_breg14: case DW_OP_breg15: case DW_OP_breg16: case DW_OP_breg17: case DW_OP_breg18: case DW_OP_breg19: case DW_OP_breg20: case DW_OP_breg21: case DW_OP_breg22: case DW_OP_breg23: case DW_OP_breg24: case DW_OP_breg25: case DW_OP_breg26: case DW_OP_breg27: case DW_OP_breg28: case DW_OP_breg29: case DW_OP_breg30: case DW_OP_breg31: printf(" (%s): %jd", dwarf_regname(re, lr->lr_atom - DW_OP_breg0), (intmax_t) lr->lr_number); break; case DW_OP_fbreg: printf(": %jd", (intmax_t) lr->lr_number); break; case DW_OP_bregx: printf(": %ju (%s) %jd", (uintmax_t) lr->lr_number, dwarf_regname(re, (unsigned int) lr->lr_number), (intmax_t) lr->lr_number2); break; case DW_OP_addr: case DW_OP_GNU_encoded_addr: printf(": %#jx", (uintmax_t) lr->lr_number); break; case DW_OP_GNU_implicit_pointer: printf(": <0x%jx> %jd", (uintmax_t) lr->lr_number, (intmax_t) lr->lr_number2); break; case DW_OP_implicit_value: printf(": %ju byte block:", (uintmax_t) lr->lr_number); b = (uint8_t *)(uintptr_t) lr->lr_number2; for (i = 0; (Dwarf_Unsigned) i < lr->lr_number; i++) printf(" %x", b[i]); break; case DW_OP_GNU_entry_value: printf(": ("); dump_dwarf_block(re, (uint8_t *)(uintptr_t) lr->lr_number2, lr->lr_number); putchar(')'); break; case DW_OP_GNU_const_type: printf(": <0x%jx> ", (uintmax_t) lr->lr_number); b = (uint8_t *)(uintptr_t) lr->lr_number2; n = *b; for (i = 1; (uint8_t) i < n; i++) printf(" %x", b[i]); break; case DW_OP_GNU_regval_type: printf(": %ju (%s) <0x%jx>", (uintmax_t) lr->lr_number, dwarf_regname(re, (unsigned int) lr->lr_number), (uintmax_t) lr->lr_number2); break; case DW_OP_GNU_convert: case DW_OP_GNU_deref_type: case DW_OP_GNU_parameter_ref: case DW_OP_GNU_reinterpret: printf(": <0x%jx>", (uintmax_t) lr->lr_number); break; default: break; } } static void dump_dwarf_block(struct readelf *re, uint8_t *b, Dwarf_Unsigned len) { Dwarf_Locdesc *llbuf; Dwarf_Signed lcnt; Dwarf_Error de; int i; if (dwarf_loclist_from_expr_b(re->dbg, b, len, re->cu_psize, re->cu_osize, re->cu_ver, &llbuf, &lcnt, &de) != DW_DLV_OK) { warnx("dwarf_loclist_form_expr_b: %s", dwarf_errmsg(de)); return; } for (i = 0; (Dwarf_Half) i < llbuf->ld_cents; i++) { dump_dwarf_loc(re, &llbuf->ld_s[i]); if (i < llbuf->ld_cents - 1) printf("; "); } dwarf_dealloc(re->dbg, llbuf->ld_s, DW_DLA_LOC_BLOCK); dwarf_dealloc(re->dbg, llbuf, DW_DLA_LOCDESC); } static void dump_dwarf_loclist(struct readelf *re) { Dwarf_Die die; Dwarf_Locdesc **llbuf; Dwarf_Unsigned lowpc; Dwarf_Signed lcnt; Dwarf_Half tag, version, pointer_size, off_size; Dwarf_Error de; struct loc_at *la_list, *left, *right, *la; size_t la_list_len, la_list_cap; unsigned int duplicates, k; int i, j, ret, has_content; la_list_len = 0; la_list_cap = 200; if ((la_list = calloc(la_list_cap, sizeof(struct loc_at))) == NULL) errx(EXIT_FAILURE, "calloc failed"); /* Search .debug_info section. */ while ((ret = dwarf_next_cu_header_b(re->dbg, NULL, &version, NULL, &pointer_size, &off_size, NULL, NULL, &de)) == DW_DLV_OK) { set_cu_context(re, pointer_size, off_size, version); die = NULL; if (dwarf_siblingof(re->dbg, die, &die, &de) != DW_DLV_OK) continue; if (dwarf_tag(die, &tag, &de) != DW_DLV_OK) { warnx("dwarf_tag failed: %s", dwarf_errmsg(de)); continue; } /* XXX: What about DW_TAG_partial_unit? */ lowpc = 0; if (tag == DW_TAG_compile_unit) { if (dwarf_attrval_unsigned(die, DW_AT_low_pc, &lowpc, &de) != DW_DLV_OK) lowpc = 0; } /* Search attributes for reference to .debug_loc section. */ search_loclist_at(re, die, lowpc, &la_list, &la_list_len, &la_list_cap); } if (ret == DW_DLV_ERROR) warnx("dwarf_next_cu_header: %s", dwarf_errmsg(de)); /* Search .debug_types section. */ do { while ((ret = dwarf_next_cu_header_c(re->dbg, 0, NULL, &version, NULL, &pointer_size, &off_size, NULL, NULL, NULL, NULL, &de)) == DW_DLV_OK) { set_cu_context(re, pointer_size, off_size, version); die = NULL; if (dwarf_siblingof(re->dbg, die, &die, &de) != DW_DLV_OK) continue; if (dwarf_tag(die, &tag, &de) != DW_DLV_OK) { warnx("dwarf_tag failed: %s", dwarf_errmsg(de)); continue; } lowpc = 0; if (tag == DW_TAG_type_unit) { if (dwarf_attrval_unsigned(die, DW_AT_low_pc, &lowpc, &de) != DW_DLV_OK) lowpc = 0; } /* * Search attributes for reference to .debug_loc * section. */ search_loclist_at(re, die, lowpc, &la_list, &la_list_len, &la_list_cap); } if (ret == DW_DLV_ERROR) warnx("dwarf_next_cu_header: %s", dwarf_errmsg(de)); } while (dwarf_next_types_section(re->dbg, &de) == DW_DLV_OK); if (la_list_len == 0) { free(la_list); return; } /* Sort la_list using loc_at_comparator. */ qsort(la_list, la_list_len, sizeof(struct loc_at), loc_at_comparator); /* Get rid of the duplicates in la_list. */ duplicates = 0; for (k = 1; k < la_list_len; ++k) { left = &la_list[k - 1 - duplicates]; right = &la_list[k]; if (left->la_off == right->la_off) duplicates++; else la_list[k - duplicates] = *right; } la_list_len -= duplicates; has_content = 0; for (k = 0; k < la_list_len; ++k) { la = &la_list[k]; if ((ret = dwarf_loclist_n(la->la_at, &llbuf, &lcnt, &de)) != DW_DLV_OK) { if (ret != DW_DLV_NO_ENTRY) warnx("dwarf_loclist_n failed: %s", dwarf_errmsg(de)); continue; } if (!has_content) { has_content = 1; printf("\nContents of section .debug_loc:\n"); printf(" Offset Begin End Expression\n"); } set_cu_context(re, la->la_cu_psize, la->la_cu_osize, la->la_cu_ver); for (i = 0; i < lcnt; i++) { printf(" %8.8jx ", (uintmax_t) la->la_off); if (llbuf[i]->ld_lopc == 0 && llbuf[i]->ld_hipc == 0) { printf("\n"); continue; } /* TODO: handle base selection entry. */ printf("%8.8jx %8.8jx ", (uintmax_t) (la->la_lowpc + llbuf[i]->ld_lopc), (uintmax_t) (la->la_lowpc + llbuf[i]->ld_hipc)); putchar('('); for (j = 0; (Dwarf_Half) j < llbuf[i]->ld_cents; j++) { dump_dwarf_loc(re, &llbuf[i]->ld_s[j]); if (j < llbuf[i]->ld_cents - 1) printf("; "); } putchar(')'); if (llbuf[i]->ld_lopc == llbuf[i]->ld_hipc) printf(" (start == end)"); putchar('\n'); } for (i = 0; i < lcnt; i++) { dwarf_dealloc(re->dbg, llbuf[i]->ld_s, DW_DLA_LOC_BLOCK); dwarf_dealloc(re->dbg, llbuf[i], DW_DLA_LOCDESC); } dwarf_dealloc(re->dbg, llbuf, DW_DLA_LIST); } if (!has_content) printf("\nSection '.debug_loc' has no debugging data.\n"); free(la_list); } /* * Retrieve a string using string table section index and the string offset. */ static const char* get_string(struct readelf *re, int strtab, size_t off) { const char *name; if ((name = elf_strptr(re->elf, strtab, off)) == NULL) return (""); return (name); } /* * Retrieve the name of a symbol using the section index of the symbol * table and the index of the symbol within that table. */ static const char * get_symbol_name(struct readelf *re, int symtab, int i) { struct section *s; const char *name; GElf_Sym sym; Elf_Data *data; int elferr; s = &re->sl[symtab]; if (s->type != SHT_SYMTAB && s->type != SHT_DYNSYM) return (""); (void) elf_errno(); if ((data = elf_getdata(s->scn, NULL)) == NULL) { elferr = elf_errno(); if (elferr != 0) warnx("elf_getdata failed: %s", elf_errmsg(elferr)); return (""); } if (gelf_getsym(data, i, &sym) != &sym) return (""); /* Return section name for STT_SECTION symbol. */ if (GELF_ST_TYPE(sym.st_info) == STT_SECTION) { if (sym.st_shndx < re->shnum && re->sl[sym.st_shndx].name != NULL) return (re->sl[sym.st_shndx].name); return (""); } if (s->link >= re->shnum || (name = elf_strptr(re->elf, s->link, sym.st_name)) == NULL) return (""); return (name); } static uint64_t get_symbol_value(struct readelf *re, int symtab, int i) { struct section *s; GElf_Sym sym; Elf_Data *data; int elferr; s = &re->sl[symtab]; if (s->type != SHT_SYMTAB && s->type != SHT_DYNSYM) return (0); (void) elf_errno(); if ((data = elf_getdata(s->scn, NULL)) == NULL) { elferr = elf_errno(); if (elferr != 0) warnx("elf_getdata failed: %s", elf_errmsg(elferr)); return (0); } if (gelf_getsym(data, i, &sym) != &sym) return (0); return (sym.st_value); } static void hex_dump(struct readelf *re) { struct section *s; Elf_Data *d; uint8_t *buf; size_t sz, nbytes; uint64_t addr; int elferr, i, j; for (i = 1; (size_t) i < re->shnum; i++) { s = &re->sl[i]; if (find_dumpop(re, (size_t) i, s->name, HEX_DUMP, -1) == NULL) continue; (void) elf_errno(); if ((d = elf_getdata(s->scn, NULL)) == NULL && (d = elf_rawdata(s->scn, NULL)) == NULL) { elferr = elf_errno(); if (elferr != 0) warnx("elf_getdata failed: %s", elf_errmsg(elferr)); continue; } (void) elf_errno(); if (d->d_size <= 0 || d->d_buf == NULL) { printf("\nSection '%s' has no data to dump.\n", s->name); continue; } buf = d->d_buf; sz = d->d_size; addr = s->addr; printf("\nHex dump of section '%s':\n", s->name); while (sz > 0) { printf(" 0x%8.8jx ", (uintmax_t)addr); nbytes = sz > 16? 16 : sz; for (j = 0; j < 16; j++) { if ((size_t)j < nbytes) printf("%2.2x", buf[j]); else printf(" "); if ((j & 3) == 3) printf(" "); } for (j = 0; (size_t)j < nbytes; j++) { if (isprint(buf[j])) printf("%c", buf[j]); else printf("."); } printf("\n"); buf += nbytes; addr += nbytes; sz -= nbytes; } } } static void str_dump(struct readelf *re) { struct section *s; Elf_Data *d; unsigned char *start, *end, *buf_end; unsigned int len; int i, j, elferr, found; for (i = 1; (size_t) i < re->shnum; i++) { s = &re->sl[i]; if (find_dumpop(re, (size_t) i, s->name, STR_DUMP, -1) == NULL) continue; (void) elf_errno(); if ((d = elf_getdata(s->scn, NULL)) == NULL && (d = elf_rawdata(s->scn, NULL)) == NULL) { elferr = elf_errno(); if (elferr != 0) warnx("elf_getdata failed: %s", elf_errmsg(elferr)); continue; } (void) elf_errno(); if (d->d_size <= 0 || d->d_buf == NULL) { printf("\nSection '%s' has no data to dump.\n", s->name); continue; } buf_end = (unsigned char *) d->d_buf + d->d_size; start = (unsigned char *) d->d_buf; found = 0; printf("\nString dump of section '%s':\n", s->name); for (;;) { while (start < buf_end && !isprint(*start)) start++; if (start >= buf_end) break; end = start + 1; while (end < buf_end && isprint(*end)) end++; printf(" [%6lx] ", (long) (start - (unsigned char *) d->d_buf)); len = end - start; for (j = 0; (unsigned int) j < len; j++) putchar(start[j]); putchar('\n'); found = 1; if (end >= buf_end) break; start = end + 1; } if (!found) printf(" No strings found in this section."); putchar('\n'); } } static void load_sections(struct readelf *re) { struct section *s; const char *name; Elf_Scn *scn; GElf_Shdr sh; size_t shstrndx, ndx; int elferr; /* Allocate storage for internal section list. */ if (!elf_getshnum(re->elf, &re->shnum)) { warnx("elf_getshnum failed: %s", elf_errmsg(-1)); return; } if (re->sl != NULL) free(re->sl); if ((re->sl = calloc(re->shnum, sizeof(*re->sl))) == NULL) err(EXIT_FAILURE, "calloc failed"); /* Get the index of .shstrtab section. */ if (!elf_getshstrndx(re->elf, &shstrndx)) { warnx("elf_getshstrndx failed: %s", elf_errmsg(-1)); return; } if ((scn = elf_getscn(re->elf, 0)) == NULL) return; (void) elf_errno(); do { if (gelf_getshdr(scn, &sh) == NULL) { warnx("gelf_getshdr failed: %s", elf_errmsg(-1)); (void) elf_errno(); continue; } if ((name = elf_strptr(re->elf, shstrndx, sh.sh_name)) == NULL) { (void) elf_errno(); name = ""; } if ((ndx = elf_ndxscn(scn)) == SHN_UNDEF) { if ((elferr = elf_errno()) != 0) { warnx("elf_ndxscn failed: %s", elf_errmsg(elferr)); continue; } } if (ndx >= re->shnum) { warnx("section index of '%s' out of range", name); continue; } if (sh.sh_link >= re->shnum) warnx("section link %llu of '%s' out of range", (unsigned long long)sh.sh_link, name); s = &re->sl[ndx]; s->name = name; s->scn = scn; s->off = sh.sh_offset; s->sz = sh.sh_size; s->entsize = sh.sh_entsize; s->align = sh.sh_addralign; s->type = sh.sh_type; s->flags = sh.sh_flags; s->addr = sh.sh_addr; s->link = sh.sh_link; s->info = sh.sh_info; } while ((scn = elf_nextscn(re->elf, scn)) != NULL); elferr = elf_errno(); if (elferr != 0) warnx("elf_nextscn failed: %s", elf_errmsg(elferr)); } static void unload_sections(struct readelf *re) { if (re->sl != NULL) { free(re->sl); re->sl = NULL; } re->shnum = 0; re->vd_s = NULL; re->vn_s = NULL; re->vs_s = NULL; re->vs = NULL; re->vs_sz = 0; if (re->ver != NULL) { free(re->ver); re->ver = NULL; re->ver_sz = 0; } } static void dump_elf(struct readelf *re) { /* Fetch ELF header. No need to continue if it fails. */ if (gelf_getehdr(re->elf, &re->ehdr) == NULL) { warnx("gelf_getehdr failed: %s", elf_errmsg(-1)); return; } if ((re->ec = gelf_getclass(re->elf)) == ELFCLASSNONE) { warnx("gelf_getclass failed: %s", elf_errmsg(-1)); return; } if (re->ehdr.e_ident[EI_DATA] == ELFDATA2MSB) { re->dw_read = _read_msb; re->dw_decode = _decode_msb; } else { re->dw_read = _read_lsb; re->dw_decode = _decode_lsb; } if (re->options & ~RE_H) load_sections(re); if ((re->options & RE_VV) || (re->options & RE_S)) search_ver(re); if (re->options & RE_H) dump_ehdr(re); if (re->options & RE_L) dump_phdr(re); if (re->options & RE_SS) dump_shdr(re); if (re->options & RE_G) dump_section_groups(re); if (re->options & RE_D) dump_dynamic(re); if (re->options & RE_R) dump_reloc(re); if (re->options & RE_S) dump_symtabs(re); if (re->options & RE_N) dump_notes(re); if (re->options & RE_II) dump_hash(re); if (re->options & RE_X) hex_dump(re); if (re->options & RE_P) str_dump(re); if (re->options & RE_VV) dump_ver(re); if (re->options & RE_AA) dump_arch_specific_info(re); if (re->options & RE_W) dump_dwarf(re); if (re->options & ~RE_H) unload_sections(re); } static void dump_dwarf(struct readelf *re) { Dwarf_Error de; int error; if (dwarf_elf_init(re->elf, DW_DLC_READ, NULL, NULL, &re->dbg, &de)) { if ((error = dwarf_errno(de)) != DW_DLE_DEBUG_INFO_NULL) errx(EXIT_FAILURE, "dwarf_elf_init failed: %s", dwarf_errmsg(de)); return; } if (re->dop & DW_A) dump_dwarf_abbrev(re); if (re->dop & DW_L) dump_dwarf_line(re); if (re->dop & DW_LL) dump_dwarf_line_decoded(re); if (re->dop & DW_I) { dump_dwarf_info(re, 0); dump_dwarf_info(re, 1); } if (re->dop & DW_P) dump_dwarf_pubnames(re); if (re->dop & DW_R) dump_dwarf_aranges(re); if (re->dop & DW_RR) dump_dwarf_ranges(re); if (re->dop & DW_M) dump_dwarf_macinfo(re); if (re->dop & DW_F) dump_dwarf_frame(re, 0); else if (re->dop & DW_FF) dump_dwarf_frame(re, 1); if (re->dop & DW_S) dump_dwarf_str(re); if (re->dop & DW_O) dump_dwarf_loclist(re); dwarf_finish(re->dbg, &de); } static void dump_ar(struct readelf *re, int fd) { Elf_Arsym *arsym; Elf_Arhdr *arhdr; Elf_Cmd cmd; Elf *e; size_t sz; off_t off; int i; re->ar = re->elf; if (re->options & RE_C) { if ((arsym = elf_getarsym(re->ar, &sz)) == NULL) { warnx("elf_getarsym() failed: %s", elf_errmsg(-1)); goto process_members; } printf("Index of archive %s: (%ju entries)\n", re->filename, (uintmax_t) sz - 1); off = 0; for (i = 0; (size_t) i < sz; i++) { if (arsym[i].as_name == NULL) break; if (arsym[i].as_off != off) { off = arsym[i].as_off; if (elf_rand(re->ar, off) != off) { warnx("elf_rand() failed: %s", elf_errmsg(-1)); continue; } if ((e = elf_begin(fd, ELF_C_READ, re->ar)) == NULL) { warnx("elf_begin() failed: %s", elf_errmsg(-1)); continue; } if ((arhdr = elf_getarhdr(e)) == NULL) { warnx("elf_getarhdr() failed: %s", elf_errmsg(-1)); elf_end(e); continue; } printf("Binary %s(%s) contains:\n", re->filename, arhdr->ar_name); } printf("\t%s\n", arsym[i].as_name); } if (elf_rand(re->ar, SARMAG) != SARMAG) { warnx("elf_rand() failed: %s", elf_errmsg(-1)); return; } } process_members: if ((re->options & ~RE_C) == 0) return; cmd = ELF_C_READ; while ((re->elf = elf_begin(fd, cmd, re->ar)) != NULL) { if ((arhdr = elf_getarhdr(re->elf)) == NULL) { warnx("elf_getarhdr() failed: %s", elf_errmsg(-1)); goto next_member; } if (strcmp(arhdr->ar_name, "/") == 0 || strcmp(arhdr->ar_name, "//") == 0 || strcmp(arhdr->ar_name, "__.SYMDEF") == 0) goto next_member; printf("\nFile: %s(%s)\n", re->filename, arhdr->ar_name); dump_elf(re); next_member: cmd = elf_next(re->elf); elf_end(re->elf); } re->elf = re->ar; } static void dump_object(struct readelf *re, int fd) { if ((re->flags & DISPLAY_FILENAME) != 0) printf("\nFile: %s\n", re->filename); if ((re->elf = elf_begin(fd, ELF_C_READ, NULL)) == NULL) { warnx("elf_begin() failed: %s", elf_errmsg(-1)); goto done; } switch (elf_kind(re->elf)) { case ELF_K_NONE: warnx("Not an ELF file."); goto done; case ELF_K_ELF: dump_elf(re); break; case ELF_K_AR: dump_ar(re, fd); break; default: warnx("Internal: libelf returned unknown elf kind."); } done: elf_end(re->elf); } static void add_dumpop(struct readelf *re, size_t si, const char *sn, int op, int t) { struct dumpop *d; if ((d = find_dumpop(re, si, sn, -1, t)) == NULL) { if ((d = calloc(1, sizeof(*d))) == NULL) err(EXIT_FAILURE, "calloc failed"); if (t == DUMP_BY_INDEX) d->u.si = si; else d->u.sn = sn; d->type = t; d->op = op; STAILQ_INSERT_TAIL(&re->v_dumpop, d, dumpop_list); } else d->op |= op; } static struct dumpop * find_dumpop(struct readelf *re, size_t si, const char *sn, int op, int t) { struct dumpop *d; STAILQ_FOREACH(d, &re->v_dumpop, dumpop_list) { if ((op == -1 || op & d->op) && (t == -1 || (unsigned) t == d->type)) { if ((d->type == DUMP_BY_INDEX && d->u.si == si) || (d->type == DUMP_BY_NAME && !strcmp(d->u.sn, sn))) return (d); } } return (NULL); } static struct { const char *ln; char sn; int value; } dwarf_op[] = { {"rawline", 'l', DW_L}, {"decodedline", 'L', DW_LL}, {"info", 'i', DW_I}, {"abbrev", 'a', DW_A}, {"pubnames", 'p', DW_P}, {"aranges", 'r', DW_R}, {"ranges", 'r', DW_R}, {"Ranges", 'R', DW_RR}, {"macro", 'm', DW_M}, {"frames", 'f', DW_F}, {"frames-interp", 'F', DW_FF}, {"str", 's', DW_S}, {"loc", 'o', DW_O}, {NULL, 0, 0} }; static void parse_dwarf_op_short(struct readelf *re, const char *op) { int i; if (op == NULL) { re->dop |= DW_DEFAULT_OPTIONS; return; } for (; *op != '\0'; op++) { for (i = 0; dwarf_op[i].ln != NULL; i++) { if (dwarf_op[i].sn == *op) { re->dop |= dwarf_op[i].value; break; } } } } static void parse_dwarf_op_long(struct readelf *re, const char *op) { char *p, *token, *bp; int i; if (op == NULL) { re->dop |= DW_DEFAULT_OPTIONS; return; } if ((p = strdup(op)) == NULL) err(EXIT_FAILURE, "strdup failed"); bp = p; while ((token = strsep(&p, ",")) != NULL) { for (i = 0; dwarf_op[i].ln != NULL; i++) { if (!strcmp(token, dwarf_op[i].ln)) { re->dop |= dwarf_op[i].value; break; } } } free(bp); } static uint64_t _read_lsb(Elf_Data *d, uint64_t *offsetp, int bytes_to_read) { uint64_t ret; uint8_t *src; src = (uint8_t *) d->d_buf + *offsetp; ret = 0; switch (bytes_to_read) { case 8: ret |= ((uint64_t) src[4]) << 32 | ((uint64_t) src[5]) << 40; ret |= ((uint64_t) src[6]) << 48 | ((uint64_t) src[7]) << 56; /* FALLTHROUGH */ case 4: ret |= ((uint64_t) src[2]) << 16 | ((uint64_t) src[3]) << 24; /* FALLTHROUGH */ case 2: ret |= ((uint64_t) src[1]) << 8; /* FALLTHROUGH */ case 1: ret |= src[0]; break; default: return (0); } *offsetp += bytes_to_read; return (ret); } static uint64_t _read_msb(Elf_Data *d, uint64_t *offsetp, int bytes_to_read) { uint64_t ret; uint8_t *src; src = (uint8_t *) d->d_buf + *offsetp; switch (bytes_to_read) { case 1: ret = src[0]; break; case 2: ret = src[1] | ((uint64_t) src[0]) << 8; break; case 4: ret = src[3] | ((uint64_t) src[2]) << 8; ret |= ((uint64_t) src[1]) << 16 | ((uint64_t) src[0]) << 24; break; case 8: ret = src[7] | ((uint64_t) src[6]) << 8; ret |= ((uint64_t) src[5]) << 16 | ((uint64_t) src[4]) << 24; ret |= ((uint64_t) src[3]) << 32 | ((uint64_t) src[2]) << 40; ret |= ((uint64_t) src[1]) << 48 | ((uint64_t) src[0]) << 56; break; default: return (0); } *offsetp += bytes_to_read; return (ret); } static uint64_t _decode_lsb(uint8_t **data, int bytes_to_read) { uint64_t ret; uint8_t *src; src = *data; ret = 0; switch (bytes_to_read) { case 8: ret |= ((uint64_t) src[4]) << 32 | ((uint64_t) src[5]) << 40; ret |= ((uint64_t) src[6]) << 48 | ((uint64_t) src[7]) << 56; /* FALLTHROUGH */ case 4: ret |= ((uint64_t) src[2]) << 16 | ((uint64_t) src[3]) << 24; /* FALLTHROUGH */ case 2: ret |= ((uint64_t) src[1]) << 8; /* FALLTHROUGH */ case 1: ret |= src[0]; break; default: return (0); } *data += bytes_to_read; return (ret); } static uint64_t _decode_msb(uint8_t **data, int bytes_to_read) { uint64_t ret; uint8_t *src; src = *data; ret = 0; switch (bytes_to_read) { case 1: ret = src[0]; break; case 2: ret = src[1] | ((uint64_t) src[0]) << 8; break; case 4: ret = src[3] | ((uint64_t) src[2]) << 8; ret |= ((uint64_t) src[1]) << 16 | ((uint64_t) src[0]) << 24; break; case 8: ret = src[7] | ((uint64_t) src[6]) << 8; ret |= ((uint64_t) src[5]) << 16 | ((uint64_t) src[4]) << 24; ret |= ((uint64_t) src[3]) << 32 | ((uint64_t) src[2]) << 40; ret |= ((uint64_t) src[1]) << 48 | ((uint64_t) src[0]) << 56; break; default: return (0); break; } *data += bytes_to_read; return (ret); } static int64_t _decode_sleb128(uint8_t **dp, uint8_t *dpe) { int64_t ret = 0; uint8_t b = 0; int shift = 0; uint8_t *src = *dp; do { if (src >= dpe) break; b = *src++; ret |= ((b & 0x7f) << shift); shift += 7; } while ((b & 0x80) != 0); if (shift < 32 && (b & 0x40) != 0) ret |= (-1 << shift); *dp = src; return (ret); } static uint64_t _decode_uleb128(uint8_t **dp, uint8_t *dpe) { uint64_t ret = 0; uint8_t b; int shift = 0; uint8_t *src = *dp; do { if (src >= dpe) break; b = *src++; ret |= ((b & 0x7f) << shift); shift += 7; } while ((b & 0x80) != 0); *dp = src; return (ret); } static void readelf_version(void) { (void) printf("%s (%s)\n", ELFTC_GETPROGNAME(), elftc_version()); exit(EXIT_SUCCESS); } #define USAGE_MESSAGE "\ Usage: %s [options] file...\n\ Display information about ELF objects and ar(1) archives.\n\n\ Options:\n\ -a | --all Equivalent to specifying options '-dhIlrsASV'.\n\ -c | --archive-index Print the archive symbol table for archives.\n\ -d | --dynamic Print the contents of SHT_DYNAMIC sections.\n\ -e | --headers Print all headers in the object.\n\ -g | --section-groups Print the contents of the section groups.\n\ -h | --file-header Print the file header for the object.\n\ -l | --program-headers Print the PHDR table for the object.\n\ -n | --notes Print the contents of SHT_NOTE sections.\n\ -p INDEX | --string-dump=INDEX\n\ Print the contents of section at index INDEX.\n\ -r | --relocs Print relocation information.\n\ -s | --syms | --symbols Print symbol tables.\n\ -t | --section-details Print additional information about sections.\n\ -v | --version Print a version identifier and exit.\n\ -w[afilmoprsFLR] | --debug-dump={abbrev,aranges,decodedline,frames,\n\ frames-interp,info,loc,macro,pubnames,\n\ ranges,Ranges,rawline,str}\n\ Display DWARF information.\n\ -x INDEX | --hex-dump=INDEX\n\ Display contents of a section as hexadecimal.\n\ -A | --arch-specific (accepted, but ignored)\n\ -D | --use-dynamic Print the symbol table specified by the DT_SYMTAB\n\ entry in the \".dynamic\" section.\n\ -H | --help Print a help message.\n\ -I | --histogram Print information on bucket list lengths for \n\ hash sections.\n\ -N | --full-section-name (accepted, but ignored)\n\ -S | --sections | --section-headers\n\ Print information about section headers.\n\ -V | --version-info Print symbol versoning information.\n\ -W | --wide Print information without wrapping long lines.\n" static void readelf_usage(int status) { fprintf(stderr, USAGE_MESSAGE, ELFTC_GETPROGNAME()); exit(status); } int main(int argc, char **argv) { cap_rights_t rights; fileargs_t *fa; struct readelf *re, re_storage; unsigned long si; int fd, opt, i; char *ep; re = &re_storage; memset(re, 0, sizeof(*re)); STAILQ_INIT(&re->v_dumpop); while ((opt = getopt_long(argc, argv, "AacDdegHhIi:lNnp:rSstuVvWw::x:", longopts, NULL)) != -1) { switch(opt) { case '?': readelf_usage(EXIT_SUCCESS); break; case 'A': re->options |= RE_AA; break; case 'a': re->options |= RE_AA | RE_D | RE_G | RE_H | RE_II | RE_L | RE_R | RE_SS | RE_S | RE_VV; break; case 'c': re->options |= RE_C; break; case 'D': re->options |= RE_DD; break; case 'd': re->options |= RE_D; break; case 'e': re->options |= RE_H | RE_L | RE_SS; break; case 'g': re->options |= RE_G; break; case 'H': readelf_usage(EXIT_SUCCESS); break; case 'h': re->options |= RE_H; break; case 'I': re->options |= RE_II; break; case 'i': /* Not implemented yet. */ break; case 'l': re->options |= RE_L; break; case 'N': re->options |= RE_NN; break; case 'n': re->options |= RE_N; break; case 'p': re->options |= RE_P; si = strtoul(optarg, &ep, 10); if (*ep == '\0') add_dumpop(re, (size_t) si, NULL, STR_DUMP, DUMP_BY_INDEX); else add_dumpop(re, 0, optarg, STR_DUMP, DUMP_BY_NAME); break; case 'r': re->options |= RE_R; break; case 'S': re->options |= RE_SS; break; case 's': re->options |= RE_S; break; case 't': re->options |= RE_SS | RE_T; break; case 'u': re->options |= RE_U; break; case 'V': re->options |= RE_VV; break; case 'v': readelf_version(); break; case 'W': re->options |= RE_WW; break; case 'w': re->options |= RE_W; parse_dwarf_op_short(re, optarg); break; case 'x': re->options |= RE_X; si = strtoul(optarg, &ep, 10); if (*ep == '\0') add_dumpop(re, (size_t) si, NULL, HEX_DUMP, DUMP_BY_INDEX); else add_dumpop(re, 0, optarg, HEX_DUMP, DUMP_BY_NAME); break; case OPTION_DEBUG_DUMP: re->options |= RE_W; parse_dwarf_op_long(re, optarg); } } argv += optind; argc -= optind; if (argc == 0 || re->options == 0) readelf_usage(EXIT_FAILURE); if (argc > 1) re->flags |= DISPLAY_FILENAME; if (elf_version(EV_CURRENT) == EV_NONE) errx(EXIT_FAILURE, "ELF library initialization failed: %s", elf_errmsg(-1)); cap_rights_init(&rights, CAP_FCNTL, CAP_FSTAT, CAP_MMAP_R, CAP_SEEK); fa = fileargs_init(argc, argv, O_RDONLY, 0, &rights, FA_OPEN); if (fa == NULL) err(1, "Unable to initialize casper fileargs"); caph_cache_catpages(); if (caph_limit_stdio() < 0) { fileargs_free(fa); err(1, "Unable to limit stdio rights"); } if (caph_enter_casper() < 0) { fileargs_free(fa); err(1, "Unable to enter capability mode"); } for (i = 0; i < argc; i++) { re->filename = argv[i]; fd = fileargs_open(fa, re->filename); if (fd < 0) { warn("open %s failed", re->filename); } else { dump_object(re, fd); close(fd); } } exit(EXIT_SUCCESS); } diff --git a/share/man/man9/counter.9 b/share/man/man9/counter.9 index 9bdb5320b71b..1eb36b571249 100644 --- a/share/man/man9/counter.9 +++ b/share/man/man9/counter.9 @@ -1,261 +1,278 @@ .\"- .\" Copyright (c) 2013 Gleb Smirnoff .\" All rights reserved. .\" .\" Redistribution and use in source and binary forms, with or without .\" modification, are permitted provided that the following conditions .\" are met: .\" 1. Redistributions of source code must retain the above copyright .\" notice, this list of conditions and the following disclaimer. .\" 2. Redistributions in binary form must reproduce the above copyright .\" notice, this list of conditions and the following disclaimer in the .\" documentation and/or other materials provided with the distribution. .\" .\" THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND .\" ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE .\" IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE .\" ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE .\" FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL .\" DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS .\" OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) .\" HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT .\" LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY .\" OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF .\" SUCH DAMAGE. .\" .\" $FreeBSD$ .\" -.Dd March 22, 2017 +.Dd March 6, 2020 .Dt COUNTER 9 .Os .Sh NAME .Nm counter .Nd "SMP-friendly kernel counter implementation" .Sh SYNOPSIS .In sys/types.h .In sys/systm.h .In sys/counter.h .Ft counter_u64_t .Fn counter_u64_alloc "int wait" .Ft void .Fn counter_u64_free "counter_u64_t c" .Ft void .Fn counter_u64_add "counter_u64_t c" "int64_t v" .Ft void .Fn counter_enter .Ft void .Fn counter_exit .Ft void .Fn counter_u64_add_protected "counter_u64_t c" "int64_t v" .Ft uint64_t .Fn counter_u64_fetch "counter_u64_t c" .Ft void .Fn counter_u64_zero "counter_u64_t c" .Ft int64_t .Fn counter_ratecheck "struct counter_rate *cr" "int64_t limit" +.Fn COUNTER_U64_SYSINIT "counter_u64_t c" +.Fn COUNTER_U64_DEFINE_EARLY "counter_u64_t c" .In sys/sysctl.h .Fn SYSCTL_COUNTER_U64 parent nbr name access ptr descr .Fn SYSCTL_ADD_COUNTER_U64 ctx parent nbr name access ptr descr .Fn SYSCTL_COUNTER_U64_ARRAY parent nbr name access ptr len descr .Fn SYSCTL_ADD_COUNTER_U64_ARRAY ctx parent nbr name access ptr len descr .Sh DESCRIPTION .Nm is a generic facility to create counters that can be utilized for any purpose (such as collecting statistical data). A .Nm is guaranteed to be lossless when several kernel threads do simultaneous updates. However, .Nm does not block the calling thread, also no .Xr atomic 9 operations are used for the update, therefore the counters can be used in any non-interrupt context. Moreover, .Nm has special optimisations for SMP environments, making .Nm update faster than simple arithmetic on the global variable. Thus .Nm is considered suitable for accounting in the performance-critical code paths. .Bl -tag -width indent .It Fn counter_u64_alloc wait Allocate a new 64-bit unsigned counter. The .Fa wait argument is the .Xr malloc 9 wait flag, should be either .Va M_NOWAIT or .Va M_WAITOK . If .Va M_NOWAIT is specified the operation may fail. .It Fn counter_u64_free c Free the previously allocated counter .Fa c . .It Fn counter_u64_add c v Add .Fa v to .Fa c . The KPI does not guarantee any protection from wraparound. .It Fn counter_enter Enter mode that would allow the safe update of several counters via .Fn counter_u64_add_protected . On some machines this expands to .Xr critical 9 section, while on other is a nop. See .Sx IMPLEMENTATION DETAILS . .It Fn counter_exit Exit mode for updating several counters. .It Fn counter_u64_add_protected c v Same as .Fn counter_u64_add , but should be preceded by .Fn counter_enter . .It Fn counter_u64_fetch c Take a snapshot of counter .Fa c . The data obtained is not guaranteed to reflect the real cumulative value for any moment. .It Fn counter_u64_zero c Clear the counter .Fa c and set it to zero. .It Fn counter_ratecheck cr limit The function is a multiprocessor-friendly version of .Fn ppsratecheck which uses .Nm internally. Returns non-negative value if the rate is not yet reached during the current second, and a negative value otherwise. If the limit was reached on previous second, but was just reset back to zero, then .Fn counter_ratecheck returns number of events since previous reset. +.It Fn COUNTER_U64_SYSINIT c +Define a +.Xr SYSINIT 9 +initializer for the global counter +.Fa c . +.It Fn COUNTER_U64_DEFINE_EARLY c +Define and initialize a global counter +.Fa c . +It is always safe to increment +.Fa c , +though updates prior to the +.Dv SI_SUB_COUNTER +.Xr SYSINIT 9 +event are lost. .It Fn SYSCTL_COUNTER_U64 parent nbr name access ptr descr Declare a static .Xr sysctl 9 oid that would represent a .Nm . The .Fa ptr argument should be a pointer to allocated .Vt counter_u64_t . A read of the oid returns value obtained through .Fn counter_u64_fetch . Any write to the oid zeroes it. .It Fn SYSCTL_ADD_COUNTER_U64 ctx parent nbr name access ptr descr Create a .Xr sysctl 9 oid that would represent a .Nm . The .Fa ptr argument should be a pointer to allocated .Vt counter_u64_t . A read of the oid returns value obtained through .Fn counter_u64_fetch . Any write to the oid zeroes it. .It Fn SYSCTL_COUNTER_U64_ARRAY parent nbr name access ptr len descr Declare a static .Xr sysctl 9 oid that would represent an array of .Nm . The .Fa ptr argument should be a pointer to allocated array of .Vt counter_u64_t's . The .Fa len argument should specify number of elements in the array. A read of the oid returns len-sized array of .Vt uint64_t values obtained through .Fn counter_u64_fetch . Any write to the oid zeroes all array elements. .It Fn SYSCTL_ADD_COUNTER_U64_ARRAY ctx parent nbr name access ptr len descr Create a .Xr sysctl 9 oid that would represent an array of .Nm . The .Fa ptr argument should be a pointer to allocated array of .Vt counter_u64_t's . The .Fa len argument should specify number of elements in the array. A read of the oid returns len-sized array of .Vt uint64_t values obtained through .Fn counter_u64_fetch . Any write to the oid zeroes all array elements. .El .Sh IMPLEMENTATION DETAILS On all architectures .Nm is implemented using per-CPU data fields that are specially aligned in memory, to avoid inter-CPU bus traffic due to shared use of the variables between CPUs. These are allocated using .Va UMA_ZONE_PCPU .Xr uma 9 zone. The update operation only touches the field that is private to current CPU. Fetch operation loops through all per-CPU fields and obtains a snapshot sum of all fields. .Pp On amd64 a .Nm counter update is implemented as a single instruction without lock semantics, operating on the private data for the current CPU, which is safe against preemption and interrupts. .Pp On i386 architecture, when machine supports the cmpxchg8 instruction, this instruction is used. The multi-instruction sequence provides the same guarantees as the amd64 single-instruction implementation. .Pp On some architectures updating a counter require a .Xr critical 9 section. .Sh EXAMPLES The following example creates a static counter array exported to userspace through a sysctl: .Bd -literal -offset indent #define MY_SIZE 8 static counter_u64_t array[MY_SIZE]; SYSCTL_COUNTER_U64_ARRAY(_debug, OID_AUTO, counter_array, CTLFLAG_RW, &array[0], MY_SIZE, "Test counter array"); .Ed .Sh SEE ALSO .Xr atomic 9 , .Xr critical 9 , .Xr locking 9 , .Xr malloc 9 , .Xr ratecheck 9 , .Xr sysctl 9 , +.Xr SYSINIT 9 , .Xr uma 9 .Sh HISTORY The .Nm facility first appeared in .Fx 10.0 . .Sh AUTHORS .An -nosplit The .Nm facility was written by .An Gleb Smirnoff and .An Konstantin Belousov . diff --git a/share/mk/bsd.compat.mk b/share/mk/bsd.compat.mk index 47c8ec65c88f..d5b096032470 100644 --- a/share/mk/bsd.compat.mk +++ b/share/mk/bsd.compat.mk @@ -1,172 +1,172 @@ # $FreeBSD$ .if !targets(__<${_this:T}>__) __<${_this:T}>__: .if defined(_LIBCOMPAT) -COMPAT_ARCH= ${TARGET_ARCH} -COMPAT_CPUTYPE= ${TARGET_CPUTYPE} +COMPAT_ARCH?= ${TARGET_ARCH} +COMPAT_CPUTYPE?= ${CPUTYPE_${_LIBCOMPAT}} .if (defined(WANT_COMPILER_TYPE) && ${WANT_COMPILER_TYPE} == gcc) || \ (defined(X_COMPILER_TYPE) && ${X_COMPILER_TYPE} == gcc) COMPAT_COMPILER_TYPE= gcc .else COMPAT_COMPILER_TYPE= clang .endif .else COMPAT_ARCH= ${MACHINE_ARCH} COMPAT_CPUTYPE= ${CPUTYPE} .include COMPAT_COMPILER_TYPE=${COMPILER_TYPE} .endif # ------------------------------------------------------------------- # 32 bit world .if ${COMPAT_ARCH} == "amd64" HAS_COMPAT=32 .if empty(COMPAT_CPUTYPE) LIB32CPUFLAGS= -march=i686 -mmmx -msse -msse2 .else LIB32CPUFLAGS= -march=${COMPAT_CPUTYPE} .endif .if ${COMPAT_COMPILER_TYPE} == gcc .else LIB32CPUFLAGS+= -target x86_64-unknown-freebsd13.0 .endif LIB32CPUFLAGS+= -m32 LIB32_MACHINE= i386 LIB32_MACHINE_ARCH= i386 LIB32WMAKEENV= MACHINE_CPU="i686 mmx sse sse2" LIB32WMAKEFLAGS= \ AS="${XAS} --32" \ LD="${XLD} -m elf_i386_fbsd -L${LIBCOMPATTMP}/usr/lib32" .elif ${COMPAT_ARCH} == "powerpc64" HAS_COMPAT=32 .if empty(COMPAT_CPUTYPE) LIB32CPUFLAGS= -mcpu=powerpc .else LIB32CPUFLAGS= -mcpu=${COMPAT_CPUTYPE} .endif .if ${COMPAT_COMPILER_TYPE} == "gcc" LIB32CPUFLAGS+= -m32 .else LIB32CPUFLAGS+= -target powerpc-unknown-freebsd13.0 # Use BFD to workaround ld.lld issues on PowerPC 32 bit LIB32CPUFLAGS+= -fuse-ld=${LD_BFD} .endif LIB32_MACHINE= powerpc LIB32_MACHINE_ARCH= powerpc LIB32WMAKEFLAGS= \ LD="${LD_BFD} -m elf32ppc_fbsd" .elif ${COMPAT_ARCH:Mmips64*} != "" HAS_COMPAT=32 .if ${COMPAT_COMPILER_TYPE} == gcc .if empty(COMPAT_CPUTYPE) LIB32CPUFLAGS= -march=mips3 .else LIB32CPUFLAGS= -march=${COMPAT_CPUTYPE} .endif .else .if ${COMPAT_ARCH:Mmips64el*} != "" LIB32CPUFLAGS= -target mipsel-unknown-freebsd13.0 .else LIB32CPUFLAGS= -target mips-unknown-freebsd13.0 .endif .endif LIB32CPUFLAGS+= -mabi=32 LIB32_MACHINE= mips LIB32_MACHINE_ARCH:= ${COMPAT_ARCH:S/64//} .if ${COMPAT_ARCH:Mmips64el*} != "" _EMULATION= elf32ltsmip_fbsd .else _EMULATION= elf32btsmip_fbsd .endif LIB32WMAKEFLAGS= LD="${XLD} -m ${_EMULATION}" LIB32LDFLAGS= -Wl,-m${_EMULATION} .endif LIB32WMAKEFLAGS+= NM="${XNM}" LIB32WMAKEFLAGS+= OBJCOPY="${XOBJCOPY}" LIB32CFLAGS= -DCOMPAT_32BIT LIB32DTRACE= ${DTRACE} -32 LIB32WMAKEFLAGS+= -DCOMPAT_32BIT # ------------------------------------------------------------------- # soft-fp world .if ${COMPAT_ARCH:Marmv[67]*} != "" HAS_COMPAT=SOFT LIBSOFTCFLAGS= -DCOMPAT_SOFTFP LIBSOFTCPUFLAGS= -mfloat-abi=softfp LIBSOFT_MACHINE= arm LIBSOFT_MACHINE_ARCH= ${COMPAT_ARCH} LIBSOFTWMAKEENV= CPUTYPE=soft LIBSOFTWMAKEFLAGS= -DCOMPAT_SOFTFP .endif # ------------------------------------------------------------------- # In the program linking case, select LIBCOMPAT .if defined(NEED_COMPAT) .ifndef HAS_COMPAT .warning NEED_COMPAT defined, but no LIBCOMPAT is available (COMPAT_ARCH == ${COMPAT_ARCH} .elif !${HAS_COMPAT:M${NEED_COMPAT}} && ${NEED_COMPAT} != "any" .error NEED_COMPAT (${NEED_COMPAT}) defined, but not in HAS_COMPAT ($HAS_COMPAT) .elif ${NEED_COMPAT} == "any" .endif .ifdef WANT_COMPAT .error Both WANT_COMPAT and NEED_COMPAT defined .endif WANT_COMPAT:= ${NEED_COMPAT} .endif .if defined(HAS_COMPAT) && defined(WANT_COMPAT) .if ${WANT_COMPAT} == "any" _LIBCOMPAT:= ${HAS_COMPAT:[1]} .else _LIBCOMPAT:= ${WANT_COMPAT} .endif .endif # ------------------------------------------------------------------- # Generic code for each type. # Set defaults based on type. libcompat= ${_LIBCOMPAT:tl} _LIBCOMPAT_MAKEVARS= _OBJTOP TMP CPUFLAGS CFLAGS CXXFLAGS LDFLAGS \ _MACHINE _MACHINE_ARCH WMAKEENV WMAKEFLAGS WMAKE .for _var in ${_LIBCOMPAT_MAKEVARS} .if !empty(LIB${_LIBCOMPAT}${_var}) LIBCOMPAT${_var}?= ${LIB${_LIBCOMPAT}${_var}} .endif .endfor # Shared flags LIBCOMPAT_OBJTOP?= ${OBJTOP}/obj-lib${libcompat} LIBCOMPATTMP?= ${LIBCOMPAT_OBJTOP}/tmp LIBCOMPATCFLAGS+= ${LIBCOMPATCPUFLAGS} \ -L${LIBCOMPATTMP}/usr/lib${libcompat} \ --sysroot=${LIBCOMPATTMP} \ ${BFLAGS} LIBCOMPATWMAKEENV+= MACHINE=${LIBCOMPAT_MACHINE} LIBCOMPATWMAKEENV+= MACHINE_ARCH=${LIBCOMPAT_MACHINE_ARCH} # -B is needed to find /usr/lib32/crti.o for GCC and /usr/libsoft/crti.o for # Clang/GCC. LIBCOMPATCFLAGS+= -B${LIBCOMPATTMP}/usr/lib${libcompat} .if defined(WANT_COMPAT) LIBDIR_BASE:= /usr/lib${libcompat} _LIB_OBJTOP= ${LIBCOMPAT_OBJTOP} CFLAGS+= ${LIBCOMPATCFLAGS} LDFLAGS+= ${CFLAGS} ${LIBCOMPATLDFLAGS} MACHINE= ${LIBCOMPAT_MACHINE} MACHINE_ARCH= ${LIBCOMPAT_MACHINE_ARCH} .endif .endif diff --git a/sys/cam/cam_periph.c b/sys/cam/cam_periph.c index 7d8f671eecd3..c671b9c75bae 100644 --- a/sys/cam/cam_periph.c +++ b/sys/cam/cam_periph.c @@ -1,2187 +1,2187 @@ /*- * Common functions for CAM "type" (peripheral) drivers. * * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 1997, 1998 Justin T. Gibbs. * Copyright (c) 1997, 1998, 1999, 2000 Kenneth D. Merry. * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions, and the following disclaimer, * without modification, immediately at the beginning of the file. * 2. The name of the author may not be used to endorse or promote products * derived from this software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE FOR * ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include static u_int camperiphnextunit(struct periph_driver *p_drv, u_int newunit, int wired, path_id_t pathid, target_id_t target, lun_id_t lun); static u_int camperiphunit(struct periph_driver *p_drv, path_id_t pathid, target_id_t target, lun_id_t lun); static void camperiphdone(struct cam_periph *periph, union ccb *done_ccb); static void camperiphfree(struct cam_periph *periph); static int camperiphscsistatuserror(union ccb *ccb, union ccb **orig_ccb, cam_flags camflags, u_int32_t sense_flags, int *openings, u_int32_t *relsim_flags, u_int32_t *timeout, u_int32_t *action, const char **action_string); static int camperiphscsisenseerror(union ccb *ccb, union ccb **orig_ccb, cam_flags camflags, u_int32_t sense_flags, int *openings, u_int32_t *relsim_flags, u_int32_t *timeout, u_int32_t *action, const char **action_string); static void cam_periph_devctl_notify(union ccb *ccb); static int nperiph_drivers; static int initialized = 0; struct periph_driver **periph_drivers; static MALLOC_DEFINE(M_CAMPERIPH, "CAM periph", "CAM peripheral buffers"); static int periph_selto_delay = 1000; TUNABLE_INT("kern.cam.periph_selto_delay", &periph_selto_delay); static int periph_noresrc_delay = 500; TUNABLE_INT("kern.cam.periph_noresrc_delay", &periph_noresrc_delay); static int periph_busy_delay = 500; TUNABLE_INT("kern.cam.periph_busy_delay", &periph_busy_delay); static u_int periph_mapmem_thresh = 65536; SYSCTL_UINT(_kern_cam, OID_AUTO, mapmem_thresh, CTLFLAG_RWTUN, &periph_mapmem_thresh, 0, "Threshold for user-space buffer mapping"); void periphdriver_register(void *data) { struct periph_driver *drv = (struct periph_driver *)data; struct periph_driver **newdrivers, **old; int ndrivers; again: ndrivers = nperiph_drivers + 2; newdrivers = malloc(sizeof(*newdrivers) * ndrivers, M_CAMPERIPH, M_WAITOK); xpt_lock_buses(); if (ndrivers != nperiph_drivers + 2) { /* * Lost race against itself; go around. */ xpt_unlock_buses(); free(newdrivers, M_CAMPERIPH); goto again; } if (periph_drivers) bcopy(periph_drivers, newdrivers, sizeof(*newdrivers) * nperiph_drivers); newdrivers[nperiph_drivers] = drv; newdrivers[nperiph_drivers + 1] = NULL; old = periph_drivers; periph_drivers = newdrivers; nperiph_drivers++; xpt_unlock_buses(); if (old) free(old, M_CAMPERIPH); /* If driver marked as early or it is late now, initialize it. */ if (((drv->flags & CAM_PERIPH_DRV_EARLY) != 0 && initialized > 0) || initialized > 1) (*drv->init)(); } int periphdriver_unregister(void *data) { struct periph_driver *drv = (struct periph_driver *)data; int error, n; /* If driver marked as early or it is late now, deinitialize it. */ if (((drv->flags & CAM_PERIPH_DRV_EARLY) != 0 && initialized > 0) || initialized > 1) { if (drv->deinit == NULL) { printf("CAM periph driver '%s' doesn't have deinit.\n", drv->driver_name); return (EOPNOTSUPP); } error = drv->deinit(); if (error != 0) return (error); } xpt_lock_buses(); for (n = 0; n < nperiph_drivers && periph_drivers[n] != drv; n++) ; KASSERT(n < nperiph_drivers, ("Periph driver '%s' was not registered", drv->driver_name)); for (; n + 1 < nperiph_drivers; n++) periph_drivers[n] = periph_drivers[n + 1]; periph_drivers[n + 1] = NULL; nperiph_drivers--; xpt_unlock_buses(); return (0); } void periphdriver_init(int level) { int i, early; initialized = max(initialized, level); for (i = 0; periph_drivers[i] != NULL; i++) { early = (periph_drivers[i]->flags & CAM_PERIPH_DRV_EARLY) ? 1 : 2; if (early == initialized) (*periph_drivers[i]->init)(); } } cam_status cam_periph_alloc(periph_ctor_t *periph_ctor, periph_oninv_t *periph_oninvalidate, periph_dtor_t *periph_dtor, periph_start_t *periph_start, char *name, cam_periph_type type, struct cam_path *path, ac_callback_t *ac_callback, ac_code code, void *arg) { struct periph_driver **p_drv; struct cam_sim *sim; struct cam_periph *periph; struct cam_periph *cur_periph; path_id_t path_id; target_id_t target_id; lun_id_t lun_id; cam_status status; u_int init_level; init_level = 0; /* * Handle Hot-Plug scenarios. If there is already a peripheral * of our type assigned to this path, we are likely waiting for * final close on an old, invalidated, peripheral. If this is * the case, queue up a deferred call to the peripheral's async * handler. If it looks like a mistaken re-allocation, complain. */ if ((periph = cam_periph_find(path, name)) != NULL) { if ((periph->flags & CAM_PERIPH_INVALID) != 0 && (periph->flags & CAM_PERIPH_NEW_DEV_FOUND) == 0) { periph->flags |= CAM_PERIPH_NEW_DEV_FOUND; periph->deferred_callback = ac_callback; periph->deferred_ac = code; return (CAM_REQ_INPROG); } else { printf("cam_periph_alloc: attempt to re-allocate " "valid device %s%d rejected flags %#x " "refcount %d\n", periph->periph_name, periph->unit_number, periph->flags, periph->refcount); } return (CAM_REQ_INVALID); } periph = (struct cam_periph *)malloc(sizeof(*periph), M_CAMPERIPH, M_NOWAIT|M_ZERO); if (periph == NULL) return (CAM_RESRC_UNAVAIL); init_level++; sim = xpt_path_sim(path); path_id = xpt_path_path_id(path); target_id = xpt_path_target_id(path); lun_id = xpt_path_lun_id(path); periph->periph_start = periph_start; periph->periph_dtor = periph_dtor; periph->periph_oninval = periph_oninvalidate; periph->type = type; periph->periph_name = name; periph->scheduled_priority = CAM_PRIORITY_NONE; periph->immediate_priority = CAM_PRIORITY_NONE; periph->refcount = 1; /* Dropped by invalidation. */ periph->sim = sim; SLIST_INIT(&periph->ccb_list); status = xpt_create_path(&path, periph, path_id, target_id, lun_id); if (status != CAM_REQ_CMP) goto failure; periph->path = path; xpt_lock_buses(); for (p_drv = periph_drivers; *p_drv != NULL; p_drv++) { if (strcmp((*p_drv)->driver_name, name) == 0) break; } if (*p_drv == NULL) { printf("cam_periph_alloc: invalid periph name '%s'\n", name); xpt_unlock_buses(); xpt_free_path(periph->path); free(periph, M_CAMPERIPH); return (CAM_REQ_INVALID); } periph->unit_number = camperiphunit(*p_drv, path_id, target_id, lun_id); cur_periph = TAILQ_FIRST(&(*p_drv)->units); while (cur_periph != NULL && cur_periph->unit_number < periph->unit_number) cur_periph = TAILQ_NEXT(cur_periph, unit_links); if (cur_periph != NULL) { KASSERT(cur_periph->unit_number != periph->unit_number, ("duplicate units on periph list")); TAILQ_INSERT_BEFORE(cur_periph, periph, unit_links); } else { TAILQ_INSERT_TAIL(&(*p_drv)->units, periph, unit_links); (*p_drv)->generation++; } xpt_unlock_buses(); init_level++; status = xpt_add_periph(periph); if (status != CAM_REQ_CMP) goto failure; init_level++; CAM_DEBUG(periph->path, CAM_DEBUG_INFO, ("Periph created\n")); status = periph_ctor(periph, arg); if (status == CAM_REQ_CMP) init_level++; failure: switch (init_level) { case 4: /* Initialized successfully */ break; case 3: CAM_DEBUG(periph->path, CAM_DEBUG_INFO, ("Periph destroyed\n")); xpt_remove_periph(periph); /* FALLTHROUGH */ case 2: xpt_lock_buses(); TAILQ_REMOVE(&(*p_drv)->units, periph, unit_links); xpt_unlock_buses(); xpt_free_path(periph->path); /* FALLTHROUGH */ case 1: free(periph, M_CAMPERIPH); /* FALLTHROUGH */ case 0: /* No cleanup to perform. */ break; default: panic("%s: Unknown init level", __func__); } return(status); } /* * Find a peripheral structure with the specified path, target, lun, * and (optionally) type. If the name is NULL, this function will return * the first peripheral driver that matches the specified path. */ struct cam_periph * cam_periph_find(struct cam_path *path, char *name) { struct periph_driver **p_drv; struct cam_periph *periph; xpt_lock_buses(); for (p_drv = periph_drivers; *p_drv != NULL; p_drv++) { if (name != NULL && (strcmp((*p_drv)->driver_name, name) != 0)) continue; TAILQ_FOREACH(periph, &(*p_drv)->units, unit_links) { if (xpt_path_comp(periph->path, path) == 0) { xpt_unlock_buses(); cam_periph_assert(periph, MA_OWNED); return(periph); } } if (name != NULL) { xpt_unlock_buses(); return(NULL); } } xpt_unlock_buses(); return(NULL); } /* * Find peripheral driver instances attached to the specified path. */ int cam_periph_list(struct cam_path *path, struct sbuf *sb) { struct sbuf local_sb; struct periph_driver **p_drv; struct cam_periph *periph; int count; int sbuf_alloc_len; sbuf_alloc_len = 16; retry: sbuf_new(&local_sb, NULL, sbuf_alloc_len, SBUF_FIXEDLEN); count = 0; xpt_lock_buses(); for (p_drv = periph_drivers; *p_drv != NULL; p_drv++) { TAILQ_FOREACH(periph, &(*p_drv)->units, unit_links) { if (xpt_path_comp(periph->path, path) != 0) continue; if (sbuf_len(&local_sb) != 0) sbuf_cat(&local_sb, ","); sbuf_printf(&local_sb, "%s%d", periph->periph_name, periph->unit_number); if (sbuf_error(&local_sb) == ENOMEM) { sbuf_alloc_len *= 2; xpt_unlock_buses(); sbuf_delete(&local_sb); goto retry; } count++; } } xpt_unlock_buses(); sbuf_finish(&local_sb); if (sbuf_len(sb) != 0) sbuf_cat(sb, ","); sbuf_cat(sb, sbuf_data(&local_sb)); sbuf_delete(&local_sb); return (count); } int cam_periph_acquire(struct cam_periph *periph) { int status; if (periph == NULL) return (EINVAL); status = ENOENT; xpt_lock_buses(); if ((periph->flags & CAM_PERIPH_INVALID) == 0) { periph->refcount++; status = 0; } xpt_unlock_buses(); return (status); } void cam_periph_doacquire(struct cam_periph *periph) { xpt_lock_buses(); KASSERT(periph->refcount >= 1, ("cam_periph_doacquire() with refcount == %d", periph->refcount)); periph->refcount++; xpt_unlock_buses(); } void cam_periph_release_locked_buses(struct cam_periph *periph) { cam_periph_assert(periph, MA_OWNED); KASSERT(periph->refcount >= 1, ("periph->refcount >= 1")); if (--periph->refcount == 0) camperiphfree(periph); } void cam_periph_release_locked(struct cam_periph *periph) { if (periph == NULL) return; xpt_lock_buses(); cam_periph_release_locked_buses(periph); xpt_unlock_buses(); } void cam_periph_release(struct cam_periph *periph) { struct mtx *mtx; if (periph == NULL) return; cam_periph_assert(periph, MA_NOTOWNED); mtx = cam_periph_mtx(periph); mtx_lock(mtx); cam_periph_release_locked(periph); mtx_unlock(mtx); } /* * hold/unhold act as mutual exclusion for sections of the code that * need to sleep and want to make sure that other sections that * will interfere are held off. This only protects exclusive sections * from each other. */ int cam_periph_hold(struct cam_periph *periph, int priority) { int error; /* * Increment the reference count on the peripheral * while we wait for our lock attempt to succeed * to ensure the peripheral doesn't disappear out * from user us while we sleep. */ if (cam_periph_acquire(periph) != 0) return (ENXIO); cam_periph_assert(periph, MA_OWNED); while ((periph->flags & CAM_PERIPH_LOCKED) != 0) { periph->flags |= CAM_PERIPH_LOCK_WANTED; if ((error = cam_periph_sleep(periph, periph, priority, "caplck", 0)) != 0) { cam_periph_release_locked(periph); return (error); } if (periph->flags & CAM_PERIPH_INVALID) { cam_periph_release_locked(periph); return (ENXIO); } } periph->flags |= CAM_PERIPH_LOCKED; return (0); } void cam_periph_unhold(struct cam_periph *periph) { cam_periph_assert(periph, MA_OWNED); periph->flags &= ~CAM_PERIPH_LOCKED; if ((periph->flags & CAM_PERIPH_LOCK_WANTED) != 0) { periph->flags &= ~CAM_PERIPH_LOCK_WANTED; wakeup(periph); } cam_periph_release_locked(periph); } /* * Look for the next unit number that is not currently in use for this * peripheral type starting at "newunit". Also exclude unit numbers that * are reserved by for future "hardwiring" unless we already know that this * is a potential wired device. Only assume that the device is "wired" the * first time through the loop since after that we'll be looking at unit * numbers that did not match a wiring entry. */ static u_int camperiphnextunit(struct periph_driver *p_drv, u_int newunit, int wired, path_id_t pathid, target_id_t target, lun_id_t lun) { struct cam_periph *periph; char *periph_name; int i, val, dunit, r; const char *dname, *strval; periph_name = p_drv->driver_name; for (;;newunit++) { for (periph = TAILQ_FIRST(&p_drv->units); periph != NULL && periph->unit_number != newunit; periph = TAILQ_NEXT(periph, unit_links)) ; if (periph != NULL && periph->unit_number == newunit) { if (wired != 0) { xpt_print(periph->path, "Duplicate Wired " "Device entry!\n"); xpt_print(periph->path, "Second device (%s " "device at scbus%d target %d lun %d) will " "not be wired\n", periph_name, pathid, target, lun); wired = 0; } continue; } if (wired) break; /* * Don't match entries like "da 4" as a wired down * device, but do match entries like "da 4 target 5" * or even "da 4 scbus 1". */ i = 0; dname = periph_name; for (;;) { r = resource_find_dev(&i, dname, &dunit, NULL, NULL); if (r != 0) break; /* if no "target" and no specific scbus, skip */ if (resource_int_value(dname, dunit, "target", &val) && (resource_string_value(dname, dunit, "at",&strval)|| strcmp(strval, "scbus") == 0)) continue; if (newunit == dunit) break; } if (r != 0) break; } return (newunit); } static u_int camperiphunit(struct periph_driver *p_drv, path_id_t pathid, target_id_t target, lun_id_t lun) { u_int unit; int wired, i, val, dunit; const char *dname, *strval; char pathbuf[32], *periph_name; periph_name = p_drv->driver_name; snprintf(pathbuf, sizeof(pathbuf), "scbus%d", pathid); unit = 0; i = 0; dname = periph_name; for (wired = 0; resource_find_dev(&i, dname, &dunit, NULL, NULL) == 0; wired = 0) { if (resource_string_value(dname, dunit, "at", &strval) == 0) { if (strcmp(strval, pathbuf) != 0) continue; wired++; } if (resource_int_value(dname, dunit, "target", &val) == 0) { if (val != target) continue; wired++; } if (resource_int_value(dname, dunit, "lun", &val) == 0) { if (val != lun) continue; wired++; } if (wired != 0) { unit = dunit; break; } } /* * Either start from 0 looking for the next unit or from * the unit number given in the resource config. This way, * if we have wildcard matches, we don't return the same * unit number twice. */ unit = camperiphnextunit(p_drv, unit, wired, pathid, target, lun); return (unit); } void cam_periph_invalidate(struct cam_periph *periph) { cam_periph_assert(periph, MA_OWNED); /* - * We only call this routine the first time a peripheral is + * We only tear down the device the first time a peripheral is * invalidated. */ if ((periph->flags & CAM_PERIPH_INVALID) != 0) return; CAM_DEBUG(periph->path, CAM_DEBUG_INFO, ("Periph invalidated\n")); if ((periph->flags & CAM_PERIPH_ANNOUNCED) && !rebooting) { struct sbuf sb; char buffer[160]; sbuf_new(&sb, buffer, 160, SBUF_FIXEDLEN); xpt_denounce_periph_sbuf(periph, &sb); sbuf_finish(&sb); sbuf_putbuf(&sb); } periph->flags |= CAM_PERIPH_INVALID; periph->flags &= ~CAM_PERIPH_NEW_DEV_FOUND; if (periph->periph_oninval != NULL) periph->periph_oninval(periph); cam_periph_release_locked(periph); } static void camperiphfree(struct cam_periph *periph) { struct periph_driver **p_drv; struct periph_driver *drv; cam_periph_assert(periph, MA_OWNED); KASSERT(periph->periph_allocating == 0, ("%s%d: freed while allocating", periph->periph_name, periph->unit_number)); for (p_drv = periph_drivers; *p_drv != NULL; p_drv++) { if (strcmp((*p_drv)->driver_name, periph->periph_name) == 0) break; } if (*p_drv == NULL) { printf("camperiphfree: attempt to free non-existant periph\n"); return; } /* * Cache a pointer to the periph_driver structure. If a * periph_driver is added or removed from the array (see * periphdriver_register()) while we drop the toplogy lock * below, p_drv may change. This doesn't protect against this * particular periph_driver going away. That will require full * reference counting in the periph_driver infrastructure. */ drv = *p_drv; /* * We need to set this flag before dropping the topology lock, to * let anyone who is traversing the list that this peripheral is * about to be freed, and there will be no more reference count * checks. */ periph->flags |= CAM_PERIPH_FREE; /* * The peripheral destructor semantics dictate calling with only the * SIM mutex held. Since it might sleep, it should not be called * with the topology lock held. */ xpt_unlock_buses(); /* * We need to call the peripheral destructor prior to removing the * peripheral from the list. Otherwise, we risk running into a * scenario where the peripheral unit number may get reused * (because it has been removed from the list), but some resources * used by the peripheral are still hanging around. In particular, * the devfs nodes used by some peripherals like the pass(4) driver * aren't fully cleaned up until the destructor is run. If the * unit number is reused before the devfs instance is fully gone, * devfs will panic. */ if (periph->periph_dtor != NULL) periph->periph_dtor(periph); /* * The peripheral list is protected by the topology lock. */ xpt_lock_buses(); TAILQ_REMOVE(&drv->units, periph, unit_links); drv->generation++; xpt_remove_periph(periph); xpt_unlock_buses(); if ((periph->flags & CAM_PERIPH_ANNOUNCED) && !rebooting) xpt_print(periph->path, "Periph destroyed\n"); else CAM_DEBUG(periph->path, CAM_DEBUG_INFO, ("Periph destroyed\n")); if (periph->flags & CAM_PERIPH_NEW_DEV_FOUND) { union ccb ccb; void *arg; switch (periph->deferred_ac) { case AC_FOUND_DEVICE: ccb.ccb_h.func_code = XPT_GDEV_TYPE; xpt_setup_ccb(&ccb.ccb_h, periph->path, CAM_PRIORITY_NORMAL); xpt_action(&ccb); arg = &ccb; break; case AC_PATH_REGISTERED: xpt_path_inq(&ccb.cpi, periph->path); arg = &ccb; break; default: arg = NULL; break; } periph->deferred_callback(NULL, periph->deferred_ac, periph->path, arg); } xpt_free_path(periph->path); free(periph, M_CAMPERIPH); xpt_lock_buses(); } /* * Map user virtual pointers into kernel virtual address space, so we can * access the memory. This is now a generic function that centralizes most * of the sanity checks on the data flags, if any. * This also only works for up to MAXPHYS memory. Since we use * buffers to map stuff in and out, we're limited to the buffer size. */ int cam_periph_mapmem(union ccb *ccb, struct cam_periph_map_info *mapinfo, u_int maxmap) { int numbufs, i; u_int8_t **data_ptrs[CAM_PERIPH_MAXMAPS]; u_int32_t lengths[CAM_PERIPH_MAXMAPS]; u_int32_t dirs[CAM_PERIPH_MAXMAPS]; bool misaligned[CAM_PERIPH_MAXMAPS]; bzero(mapinfo, sizeof(*mapinfo)); if (maxmap == 0) maxmap = DFLTPHYS; /* traditional default */ else if (maxmap > MAXPHYS) maxmap = MAXPHYS; /* for safety */ switch(ccb->ccb_h.func_code) { case XPT_DEV_MATCH: if (ccb->cdm.match_buf_len == 0) { printf("cam_periph_mapmem: invalid match buffer " "length 0\n"); return(EINVAL); } if (ccb->cdm.pattern_buf_len > 0) { data_ptrs[0] = (u_int8_t **)&ccb->cdm.patterns; lengths[0] = ccb->cdm.pattern_buf_len; dirs[0] = CAM_DIR_OUT; data_ptrs[1] = (u_int8_t **)&ccb->cdm.matches; lengths[1] = ccb->cdm.match_buf_len; dirs[1] = CAM_DIR_IN; numbufs = 2; } else { data_ptrs[0] = (u_int8_t **)&ccb->cdm.matches; lengths[0] = ccb->cdm.match_buf_len; dirs[0] = CAM_DIR_IN; numbufs = 1; } /* * This request will not go to the hardware, no reason * to be so strict. vmapbuf() is able to map up to MAXPHYS. */ maxmap = MAXPHYS; break; case XPT_SCSI_IO: case XPT_CONT_TARGET_IO: if ((ccb->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_NONE) return(0); if ((ccb->ccb_h.flags & CAM_DATA_MASK) != CAM_DATA_VADDR) return (EINVAL); data_ptrs[0] = &ccb->csio.data_ptr; lengths[0] = ccb->csio.dxfer_len; dirs[0] = ccb->ccb_h.flags & CAM_DIR_MASK; numbufs = 1; break; case XPT_ATA_IO: if ((ccb->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_NONE) return(0); if ((ccb->ccb_h.flags & CAM_DATA_MASK) != CAM_DATA_VADDR) return (EINVAL); data_ptrs[0] = &ccb->ataio.data_ptr; lengths[0] = ccb->ataio.dxfer_len; dirs[0] = ccb->ccb_h.flags & CAM_DIR_MASK; numbufs = 1; break; case XPT_MMC_IO: if ((ccb->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_NONE) return(0); /* Two mappings: one for cmd->data and one for cmd->data->data */ data_ptrs[0] = (unsigned char **)&ccb->mmcio.cmd.data; lengths[0] = sizeof(struct mmc_data *); dirs[0] = ccb->ccb_h.flags & CAM_DIR_MASK; data_ptrs[1] = (unsigned char **)&ccb->mmcio.cmd.data->data; lengths[1] = ccb->mmcio.cmd.data->len; dirs[1] = ccb->ccb_h.flags & CAM_DIR_MASK; numbufs = 2; break; case XPT_SMP_IO: data_ptrs[0] = &ccb->smpio.smp_request; lengths[0] = ccb->smpio.smp_request_len; dirs[0] = CAM_DIR_OUT; data_ptrs[1] = &ccb->smpio.smp_response; lengths[1] = ccb->smpio.smp_response_len; dirs[1] = CAM_DIR_IN; numbufs = 2; break; case XPT_NVME_IO: case XPT_NVME_ADMIN: if ((ccb->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_NONE) return (0); if ((ccb->ccb_h.flags & CAM_DATA_MASK) != CAM_DATA_VADDR) return (EINVAL); data_ptrs[0] = &ccb->nvmeio.data_ptr; lengths[0] = ccb->nvmeio.dxfer_len; dirs[0] = ccb->ccb_h.flags & CAM_DIR_MASK; numbufs = 1; break; case XPT_DEV_ADVINFO: if (ccb->cdai.bufsiz == 0) return (0); data_ptrs[0] = (uint8_t **)&ccb->cdai.buf; lengths[0] = ccb->cdai.bufsiz; dirs[0] = CAM_DIR_IN; numbufs = 1; /* * This request will not go to the hardware, no reason * to be so strict. vmapbuf() is able to map up to MAXPHYS. */ maxmap = MAXPHYS; break; default: return(EINVAL); break; /* NOTREACHED */ } /* * Check the transfer length and permissions first, so we don't * have to unmap any previously mapped buffers. */ for (i = 0; i < numbufs; i++) { if (lengths[i] > maxmap) { printf("cam_periph_mapmem: attempt to map %lu bytes, " "which is greater than %lu\n", (long)(lengths[i]), (u_long)maxmap); return (E2BIG); } /* * The userland data pointer passed in may not be page * aligned. vmapbuf() truncates the address to a page * boundary, so if the address isn't page aligned, we'll * need enough space for the given transfer length, plus * whatever extra space is necessary to make it to the page * boundary. */ misaligned[i] = (lengths[i] + (((vm_offset_t)(*data_ptrs[i])) & PAGE_MASK) > MAXPHYS); } /* * This keeps the kernel stack of current thread from getting * swapped. In low-memory situations where the kernel stack might * otherwise get swapped out, this holds it and allows the thread * to make progress and release the kernel mapped pages sooner. * * XXX KDM should I use P_NOSWAP instead? */ PHOLD(curproc); for (i = 0; i < numbufs; i++) { /* Save the user's data address. */ mapinfo->orig[i] = *data_ptrs[i]; /* * For small buffers use malloc+copyin/copyout instead of * mapping to KVA to avoid expensive TLB shootdowns. For * small allocations malloc is backed by UMA, and so much * cheaper on SMP systems. */ if ((lengths[i] <= periph_mapmem_thresh || misaligned[i]) && ccb->ccb_h.func_code != XPT_MMC_IO) { *data_ptrs[i] = malloc(lengths[i], M_CAMPERIPH, M_WAITOK); if (dirs[i] != CAM_DIR_IN) { if (copyin(mapinfo->orig[i], *data_ptrs[i], lengths[i]) != 0) { free(*data_ptrs[i], M_CAMPERIPH); *data_ptrs[i] = mapinfo->orig[i]; goto fail; } } else bzero(*data_ptrs[i], lengths[i]); continue; } /* * Get the buffer. */ mapinfo->bp[i] = uma_zalloc(pbuf_zone, M_WAITOK); /* put our pointer in the data slot */ mapinfo->bp[i]->b_data = *data_ptrs[i]; /* set the transfer length, we know it's < MAXPHYS */ mapinfo->bp[i]->b_bufsize = lengths[i]; /* set the direction */ mapinfo->bp[i]->b_iocmd = (dirs[i] == CAM_DIR_OUT) ? BIO_WRITE : BIO_READ; /* * Map the buffer into kernel memory. * * Note that useracc() alone is not a sufficient test. * vmapbuf() can still fail due to a smaller file mapped * into a larger area of VM, or if userland races against * vmapbuf() after the useracc() check. */ if (vmapbuf(mapinfo->bp[i], 1) < 0) { uma_zfree(pbuf_zone, mapinfo->bp[i]); goto fail; } /* set our pointer to the new mapped area */ *data_ptrs[i] = mapinfo->bp[i]->b_data; } /* * Now that we've gotten this far, change ownership to the kernel * of the buffers so that we don't run afoul of returning to user * space with locks (on the buffer) held. */ for (i = 0; i < numbufs; i++) { if (mapinfo->bp[i]) BUF_KERNPROC(mapinfo->bp[i]); } mapinfo->num_bufs_used = numbufs; return(0); fail: for (i--; i >= 0; i--) { if (mapinfo->bp[i]) { vunmapbuf(mapinfo->bp[i]); uma_zfree(pbuf_zone, mapinfo->bp[i]); } else free(*data_ptrs[i], M_CAMPERIPH); *data_ptrs[i] = mapinfo->orig[i]; } PRELE(curproc); return(EACCES); } /* * Unmap memory segments mapped into kernel virtual address space by * cam_periph_mapmem(). */ void cam_periph_unmapmem(union ccb *ccb, struct cam_periph_map_info *mapinfo) { int numbufs, i; u_int8_t **data_ptrs[CAM_PERIPH_MAXMAPS]; u_int32_t lengths[CAM_PERIPH_MAXMAPS]; u_int32_t dirs[CAM_PERIPH_MAXMAPS]; if (mapinfo->num_bufs_used <= 0) { /* nothing to free and the process wasn't held. */ return; } switch (ccb->ccb_h.func_code) { case XPT_DEV_MATCH: if (ccb->cdm.pattern_buf_len > 0) { data_ptrs[0] = (u_int8_t **)&ccb->cdm.patterns; lengths[0] = ccb->cdm.pattern_buf_len; dirs[0] = CAM_DIR_OUT; data_ptrs[1] = (u_int8_t **)&ccb->cdm.matches; lengths[1] = ccb->cdm.match_buf_len; dirs[1] = CAM_DIR_IN; numbufs = 2; } else { data_ptrs[0] = (u_int8_t **)&ccb->cdm.matches; lengths[0] = ccb->cdm.match_buf_len; dirs[0] = CAM_DIR_IN; numbufs = 1; } break; case XPT_SCSI_IO: case XPT_CONT_TARGET_IO: data_ptrs[0] = &ccb->csio.data_ptr; lengths[0] = ccb->csio.dxfer_len; dirs[0] = ccb->ccb_h.flags & CAM_DIR_MASK; numbufs = 1; break; case XPT_ATA_IO: data_ptrs[0] = &ccb->ataio.data_ptr; lengths[0] = ccb->ataio.dxfer_len; dirs[0] = ccb->ccb_h.flags & CAM_DIR_MASK; numbufs = 1; break; case XPT_MMC_IO: data_ptrs[0] = (u_int8_t **)&ccb->mmcio.cmd.data; lengths[0] = sizeof(struct mmc_data *); dirs[0] = ccb->ccb_h.flags & CAM_DIR_MASK; data_ptrs[1] = (u_int8_t **)&ccb->mmcio.cmd.data->data; lengths[1] = ccb->mmcio.cmd.data->len; dirs[1] = ccb->ccb_h.flags & CAM_DIR_MASK; numbufs = 2; break; case XPT_SMP_IO: data_ptrs[0] = &ccb->smpio.smp_request; lengths[0] = ccb->smpio.smp_request_len; dirs[0] = CAM_DIR_OUT; data_ptrs[1] = &ccb->smpio.smp_response; lengths[1] = ccb->smpio.smp_response_len; dirs[1] = CAM_DIR_IN; numbufs = 2; break; case XPT_NVME_IO: case XPT_NVME_ADMIN: data_ptrs[0] = &ccb->nvmeio.data_ptr; lengths[0] = ccb->nvmeio.dxfer_len; dirs[0] = ccb->ccb_h.flags & CAM_DIR_MASK; numbufs = 1; break; case XPT_DEV_ADVINFO: data_ptrs[0] = (uint8_t **)&ccb->cdai.buf; lengths[0] = ccb->cdai.bufsiz; dirs[0] = CAM_DIR_IN; numbufs = 1; break; default: /* allow ourselves to be swapped once again */ PRELE(curproc); return; break; /* NOTREACHED */ } for (i = 0; i < numbufs; i++) { if (mapinfo->bp[i]) { /* unmap the buffer */ vunmapbuf(mapinfo->bp[i]); /* release the buffer */ uma_zfree(pbuf_zone, mapinfo->bp[i]); } else { if (dirs[i] != CAM_DIR_OUT) { copyout(*data_ptrs[i], mapinfo->orig[i], lengths[i]); } free(*data_ptrs[i], M_CAMPERIPH); } /* Set the user's pointer back to the original value */ *data_ptrs[i] = mapinfo->orig[i]; } /* allow ourselves to be swapped once again */ PRELE(curproc); } int cam_periph_ioctl(struct cam_periph *periph, u_long cmd, caddr_t addr, int (*error_routine)(union ccb *ccb, cam_flags camflags, u_int32_t sense_flags)) { union ccb *ccb; int error; int found; error = found = 0; switch(cmd){ case CAMGETPASSTHRU: ccb = cam_periph_getccb(periph, CAM_PRIORITY_NORMAL); xpt_setup_ccb(&ccb->ccb_h, ccb->ccb_h.path, CAM_PRIORITY_NORMAL); ccb->ccb_h.func_code = XPT_GDEVLIST; /* * Basically, the point of this is that we go through * getting the list of devices, until we find a passthrough * device. In the current version of the CAM code, the * only way to determine what type of device we're dealing * with is by its name. */ while (found == 0) { ccb->cgdl.index = 0; ccb->cgdl.status = CAM_GDEVLIST_MORE_DEVS; while (ccb->cgdl.status == CAM_GDEVLIST_MORE_DEVS) { /* we want the next device in the list */ xpt_action(ccb); if (strncmp(ccb->cgdl.periph_name, "pass", 4) == 0){ found = 1; break; } } if ((ccb->cgdl.status == CAM_GDEVLIST_LAST_DEVICE) && (found == 0)) { ccb->cgdl.periph_name[0] = '\0'; ccb->cgdl.unit_number = 0; break; } } /* copy the result back out */ bcopy(ccb, addr, sizeof(union ccb)); /* and release the ccb */ xpt_release_ccb(ccb); break; default: error = ENOTTY; break; } return(error); } static void cam_periph_done_panic(struct cam_periph *periph, union ccb *done_ccb) { panic("%s: already done with ccb %p", __func__, done_ccb); } static void cam_periph_done(struct cam_periph *periph, union ccb *done_ccb) { /* Caller will release the CCB */ xpt_path_assert(done_ccb->ccb_h.path, MA_OWNED); done_ccb->ccb_h.cbfcnp = cam_periph_done_panic; wakeup(&done_ccb->ccb_h.cbfcnp); } static void cam_periph_ccbwait(union ccb *ccb) { if ((ccb->ccb_h.func_code & XPT_FC_QUEUED) != 0) { while (ccb->ccb_h.cbfcnp != cam_periph_done_panic) xpt_path_sleep(ccb->ccb_h.path, &ccb->ccb_h.cbfcnp, PRIBIO, "cbwait", 0); } KASSERT(ccb->ccb_h.pinfo.index == CAM_UNQUEUED_INDEX && (ccb->ccb_h.status & CAM_STATUS_MASK) != CAM_REQ_INPROG, ("%s: proceeding with incomplete ccb: ccb=%p, func_code=%#x, " "status=%#x, index=%d", __func__, ccb, ccb->ccb_h.func_code, ccb->ccb_h.status, ccb->ccb_h.pinfo.index)); } /* * Dispatch a CCB and wait for it to complete. If the CCB has set a * callback function (ccb->ccb_h.cbfcnp), it will be overwritten and lost. */ int cam_periph_runccb(union ccb *ccb, int (*error_routine)(union ccb *ccb, cam_flags camflags, u_int32_t sense_flags), cam_flags camflags, u_int32_t sense_flags, struct devstat *ds) { struct bintime *starttime; struct bintime ltime; int error; bool must_poll; uint32_t timeout = 1; starttime = NULL; xpt_path_assert(ccb->ccb_h.path, MA_OWNED); KASSERT((ccb->ccb_h.flags & CAM_UNLOCKED) == 0, ("%s: ccb=%p, func_code=%#x, flags=%#x", __func__, ccb, ccb->ccb_h.func_code, ccb->ccb_h.flags)); /* * If the user has supplied a stats structure, and if we understand * this particular type of ccb, record the transaction start. */ if (ds != NULL && (ccb->ccb_h.func_code == XPT_SCSI_IO || ccb->ccb_h.func_code == XPT_ATA_IO || ccb->ccb_h.func_code == XPT_NVME_IO)) { starttime = <ime; binuptime(starttime); devstat_start_transaction(ds, starttime); } /* * We must poll the I/O while we're dumping. The scheduler is normally * stopped for dumping, except when we call doadump from ddb. While the * scheduler is running in this case, we still need to poll the I/O to * avoid sleeping waiting for the ccb to complete. * * A panic triggered dump stops the scheduler, any callback from the * shutdown_post_sync event will run with the scheduler stopped, but * before we're officially dumping. To avoid hanging in adashutdown * initiated commands (or other similar situations), we have to test for * either SCHEDULER_STOPPED() here as well. * * To avoid locking problems, dumping/polling callers must call * without a periph lock held. */ must_poll = dumping || SCHEDULER_STOPPED(); ccb->ccb_h.cbfcnp = cam_periph_done; /* * If we're polling, then we need to ensure that we have ample resources * in the periph. cam_periph_error can reschedule the ccb by calling * xpt_action and returning ERESTART, so we have to effect the polling * in the do loop below. */ if (must_poll) { timeout = xpt_poll_setup(ccb); } if (timeout == 0) { ccb->ccb_h.status = CAM_RESRC_UNAVAIL; error = EBUSY; } else { xpt_action(ccb); do { if (must_poll) { xpt_pollwait(ccb, timeout); timeout = ccb->ccb_h.timeout * 10; } else { cam_periph_ccbwait(ccb); } if ((ccb->ccb_h.status & CAM_STATUS_MASK) == CAM_REQ_CMP) error = 0; else if (error_routine != NULL) { ccb->ccb_h.cbfcnp = cam_periph_done; error = (*error_routine)(ccb, camflags, sense_flags); } else error = 0; } while (error == ERESTART); } if ((ccb->ccb_h.status & CAM_DEV_QFRZN) != 0) { cam_release_devq(ccb->ccb_h.path, /* relsim_flags */0, /* openings */0, /* timeout */0, /* getcount_only */ FALSE); ccb->ccb_h.status &= ~CAM_DEV_QFRZN; } if (ds != NULL) { uint32_t bytes; devstat_tag_type tag; bool valid = true; if (ccb->ccb_h.func_code == XPT_SCSI_IO) { bytes = ccb->csio.dxfer_len - ccb->csio.resid; tag = (devstat_tag_type)(ccb->csio.tag_action & 0x3); } else if (ccb->ccb_h.func_code == XPT_ATA_IO) { bytes = ccb->ataio.dxfer_len - ccb->ataio.resid; tag = (devstat_tag_type)0; } else if (ccb->ccb_h.func_code == XPT_NVME_IO) { bytes = ccb->nvmeio.dxfer_len; /* NB: resid no possible */ tag = (devstat_tag_type)0; } else { valid = false; } if (valid) devstat_end_transaction(ds, bytes, tag, ((ccb->ccb_h.flags & CAM_DIR_MASK) == CAM_DIR_NONE) ? DEVSTAT_NO_DATA : (ccb->ccb_h.flags & CAM_DIR_OUT) ? DEVSTAT_WRITE : DEVSTAT_READ, NULL, starttime); } return(error); } void cam_freeze_devq(struct cam_path *path) { struct ccb_hdr ccb_h; CAM_DEBUG(path, CAM_DEBUG_TRACE, ("cam_freeze_devq\n")); xpt_setup_ccb(&ccb_h, path, /*priority*/1); ccb_h.func_code = XPT_NOOP; ccb_h.flags = CAM_DEV_QFREEZE; xpt_action((union ccb *)&ccb_h); } u_int32_t cam_release_devq(struct cam_path *path, u_int32_t relsim_flags, u_int32_t openings, u_int32_t arg, int getcount_only) { struct ccb_relsim crs; CAM_DEBUG(path, CAM_DEBUG_TRACE, ("cam_release_devq(%u, %u, %u, %d)\n", relsim_flags, openings, arg, getcount_only)); xpt_setup_ccb(&crs.ccb_h, path, CAM_PRIORITY_NORMAL); crs.ccb_h.func_code = XPT_REL_SIMQ; crs.ccb_h.flags = getcount_only ? CAM_DEV_QFREEZE : 0; crs.release_flags = relsim_flags; crs.openings = openings; crs.release_timeout = arg; xpt_action((union ccb *)&crs); return (crs.qfrozen_cnt); } #define saved_ccb_ptr ppriv_ptr0 static void camperiphdone(struct cam_periph *periph, union ccb *done_ccb) { union ccb *saved_ccb; cam_status status; struct scsi_start_stop_unit *scsi_cmd; int error = 0, error_code, sense_key, asc, ascq; scsi_cmd = (struct scsi_start_stop_unit *) &done_ccb->csio.cdb_io.cdb_bytes; status = done_ccb->ccb_h.status; if ((status & CAM_STATUS_MASK) != CAM_REQ_CMP) { if (scsi_extract_sense_ccb(done_ccb, &error_code, &sense_key, &asc, &ascq)) { /* * If the error is "invalid field in CDB", * and the load/eject flag is set, turn the * flag off and try again. This is just in * case the drive in question barfs on the * load eject flag. The CAM code should set * the load/eject flag by default for * removable media. */ if ((scsi_cmd->opcode == START_STOP_UNIT) && ((scsi_cmd->how & SSS_LOEJ) != 0) && (asc == 0x24) && (ascq == 0x00)) { scsi_cmd->how &= ~SSS_LOEJ; if (status & CAM_DEV_QFRZN) { cam_release_devq(done_ccb->ccb_h.path, 0, 0, 0, 0); done_ccb->ccb_h.status &= ~CAM_DEV_QFRZN; } xpt_action(done_ccb); goto out; } } error = cam_periph_error(done_ccb, 0, SF_RETRY_UA | SF_NO_PRINT); if (error == ERESTART) goto out; if (done_ccb->ccb_h.status & CAM_DEV_QFRZN) { cam_release_devq(done_ccb->ccb_h.path, 0, 0, 0, 0); done_ccb->ccb_h.status &= ~CAM_DEV_QFRZN; } } else { /* * If we have successfully taken a device from the not * ready to ready state, re-scan the device and re-get * the inquiry information. Many devices (mostly disks) * don't properly report their inquiry information unless * they are spun up. */ if (scsi_cmd->opcode == START_STOP_UNIT) xpt_async(AC_INQ_CHANGED, done_ccb->ccb_h.path, NULL); } /* If we tried long wait and still failed, remember that. */ if ((periph->flags & CAM_PERIPH_RECOVERY_WAIT) && (done_ccb->csio.cdb_io.cdb_bytes[0] == TEST_UNIT_READY)) { periph->flags &= ~CAM_PERIPH_RECOVERY_WAIT; if (error != 0 && done_ccb->ccb_h.retry_count == 0) periph->flags |= CAM_PERIPH_RECOVERY_WAIT_FAILED; } /* * After recovery action(s) completed, return to the original CCB. * If the recovery CCB has failed, considering its own possible * retries and recovery, assume we are back in state where we have * been originally, but without recovery hopes left. In such case, * after the final attempt below, we cancel any further retries, * blocking by that also any new recovery attempts for this CCB, * and the result will be the final one returned to the CCB owher. */ saved_ccb = (union ccb *)done_ccb->ccb_h.saved_ccb_ptr; bcopy(saved_ccb, done_ccb, sizeof(*done_ccb)); xpt_free_ccb(saved_ccb); if (done_ccb->ccb_h.cbfcnp != camperiphdone) periph->flags &= ~CAM_PERIPH_RECOVERY_INPROG; if (error != 0) done_ccb->ccb_h.retry_count = 0; xpt_action(done_ccb); out: /* Drop freeze taken due to CAM_DEV_QFREEZE flag set. */ cam_release_devq(done_ccb->ccb_h.path, 0, 0, 0, 0); } /* * Generic Async Event handler. Peripheral drivers usually * filter out the events that require personal attention, * and leave the rest to this function. */ void cam_periph_async(struct cam_periph *periph, u_int32_t code, struct cam_path *path, void *arg) { switch (code) { case AC_LOST_DEVICE: cam_periph_invalidate(periph); break; default: break; } } void cam_periph_bus_settle(struct cam_periph *periph, u_int bus_settle) { struct ccb_getdevstats cgds; xpt_setup_ccb(&cgds.ccb_h, periph->path, CAM_PRIORITY_NORMAL); cgds.ccb_h.func_code = XPT_GDEV_STATS; xpt_action((union ccb *)&cgds); cam_periph_freeze_after_event(periph, &cgds.last_reset, bus_settle); } void cam_periph_freeze_after_event(struct cam_periph *periph, struct timeval* event_time, u_int duration_ms) { struct timeval delta; struct timeval duration_tv; if (!timevalisset(event_time)) return; microtime(&delta); timevalsub(&delta, event_time); duration_tv.tv_sec = duration_ms / 1000; duration_tv.tv_usec = (duration_ms % 1000) * 1000; if (timevalcmp(&delta, &duration_tv, <)) { timevalsub(&duration_tv, &delta); duration_ms = duration_tv.tv_sec * 1000; duration_ms += duration_tv.tv_usec / 1000; cam_freeze_devq(periph->path); cam_release_devq(periph->path, RELSIM_RELEASE_AFTER_TIMEOUT, /*reduction*/0, /*timeout*/duration_ms, /*getcount_only*/0); } } static int camperiphscsistatuserror(union ccb *ccb, union ccb **orig_ccb, cam_flags camflags, u_int32_t sense_flags, int *openings, u_int32_t *relsim_flags, u_int32_t *timeout, u_int32_t *action, const char **action_string) { struct cam_periph *periph; int error; switch (ccb->csio.scsi_status) { case SCSI_STATUS_OK: case SCSI_STATUS_COND_MET: case SCSI_STATUS_INTERMED: case SCSI_STATUS_INTERMED_COND_MET: error = 0; break; case SCSI_STATUS_CMD_TERMINATED: case SCSI_STATUS_CHECK_COND: error = camperiphscsisenseerror(ccb, orig_ccb, camflags, sense_flags, openings, relsim_flags, timeout, action, action_string); break; case SCSI_STATUS_QUEUE_FULL: { /* no decrement */ struct ccb_getdevstats cgds; /* * First off, find out what the current * transaction counts are. */ xpt_setup_ccb(&cgds.ccb_h, ccb->ccb_h.path, CAM_PRIORITY_NORMAL); cgds.ccb_h.func_code = XPT_GDEV_STATS; xpt_action((union ccb *)&cgds); /* * If we were the only transaction active, treat * the QUEUE FULL as if it were a BUSY condition. */ if (cgds.dev_active != 0) { int total_openings; /* * Reduce the number of openings to * be 1 less than the amount it took * to get a queue full bounded by the * minimum allowed tag count for this * device. */ total_openings = cgds.dev_active + cgds.dev_openings; *openings = cgds.dev_active; if (*openings < cgds.mintags) *openings = cgds.mintags; if (*openings < total_openings) *relsim_flags = RELSIM_ADJUST_OPENINGS; else { /* * Some devices report queue full for * temporary resource shortages. For * this reason, we allow a minimum * tag count to be entered via a * quirk entry to prevent the queue * count on these devices from falling * to a pessimisticly low value. We * still wait for the next successful * completion, however, before queueing * more transactions to the device. */ *relsim_flags = RELSIM_RELEASE_AFTER_CMDCMPLT; } *timeout = 0; error = ERESTART; *action &= ~SSQ_PRINT_SENSE; break; } /* FALLTHROUGH */ } case SCSI_STATUS_BUSY: /* * Restart the queue after either another * command completes or a 1 second timeout. */ periph = xpt_path_periph(ccb->ccb_h.path); if (periph->flags & CAM_PERIPH_INVALID) { error = EIO; *action_string = "Periph was invalidated"; } else if ((sense_flags & SF_RETRY_BUSY) != 0 || ccb->ccb_h.retry_count > 0) { if ((sense_flags & SF_RETRY_BUSY) == 0) ccb->ccb_h.retry_count--; error = ERESTART; *relsim_flags = RELSIM_RELEASE_AFTER_TIMEOUT | RELSIM_RELEASE_AFTER_CMDCMPLT; *timeout = 1000; } else { error = EIO; *action_string = "Retries exhausted"; } break; case SCSI_STATUS_RESERV_CONFLICT: default: error = EIO; break; } return (error); } static int camperiphscsisenseerror(union ccb *ccb, union ccb **orig, cam_flags camflags, u_int32_t sense_flags, int *openings, u_int32_t *relsim_flags, u_int32_t *timeout, u_int32_t *action, const char **action_string) { struct cam_periph *periph; union ccb *orig_ccb = ccb; int error, recoveryccb; #if defined(BUF_TRACKING) || defined(FULL_BUF_TRACKING) if (ccb->ccb_h.func_code == XPT_SCSI_IO && ccb->csio.bio != NULL) biotrack(ccb->csio.bio, __func__); #endif periph = xpt_path_periph(ccb->ccb_h.path); recoveryccb = (ccb->ccb_h.cbfcnp == camperiphdone); if ((periph->flags & CAM_PERIPH_RECOVERY_INPROG) && !recoveryccb) { /* * If error recovery is already in progress, don't attempt * to process this error, but requeue it unconditionally * and attempt to process it once error recovery has * completed. This failed command is probably related to * the error that caused the currently active error recovery * action so our current recovery efforts should also * address this command. Be aware that the error recovery * code assumes that only one recovery action is in progress * on a particular peripheral instance at any given time * (e.g. only one saved CCB for error recovery) so it is * imperitive that we don't violate this assumption. */ error = ERESTART; *action &= ~SSQ_PRINT_SENSE; } else { scsi_sense_action err_action; struct ccb_getdev cgd; /* * Grab the inquiry data for this device. */ xpt_setup_ccb(&cgd.ccb_h, ccb->ccb_h.path, CAM_PRIORITY_NORMAL); cgd.ccb_h.func_code = XPT_GDEV_TYPE; xpt_action((union ccb *)&cgd); err_action = scsi_error_action(&ccb->csio, &cgd.inq_data, sense_flags); error = err_action & SS_ERRMASK; /* * Do not autostart sequential access devices * to avoid unexpected tape loading. */ if ((err_action & SS_MASK) == SS_START && SID_TYPE(&cgd.inq_data) == T_SEQUENTIAL) { *action_string = "Will not autostart a " "sequential access device"; goto sense_error_done; } /* * Avoid recovery recursion if recovery action is the same. */ if ((err_action & SS_MASK) >= SS_START && recoveryccb) { if (((err_action & SS_MASK) == SS_START && ccb->csio.cdb_io.cdb_bytes[0] == START_STOP_UNIT) || ((err_action & SS_MASK) == SS_TUR && (ccb->csio.cdb_io.cdb_bytes[0] == TEST_UNIT_READY))) { err_action = SS_RETRY|SSQ_DECREMENT_COUNT|EIO; *relsim_flags = RELSIM_RELEASE_AFTER_TIMEOUT; *timeout = 500; } } /* * If the recovery action will consume a retry, * make sure we actually have retries available. */ if ((err_action & SSQ_DECREMENT_COUNT) != 0) { if (ccb->ccb_h.retry_count > 0 && (periph->flags & CAM_PERIPH_INVALID) == 0) ccb->ccb_h.retry_count--; else { *action_string = "Retries exhausted"; goto sense_error_done; } } if ((err_action & SS_MASK) >= SS_START) { /* * Do common portions of commands that * use recovery CCBs. */ orig_ccb = xpt_alloc_ccb_nowait(); if (orig_ccb == NULL) { *action_string = "Can't allocate recovery CCB"; goto sense_error_done; } /* * Clear freeze flag for original request here, as * this freeze will be dropped as part of ERESTART. */ ccb->ccb_h.status &= ~CAM_DEV_QFRZN; bcopy(ccb, orig_ccb, sizeof(*orig_ccb)); } switch (err_action & SS_MASK) { case SS_NOP: *action_string = "No recovery action needed"; error = 0; break; case SS_RETRY: *action_string = "Retrying command (per sense data)"; error = ERESTART; break; case SS_FAIL: *action_string = "Unretryable error"; break; case SS_START: { int le; /* * Send a start unit command to the device, and * then retry the command. */ *action_string = "Attempting to start unit"; periph->flags |= CAM_PERIPH_RECOVERY_INPROG; /* * Check for removable media and set * load/eject flag appropriately. */ if (SID_IS_REMOVABLE(&cgd.inq_data)) le = TRUE; else le = FALSE; scsi_start_stop(&ccb->csio, /*retries*/1, camperiphdone, MSG_SIMPLE_Q_TAG, /*start*/TRUE, /*load/eject*/le, /*immediate*/FALSE, SSD_FULL_SIZE, /*timeout*/50000); break; } case SS_TUR: { /* * Send a Test Unit Ready to the device. * If the 'many' flag is set, we send 120 * test unit ready commands, one every half * second. Otherwise, we just send one TUR. * We only want to do this if the retry * count has not been exhausted. */ int retries; if ((err_action & SSQ_MANY) != 0 && (periph->flags & CAM_PERIPH_RECOVERY_WAIT_FAILED) == 0) { periph->flags |= CAM_PERIPH_RECOVERY_WAIT; *action_string = "Polling device for readiness"; retries = 120; } else { *action_string = "Testing device for readiness"; retries = 1; } periph->flags |= CAM_PERIPH_RECOVERY_INPROG; scsi_test_unit_ready(&ccb->csio, retries, camperiphdone, MSG_SIMPLE_Q_TAG, SSD_FULL_SIZE, /*timeout*/5000); /* * Accomplish our 500ms delay by deferring * the release of our device queue appropriately. */ *relsim_flags = RELSIM_RELEASE_AFTER_TIMEOUT; *timeout = 500; break; } default: panic("Unhandled error action %x", err_action); } if ((err_action & SS_MASK) >= SS_START) { /* * Drop the priority, so that the recovery * CCB is the first to execute. Freeze the queue * after this command is sent so that we can * restore the old csio and have it queued in * the proper order before we release normal * transactions to the device. */ ccb->ccb_h.pinfo.priority--; ccb->ccb_h.flags |= CAM_DEV_QFREEZE; ccb->ccb_h.saved_ccb_ptr = orig_ccb; error = ERESTART; *orig = orig_ccb; } sense_error_done: *action = err_action; } return (error); } /* * Generic error handler. Peripheral drivers usually filter * out the errors that they handle in a unique manner, then * call this function. */ int cam_periph_error(union ccb *ccb, cam_flags camflags, u_int32_t sense_flags) { struct cam_path *newpath; union ccb *orig_ccb, *scan_ccb; struct cam_periph *periph; const char *action_string; cam_status status; int frozen, error, openings, devctl_err; u_int32_t action, relsim_flags, timeout; action = SSQ_PRINT_SENSE; periph = xpt_path_periph(ccb->ccb_h.path); action_string = NULL; status = ccb->ccb_h.status; frozen = (status & CAM_DEV_QFRZN) != 0; status &= CAM_STATUS_MASK; devctl_err = openings = relsim_flags = timeout = 0; orig_ccb = ccb; /* Filter the errors that should be reported via devctl */ switch (ccb->ccb_h.status & CAM_STATUS_MASK) { case CAM_CMD_TIMEOUT: case CAM_REQ_ABORTED: case CAM_REQ_CMP_ERR: case CAM_REQ_TERMIO: case CAM_UNREC_HBA_ERROR: case CAM_DATA_RUN_ERR: case CAM_SCSI_STATUS_ERROR: case CAM_ATA_STATUS_ERROR: case CAM_SMP_STATUS_ERROR: devctl_err++; break; default: break; } switch (status) { case CAM_REQ_CMP: error = 0; action &= ~SSQ_PRINT_SENSE; break; case CAM_SCSI_STATUS_ERROR: error = camperiphscsistatuserror(ccb, &orig_ccb, camflags, sense_flags, &openings, &relsim_flags, &timeout, &action, &action_string); break; case CAM_AUTOSENSE_FAIL: error = EIO; /* we have to kill the command */ break; case CAM_UA_ABORT: case CAM_UA_TERMIO: case CAM_MSG_REJECT_REC: /* XXX Don't know that these are correct */ error = EIO; break; case CAM_SEL_TIMEOUT: if ((camflags & CAM_RETRY_SELTO) != 0) { if (ccb->ccb_h.retry_count > 0 && (periph->flags & CAM_PERIPH_INVALID) == 0) { ccb->ccb_h.retry_count--; error = ERESTART; /* * Wait a bit to give the device * time to recover before we try again. */ relsim_flags = RELSIM_RELEASE_AFTER_TIMEOUT; timeout = periph_selto_delay; break; } action_string = "Retries exhausted"; } /* FALLTHROUGH */ case CAM_DEV_NOT_THERE: error = ENXIO; action = SSQ_LOST; break; case CAM_REQ_INVALID: case CAM_PATH_INVALID: case CAM_NO_HBA: case CAM_PROVIDE_FAIL: case CAM_REQ_TOO_BIG: case CAM_LUN_INVALID: case CAM_TID_INVALID: case CAM_FUNC_NOTAVAIL: error = EINVAL; break; case CAM_SCSI_BUS_RESET: case CAM_BDR_SENT: /* * Commands that repeatedly timeout and cause these * kinds of error recovery actions, should return * CAM_CMD_TIMEOUT, which allows us to safely assume * that this command was an innocent bystander to * these events and should be unconditionally * retried. */ case CAM_REQUEUE_REQ: /* Unconditional requeue if device is still there */ if (periph->flags & CAM_PERIPH_INVALID) { action_string = "Periph was invalidated"; error = EIO; } else if (sense_flags & SF_NO_RETRY) { error = EIO; action_string = "Retry was blocked"; } else { error = ERESTART; action &= ~SSQ_PRINT_SENSE; } break; case CAM_RESRC_UNAVAIL: /* Wait a bit for the resource shortage to abate. */ timeout = periph_noresrc_delay; /* FALLTHROUGH */ case CAM_BUSY: if (timeout == 0) { /* Wait a bit for the busy condition to abate. */ timeout = periph_busy_delay; } relsim_flags = RELSIM_RELEASE_AFTER_TIMEOUT; /* FALLTHROUGH */ case CAM_ATA_STATUS_ERROR: case CAM_REQ_CMP_ERR: case CAM_CMD_TIMEOUT: case CAM_UNEXP_BUSFREE: case CAM_UNCOR_PARITY: case CAM_DATA_RUN_ERR: default: if (periph->flags & CAM_PERIPH_INVALID) { error = EIO; action_string = "Periph was invalidated"; } else if (ccb->ccb_h.retry_count == 0) { error = EIO; action_string = "Retries exhausted"; } else if (sense_flags & SF_NO_RETRY) { error = EIO; action_string = "Retry was blocked"; } else { ccb->ccb_h.retry_count--; error = ERESTART; } break; } if ((sense_flags & SF_PRINT_ALWAYS) || CAM_DEBUGGED(ccb->ccb_h.path, CAM_DEBUG_INFO)) action |= SSQ_PRINT_SENSE; else if (sense_flags & SF_NO_PRINT) action &= ~SSQ_PRINT_SENSE; if ((action & SSQ_PRINT_SENSE) != 0) cam_error_print(orig_ccb, CAM_ESF_ALL, CAM_EPF_ALL); if (error != 0 && (action & SSQ_PRINT_SENSE) != 0) { if (error != ERESTART) { if (action_string == NULL) action_string = "Unretryable error"; xpt_print(ccb->ccb_h.path, "Error %d, %s\n", error, action_string); } else if (action_string != NULL) xpt_print(ccb->ccb_h.path, "%s\n", action_string); else { xpt_print(ccb->ccb_h.path, "Retrying command, %d more tries remain\n", ccb->ccb_h.retry_count); } } if (devctl_err && (error != 0 || (action & SSQ_PRINT_SENSE) != 0)) cam_periph_devctl_notify(orig_ccb); if ((action & SSQ_LOST) != 0) { lun_id_t lun_id; /* * For a selection timeout, we consider all of the LUNs on * the target to be gone. If the status is CAM_DEV_NOT_THERE, * then we only get rid of the device(s) specified by the * path in the original CCB. */ if (status == CAM_SEL_TIMEOUT) lun_id = CAM_LUN_WILDCARD; else lun_id = xpt_path_lun_id(ccb->ccb_h.path); /* Should we do more if we can't create the path?? */ if (xpt_create_path(&newpath, periph, xpt_path_path_id(ccb->ccb_h.path), xpt_path_target_id(ccb->ccb_h.path), lun_id) == CAM_REQ_CMP) { /* * Let peripheral drivers know that this * device has gone away. */ xpt_async(AC_LOST_DEVICE, newpath, NULL); xpt_free_path(newpath); } } /* Broadcast UNIT ATTENTIONs to all periphs. */ if ((action & SSQ_UA) != 0) xpt_async(AC_UNIT_ATTENTION, orig_ccb->ccb_h.path, orig_ccb); /* Rescan target on "Reported LUNs data has changed" */ if ((action & SSQ_RESCAN) != 0) { if (xpt_create_path(&newpath, NULL, xpt_path_path_id(ccb->ccb_h.path), xpt_path_target_id(ccb->ccb_h.path), CAM_LUN_WILDCARD) == CAM_REQ_CMP) { scan_ccb = xpt_alloc_ccb_nowait(); if (scan_ccb != NULL) { scan_ccb->ccb_h.path = newpath; scan_ccb->ccb_h.func_code = XPT_SCAN_TGT; scan_ccb->crcn.flags = 0; xpt_rescan(scan_ccb); } else { xpt_print(newpath, "Can't allocate CCB to rescan target\n"); xpt_free_path(newpath); } } } /* Attempt a retry */ if (error == ERESTART || error == 0) { if (frozen != 0) ccb->ccb_h.status &= ~CAM_DEV_QFRZN; if (error == ERESTART) xpt_action(ccb); if (frozen != 0) cam_release_devq(ccb->ccb_h.path, relsim_flags, openings, timeout, /*getcount_only*/0); } return (error); } #define CAM_PERIPH_DEVD_MSG_SIZE 256 static void cam_periph_devctl_notify(union ccb *ccb) { struct cam_periph *periph; struct ccb_getdev *cgd; struct sbuf sb; int serr, sk, asc, ascq; char *sbmsg, *type; sbmsg = malloc(CAM_PERIPH_DEVD_MSG_SIZE, M_CAMPERIPH, M_NOWAIT); if (sbmsg == NULL) return; sbuf_new(&sb, sbmsg, CAM_PERIPH_DEVD_MSG_SIZE, SBUF_FIXEDLEN); periph = xpt_path_periph(ccb->ccb_h.path); sbuf_printf(&sb, "device=%s%d ", periph->periph_name, periph->unit_number); sbuf_printf(&sb, "serial=\""); if ((cgd = (struct ccb_getdev *)xpt_alloc_ccb_nowait()) != NULL) { xpt_setup_ccb(&cgd->ccb_h, ccb->ccb_h.path, CAM_PRIORITY_NORMAL); cgd->ccb_h.func_code = XPT_GDEV_TYPE; xpt_action((union ccb *)cgd); if (cgd->ccb_h.status == CAM_REQ_CMP) sbuf_bcat(&sb, cgd->serial_num, cgd->serial_num_len); xpt_free_ccb((union ccb *)cgd); } sbuf_printf(&sb, "\" "); sbuf_printf(&sb, "cam_status=\"0x%x\" ", ccb->ccb_h.status); switch (ccb->ccb_h.status & CAM_STATUS_MASK) { case CAM_CMD_TIMEOUT: sbuf_printf(&sb, "timeout=%d ", ccb->ccb_h.timeout); type = "timeout"; break; case CAM_SCSI_STATUS_ERROR: sbuf_printf(&sb, "scsi_status=%d ", ccb->csio.scsi_status); if (scsi_extract_sense_ccb(ccb, &serr, &sk, &asc, &ascq)) sbuf_printf(&sb, "scsi_sense=\"%02x %02x %02x %02x\" ", serr, sk, asc, ascq); type = "error"; break; case CAM_ATA_STATUS_ERROR: sbuf_printf(&sb, "RES=\""); ata_res_sbuf(&ccb->ataio.res, &sb); sbuf_printf(&sb, "\" "); type = "error"; break; default: type = "error"; break; } if (ccb->ccb_h.func_code == XPT_SCSI_IO) { sbuf_printf(&sb, "CDB=\""); scsi_cdb_sbuf(scsiio_cdb_ptr(&ccb->csio), &sb); sbuf_printf(&sb, "\" "); } else if (ccb->ccb_h.func_code == XPT_ATA_IO) { sbuf_printf(&sb, "ACB=\""); ata_cmd_sbuf(&ccb->ataio.cmd, &sb); sbuf_printf(&sb, "\" "); } if (sbuf_finish(&sb) == 0) devctl_notify("CAM", "periph", type, sbuf_data(&sb)); sbuf_delete(&sb); free(sbmsg, M_CAMPERIPH); } /* * Sysctl to force an invalidation of the drive right now. Can be * called with CTLFLAG_MPSAFE since we take periph lock. */ int cam_periph_invalidate_sysctl(SYSCTL_HANDLER_ARGS) { struct cam_periph *periph; int error, value; periph = arg1; value = 0; error = sysctl_handle_int(oidp, &value, 0, req); if (error != 0 || req->newptr == NULL || value != 1) return (error); cam_periph_lock(periph); cam_periph_invalidate(periph); cam_periph_unlock(periph); return (0); } diff --git a/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dmu.c b/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dmu.c index 99db969b7181..59e551e75d43 100644 --- a/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dmu.c +++ b/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/dmu.c @@ -1,2748 +1,2748 @@ /* * CDDL HEADER START * * The contents of this file are subject to the terms of the * Common Development and Distribution License (the "License"). * You may not use this file except in compliance with the License. * * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE * or http://www.opensolaris.org/os/licensing. * See the License for the specific language governing permissions * and limitations under the License. * * When distributing Covered Code, include this CDDL HEADER in each * file and include the License file at usr/src/OPENSOLARIS.LICENSE. * If applicable, add the following below this CDDL HEADER, with the * fields enclosed by brackets "[]" replaced with your own identifying * information: Portions Copyright [yyyy] [name of copyright owner] * * CDDL HEADER END */ /* * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. * Copyright (c) 2011, 2017 by Delphix. All rights reserved. * Copyright (c) 2019 Datto Inc. */ /* Copyright (c) 2013 by Saso Kiselkov. All rights reserved. */ /* Copyright (c) 2013, Joyent, Inc. All rights reserved. */ /* Copyright 2016 Nexenta Systems, Inc. All rights reserved. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef _KERNEL #include #include #include #endif /* * Enable/disable nopwrite feature. */ int zfs_nopwrite_enabled = 1; SYSCTL_DECL(_vfs_zfs); SYSCTL_INT(_vfs_zfs, OID_AUTO, nopwrite_enabled, CTLFLAG_RDTUN, &zfs_nopwrite_enabled, 0, "Enable nopwrite feature"); /* * Tunable to control percentage of dirtied L1 blocks from frees allowed into * one TXG. After this threshold is crossed, additional dirty blocks from frees * will wait until the next TXG. * A value of zero will disable this throttle. */ uint32_t zfs_per_txg_dirty_frees_percent = 5; SYSCTL_INT(_vfs_zfs, OID_AUTO, per_txg_dirty_frees_percent, CTLFLAG_RWTUN, &zfs_per_txg_dirty_frees_percent, 0, "Percentage of dirtied indirect blocks from frees allowed in one txg"); /* * This can be used for testing, to ensure that certain actions happen * while in the middle of a remap (which might otherwise complete too * quickly). */ int zfs_object_remap_one_indirect_delay_ticks = 0; /* * Limit the amount we can prefetch with one call to this amount. This * helps to limit the amount of memory that can be used by prefetching. * Larger objects should be prefetched a bit at a time. */ uint64_t dmu_prefetch_max = 8 * SPA_MAXBLOCKSIZE; const dmu_object_type_info_t dmu_ot[DMU_OT_NUMTYPES] = { { DMU_BSWAP_UINT8, TRUE, FALSE, "unallocated" }, { DMU_BSWAP_ZAP, TRUE, TRUE, "object directory" }, { DMU_BSWAP_UINT64, TRUE, TRUE, "object array" }, { DMU_BSWAP_UINT8, TRUE, FALSE, "packed nvlist" }, { DMU_BSWAP_UINT64, TRUE, FALSE, "packed nvlist size" }, { DMU_BSWAP_UINT64, TRUE, FALSE, "bpobj" }, { DMU_BSWAP_UINT64, TRUE, FALSE, "bpobj header" }, { DMU_BSWAP_UINT64, TRUE, FALSE, "SPA space map header" }, { DMU_BSWAP_UINT64, TRUE, FALSE, "SPA space map" }, { DMU_BSWAP_UINT64, TRUE, FALSE, "ZIL intent log" }, { DMU_BSWAP_DNODE, TRUE, FALSE, "DMU dnode" }, { DMU_BSWAP_OBJSET, TRUE, TRUE, "DMU objset" }, { DMU_BSWAP_UINT64, TRUE, TRUE, "DSL directory" }, { DMU_BSWAP_ZAP, TRUE, TRUE, "DSL directory child map" }, { DMU_BSWAP_ZAP, TRUE, TRUE, "DSL dataset snap map" }, { DMU_BSWAP_ZAP, TRUE, TRUE, "DSL props" }, { DMU_BSWAP_UINT64, TRUE, TRUE, "DSL dataset" }, { DMU_BSWAP_ZNODE, TRUE, FALSE, "ZFS znode" }, { DMU_BSWAP_OLDACL, TRUE, FALSE, "ZFS V0 ACL" }, { DMU_BSWAP_UINT8, FALSE, FALSE, "ZFS plain file" }, { DMU_BSWAP_ZAP, TRUE, FALSE, "ZFS directory" }, { DMU_BSWAP_ZAP, TRUE, FALSE, "ZFS master node" }, { DMU_BSWAP_ZAP, TRUE, FALSE, "ZFS delete queue" }, { DMU_BSWAP_UINT8, FALSE, FALSE, "zvol object" }, { DMU_BSWAP_ZAP, TRUE, FALSE, "zvol prop" }, { DMU_BSWAP_UINT8, FALSE, FALSE, "other uint8[]" }, { DMU_BSWAP_UINT64, FALSE, FALSE, "other uint64[]" }, { DMU_BSWAP_ZAP, TRUE, FALSE, "other ZAP" }, { DMU_BSWAP_ZAP, TRUE, FALSE, "persistent error log" }, { DMU_BSWAP_UINT8, TRUE, FALSE, "SPA history" }, { DMU_BSWAP_UINT64, TRUE, FALSE, "SPA history offsets" }, { DMU_BSWAP_ZAP, TRUE, TRUE, "Pool properties" }, { DMU_BSWAP_ZAP, TRUE, TRUE, "DSL permissions" }, { DMU_BSWAP_ACL, TRUE, FALSE, "ZFS ACL" }, { DMU_BSWAP_UINT8, TRUE, FALSE, "ZFS SYSACL" }, { DMU_BSWAP_UINT8, TRUE, FALSE, "FUID table" }, { DMU_BSWAP_UINT64, TRUE, FALSE, "FUID table size" }, { DMU_BSWAP_ZAP, TRUE, TRUE, "DSL dataset next clones" }, { DMU_BSWAP_ZAP, TRUE, FALSE, "scan work queue" }, { DMU_BSWAP_ZAP, TRUE, FALSE, "ZFS user/group used" }, { DMU_BSWAP_ZAP, TRUE, FALSE, "ZFS user/group quota" }, { DMU_BSWAP_ZAP, TRUE, TRUE, "snapshot refcount tags" }, { DMU_BSWAP_ZAP, TRUE, FALSE, "DDT ZAP algorithm" }, { DMU_BSWAP_ZAP, TRUE, FALSE, "DDT statistics" }, { DMU_BSWAP_UINT8, TRUE, FALSE, "System attributes" }, { DMU_BSWAP_ZAP, TRUE, FALSE, "SA master node" }, { DMU_BSWAP_ZAP, TRUE, FALSE, "SA attr registration" }, { DMU_BSWAP_ZAP, TRUE, FALSE, "SA attr layouts" }, { DMU_BSWAP_ZAP, TRUE, FALSE, "scan translations" }, { DMU_BSWAP_UINT8, FALSE, FALSE, "deduplicated block" }, { DMU_BSWAP_ZAP, TRUE, TRUE, "DSL deadlist map" }, { DMU_BSWAP_UINT64, TRUE, TRUE, "DSL deadlist map hdr" }, { DMU_BSWAP_ZAP, TRUE, TRUE, "DSL dir clones" }, { DMU_BSWAP_UINT64, TRUE, FALSE, "bpobj subobj" } }; const dmu_object_byteswap_info_t dmu_ot_byteswap[DMU_BSWAP_NUMFUNCS] = { { byteswap_uint8_array, "uint8" }, { byteswap_uint16_array, "uint16" }, { byteswap_uint32_array, "uint32" }, { byteswap_uint64_array, "uint64" }, { zap_byteswap, "zap" }, { dnode_buf_byteswap, "dnode" }, { dmu_objset_byteswap, "objset" }, { zfs_znode_byteswap, "znode" }, { zfs_oldacl_byteswap, "oldacl" }, { zfs_acl_byteswap, "acl" } }; int dmu_buf_hold_noread_by_dnode(dnode_t *dn, uint64_t offset, void *tag, dmu_buf_t **dbp) { uint64_t blkid; dmu_buf_impl_t *db; blkid = dbuf_whichblock(dn, 0, offset); rw_enter(&dn->dn_struct_rwlock, RW_READER); db = dbuf_hold(dn, blkid, tag); rw_exit(&dn->dn_struct_rwlock); if (db == NULL) { *dbp = NULL; return (SET_ERROR(EIO)); } *dbp = &db->db; return (0); } int dmu_buf_hold_noread(objset_t *os, uint64_t object, uint64_t offset, void *tag, dmu_buf_t **dbp) { dnode_t *dn; uint64_t blkid; dmu_buf_impl_t *db; int err; err = dnode_hold(os, object, FTAG, &dn); if (err) return (err); blkid = dbuf_whichblock(dn, 0, offset); rw_enter(&dn->dn_struct_rwlock, RW_READER); db = dbuf_hold(dn, blkid, tag); rw_exit(&dn->dn_struct_rwlock); dnode_rele(dn, FTAG); if (db == NULL) { *dbp = NULL; return (SET_ERROR(EIO)); } *dbp = &db->db; return (err); } int dmu_buf_hold_by_dnode(dnode_t *dn, uint64_t offset, void *tag, dmu_buf_t **dbp, int flags) { int err; int db_flags = DB_RF_CANFAIL; if (flags & DMU_READ_NO_PREFETCH) db_flags |= DB_RF_NOPREFETCH; err = dmu_buf_hold_noread_by_dnode(dn, offset, tag, dbp); if (err == 0) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)(*dbp); err = dbuf_read(db, NULL, db_flags); if (err != 0) { dbuf_rele(db, tag); *dbp = NULL; } } return (err); } int dmu_buf_hold(objset_t *os, uint64_t object, uint64_t offset, void *tag, dmu_buf_t **dbp, int flags) { int err; int db_flags = DB_RF_CANFAIL; if (flags & DMU_READ_NO_PREFETCH) db_flags |= DB_RF_NOPREFETCH; err = dmu_buf_hold_noread(os, object, offset, tag, dbp); if (err == 0) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)(*dbp); err = dbuf_read(db, NULL, db_flags); if (err != 0) { dbuf_rele(db, tag); *dbp = NULL; } } return (err); } int dmu_bonus_max(void) { return (DN_OLD_MAX_BONUSLEN); } int dmu_set_bonus(dmu_buf_t *db_fake, int newsize, dmu_tx_t *tx) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; dnode_t *dn; int error; DB_DNODE_ENTER(db); dn = DB_DNODE(db); if (dn->dn_bonus != db) { error = SET_ERROR(EINVAL); } else if (newsize < 0 || newsize > db_fake->db_size) { error = SET_ERROR(EINVAL); } else { dnode_setbonuslen(dn, newsize, tx); error = 0; } DB_DNODE_EXIT(db); return (error); } int dmu_set_bonustype(dmu_buf_t *db_fake, dmu_object_type_t type, dmu_tx_t *tx) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; dnode_t *dn; int error; DB_DNODE_ENTER(db); dn = DB_DNODE(db); if (!DMU_OT_IS_VALID(type)) { error = SET_ERROR(EINVAL); } else if (dn->dn_bonus != db) { error = SET_ERROR(EINVAL); } else { dnode_setbonus_type(dn, type, tx); error = 0; } DB_DNODE_EXIT(db); return (error); } dmu_object_type_t dmu_get_bonustype(dmu_buf_t *db_fake) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; dnode_t *dn; dmu_object_type_t type; DB_DNODE_ENTER(db); dn = DB_DNODE(db); type = dn->dn_bonustype; DB_DNODE_EXIT(db); return (type); } int dmu_rm_spill(objset_t *os, uint64_t object, dmu_tx_t *tx) { dnode_t *dn; int error; error = dnode_hold(os, object, FTAG, &dn); dbuf_rm_spill(dn, tx); rw_enter(&dn->dn_struct_rwlock, RW_WRITER); dnode_rm_spill(dn, tx); rw_exit(&dn->dn_struct_rwlock); dnode_rele(dn, FTAG); return (error); } /* * returns ENOENT, EIO, or 0. */ int dmu_bonus_hold(objset_t *os, uint64_t object, void *tag, dmu_buf_t **dbp) { dnode_t *dn; dmu_buf_impl_t *db; int error; error = dnode_hold(os, object, FTAG, &dn); if (error) return (error); rw_enter(&dn->dn_struct_rwlock, RW_READER); if (dn->dn_bonus == NULL) { rw_exit(&dn->dn_struct_rwlock); rw_enter(&dn->dn_struct_rwlock, RW_WRITER); if (dn->dn_bonus == NULL) dbuf_create_bonus(dn); } db = dn->dn_bonus; /* as long as the bonus buf is held, the dnode will be held */ if (zfs_refcount_add(&db->db_holds, tag) == 1) { VERIFY(dnode_add_ref(dn, db)); atomic_inc_32(&dn->dn_dbufs_count); } /* * Wait to drop dn_struct_rwlock until after adding the bonus dbuf's * hold and incrementing the dbuf count to ensure that dnode_move() sees * a dnode hold for every dbuf. */ rw_exit(&dn->dn_struct_rwlock); dnode_rele(dn, FTAG); VERIFY(0 == dbuf_read(db, NULL, DB_RF_MUST_SUCCEED | DB_RF_NOPREFETCH)); *dbp = &db->db; return (0); } /* * returns ENOENT, EIO, or 0. * * This interface will allocate a blank spill dbuf when a spill blk * doesn't already exist on the dnode. * * if you only want to find an already existing spill db, then * dmu_spill_hold_existing() should be used. */ int dmu_spill_hold_by_dnode(dnode_t *dn, uint32_t flags, void *tag, dmu_buf_t **dbp) { dmu_buf_impl_t *db = NULL; int err; if ((flags & DB_RF_HAVESTRUCT) == 0) rw_enter(&dn->dn_struct_rwlock, RW_READER); db = dbuf_hold(dn, DMU_SPILL_BLKID, tag); if ((flags & DB_RF_HAVESTRUCT) == 0) rw_exit(&dn->dn_struct_rwlock); ASSERT(db != NULL); err = dbuf_read(db, NULL, flags); if (err == 0) *dbp = &db->db; else dbuf_rele(db, tag); return (err); } int dmu_spill_hold_existing(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus; dnode_t *dn; int err; DB_DNODE_ENTER(db); dn = DB_DNODE(db); if (spa_version(dn->dn_objset->os_spa) < SPA_VERSION_SA) { err = SET_ERROR(EINVAL); } else { rw_enter(&dn->dn_struct_rwlock, RW_READER); if (!dn->dn_have_spill) { err = SET_ERROR(ENOENT); } else { err = dmu_spill_hold_by_dnode(dn, DB_RF_HAVESTRUCT | DB_RF_CANFAIL, tag, dbp); } rw_exit(&dn->dn_struct_rwlock); } DB_DNODE_EXIT(db); return (err); } int dmu_spill_hold_by_bonus(dmu_buf_t *bonus, void *tag, dmu_buf_t **dbp) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)bonus; dnode_t *dn; int err; DB_DNODE_ENTER(db); dn = DB_DNODE(db); err = dmu_spill_hold_by_dnode(dn, DB_RF_CANFAIL, tag, dbp); DB_DNODE_EXIT(db); return (err); } /* * Note: longer-term, we should modify all of the dmu_buf_*() interfaces * to take a held dnode rather than -- the lookup is wasteful, * and can induce severe lock contention when writing to several files * whose dnodes are in the same block. */ int dmu_buf_hold_array_by_dnode(dnode_t *dn, uint64_t offset, uint64_t length, boolean_t read, void *tag, int *numbufsp, dmu_buf_t ***dbpp, uint32_t flags) { dmu_buf_t **dbp; uint64_t blkid, nblks, i; uint32_t dbuf_flags; int err; zio_t *zio; ASSERT(length <= DMU_MAX_ACCESS); /* * Note: We directly notify the prefetch code of this read, so that * we can tell it about the multi-block read. dbuf_read() only knows * about the one block it is accessing. */ dbuf_flags = DB_RF_CANFAIL | DB_RF_NEVERWAIT | DB_RF_HAVESTRUCT | DB_RF_NOPREFETCH; rw_enter(&dn->dn_struct_rwlock, RW_READER); if (dn->dn_datablkshift) { int blkshift = dn->dn_datablkshift; nblks = (P2ROUNDUP(offset + length, 1ULL << blkshift) - P2ALIGN(offset, 1ULL << blkshift)) >> blkshift; } else { if (offset + length > dn->dn_datablksz) { zfs_panic_recover("zfs: accessing past end of object " "%llx/%llx (size=%u access=%llu+%llu)", (longlong_t)dn->dn_objset-> os_dsl_dataset->ds_object, (longlong_t)dn->dn_object, dn->dn_datablksz, (longlong_t)offset, (longlong_t)length); rw_exit(&dn->dn_struct_rwlock); return (SET_ERROR(EIO)); } nblks = 1; } dbp = kmem_zalloc(sizeof (dmu_buf_t *) * nblks, KM_SLEEP); #if defined(_KERNEL) && defined(RACCT) if (racct_enable && !read) { PROC_LOCK(curproc); racct_add_force(curproc, RACCT_WRITEBPS, length); racct_add_force(curproc, RACCT_WRITEIOPS, nblks); PROC_UNLOCK(curproc); } #endif zio = zio_root(dn->dn_objset->os_spa, NULL, NULL, ZIO_FLAG_CANFAIL); blkid = dbuf_whichblock(dn, 0, offset); for (i = 0; i < nblks; i++) { dmu_buf_impl_t *db = dbuf_hold(dn, blkid + i, tag); if (db == NULL) { rw_exit(&dn->dn_struct_rwlock); dmu_buf_rele_array(dbp, nblks, tag); zio_nowait(zio); return (SET_ERROR(EIO)); } /* initiate async i/o */ if (read) (void) dbuf_read(db, zio, dbuf_flags); #ifdef _KERNEL else curthread->td_ru.ru_oublock++; #endif dbp[i] = &db->db; } if ((flags & DMU_READ_NO_PREFETCH) == 0 && DNODE_META_IS_CACHEABLE(dn) && length <= zfetch_array_rd_sz) { dmu_zfetch(&dn->dn_zfetch, blkid, nblks, read && DNODE_IS_CACHEABLE(dn)); } rw_exit(&dn->dn_struct_rwlock); /* wait for async i/o */ err = zio_wait(zio); if (err) { dmu_buf_rele_array(dbp, nblks, tag); return (err); } /* wait for other io to complete */ if (read) { for (i = 0; i < nblks; i++) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)dbp[i]; mutex_enter(&db->db_mtx); while (db->db_state == DB_READ || db->db_state == DB_FILL) cv_wait(&db->db_changed, &db->db_mtx); if (db->db_state == DB_UNCACHED) err = SET_ERROR(EIO); mutex_exit(&db->db_mtx); if (err) { dmu_buf_rele_array(dbp, nblks, tag); return (err); } } } *numbufsp = nblks; *dbpp = dbp; return (0); } static int dmu_buf_hold_array(objset_t *os, uint64_t object, uint64_t offset, uint64_t length, int read, void *tag, int *numbufsp, dmu_buf_t ***dbpp) { dnode_t *dn; int err; err = dnode_hold(os, object, FTAG, &dn); if (err) return (err); err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag, numbufsp, dbpp, DMU_READ_PREFETCH); dnode_rele(dn, FTAG); return (err); } int dmu_buf_hold_array_by_bonus(dmu_buf_t *db_fake, uint64_t offset, uint64_t length, boolean_t read, void *tag, int *numbufsp, dmu_buf_t ***dbpp) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; dnode_t *dn; int err; DB_DNODE_ENTER(db); dn = DB_DNODE(db); err = dmu_buf_hold_array_by_dnode(dn, offset, length, read, tag, numbufsp, dbpp, DMU_READ_PREFETCH); DB_DNODE_EXIT(db); return (err); } void dmu_buf_rele_array(dmu_buf_t **dbp_fake, int numbufs, void *tag) { int i; dmu_buf_impl_t **dbp = (dmu_buf_impl_t **)dbp_fake; if (numbufs == 0) return; for (i = 0; i < numbufs; i++) { if (dbp[i]) dbuf_rele(dbp[i], tag); } kmem_free(dbp, sizeof (dmu_buf_t *) * numbufs); } /* * Issue prefetch i/os for the given blocks. If level is greater than 0, the * indirect blocks prefeteched will be those that point to the blocks containing * the data starting at offset, and continuing to offset + len. * * Note that if the indirect blocks above the blocks being prefetched are not in * cache, they will be asychronously read in. */ void dmu_prefetch(objset_t *os, uint64_t object, int64_t level, uint64_t offset, uint64_t len, zio_priority_t pri) { dnode_t *dn; uint64_t blkid; int nblks, err; if (len == 0) { /* they're interested in the bonus buffer */ dn = DMU_META_DNODE(os); if (object == 0 || object >= DN_MAX_OBJECT) return; rw_enter(&dn->dn_struct_rwlock, RW_READER); blkid = dbuf_whichblock(dn, level, object * sizeof (dnode_phys_t)); dbuf_prefetch(dn, level, blkid, pri, 0); rw_exit(&dn->dn_struct_rwlock); return; } /* * See comment before the definition of dmu_prefetch_max. */ len = MIN(len, dmu_prefetch_max); /* * XXX - Note, if the dnode for the requested object is not * already cached, we will do a *synchronous* read in the * dnode_hold() call. The same is true for any indirects. */ err = dnode_hold(os, object, FTAG, &dn); if (err != 0) return; rw_enter(&dn->dn_struct_rwlock, RW_READER); /* * offset + len - 1 is the last byte we want to prefetch for, and offset * is the first. Then dbuf_whichblk(dn, level, off + len - 1) is the * last block we want to prefetch, and dbuf_whichblock(dn, level, * offset) is the first. Then the number we need to prefetch is the * last - first + 1. */ if (level > 0 || dn->dn_datablkshift != 0) { nblks = dbuf_whichblock(dn, level, offset + len - 1) - dbuf_whichblock(dn, level, offset) + 1; } else { nblks = (offset < dn->dn_datablksz); } if (nblks != 0) { blkid = dbuf_whichblock(dn, level, offset); for (int i = 0; i < nblks; i++) dbuf_prefetch(dn, level, blkid + i, pri, 0); } rw_exit(&dn->dn_struct_rwlock); dnode_rele(dn, FTAG); } /* * Get the next "chunk" of file data to free. We traverse the file from * the end so that the file gets shorter over time (if we crashes in the * middle, this will leave us in a better state). We find allocated file * data by simply searching the allocated level 1 indirects. * * On input, *start should be the first offset that does not need to be * freed (e.g. "offset + length"). On return, *start will be the first * offset that should be freed and l1blks is set to the number of level 1 * indirect blocks found within the chunk. */ static int get_next_chunk(dnode_t *dn, uint64_t *start, uint64_t minimum, uint64_t *l1blks) { uint64_t blks; uint64_t maxblks = DMU_MAX_ACCESS >> (dn->dn_indblkshift + 1); /* bytes of data covered by a level-1 indirect block */ uint64_t iblkrange = dn->dn_datablksz * EPB(dn->dn_indblkshift, SPA_BLKPTRSHIFT); ASSERT3U(minimum, <=, *start); /* * Check if we can free the entire range assuming that all of the * L1 blocks in this range have data. If we can, we use this * worst case value as an estimate so we can avoid having to look * at the object's actual data. */ uint64_t total_l1blks = (roundup(*start, iblkrange) - (minimum / iblkrange * iblkrange)) / iblkrange; if (total_l1blks <= maxblks) { *l1blks = total_l1blks; *start = minimum; return (0); } ASSERT(ISP2(iblkrange)); for (blks = 0; *start > minimum && blks < maxblks; blks++) { int err; /* * dnode_next_offset(BACKWARDS) will find an allocated L1 * indirect block at or before the input offset. We must * decrement *start so that it is at the end of the region * to search. */ (*start)--; err = dnode_next_offset(dn, DNODE_FIND_BACKWARDS, start, 2, 1, 0); /* if there are no indirect blocks before start, we are done */ if (err == ESRCH) { *start = minimum; break; } else if (err != 0) { *l1blks = blks; return (err); } /* set start to the beginning of this L1 indirect */ *start = P2ALIGN(*start, iblkrange); } if (*start < minimum) *start = minimum; *l1blks = blks; return (0); } /* * If this objset is of type OST_ZFS return true if vfs's unmounted flag is set, * otherwise return false. * Used below in dmu_free_long_range_impl() to enable abort when unmounting */ /*ARGSUSED*/ static boolean_t dmu_objset_zfs_unmounting(objset_t *os) { #ifdef _KERNEL if (dmu_objset_type(os) == DMU_OST_ZFS) return (zfs_get_vfs_flag_unmounted(os)); #endif return (B_FALSE); } static int dmu_free_long_range_impl(objset_t *os, dnode_t *dn, uint64_t offset, uint64_t length) { uint64_t object_size = (dn->dn_maxblkid + 1) * dn->dn_datablksz; int err; uint64_t dirty_frees_threshold; dsl_pool_t *dp = dmu_objset_pool(os); if (offset >= object_size) return (0); if (zfs_per_txg_dirty_frees_percent <= 100) dirty_frees_threshold = zfs_per_txg_dirty_frees_percent * zfs_dirty_data_max / 100; else dirty_frees_threshold = zfs_dirty_data_max / 20; if (length == DMU_OBJECT_END || offset + length > object_size) length = object_size - offset; while (length != 0) { uint64_t chunk_end, chunk_begin, chunk_len; uint64_t l1blks; dmu_tx_t *tx; if (dmu_objset_zfs_unmounting(dn->dn_objset)) return (SET_ERROR(EINTR)); chunk_end = chunk_begin = offset + length; /* move chunk_begin backwards to the beginning of this chunk */ err = get_next_chunk(dn, &chunk_begin, offset, &l1blks); if (err) return (err); ASSERT3U(chunk_begin, >=, offset); ASSERT3U(chunk_begin, <=, chunk_end); chunk_len = chunk_end - chunk_begin; tx = dmu_tx_create(os); dmu_tx_hold_free(tx, dn->dn_object, chunk_begin, chunk_len); /* * Mark this transaction as typically resulting in a net * reduction in space used. */ dmu_tx_mark_netfree(tx); err = dmu_tx_assign(tx, TXG_WAIT); if (err) { dmu_tx_abort(tx); return (err); } uint64_t txg = dmu_tx_get_txg(tx); mutex_enter(&dp->dp_lock); uint64_t long_free_dirty = dp->dp_long_free_dirty_pertxg[txg & TXG_MASK]; mutex_exit(&dp->dp_lock); /* * To avoid filling up a TXG with just frees, wait for * the next TXG to open before freeing more chunks if * we have reached the threshold of frees. */ if (dirty_frees_threshold != 0 && long_free_dirty >= dirty_frees_threshold) { dmu_tx_commit(tx); txg_wait_open(dp, 0); continue; } /* * In order to prevent unnecessary write throttling, for each * TXG, we track the cumulative size of L1 blocks being dirtied * in dnode_free_range() below. We compare this number to a * tunable threshold, past which we prevent new L1 dirty freeing * blocks from being added into the open TXG. See * dmu_free_long_range_impl() for details. The threshold * prevents write throttle activation due to dirty freeing L1 * blocks taking up a large percentage of zfs_dirty_data_max. */ mutex_enter(&dp->dp_lock); dp->dp_long_free_dirty_pertxg[txg & TXG_MASK] += l1blks << dn->dn_indblkshift; mutex_exit(&dp->dp_lock); DTRACE_PROBE3(free__long__range, uint64_t, long_free_dirty, uint64_t, chunk_len, uint64_t, txg); dnode_free_range(dn, chunk_begin, chunk_len, tx); dmu_tx_commit(tx); length -= chunk_len; } return (0); } int dmu_free_long_range(objset_t *os, uint64_t object, uint64_t offset, uint64_t length) { dnode_t *dn; int err; err = dnode_hold(os, object, FTAG, &dn); if (err != 0) return (err); err = dmu_free_long_range_impl(os, dn, offset, length); /* * It is important to zero out the maxblkid when freeing the entire * file, so that (a) subsequent calls to dmu_free_long_range_impl() * will take the fast path, and (b) dnode_reallocate() can verify * that the entire file has been freed. */ if (err == 0 && offset == 0 && length == DMU_OBJECT_END) dn->dn_maxblkid = 0; dnode_rele(dn, FTAG); return (err); } int dmu_free_long_object(objset_t *os, uint64_t object) { dmu_tx_t *tx; int err; err = dmu_free_long_range(os, object, 0, DMU_OBJECT_END); if (err != 0) return (err); tx = dmu_tx_create(os); dmu_tx_hold_bonus(tx, object); dmu_tx_hold_free(tx, object, 0, DMU_OBJECT_END); dmu_tx_mark_netfree(tx); err = dmu_tx_assign(tx, TXG_WAIT); if (err == 0) { err = dmu_object_free(os, object, tx); dmu_tx_commit(tx); } else { dmu_tx_abort(tx); } return (err); } int dmu_free_range(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, dmu_tx_t *tx) { dnode_t *dn; int err = dnode_hold(os, object, FTAG, &dn); if (err) return (err); ASSERT(offset < UINT64_MAX); ASSERT(size == -1ULL || size <= UINT64_MAX - offset); dnode_free_range(dn, offset, size, tx); dnode_rele(dn, FTAG); return (0); } static int dmu_read_impl(dnode_t *dn, uint64_t offset, uint64_t size, void *buf, uint32_t flags) { dmu_buf_t **dbp; int numbufs, err = 0; /* * Deal with odd block sizes, where there can't be data past the first * block. If we ever do the tail block optimization, we will need to * handle that here as well. */ if (dn->dn_maxblkid == 0) { int newsz = offset > dn->dn_datablksz ? 0 : MIN(size, dn->dn_datablksz - offset); bzero((char *)buf + newsz, size - newsz); size = newsz; } while (size > 0) { uint64_t mylen = MIN(size, DMU_MAX_ACCESS / 2); int i; /* * NB: we could do this block-at-a-time, but it's nice * to be reading in parallel. */ err = dmu_buf_hold_array_by_dnode(dn, offset, mylen, TRUE, FTAG, &numbufs, &dbp, flags); if (err) break; for (i = 0; i < numbufs; i++) { int tocpy; int bufoff; dmu_buf_t *db = dbp[i]; ASSERT(size > 0); bufoff = offset - db->db_offset; tocpy = (int)MIN(db->db_size - bufoff, size); bcopy((char *)db->db_data + bufoff, buf, tocpy); offset += tocpy; size -= tocpy; buf = (char *)buf + tocpy; } dmu_buf_rele_array(dbp, numbufs, FTAG); } return (err); } int dmu_read(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, void *buf, uint32_t flags) { dnode_t *dn; int err; err = dnode_hold(os, object, FTAG, &dn); if (err != 0) return (err); err = dmu_read_impl(dn, offset, size, buf, flags); dnode_rele(dn, FTAG); return (err); } int dmu_read_by_dnode(dnode_t *dn, uint64_t offset, uint64_t size, void *buf, uint32_t flags) { return (dmu_read_impl(dn, offset, size, buf, flags)); } static void dmu_write_impl(dmu_buf_t **dbp, int numbufs, uint64_t offset, uint64_t size, const void *buf, dmu_tx_t *tx) { int i; for (i = 0; i < numbufs; i++) { int tocpy; int bufoff; dmu_buf_t *db = dbp[i]; ASSERT(size > 0); bufoff = offset - db->db_offset; tocpy = (int)MIN(db->db_size - bufoff, size); ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); if (tocpy == db->db_size) dmu_buf_will_fill(db, tx); else dmu_buf_will_dirty(db, tx); bcopy(buf, (char *)db->db_data + bufoff, tocpy); if (tocpy == db->db_size) dmu_buf_fill_done(db, tx); offset += tocpy; size -= tocpy; buf = (char *)buf + tocpy; } } void dmu_write(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, const void *buf, dmu_tx_t *tx) { dmu_buf_t **dbp; int numbufs; if (size == 0) return; VERIFY0(dmu_buf_hold_array(os, object, offset, size, FALSE, FTAG, &numbufs, &dbp)); dmu_write_impl(dbp, numbufs, offset, size, buf, tx); dmu_buf_rele_array(dbp, numbufs, FTAG); } void dmu_write_by_dnode(dnode_t *dn, uint64_t offset, uint64_t size, const void *buf, dmu_tx_t *tx) { dmu_buf_t **dbp; int numbufs; if (size == 0) return; VERIFY0(dmu_buf_hold_array_by_dnode(dn, offset, size, FALSE, FTAG, &numbufs, &dbp, DMU_READ_PREFETCH)); dmu_write_impl(dbp, numbufs, offset, size, buf, tx); dmu_buf_rele_array(dbp, numbufs, FTAG); } static int dmu_object_remap_one_indirect(objset_t *os, dnode_t *dn, uint64_t last_removal_txg, uint64_t offset) { uint64_t l1blkid = dbuf_whichblock(dn, 1, offset); int err = 0; rw_enter(&dn->dn_struct_rwlock, RW_READER); dmu_buf_impl_t *dbuf = dbuf_hold_level(dn, 1, l1blkid, FTAG); ASSERT3P(dbuf, !=, NULL); /* * If the block hasn't been written yet, this default will ensure * we don't try to remap it. */ uint64_t birth = UINT64_MAX; ASSERT3U(last_removal_txg, !=, UINT64_MAX); if (dbuf->db_blkptr != NULL) birth = dbuf->db_blkptr->blk_birth; rw_exit(&dn->dn_struct_rwlock); /* * If this L1 was already written after the last removal, then we've * already tried to remap it. */ if (birth <= last_removal_txg && dbuf_read(dbuf, NULL, DB_RF_MUST_SUCCEED) == 0 && dbuf_can_remap(dbuf)) { dmu_tx_t *tx = dmu_tx_create(os); dmu_tx_hold_remap_l1indirect(tx, dn->dn_object); err = dmu_tx_assign(tx, TXG_WAIT); if (err == 0) { (void) dbuf_dirty(dbuf, tx); dmu_tx_commit(tx); } else { dmu_tx_abort(tx); } } dbuf_rele(dbuf, FTAG); delay(zfs_object_remap_one_indirect_delay_ticks); return (err); } /* * Remap all blockpointers in the object, if possible, so that they reference * only concrete vdevs. * * To do this, iterate over the L0 blockpointers and remap any that reference * an indirect vdev. Note that we only examine L0 blockpointers; since we * cannot guarantee that we can remap all blockpointer anyways (due to split * blocks), we do not want to make the code unnecessarily complicated to * catch the unlikely case that there is an L1 block on an indirect vdev that * contains no indirect blockpointers. */ int dmu_object_remap_indirects(objset_t *os, uint64_t object, uint64_t last_removal_txg) { uint64_t offset, l1span; int err; dnode_t *dn; err = dnode_hold(os, object, FTAG, &dn); if (err != 0) { return (err); } if (dn->dn_nlevels <= 1) { if (issig(JUSTLOOKING) && issig(FORREAL)) { err = SET_ERROR(EINTR); } /* * If the dnode has no indirect blocks, we cannot dirty them. * We still want to remap the blkptr(s) in the dnode if * appropriate, so mark it as dirty. */ if (err == 0 && dnode_needs_remap(dn)) { dmu_tx_t *tx = dmu_tx_create(os); dmu_tx_hold_bonus(tx, dn->dn_object); if ((err = dmu_tx_assign(tx, TXG_WAIT)) == 0) { dnode_setdirty(dn, tx); dmu_tx_commit(tx); } else { dmu_tx_abort(tx); } } dnode_rele(dn, FTAG); return (err); } offset = 0; l1span = 1ULL << (dn->dn_indblkshift - SPA_BLKPTRSHIFT + dn->dn_datablkshift); /* * Find the next L1 indirect that is not a hole. */ while (dnode_next_offset(dn, 0, &offset, 2, 1, 0) == 0) { if (issig(JUSTLOOKING) && issig(FORREAL)) { err = SET_ERROR(EINTR); break; } if ((err = dmu_object_remap_one_indirect(os, dn, last_removal_txg, offset)) != 0) { break; } offset += l1span; } dnode_rele(dn, FTAG); return (err); } void dmu_prealloc(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, dmu_tx_t *tx) { dmu_buf_t **dbp; int numbufs, i; if (size == 0) return; VERIFY(0 == dmu_buf_hold_array(os, object, offset, size, FALSE, FTAG, &numbufs, &dbp)); for (i = 0; i < numbufs; i++) { dmu_buf_t *db = dbp[i]; dmu_buf_will_not_fill(db, tx); } dmu_buf_rele_array(dbp, numbufs, FTAG); } void dmu_write_embedded(objset_t *os, uint64_t object, uint64_t offset, void *data, uint8_t etype, uint8_t comp, int uncompressed_size, int compressed_size, int byteorder, dmu_tx_t *tx) { dmu_buf_t *db; ASSERT3U(etype, <, NUM_BP_EMBEDDED_TYPES); ASSERT3U(comp, <, ZIO_COMPRESS_FUNCTIONS); VERIFY0(dmu_buf_hold_noread(os, object, offset, FTAG, &db)); dmu_buf_write_embedded(db, data, (bp_embedded_type_t)etype, (enum zio_compress)comp, uncompressed_size, compressed_size, byteorder, tx); dmu_buf_rele(db, FTAG); } /* * DMU support for xuio */ kstat_t *xuio_ksp = NULL; int dmu_xuio_init(xuio_t *xuio, int nblk) { dmu_xuio_t *priv; uio_t *uio = &xuio->xu_uio; uio->uio_iovcnt = nblk; uio->uio_iov = kmem_zalloc(nblk * sizeof (iovec_t), KM_SLEEP); priv = kmem_zalloc(sizeof (dmu_xuio_t), KM_SLEEP); priv->cnt = nblk; priv->bufs = kmem_zalloc(nblk * sizeof (arc_buf_t *), KM_SLEEP); priv->iovp = uio->uio_iov; XUIO_XUZC_PRIV(xuio) = priv; if (XUIO_XUZC_RW(xuio) == UIO_READ) XUIOSTAT_INCR(xuiostat_onloan_rbuf, nblk); else XUIOSTAT_INCR(xuiostat_onloan_wbuf, nblk); return (0); } void dmu_xuio_fini(xuio_t *xuio) { dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); int nblk = priv->cnt; kmem_free(priv->iovp, nblk * sizeof (iovec_t)); kmem_free(priv->bufs, nblk * sizeof (arc_buf_t *)); kmem_free(priv, sizeof (dmu_xuio_t)); if (XUIO_XUZC_RW(xuio) == UIO_READ) XUIOSTAT_INCR(xuiostat_onloan_rbuf, -nblk); else XUIOSTAT_INCR(xuiostat_onloan_wbuf, -nblk); } /* * Initialize iov[priv->next] and priv->bufs[priv->next] with { off, n, abuf } * and increase priv->next by 1. */ int dmu_xuio_add(xuio_t *xuio, arc_buf_t *abuf, offset_t off, size_t n) { struct iovec *iov; uio_t *uio = &xuio->xu_uio; dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); int i = priv->next++; ASSERT(i < priv->cnt); ASSERT(off + n <= arc_buf_lsize(abuf)); iov = uio->uio_iov + i; iov->iov_base = (char *)abuf->b_data + off; iov->iov_len = n; priv->bufs[i] = abuf; return (0); } int dmu_xuio_cnt(xuio_t *xuio) { dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); return (priv->cnt); } arc_buf_t * dmu_xuio_arcbuf(xuio_t *xuio, int i) { dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); ASSERT(i < priv->cnt); return (priv->bufs[i]); } void dmu_xuio_clear(xuio_t *xuio, int i) { dmu_xuio_t *priv = XUIO_XUZC_PRIV(xuio); ASSERT(i < priv->cnt); priv->bufs[i] = NULL; } static void xuio_stat_init(void) { xuio_ksp = kstat_create("zfs", 0, "xuio_stats", "misc", KSTAT_TYPE_NAMED, sizeof (xuio_stats) / sizeof (kstat_named_t), KSTAT_FLAG_VIRTUAL); if (xuio_ksp != NULL) { xuio_ksp->ks_data = &xuio_stats; kstat_install(xuio_ksp); } } static void xuio_stat_fini(void) { if (xuio_ksp != NULL) { kstat_delete(xuio_ksp); xuio_ksp = NULL; } } void xuio_stat_wbuf_copied(void) { XUIOSTAT_BUMP(xuiostat_wbuf_copied); } void xuio_stat_wbuf_nocopy(void) { XUIOSTAT_BUMP(xuiostat_wbuf_nocopy); } #ifdef _KERNEL int dmu_read_uio_dnode(dnode_t *dn, uio_t *uio, uint64_t size) { dmu_buf_t **dbp; int numbufs, i, err; xuio_t *xuio = NULL; /* * NB: we could do this block-at-a-time, but it's nice * to be reading in parallel. */ err = dmu_buf_hold_array_by_dnode(dn, uio->uio_loffset, size, TRUE, FTAG, &numbufs, &dbp, 0); if (err) return (err); #ifdef UIO_XUIO if (uio->uio_extflg == UIO_XUIO) xuio = (xuio_t *)uio; #endif for (i = 0; i < numbufs; i++) { int tocpy; int bufoff; dmu_buf_t *db = dbp[i]; ASSERT(size > 0); bufoff = uio->uio_loffset - db->db_offset; tocpy = (int)MIN(db->db_size - bufoff, size); if (xuio) { dmu_buf_impl_t *dbi = (dmu_buf_impl_t *)db; arc_buf_t *dbuf_abuf = dbi->db_buf; arc_buf_t *abuf = dbuf_loan_arcbuf(dbi); err = dmu_xuio_add(xuio, abuf, bufoff, tocpy); if (!err) { uio->uio_resid -= tocpy; uio->uio_loffset += tocpy; } if (abuf == dbuf_abuf) XUIOSTAT_BUMP(xuiostat_rbuf_nocopy); else XUIOSTAT_BUMP(xuiostat_rbuf_copied); } else { #ifdef illumos err = uiomove((char *)db->db_data + bufoff, tocpy, UIO_READ, uio); #else err = vn_io_fault_uiomove((char *)db->db_data + bufoff, tocpy, uio); #endif } if (err) break; size -= tocpy; } dmu_buf_rele_array(dbp, numbufs, FTAG); return (err); } /* * Read 'size' bytes into the uio buffer. * From object zdb->db_object. * Starting at offset uio->uio_loffset. * * If the caller already has a dbuf in the target object * (e.g. its bonus buffer), this routine is faster than dmu_read_uio(), * because we don't have to find the dnode_t for the object. */ int dmu_read_uio_dbuf(dmu_buf_t *zdb, uio_t *uio, uint64_t size) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb; dnode_t *dn; int err; if (size == 0) return (0); DB_DNODE_ENTER(db); dn = DB_DNODE(db); err = dmu_read_uio_dnode(dn, uio, size); DB_DNODE_EXIT(db); return (err); } /* * Read 'size' bytes into the uio buffer. * From the specified object * Starting at offset uio->uio_loffset. */ int dmu_read_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size) { dnode_t *dn; int err; if (size == 0) return (0); err = dnode_hold(os, object, FTAG, &dn); if (err) return (err); err = dmu_read_uio_dnode(dn, uio, size); dnode_rele(dn, FTAG); return (err); } int dmu_write_uio_dnode(dnode_t *dn, uio_t *uio, uint64_t size, dmu_tx_t *tx) { dmu_buf_t **dbp; int numbufs; int err = 0; int i; err = dmu_buf_hold_array_by_dnode(dn, uio->uio_loffset, size, FALSE, FTAG, &numbufs, &dbp, DMU_READ_PREFETCH); if (err) return (err); for (i = 0; i < numbufs; i++) { int tocpy; int bufoff; dmu_buf_t *db = dbp[i]; ASSERT(size > 0); bufoff = uio->uio_loffset - db->db_offset; tocpy = (int)MIN(db->db_size - bufoff, size); ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); if (tocpy == db->db_size) dmu_buf_will_fill(db, tx); else dmu_buf_will_dirty(db, tx); #ifdef illumos /* * XXX uiomove could block forever (eg. nfs-backed * pages). There needs to be a uiolockdown() function * to lock the pages in memory, so that uiomove won't * block. */ err = uiomove((char *)db->db_data + bufoff, tocpy, UIO_WRITE, uio); #else err = vn_io_fault_uiomove((char *)db->db_data + bufoff, tocpy, uio); #endif if (tocpy == db->db_size) dmu_buf_fill_done(db, tx); if (err) break; size -= tocpy; } dmu_buf_rele_array(dbp, numbufs, FTAG); return (err); } /* * Write 'size' bytes from the uio buffer. * To object zdb->db_object. * Starting at offset uio->uio_loffset. * * If the caller already has a dbuf in the target object * (e.g. its bonus buffer), this routine is faster than dmu_write_uio(), * because we don't have to find the dnode_t for the object. */ int dmu_write_uio_dbuf(dmu_buf_t *zdb, uio_t *uio, uint64_t size, dmu_tx_t *tx) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)zdb; dnode_t *dn; int err; if (size == 0) return (0); DB_DNODE_ENTER(db); dn = DB_DNODE(db); err = dmu_write_uio_dnode(dn, uio, size, tx); DB_DNODE_EXIT(db); return (err); } /* * Write 'size' bytes from the uio buffer. * To the specified object. * Starting at offset uio->uio_loffset. */ int dmu_write_uio(objset_t *os, uint64_t object, uio_t *uio, uint64_t size, dmu_tx_t *tx) { dnode_t *dn; int err; if (size == 0) return (0); err = dnode_hold(os, object, FTAG, &dn); if (err) return (err); err = dmu_write_uio_dnode(dn, uio, size, tx); dnode_rele(dn, FTAG); return (err); } #ifdef illumos int dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, page_t *pp, dmu_tx_t *tx) { dmu_buf_t **dbp; int numbufs, i; int err; if (size == 0) return (0); err = dmu_buf_hold_array(os, object, offset, size, FALSE, FTAG, &numbufs, &dbp); if (err) return (err); for (i = 0; i < numbufs; i++) { int tocpy, copied, thiscpy; int bufoff; dmu_buf_t *db = dbp[i]; caddr_t va; ASSERT(size > 0); ASSERT3U(db->db_size, >=, PAGESIZE); bufoff = offset - db->db_offset; tocpy = (int)MIN(db->db_size - bufoff, size); ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); if (tocpy == db->db_size) dmu_buf_will_fill(db, tx); else dmu_buf_will_dirty(db, tx); for (copied = 0; copied < tocpy; copied += PAGESIZE) { ASSERT3U(pp->p_offset, ==, db->db_offset + bufoff); thiscpy = MIN(PAGESIZE, tocpy - copied); va = zfs_map_page(pp, S_READ); bcopy(va, (char *)db->db_data + bufoff, thiscpy); zfs_unmap_page(pp, va); pp = pp->p_next; bufoff += PAGESIZE; } if (tocpy == db->db_size) dmu_buf_fill_done(db, tx); offset += tocpy; size -= tocpy; } dmu_buf_rele_array(dbp, numbufs, FTAG); return (err); } #else /* !illumos */ int dmu_write_pages(objset_t *os, uint64_t object, uint64_t offset, uint64_t size, vm_page_t *ma, dmu_tx_t *tx) { dmu_buf_t **dbp; struct sf_buf *sf; int numbufs, i; int err; if (size == 0) return (0); err = dmu_buf_hold_array(os, object, offset, size, FALSE, FTAG, &numbufs, &dbp); if (err) return (err); for (i = 0; i < numbufs; i++) { int tocpy, copied, thiscpy; int bufoff; dmu_buf_t *db = dbp[i]; caddr_t va; ASSERT(size > 0); ASSERT3U(db->db_size, >=, PAGESIZE); bufoff = offset - db->db_offset; tocpy = (int)MIN(db->db_size - bufoff, size); ASSERT(i == 0 || i == numbufs-1 || tocpy == db->db_size); if (tocpy == db->db_size) dmu_buf_will_fill(db, tx); else dmu_buf_will_dirty(db, tx); for (copied = 0; copied < tocpy; copied += PAGESIZE) { ASSERT3U(ptoa((*ma)->pindex), ==, db->db_offset + bufoff); thiscpy = MIN(PAGESIZE, tocpy - copied); va = zfs_map_page(*ma, &sf); bcopy(va, (char *)db->db_data + bufoff, thiscpy); zfs_unmap_page(sf); ma += 1; bufoff += PAGESIZE; } if (tocpy == db->db_size) dmu_buf_fill_done(db, tx); offset += tocpy; size -= tocpy; } dmu_buf_rele_array(dbp, numbufs, FTAG); return (err); } int dmu_read_pages(objset_t *os, uint64_t object, vm_page_t *ma, int count, int *rbehind, int *rahead, int last_size) { struct sf_buf *sf; vm_object_t vmobj; vm_page_t m; dmu_buf_t **dbp; dmu_buf_t *db; caddr_t va; int numbufs, i; int bufoff, pgoff, tocpy; int mi, di; int err; ASSERT3U(ma[0]->pindex + count - 1, ==, ma[count - 1]->pindex); ASSERT(last_size <= PAGE_SIZE); err = dmu_buf_hold_array(os, object, IDX_TO_OFF(ma[0]->pindex), IDX_TO_OFF(count - 1) + last_size, TRUE, FTAG, &numbufs, &dbp); if (err != 0) return (err); #ifdef DEBUG IMPLY(last_size < PAGE_SIZE, *rahead == 0); if (dbp[0]->db_offset != 0 || numbufs > 1) { for (i = 0; i < numbufs; i++) { ASSERT(ISP2(dbp[i]->db_size)); ASSERT((dbp[i]->db_offset % dbp[i]->db_size) == 0); ASSERT3U(dbp[i]->db_size, ==, dbp[0]->db_size); } } #endif vmobj = ma[0]->object; db = dbp[0]; for (i = 0; i < *rbehind; i++) { m = vm_page_grab_unlocked(vmobj, ma[0]->pindex - 1 - i, VM_ALLOC_NORMAL | VM_ALLOC_NOWAIT | VM_ALLOC_SBUSY | VM_ALLOC_IGN_SBUSY); if (m == NULL) break; if (!vm_page_none_valid(m)) { ASSERT3U(m->valid, ==, VM_PAGE_BITS_ALL); vm_page_sunbusy(m); break; } ASSERT(m->dirty == 0); - ASSERT(!pmap_page_is_mapped(m)); + ASSERT(!pmap_page_is_write_mapped(m)); ASSERT(db->db_size > PAGE_SIZE); bufoff = IDX_TO_OFF(m->pindex) % db->db_size; va = zfs_map_page(m, &sf); bcopy((char *)db->db_data + bufoff, va, PAGESIZE); zfs_unmap_page(sf); vm_page_valid(m); if ((m->busy_lock & VPB_BIT_WAITERS) != 0) vm_page_activate(m); else vm_page_deactivate(m); vm_page_sunbusy(m); } *rbehind = i; bufoff = IDX_TO_OFF(ma[0]->pindex) % db->db_size; pgoff = 0; for (mi = 0, di = 0; mi < count && di < numbufs; ) { if (pgoff == 0) { m = ma[mi]; if (m != bogus_page) { vm_page_assert_xbusied(m); ASSERT(vm_page_none_valid(m)); ASSERT(m->dirty == 0); ASSERT(!pmap_page_is_mapped(m)); va = zfs_map_page(m, &sf); } } if (bufoff == 0) db = dbp[di]; if (m != bogus_page) { ASSERT3U(IDX_TO_OFF(m->pindex) + pgoff, ==, db->db_offset + bufoff); } /* * We do not need to clamp the copy size by the file * size as the last block is zero-filled beyond the * end of file anyway. */ tocpy = MIN(db->db_size - bufoff, PAGESIZE - pgoff); if (m != bogus_page) bcopy((char *)db->db_data + bufoff, va + pgoff, tocpy); pgoff += tocpy; ASSERT(pgoff <= PAGESIZE); if (pgoff == PAGESIZE) { if (m != bogus_page) { zfs_unmap_page(sf); vm_page_valid(m); } ASSERT(mi < count); mi++; pgoff = 0; } bufoff += tocpy; ASSERT(bufoff <= db->db_size); if (bufoff == db->db_size) { ASSERT(di < numbufs); di++; bufoff = 0; } } #ifdef DEBUG /* * Three possibilities: * - last requested page ends at a buffer boundary and , thus, * all pages and buffers have been iterated; * - all requested pages are filled, but the last buffer * has not been exhausted; * the read-ahead is possible only in this case; * - all buffers have been read, but the last page has not been * fully filled; * this is only possible if the file has only a single buffer * with a size that is not a multiple of the page size. */ if (mi == count) { ASSERT(di >= numbufs - 1); IMPLY(*rahead != 0, di == numbufs - 1); IMPLY(*rahead != 0, bufoff != 0); ASSERT(pgoff == 0); } if (di == numbufs) { ASSERT(mi >= count - 1); ASSERT(*rahead == 0); IMPLY(pgoff == 0, mi == count); if (pgoff != 0) { ASSERT(mi == count - 1); ASSERT((dbp[0]->db_size & PAGE_MASK) != 0); } } #endif if (pgoff != 0) { ASSERT(m != bogus_page); bzero(va + pgoff, PAGESIZE - pgoff); zfs_unmap_page(sf); vm_page_valid(m); } for (i = 0; i < *rahead; i++) { m = vm_page_grab_unlocked(vmobj, ma[count - 1]->pindex + 1 + i, VM_ALLOC_NORMAL | VM_ALLOC_NOWAIT | VM_ALLOC_SBUSY | VM_ALLOC_IGN_SBUSY); if (m == NULL) break; if (!vm_page_none_valid(m)) { ASSERT3U(m->valid, ==, VM_PAGE_BITS_ALL); vm_page_sunbusy(m); break; } ASSERT(m->dirty == 0); - ASSERT(!pmap_page_is_mapped(m)); + ASSERT(!pmap_page_is_write_mapped(m)); ASSERT(db->db_size > PAGE_SIZE); bufoff = IDX_TO_OFF(m->pindex) % db->db_size; tocpy = MIN(db->db_size - bufoff, PAGESIZE); va = zfs_map_page(m, &sf); bcopy((char *)db->db_data + bufoff, va, tocpy); if (tocpy < PAGESIZE) { ASSERT(i == *rahead - 1); ASSERT((db->db_size & PAGE_MASK) != 0); bzero(va + tocpy, PAGESIZE - tocpy); } zfs_unmap_page(sf); vm_page_valid(m); if ((m->busy_lock & VPB_BIT_WAITERS) != 0) vm_page_activate(m); else vm_page_deactivate(m); vm_page_sunbusy(m); } *rahead = i; dmu_buf_rele_array(dbp, numbufs, FTAG); return (0); } #endif /* illumos */ #endif /* _KERNEL */ /* * Allocate a loaned anonymous arc buffer. */ arc_buf_t * dmu_request_arcbuf(dmu_buf_t *handle, int size) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)handle; return (arc_loan_buf(db->db_objset->os_spa, B_FALSE, size)); } /* * Free a loaned arc buffer. */ void dmu_return_arcbuf(arc_buf_t *buf) { arc_return_buf(buf, FTAG); arc_buf_destroy(buf, FTAG); } /* * When possible directly assign passed loaned arc buffer to a dbuf. * If this is not possible copy the contents of passed arc buf via * dmu_write(). */ void dmu_assign_arcbuf_dnode(dnode_t *dn, uint64_t offset, arc_buf_t *buf, dmu_tx_t *tx) { dmu_buf_impl_t *db; uint32_t blksz = (uint32_t)arc_buf_lsize(buf); uint64_t blkid; rw_enter(&dn->dn_struct_rwlock, RW_READER); blkid = dbuf_whichblock(dn, 0, offset); VERIFY((db = dbuf_hold(dn, blkid, FTAG)) != NULL); rw_exit(&dn->dn_struct_rwlock); /* * We can only assign if the offset is aligned, the arc buf is the * same size as the dbuf, and the dbuf is not metadata. */ if (offset == db->db.db_offset && blksz == db->db.db_size) { #ifdef _KERNEL curthread->td_ru.ru_oublock++; #ifdef RACCT if (racct_enable) { PROC_LOCK(curproc); racct_add_force(curproc, RACCT_WRITEBPS, blksz); racct_add_force(curproc, RACCT_WRITEIOPS, 1); PROC_UNLOCK(curproc); } #endif /* RACCT */ #endif /* _KERNEL */ dbuf_assign_arcbuf(db, buf, tx); dbuf_rele(db, FTAG); } else { objset_t *os; uint64_t object; /* compressed bufs must always be assignable to their dbuf */ ASSERT3U(arc_get_compression(buf), ==, ZIO_COMPRESS_OFF); ASSERT(!(buf->b_flags & ARC_BUF_FLAG_COMPRESSED)); os = dn->dn_objset; object = dn->dn_object; dbuf_rele(db, FTAG); dmu_write(os, object, offset, blksz, buf->b_data, tx); dmu_return_arcbuf(buf); XUIOSTAT_BUMP(xuiostat_wbuf_copied); } } void dmu_assign_arcbuf(dmu_buf_t *handle, uint64_t offset, arc_buf_t *buf, dmu_tx_t *tx) { dmu_buf_impl_t *dbuf = (dmu_buf_impl_t *)handle; DB_DNODE_ENTER(dbuf); dmu_assign_arcbuf_dnode(DB_DNODE(dbuf), offset, buf, tx); DB_DNODE_EXIT(dbuf); } typedef struct { dbuf_dirty_record_t *dsa_dr; dmu_sync_cb_t *dsa_done; zgd_t *dsa_zgd; dmu_tx_t *dsa_tx; } dmu_sync_arg_t; /* ARGSUSED */ static void dmu_sync_ready(zio_t *zio, arc_buf_t *buf, void *varg) { dmu_sync_arg_t *dsa = varg; dmu_buf_t *db = dsa->dsa_zgd->zgd_db; blkptr_t *bp = zio->io_bp; if (zio->io_error == 0) { if (BP_IS_HOLE(bp)) { /* * A block of zeros may compress to a hole, but the * block size still needs to be known for replay. */ BP_SET_LSIZE(bp, db->db_size); } else if (!BP_IS_EMBEDDED(bp)) { ASSERT(BP_GET_LEVEL(bp) == 0); bp->blk_fill = 1; } } } static void dmu_sync_late_arrival_ready(zio_t *zio) { dmu_sync_ready(zio, NULL, zio->io_private); } /* ARGSUSED */ static void dmu_sync_done(zio_t *zio, arc_buf_t *buf, void *varg) { dmu_sync_arg_t *dsa = varg; dbuf_dirty_record_t *dr = dsa->dsa_dr; dmu_buf_impl_t *db = dr->dr_dbuf; zgd_t *zgd = dsa->dsa_zgd; /* * Record the vdev(s) backing this blkptr so they can be flushed after * the writes for the lwb have completed. */ if (zio->io_error == 0) { zil_lwb_add_block(zgd->zgd_lwb, zgd->zgd_bp); } mutex_enter(&db->db_mtx); ASSERT(dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC); if (zio->io_error == 0) { dr->dt.dl.dr_nopwrite = !!(zio->io_flags & ZIO_FLAG_NOPWRITE); if (dr->dt.dl.dr_nopwrite) { blkptr_t *bp = zio->io_bp; blkptr_t *bp_orig = &zio->io_bp_orig; uint8_t chksum = BP_GET_CHECKSUM(bp_orig); ASSERT(BP_EQUAL(bp, bp_orig)); VERIFY(BP_EQUAL(bp, db->db_blkptr)); ASSERT(zio->io_prop.zp_compress != ZIO_COMPRESS_OFF); ASSERT(zio_checksum_table[chksum].ci_flags & ZCHECKSUM_FLAG_NOPWRITE); } dr->dt.dl.dr_overridden_by = *zio->io_bp; dr->dt.dl.dr_override_state = DR_OVERRIDDEN; dr->dt.dl.dr_copies = zio->io_prop.zp_copies; /* * Old style holes are filled with all zeros, whereas * new-style holes maintain their lsize, type, level, * and birth time (see zio_write_compress). While we * need to reset the BP_SET_LSIZE() call that happened * in dmu_sync_ready for old style holes, we do *not* * want to wipe out the information contained in new * style holes. Thus, only zero out the block pointer if * it's an old style hole. */ if (BP_IS_HOLE(&dr->dt.dl.dr_overridden_by) && dr->dt.dl.dr_overridden_by.blk_birth == 0) BP_ZERO(&dr->dt.dl.dr_overridden_by); } else { dr->dt.dl.dr_override_state = DR_NOT_OVERRIDDEN; } cv_broadcast(&db->db_changed); mutex_exit(&db->db_mtx); dsa->dsa_done(dsa->dsa_zgd, zio->io_error); kmem_free(dsa, sizeof (*dsa)); } static void dmu_sync_late_arrival_done(zio_t *zio) { blkptr_t *bp = zio->io_bp; dmu_sync_arg_t *dsa = zio->io_private; blkptr_t *bp_orig = &zio->io_bp_orig; zgd_t *zgd = dsa->dsa_zgd; if (zio->io_error == 0) { /* * Record the vdev(s) backing this blkptr so they can be * flushed after the writes for the lwb have completed. */ zil_lwb_add_block(zgd->zgd_lwb, zgd->zgd_bp); if (!BP_IS_HOLE(bp)) { ASSERT(!(zio->io_flags & ZIO_FLAG_NOPWRITE)); ASSERT(BP_IS_HOLE(bp_orig) || !BP_EQUAL(bp, bp_orig)); ASSERT(zio->io_bp->blk_birth == zio->io_txg); ASSERT(zio->io_txg > spa_syncing_txg(zio->io_spa)); zio_free(zio->io_spa, zio->io_txg, zio->io_bp); } } dmu_tx_commit(dsa->dsa_tx); dsa->dsa_done(dsa->dsa_zgd, zio->io_error); abd_put(zio->io_abd); kmem_free(dsa, sizeof (*dsa)); } static int dmu_sync_late_arrival(zio_t *pio, objset_t *os, dmu_sync_cb_t *done, zgd_t *zgd, zio_prop_t *zp, zbookmark_phys_t *zb) { dmu_sync_arg_t *dsa; dmu_tx_t *tx; tx = dmu_tx_create(os); dmu_tx_hold_space(tx, zgd->zgd_db->db_size); if (dmu_tx_assign(tx, TXG_WAIT) != 0) { dmu_tx_abort(tx); /* Make zl_get_data do txg_waited_synced() */ return (SET_ERROR(EIO)); } /* * In order to prevent the zgd's lwb from being free'd prior to * dmu_sync_late_arrival_done() being called, we have to ensure * the lwb's "max txg" takes this tx's txg into account. */ zil_lwb_add_txg(zgd->zgd_lwb, dmu_tx_get_txg(tx)); dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP); dsa->dsa_dr = NULL; dsa->dsa_done = done; dsa->dsa_zgd = zgd; dsa->dsa_tx = tx; /* * Since we are currently syncing this txg, it's nontrivial to * determine what BP to nopwrite against, so we disable nopwrite. * * When syncing, the db_blkptr is initially the BP of the previous * txg. We can not nopwrite against it because it will be changed * (this is similar to the non-late-arrival case where the dbuf is * dirty in a future txg). * * Then dbuf_write_ready() sets bp_blkptr to the location we will write. * We can not nopwrite against it because although the BP will not * (typically) be changed, the data has not yet been persisted to this * location. * * Finally, when dbuf_write_done() is called, it is theoretically * possible to always nopwrite, because the data that was written in * this txg is the same data that we are trying to write. However we * would need to check that this dbuf is not dirty in any future * txg's (as we do in the normal dmu_sync() path). For simplicity, we * don't nopwrite in this case. */ zp->zp_nopwrite = B_FALSE; zio_nowait(zio_write(pio, os->os_spa, dmu_tx_get_txg(tx), zgd->zgd_bp, abd_get_from_buf(zgd->zgd_db->db_data, zgd->zgd_db->db_size), zgd->zgd_db->db_size, zgd->zgd_db->db_size, zp, dmu_sync_late_arrival_ready, NULL, NULL, dmu_sync_late_arrival_done, dsa, ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, zb)); return (0); } /* * Intent log support: sync the block associated with db to disk. * N.B. and XXX: the caller is responsible for making sure that the * data isn't changing while dmu_sync() is writing it. * * Return values: * * EEXIST: this txg has already been synced, so there's nothing to do. * The caller should not log the write. * * ENOENT: the block was dbuf_free_range()'d, so there's nothing to do. * The caller should not log the write. * * EALREADY: this block is already in the process of being synced. * The caller should track its progress (somehow). * * EIO: could not do the I/O. * The caller should do a txg_wait_synced(). * * 0: the I/O has been initiated. * The caller should log this blkptr in the done callback. * It is possible that the I/O will fail, in which case * the error will be reported to the done callback and * propagated to pio from zio_done(). */ int dmu_sync(zio_t *pio, uint64_t txg, dmu_sync_cb_t *done, zgd_t *zgd) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)zgd->zgd_db; objset_t *os = db->db_objset; dsl_dataset_t *ds = os->os_dsl_dataset; dbuf_dirty_record_t *dr; dmu_sync_arg_t *dsa; zbookmark_phys_t zb; zio_prop_t zp; dnode_t *dn; ASSERT(pio != NULL); ASSERT(txg != 0); SET_BOOKMARK(&zb, ds->ds_object, db->db.db_object, db->db_level, db->db_blkid); DB_DNODE_ENTER(db); dn = DB_DNODE(db); dmu_write_policy(os, dn, db->db_level, WP_DMU_SYNC, &zp); DB_DNODE_EXIT(db); /* * If we're frozen (running ziltest), we always need to generate a bp. */ if (txg > spa_freeze_txg(os->os_spa)) return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb)); /* * Grabbing db_mtx now provides a barrier between dbuf_sync_leaf() * and us. If we determine that this txg is not yet syncing, * but it begins to sync a moment later, that's OK because the * sync thread will block in dbuf_sync_leaf() until we drop db_mtx. */ mutex_enter(&db->db_mtx); if (txg <= spa_last_synced_txg(os->os_spa)) { /* * This txg has already synced. There's nothing to do. */ mutex_exit(&db->db_mtx); return (SET_ERROR(EEXIST)); } if (txg <= spa_syncing_txg(os->os_spa)) { /* * This txg is currently syncing, so we can't mess with * the dirty record anymore; just write a new log block. */ mutex_exit(&db->db_mtx); return (dmu_sync_late_arrival(pio, os, done, zgd, &zp, &zb)); } dr = db->db_last_dirty; while (dr && dr->dr_txg != txg) dr = dr->dr_next; if (dr == NULL) { /* * There's no dr for this dbuf, so it must have been freed. * There's no need to log writes to freed blocks, so we're done. */ mutex_exit(&db->db_mtx); return (SET_ERROR(ENOENT)); } ASSERT(dr->dr_next == NULL || dr->dr_next->dr_txg < txg); if (db->db_blkptr != NULL) { /* * We need to fill in zgd_bp with the current blkptr so that * the nopwrite code can check if we're writing the same * data that's already on disk. We can only nopwrite if we * are sure that after making the copy, db_blkptr will not * change until our i/o completes. We ensure this by * holding the db_mtx, and only allowing nopwrite if the * block is not already dirty (see below). This is verified * by dmu_sync_done(), which VERIFYs that the db_blkptr has * not changed. */ *zgd->zgd_bp = *db->db_blkptr; } /* * Assume the on-disk data is X, the current syncing data (in * txg - 1) is Y, and the current in-memory data is Z (currently * in dmu_sync). * * We usually want to perform a nopwrite if X and Z are the * same. However, if Y is different (i.e. the BP is going to * change before this write takes effect), then a nopwrite will * be incorrect - we would override with X, which could have * been freed when Y was written. * * (Note that this is not a concern when we are nop-writing from * syncing context, because X and Y must be identical, because * all previous txgs have been synced.) * * Therefore, we disable nopwrite if the current BP could change * before this TXG. There are two ways it could change: by * being dirty (dr_next is non-NULL), or by being freed * (dnode_block_freed()). This behavior is verified by * zio_done(), which VERIFYs that the override BP is identical * to the on-disk BP. */ DB_DNODE_ENTER(db); dn = DB_DNODE(db); if (dr->dr_next != NULL || dnode_block_freed(dn, db->db_blkid)) zp.zp_nopwrite = B_FALSE; DB_DNODE_EXIT(db); ASSERT(dr->dr_txg == txg); if (dr->dt.dl.dr_override_state == DR_IN_DMU_SYNC || dr->dt.dl.dr_override_state == DR_OVERRIDDEN) { /* * We have already issued a sync write for this buffer, * or this buffer has already been synced. It could not * have been dirtied since, or we would have cleared the state. */ mutex_exit(&db->db_mtx); return (SET_ERROR(EALREADY)); } ASSERT(dr->dt.dl.dr_override_state == DR_NOT_OVERRIDDEN); dr->dt.dl.dr_override_state = DR_IN_DMU_SYNC; mutex_exit(&db->db_mtx); dsa = kmem_alloc(sizeof (dmu_sync_arg_t), KM_SLEEP); dsa->dsa_dr = dr; dsa->dsa_done = done; dsa->dsa_zgd = zgd; dsa->dsa_tx = NULL; zio_nowait(arc_write(pio, os->os_spa, txg, zgd->zgd_bp, dr->dt.dl.dr_data, DBUF_IS_L2CACHEABLE(db), &zp, dmu_sync_ready, NULL, NULL, dmu_sync_done, dsa, ZIO_PRIORITY_SYNC_WRITE, ZIO_FLAG_CANFAIL, &zb)); return (0); } int dmu_object_set_blocksize(objset_t *os, uint64_t object, uint64_t size, int ibs, dmu_tx_t *tx) { dnode_t *dn; int err; err = dnode_hold(os, object, FTAG, &dn); if (err) return (err); err = dnode_set_blksz(dn, size, ibs, tx); dnode_rele(dn, FTAG); return (err); } void dmu_object_set_checksum(objset_t *os, uint64_t object, uint8_t checksum, dmu_tx_t *tx) { dnode_t *dn; /* * Send streams include each object's checksum function. This * check ensures that the receiving system can understand the * checksum function transmitted. */ ASSERT3U(checksum, <, ZIO_CHECKSUM_LEGACY_FUNCTIONS); VERIFY0(dnode_hold(os, object, FTAG, &dn)); ASSERT3U(checksum, <, ZIO_CHECKSUM_FUNCTIONS); dn->dn_checksum = checksum; dnode_setdirty(dn, tx); dnode_rele(dn, FTAG); } void dmu_object_set_compress(objset_t *os, uint64_t object, uint8_t compress, dmu_tx_t *tx) { dnode_t *dn; /* * Send streams include each object's compression function. This * check ensures that the receiving system can understand the * compression function transmitted. */ ASSERT3U(compress, <, ZIO_COMPRESS_LEGACY_FUNCTIONS); VERIFY0(dnode_hold(os, object, FTAG, &dn)); dn->dn_compress = compress; dnode_setdirty(dn, tx); dnode_rele(dn, FTAG); } int zfs_mdcomp_disable = 0; SYSCTL_INT(_vfs_zfs, OID_AUTO, mdcomp_disable, CTLFLAG_RWTUN, &zfs_mdcomp_disable, 0, "Disable metadata compression"); /* * When the "redundant_metadata" property is set to "most", only indirect * blocks of this level and higher will have an additional ditto block. */ int zfs_redundant_metadata_most_ditto_level = 2; void dmu_write_policy(objset_t *os, dnode_t *dn, int level, int wp, zio_prop_t *zp) { dmu_object_type_t type = dn ? dn->dn_type : DMU_OT_OBJSET; boolean_t ismd = (level > 0 || DMU_OT_IS_METADATA(type) || (wp & WP_SPILL)); enum zio_checksum checksum = os->os_checksum; enum zio_compress compress = os->os_compress; enum zio_checksum dedup_checksum = os->os_dedup_checksum; boolean_t dedup = B_FALSE; boolean_t nopwrite = B_FALSE; boolean_t dedup_verify = os->os_dedup_verify; int copies = os->os_copies; /* * We maintain different write policies for each of the following * types of data: * 1. metadata * 2. preallocated blocks (i.e. level-0 blocks of a dump device) * 3. all other level 0 blocks */ if (ismd) { if (zfs_mdcomp_disable) { compress = ZIO_COMPRESS_EMPTY; } else { /* * XXX -- we should design a compression algorithm * that specializes in arrays of bps. */ compress = zio_compress_select(os->os_spa, ZIO_COMPRESS_ON, ZIO_COMPRESS_ON); } /* * Metadata always gets checksummed. If the data * checksum is multi-bit correctable, and it's not a * ZBT-style checksum, then it's suitable for metadata * as well. Otherwise, the metadata checksum defaults * to fletcher4. */ if (!(zio_checksum_table[checksum].ci_flags & ZCHECKSUM_FLAG_METADATA) || (zio_checksum_table[checksum].ci_flags & ZCHECKSUM_FLAG_EMBEDDED)) checksum = ZIO_CHECKSUM_FLETCHER_4; if (os->os_redundant_metadata == ZFS_REDUNDANT_METADATA_ALL || (os->os_redundant_metadata == ZFS_REDUNDANT_METADATA_MOST && (level >= zfs_redundant_metadata_most_ditto_level || DMU_OT_IS_METADATA(type) || (wp & WP_SPILL)))) copies++; } else if (wp & WP_NOFILL) { ASSERT(level == 0); /* * If we're writing preallocated blocks, we aren't actually * writing them so don't set any policy properties. These * blocks are currently only used by an external subsystem * outside of zfs (i.e. dump) and not written by the zio * pipeline. */ compress = ZIO_COMPRESS_OFF; checksum = ZIO_CHECKSUM_NOPARITY; } else { compress = zio_compress_select(os->os_spa, dn->dn_compress, compress); checksum = (dedup_checksum == ZIO_CHECKSUM_OFF) ? zio_checksum_select(dn->dn_checksum, checksum) : dedup_checksum; /* * Determine dedup setting. If we are in dmu_sync(), * we won't actually dedup now because that's all * done in syncing context; but we do want to use the * dedup checkum. If the checksum is not strong * enough to ensure unique signatures, force * dedup_verify. */ if (dedup_checksum != ZIO_CHECKSUM_OFF) { dedup = (wp & WP_DMU_SYNC) ? B_FALSE : B_TRUE; if (!(zio_checksum_table[checksum].ci_flags & ZCHECKSUM_FLAG_DEDUP)) dedup_verify = B_TRUE; } /* * Enable nopwrite if we have secure enough checksum * algorithm (see comment in zio_nop_write) and * compression is enabled. We don't enable nopwrite if * dedup is enabled as the two features are mutually * exclusive. */ nopwrite = (!dedup && (zio_checksum_table[checksum].ci_flags & ZCHECKSUM_FLAG_NOPWRITE) && compress != ZIO_COMPRESS_OFF && zfs_nopwrite_enabled); } zp->zp_checksum = checksum; zp->zp_compress = compress; ASSERT3U(zp->zp_compress, !=, ZIO_COMPRESS_INHERIT); zp->zp_type = (wp & WP_SPILL) ? dn->dn_bonustype : type; zp->zp_level = level; zp->zp_copies = MIN(copies, spa_max_replication(os->os_spa)); zp->zp_dedup = dedup; zp->zp_dedup_verify = dedup && dedup_verify; zp->zp_nopwrite = nopwrite; zp->zp_zpl_smallblk = DMU_OT_IS_FILE(zp->zp_type) ? os->os_zpl_special_smallblock : 0; } int dmu_offset_next(objset_t *os, uint64_t object, boolean_t hole, uint64_t *off) { dnode_t *dn; int err; /* * Sync any current changes before * we go trundling through the block pointers. */ err = dmu_object_wait_synced(os, object); if (err) { return (err); } err = dnode_hold(os, object, FTAG, &dn); if (err) { return (err); } err = dnode_next_offset(dn, (hole ? DNODE_FIND_HOLE : 0), off, 1, 1, 0); dnode_rele(dn, FTAG); return (err); } /* * Given the ZFS object, if it contains any dirty nodes * this function flushes all dirty blocks to disk. This * ensures the DMU object info is updated. A more efficient * future version might just find the TXG with the maximum * ID and wait for that to be synced. */ int dmu_object_wait_synced(objset_t *os, uint64_t object) { dnode_t *dn; int error, i; error = dnode_hold(os, object, FTAG, &dn); if (error) { return (error); } for (i = 0; i < TXG_SIZE; i++) { if (list_link_active(&dn->dn_dirty_link[i])) { break; } } dnode_rele(dn, FTAG); if (i != TXG_SIZE) { txg_wait_synced(dmu_objset_pool(os), 0); } return (0); } void __dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi) { dnode_phys_t *dnp = dn->dn_phys; doi->doi_data_block_size = dn->dn_datablksz; doi->doi_metadata_block_size = dn->dn_indblkshift ? 1ULL << dn->dn_indblkshift : 0; doi->doi_type = dn->dn_type; doi->doi_bonus_type = dn->dn_bonustype; doi->doi_bonus_size = dn->dn_bonuslen; doi->doi_dnodesize = dn->dn_num_slots << DNODE_SHIFT; doi->doi_indirection = dn->dn_nlevels; doi->doi_checksum = dn->dn_checksum; doi->doi_compress = dn->dn_compress; doi->doi_nblkptr = dn->dn_nblkptr; doi->doi_physical_blocks_512 = (DN_USED_BYTES(dnp) + 256) >> 9; doi->doi_max_offset = (dn->dn_maxblkid + 1) * dn->dn_datablksz; doi->doi_fill_count = 0; for (int i = 0; i < dnp->dn_nblkptr; i++) doi->doi_fill_count += BP_GET_FILL(&dnp->dn_blkptr[i]); } void dmu_object_info_from_dnode(dnode_t *dn, dmu_object_info_t *doi) { rw_enter(&dn->dn_struct_rwlock, RW_READER); mutex_enter(&dn->dn_mtx); __dmu_object_info_from_dnode(dn, doi); mutex_exit(&dn->dn_mtx); rw_exit(&dn->dn_struct_rwlock); } /* * Get information on a DMU object. * If doi is NULL, just indicates whether the object exists. */ int dmu_object_info(objset_t *os, uint64_t object, dmu_object_info_t *doi) { dnode_t *dn; int err = dnode_hold(os, object, FTAG, &dn); if (err) return (err); if (doi != NULL) dmu_object_info_from_dnode(dn, doi); dnode_rele(dn, FTAG); return (0); } /* * As above, but faster; can be used when you have a held dbuf in hand. */ void dmu_object_info_from_db(dmu_buf_t *db_fake, dmu_object_info_t *doi) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; DB_DNODE_ENTER(db); dmu_object_info_from_dnode(DB_DNODE(db), doi); DB_DNODE_EXIT(db); } /* * Faster still when you only care about the size. * This is specifically optimized for zfs_getattr(). */ void dmu_object_size_from_db(dmu_buf_t *db_fake, uint32_t *blksize, u_longlong_t *nblk512) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; dnode_t *dn; DB_DNODE_ENTER(db); dn = DB_DNODE(db); *blksize = dn->dn_datablksz; /* add in number of slots used for the dnode itself */ *nblk512 = ((DN_USED_BYTES(dn->dn_phys) + SPA_MINBLOCKSIZE/2) >> SPA_MINBLOCKSHIFT) + dn->dn_num_slots; DB_DNODE_EXIT(db); } void dmu_object_dnsize_from_db(dmu_buf_t *db_fake, int *dnsize) { dmu_buf_impl_t *db = (dmu_buf_impl_t *)db_fake; dnode_t *dn; DB_DNODE_ENTER(db); dn = DB_DNODE(db); *dnsize = dn->dn_num_slots << DNODE_SHIFT; DB_DNODE_EXIT(db); } void byteswap_uint64_array(void *vbuf, size_t size) { uint64_t *buf = vbuf; size_t count = size >> 3; int i; ASSERT((size & 7) == 0); for (i = 0; i < count; i++) buf[i] = BSWAP_64(buf[i]); } void byteswap_uint32_array(void *vbuf, size_t size) { uint32_t *buf = vbuf; size_t count = size >> 2; int i; ASSERT((size & 3) == 0); for (i = 0; i < count; i++) buf[i] = BSWAP_32(buf[i]); } void byteswap_uint16_array(void *vbuf, size_t size) { uint16_t *buf = vbuf; size_t count = size >> 1; int i; ASSERT((size & 1) == 0); for (i = 0; i < count; i++) buf[i] = BSWAP_16(buf[i]); } /* ARGSUSED */ void byteswap_uint8_array(void *vbuf, size_t size) { } void dmu_init(void) { abd_init(); zfs_dbgmsg_init(); sa_cache_init(); xuio_stat_init(); dmu_objset_init(); dnode_init(); zfetch_init(); zio_compress_init(); l2arc_init(); arc_init(); dbuf_init(); } void dmu_fini(void) { arc_fini(); /* arc depends on l2arc, so arc must go first */ l2arc_fini(); zfetch_fini(); zio_compress_fini(); dbuf_fini(); dnode_fini(); dmu_objset_fini(); xuio_stat_fini(); sa_cache_fini(); zfs_dbgmsg_fini(); abd_fini(); } diff --git a/sys/conf/files b/sys/conf/files index 6fbb76e7a056..f000bc7a383e 100644 --- a/sys/conf/files +++ b/sys/conf/files @@ -1,4986 +1,4987 @@ # $FreeBSD$ # # The long compile-with and dependency lines are required because of # limitations in config: backslash-newline doesn't work in strings, and # dependency lines other than the first are silently ignored. # acpi_quirks.h optional acpi \ dependency "$S/tools/acpi_quirks2h.awk $S/dev/acpica/acpi_quirks" \ compile-with "${AWK} -f $S/tools/acpi_quirks2h.awk $S/dev/acpica/acpi_quirks" \ no-obj no-implicit-rule before-depend \ clean "acpi_quirks.h" bhnd_nvram_map.h optional bhnd \ dependency "$S/dev/bhnd/tools/nvram_map_gen.sh $S/dev/bhnd/tools/nvram_map_gen.awk $S/dev/bhnd/nvram/nvram_map" \ compile-with "sh $S/dev/bhnd/tools/nvram_map_gen.sh $S/dev/bhnd/nvram/nvram_map -h" \ no-obj no-implicit-rule before-depend \ clean "bhnd_nvram_map.h" bhnd_nvram_map_data.h optional bhnd \ dependency "$S/dev/bhnd/tools/nvram_map_gen.sh $S/dev/bhnd/tools/nvram_map_gen.awk $S/dev/bhnd/nvram/nvram_map" \ compile-with "sh $S/dev/bhnd/tools/nvram_map_gen.sh $S/dev/bhnd/nvram/nvram_map -d" \ no-obj no-implicit-rule before-depend \ clean "bhnd_nvram_map_data.h" fdt_static_dtb.h optional fdt fdt_dtb_static \ compile-with "sh -c 'MACHINE=${MACHINE} $S/tools/fdt/make_dtbh.sh ${FDT_DTS_FILE} ${.CURDIR}'" \ dependency "${FDT_DTS_FILE:T:R}.dtb" \ no-obj no-implicit-rule before-depend \ clean "fdt_static_dtb.h" feeder_eq_gen.h optional sound \ dependency "$S/tools/sound/feeder_eq_mkfilter.awk" \ compile-with "${AWK} -f $S/tools/sound/feeder_eq_mkfilter.awk -- ${FEEDER_EQ_PRESETS} > feeder_eq_gen.h" \ no-obj no-implicit-rule before-depend \ clean "feeder_eq_gen.h" feeder_rate_gen.h optional sound \ dependency "$S/tools/sound/feeder_rate_mkfilter.awk" \ compile-with "${AWK} -f $S/tools/sound/feeder_rate_mkfilter.awk -- ${FEEDER_RATE_PRESETS} > feeder_rate_gen.h" \ no-obj no-implicit-rule before-depend \ clean "feeder_rate_gen.h" font.h optional sc_dflt_font \ compile-with "uudecode < ${SRCTOP}/share/syscons/fonts/${SC_DFLT_FONT}-8x16.fnt && file2c 'u_char dflt_font_16[16*256] = {' '};' < ${SC_DFLT_FONT}-8x16 > font.h && uudecode < ${SRCTOP}/share/syscons/fonts/${SC_DFLT_FONT}-8x14.fnt && file2c 'u_char dflt_font_14[14*256] = {' '};' < ${SC_DFLT_FONT}-8x14 >> font.h && uudecode < ${SRCTOP}/share/syscons/fonts/${SC_DFLT_FONT}-8x8.fnt && file2c 'u_char dflt_font_8[8*256] = {' '};' < ${SC_DFLT_FONT}-8x8 >> font.h" \ no-obj no-implicit-rule before-depend \ clean "font.h ${SC_DFLT_FONT}-8x14 ${SC_DFLT_FONT}-8x16 ${SC_DFLT_FONT}-8x8" snd_fxdiv_gen.h optional sound \ dependency "$S/tools/sound/snd_fxdiv_gen.awk" \ compile-with "${AWK} -f $S/tools/sound/snd_fxdiv_gen.awk -- > snd_fxdiv_gen.h" \ no-obj no-implicit-rule before-depend \ clean "snd_fxdiv_gen.h" miidevs.h optional miibus | mii \ dependency "$S/tools/miidevs2h.awk $S/dev/mii/miidevs" \ compile-with "${AWK} -f $S/tools/miidevs2h.awk $S/dev/mii/miidevs" \ no-obj no-implicit-rule before-depend \ clean "miidevs.h" pccarddevs.h standard \ dependency "$S/tools/pccarddevs2h.awk $S/dev/pccard/pccarddevs" \ compile-with "${AWK} -f $S/tools/pccarddevs2h.awk $S/dev/pccard/pccarddevs" \ no-obj no-implicit-rule before-depend \ clean "pccarddevs.h" kbdmuxmap.h optional kbdmux_dflt_keymap \ compile-with "${KEYMAP} -L ${KBDMUX_DFLT_KEYMAP} | ${KEYMAP_FIX} > ${.TARGET}" \ no-obj no-implicit-rule before-depend \ clean "kbdmuxmap.h" teken_state.h optional sc | vt \ dependency "$S/teken/gensequences $S/teken/sequences" \ compile-with "${AWK} -f $S/teken/gensequences $S/teken/sequences > teken_state.h" \ no-obj no-implicit-rule before-depend \ clean "teken_state.h" ukbdmap.h optional ukbd_dflt_keymap \ compile-with "${KEYMAP} -L ${UKBD_DFLT_KEYMAP} | ${KEYMAP_FIX} > ${.TARGET}" \ no-obj no-implicit-rule before-depend \ clean "ukbdmap.h" usbdevs.h optional usb \ dependency "$S/tools/usbdevs2h.awk $S/dev/usb/usbdevs" \ compile-with "${AWK} -f $S/tools/usbdevs2h.awk $S/dev/usb/usbdevs -h" \ no-obj no-implicit-rule before-depend \ clean "usbdevs.h" usbdevs_data.h optional usb \ dependency "$S/tools/usbdevs2h.awk $S/dev/usb/usbdevs" \ compile-with "${AWK} -f $S/tools/usbdevs2h.awk $S/dev/usb/usbdevs -d" \ no-obj no-implicit-rule before-depend \ clean "usbdevs_data.h" sdiodevs.h optional mmccam \ dependency "$S/tools/sdiodevs2h.awk $S/dev/sdio/sdiodevs" \ compile-with "${AWK} -f $S/tools/sdiodevs2h.awk $S/dev/sdio/sdiodevs -h" \ no-obj no-implicit-rule before-depend \ clean "sdiodevs.h" sdiodevs_data.h optional mmccam \ dependency "$S/tools/sdiodevs2h.awk $S/dev/sdio/sdiodevs" \ compile-with "${AWK} -f $S/tools/sdiodevs2h.awk $S/dev/sdio/sdiodevs -d" \ no-obj no-implicit-rule before-depend \ clean "sdiodevs_data.h" cam/cam.c optional scbus cam/cam_compat.c optional scbus cam/cam_iosched.c optional scbus cam/cam_periph.c optional scbus cam/cam_queue.c optional scbus cam/cam_sim.c optional scbus cam/cam_xpt.c optional scbus cam/ata/ata_all.c optional scbus cam/ata/ata_xpt.c optional scbus cam/ata/ata_pmp.c optional scbus cam/nvme/nvme_all.c optional scbus cam/nvme/nvme_da.c optional nda | da cam/nvme/nvme_xpt.c optional scbus cam/scsi/scsi_xpt.c optional scbus cam/scsi/scsi_all.c optional scbus cam/scsi/scsi_cd.c optional cd cam/scsi/scsi_ch.c optional ch cam/ata/ata_da.c optional ada | da cam/ctl/ctl.c optional ctl cam/ctl/ctl_backend.c optional ctl cam/ctl/ctl_backend_block.c optional ctl cam/ctl/ctl_backend_ramdisk.c optional ctl cam/ctl/ctl_cmd_table.c optional ctl cam/ctl/ctl_frontend.c optional ctl cam/ctl/ctl_frontend_cam_sim.c optional ctl cam/ctl/ctl_frontend_ioctl.c optional ctl cam/ctl/ctl_frontend_iscsi.c optional ctl cfiscsi cam/ctl/ctl_ha.c optional ctl cam/ctl/ctl_scsi_all.c optional ctl cam/ctl/ctl_tpc.c optional ctl cam/ctl/ctl_tpc_local.c optional ctl cam/ctl/ctl_error.c optional ctl cam/ctl/ctl_util.c optional ctl cam/ctl/scsi_ctl.c optional ctl cam/mmc/mmc_xpt.c optional scbus mmccam cam/mmc/mmc_da.c optional scbus mmccam da cam/scsi/scsi_da.c optional da cam/scsi/scsi_pass.c optional pass cam/scsi/scsi_pt.c optional pt cam/scsi/scsi_sa.c optional sa cam/scsi/scsi_enc.c optional ses cam/scsi/scsi_enc_ses.c optional ses cam/scsi/scsi_enc_safte.c optional ses cam/scsi/scsi_sg.c optional sg cam/scsi/scsi_targ_bh.c optional targbh cam/scsi/scsi_target.c optional targ cam/scsi/smp_all.c optional scbus # shared between zfs and dtrace cddl/compat/opensolaris/kern/opensolaris.c optional zfs | dtrace compile-with "${CDDL_C}" cddl/compat/opensolaris/kern/opensolaris_cmn_err.c optional zfs | dtrace compile-with "${CDDL_C}" cddl/compat/opensolaris/kern/opensolaris_kmem.c optional zfs | dtrace compile-with "${CDDL_C}" cddl/compat/opensolaris/kern/opensolaris_misc.c optional zfs | dtrace compile-with "${CDDL_C}" cddl/compat/opensolaris/kern/opensolaris_proc.c optional zfs | dtrace compile-with "${CDDL_C}" cddl/compat/opensolaris/kern/opensolaris_sunddi.c optional zfs | dtrace compile-with "${CDDL_C}" cddl/compat/opensolaris/kern/opensolaris_taskq.c optional zfs | dtrace compile-with "${CDDL_C}" # zfs specific cddl/compat/opensolaris/kern/opensolaris_acl.c optional zfs compile-with "${ZFS_C}" cddl/compat/opensolaris/kern/opensolaris_dtrace.c optional zfs compile-with "${ZFS_C}" cddl/compat/opensolaris/kern/opensolaris_kobj.c optional zfs compile-with "${ZFS_C}" cddl/compat/opensolaris/kern/opensolaris_kstat.c optional zfs compile-with "${ZFS_C}" cddl/compat/opensolaris/kern/opensolaris_lookup.c optional zfs compile-with "${ZFS_C}" cddl/compat/opensolaris/kern/opensolaris_policy.c optional zfs compile-with "${ZFS_C}" cddl/compat/opensolaris/kern/opensolaris_string.c optional zfs compile-with "${ZFS_C}" cddl/compat/opensolaris/kern/opensolaris_sysevent.c optional zfs compile-with "${ZFS_C}" cddl/compat/opensolaris/kern/opensolaris_uio.c optional zfs compile-with "${ZFS_C}" cddl/compat/opensolaris/kern/opensolaris_vfs.c optional zfs compile-with "${ZFS_C}" cddl/compat/opensolaris/kern/opensolaris_vm.c optional zfs compile-with "${ZFS_C}" cddl/compat/opensolaris/kern/opensolaris_zone.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/common/acl/acl_common.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/common/avl/avl.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/common/lz4/lz4.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/common/nvpair/opensolaris_fnvpair.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/common/nvpair/opensolaris_nvpair.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/common/nvpair/opensolaris_nvpair_alloc_fixed.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/common/unicode/u8_textprep.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/common/zfs/zfeature_common.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/common/zfs/zfs_comutil.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/common/zfs/zfs_deleg.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/common/zfs/zfs_fletcher.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/common/zfs/zfs_ioctl_compat.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/common/zfs/zfs_namecheck.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/common/zfs/zfs_prop.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/common/zfs/zpool_prop.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/common/zfs/zprop_common.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/vnode.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/abd.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/aggsum.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/arc.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/blkptr.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/bplist.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/bpobj.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/bptree.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/bqueue.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/cityhash.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/dbuf.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/dbuf_stats.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/ddt.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/ddt_zap.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/dmu.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/dmu_diff.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/dmu_object.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/dmu_objset.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/dmu_send.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/dmu_traverse.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/dmu_tx.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/dmu_zfetch.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/dnode.c optional zfs compile-with "${ZFS_C}" \ warning "kernel contains CDDL licensed ZFS filesystem" cddl/contrib/opensolaris/uts/common/fs/zfs/dnode_sync.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/dsl_bookmark.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/dsl_dataset.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/dsl_deadlist.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/dsl_deleg.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/dsl_destroy.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/dsl_dir.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/dsl_pool.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/dsl_prop.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/dsl_scan.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/dsl_userhold.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/dsl_synctask.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/gzip.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/lzjb.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/metaslab.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/mmp.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/multilist.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/range_tree.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/refcount.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/rrwlock.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/sa.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/sha256.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/skein_zfs.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/spa.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/spa_checkpoint.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/spa_config.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/spa_errlog.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/spa_history.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/spa_misc.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/space_map.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/space_reftree.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/trim_map.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/txg.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/uberblock.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/unique.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/vdev.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/vdev_cache.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/vdev_file.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/vdev_indirect.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/vdev_indirect_births.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/vdev_indirect_mapping.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/vdev_initialize.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/vdev_geom.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/vdev_label.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/vdev_mirror.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/vdev_missing.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/vdev_queue.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/vdev_raidz.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/vdev_removal.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/vdev_root.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/zap.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/zap_leaf.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/zap_micro.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/zcp.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/zcp_get.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/zcp_global.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/zcp_iter.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/zcp_synctask.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/zfeature.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/zfs_acl.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/zfs_byteswap.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/zfs_ctldir.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/zfs_debug.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/zfs_dir.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/zfs_fm.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/zfs_fuid.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/zfs_ioctl.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/zfs_log.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/zfs_onexit.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/zfs_replay.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/zfs_rlock.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/zfs_sa.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/zfs_vfsops.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/zfs_vnops.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/zfs_znode.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/zil.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/zio.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/zio_checksum.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/zio_compress.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/zio_inject.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/zle.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/zrlock.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/zthr.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/zvol.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/os/callb.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/os/fm.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/os/list.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/os/nvpair_alloc_system.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/zmod/zmod.c optional zfs compile-with "${ZFS_C}" # zfs lua support cddl/contrib/opensolaris/uts/common/fs/zfs/lua/lapi.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/lua/lauxlib.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/lua/lbaselib.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/lua/lbitlib.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/lua/lcode.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/lua/lcompat.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/lua/lcorolib.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/lua/lctype.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/lua/ldebug.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/lua/ldo.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/lua/ldump.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/lua/lfunc.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/lua/lgc.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/lua/llex.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/lua/lmem.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/lua/lobject.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/lua/lopcodes.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/lua/lparser.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/lua/lstate.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/lua/lstring.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/lua/lstrlib.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/lua/ltable.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/lua/ltablib.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/lua/ltm.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/lua/lundump.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/lua/lvm.c optional zfs compile-with "${ZFS_C}" cddl/contrib/opensolaris/uts/common/fs/zfs/lua/lzio.c optional zfs compile-with "${ZFS_C}" # dtrace specific cddl/contrib/opensolaris/uts/common/dtrace/dtrace.c optional dtrace compile-with "${DTRACE_C}" \ warning "kernel contains CDDL licensed DTRACE" cddl/contrib/opensolaris/uts/common/dtrace/dtrace_xoroshiro128_plus.c optional dtrace compile-with "${DTRACE_C}" cddl/dev/dtmalloc/dtmalloc.c optional dtmalloc | dtraceall compile-with "${CDDL_C}" cddl/dev/profile/profile.c optional dtrace_profile | dtraceall compile-with "${CDDL_C}" cddl/dev/sdt/sdt.c optional dtrace_sdt | dtraceall compile-with "${CDDL_C}" cddl/dev/fbt/fbt.c optional dtrace_fbt | dtraceall compile-with "${FBT_C}" cddl/dev/systrace/systrace.c optional dtrace_systrace | dtraceall compile-with "${CDDL_C}" cddl/dev/prototype.c optional dtrace_prototype | dtraceall compile-with "${CDDL_C}" fs/nfsclient/nfs_clkdtrace.c optional dtnfscl nfscl | dtraceall nfscl compile-with "${CDDL_C}" compat/cloudabi/cloudabi_clock.c optional compat_cloudabi32 | compat_cloudabi64 compat/cloudabi/cloudabi_errno.c optional compat_cloudabi32 | compat_cloudabi64 compat/cloudabi/cloudabi_fd.c optional compat_cloudabi32 | compat_cloudabi64 compat/cloudabi/cloudabi_file.c optional compat_cloudabi32 | compat_cloudabi64 compat/cloudabi/cloudabi_futex.c optional compat_cloudabi32 | compat_cloudabi64 compat/cloudabi/cloudabi_mem.c optional compat_cloudabi32 | compat_cloudabi64 compat/cloudabi/cloudabi_proc.c optional compat_cloudabi32 | compat_cloudabi64 compat/cloudabi/cloudabi_random.c optional compat_cloudabi32 | compat_cloudabi64 compat/cloudabi/cloudabi_sock.c optional compat_cloudabi32 | compat_cloudabi64 compat/cloudabi/cloudabi_thread.c optional compat_cloudabi32 | compat_cloudabi64 compat/cloudabi/cloudabi_vdso.c optional compat_cloudabi32 | compat_cloudabi64 compat/cloudabi32/cloudabi32_fd.c optional compat_cloudabi32 compat/cloudabi32/cloudabi32_module.c optional compat_cloudabi32 compat/cloudabi32/cloudabi32_poll.c optional compat_cloudabi32 compat/cloudabi32/cloudabi32_sock.c optional compat_cloudabi32 compat/cloudabi32/cloudabi32_syscalls.c optional compat_cloudabi32 compat/cloudabi32/cloudabi32_sysent.c optional compat_cloudabi32 compat/cloudabi32/cloudabi32_thread.c optional compat_cloudabi32 compat/cloudabi64/cloudabi64_fd.c optional compat_cloudabi64 compat/cloudabi64/cloudabi64_module.c optional compat_cloudabi64 compat/cloudabi64/cloudabi64_poll.c optional compat_cloudabi64 compat/cloudabi64/cloudabi64_sock.c optional compat_cloudabi64 compat/cloudabi64/cloudabi64_syscalls.c optional compat_cloudabi64 compat/cloudabi64/cloudabi64_sysent.c optional compat_cloudabi64 compat/cloudabi64/cloudabi64_thread.c optional compat_cloudabi64 compat/freebsd32/freebsd32_capability.c optional compat_freebsd32 compat/freebsd32/freebsd32_ioctl.c optional compat_freebsd32 compat/freebsd32/freebsd32_misc.c optional compat_freebsd32 compat/freebsd32/freebsd32_syscalls.c optional compat_freebsd32 compat/freebsd32/freebsd32_sysent.c optional compat_freebsd32 contrib/ck/src/ck_array.c standard compile-with "${NORMAL_C} -I$S/contrib/ck/include" contrib/ck/src/ck_barrier_centralized.c standard compile-with "${NORMAL_C} -I$S/contrib/ck/include" contrib/ck/src/ck_barrier_combining.c standard compile-with "${NORMAL_C} -I$S/contrib/ck/include" contrib/ck/src/ck_barrier_dissemination.c standard compile-with "${NORMAL_C} -I$S/contrib/ck/include" contrib/ck/src/ck_barrier_mcs.c standard compile-with "${NORMAL_C} -I$S/contrib/ck/include" contrib/ck/src/ck_barrier_tournament.c standard compile-with "${NORMAL_C} -I$S/contrib/ck/include" contrib/ck/src/ck_epoch.c standard compile-with "${NORMAL_C} -I$S/contrib/ck/include" contrib/ck/src/ck_hp.c standard compile-with "${NORMAL_C} -I$S/contrib/ck/include" contrib/ck/src/ck_hs.c standard compile-with "${NORMAL_C} -I$S/contrib/ck/include" contrib/ck/src/ck_ht.c standard compile-with "${NORMAL_C} -I$S/contrib/ck/include" contrib/ck/src/ck_rhs.c standard compile-with "${NORMAL_C} -I$S/contrib/ck/include" contrib/dev/acpica/common/ahids.c optional acpi acpi_debug contrib/dev/acpica/common/ahuuids.c optional acpi acpi_debug contrib/dev/acpica/components/debugger/dbcmds.c optional acpi acpi_debug contrib/dev/acpica/components/debugger/dbconvert.c optional acpi acpi_debug contrib/dev/acpica/components/debugger/dbdisply.c optional acpi acpi_debug contrib/dev/acpica/components/debugger/dbexec.c optional acpi acpi_debug contrib/dev/acpica/components/debugger/dbhistry.c optional acpi acpi_debug contrib/dev/acpica/components/debugger/dbinput.c optional acpi acpi_debug contrib/dev/acpica/components/debugger/dbmethod.c optional acpi acpi_debug contrib/dev/acpica/components/debugger/dbnames.c optional acpi acpi_debug contrib/dev/acpica/components/debugger/dbobject.c optional acpi acpi_debug contrib/dev/acpica/components/debugger/dbstats.c optional acpi acpi_debug contrib/dev/acpica/components/debugger/dbtest.c optional acpi acpi_debug contrib/dev/acpica/components/debugger/dbutils.c optional acpi acpi_debug contrib/dev/acpica/components/debugger/dbxface.c optional acpi acpi_debug contrib/dev/acpica/components/disassembler/dmbuffer.c optional acpi acpi_debug contrib/dev/acpica/components/disassembler/dmcstyle.c optional acpi acpi_debug contrib/dev/acpica/components/disassembler/dmdeferred.c optional acpi acpi_debug contrib/dev/acpica/components/disassembler/dmnames.c optional acpi acpi_debug contrib/dev/acpica/components/disassembler/dmopcode.c optional acpi acpi_debug contrib/dev/acpica/components/disassembler/dmresrc.c optional acpi acpi_debug contrib/dev/acpica/components/disassembler/dmresrcl.c optional acpi acpi_debug contrib/dev/acpica/components/disassembler/dmresrcl2.c optional acpi acpi_debug contrib/dev/acpica/components/disassembler/dmresrcs.c optional acpi acpi_debug contrib/dev/acpica/components/disassembler/dmutils.c optional acpi acpi_debug contrib/dev/acpica/components/disassembler/dmwalk.c optional acpi acpi_debug contrib/dev/acpica/components/dispatcher/dsargs.c optional acpi contrib/dev/acpica/components/dispatcher/dscontrol.c optional acpi contrib/dev/acpica/components/dispatcher/dsdebug.c optional acpi contrib/dev/acpica/components/dispatcher/dsfield.c optional acpi contrib/dev/acpica/components/dispatcher/dsinit.c optional acpi contrib/dev/acpica/components/dispatcher/dsmethod.c optional acpi contrib/dev/acpica/components/dispatcher/dsmthdat.c optional acpi contrib/dev/acpica/components/dispatcher/dsobject.c optional acpi contrib/dev/acpica/components/dispatcher/dsopcode.c optional acpi contrib/dev/acpica/components/dispatcher/dspkginit.c optional acpi contrib/dev/acpica/components/dispatcher/dsutils.c optional acpi contrib/dev/acpica/components/dispatcher/dswexec.c optional acpi contrib/dev/acpica/components/dispatcher/dswload.c optional acpi contrib/dev/acpica/components/dispatcher/dswload2.c optional acpi contrib/dev/acpica/components/dispatcher/dswscope.c optional acpi contrib/dev/acpica/components/dispatcher/dswstate.c optional acpi contrib/dev/acpica/components/events/evevent.c optional acpi contrib/dev/acpica/components/events/evglock.c optional acpi contrib/dev/acpica/components/events/evgpe.c optional acpi contrib/dev/acpica/components/events/evgpeblk.c optional acpi contrib/dev/acpica/components/events/evgpeinit.c optional acpi contrib/dev/acpica/components/events/evgpeutil.c optional acpi contrib/dev/acpica/components/events/evhandler.c optional acpi contrib/dev/acpica/components/events/evmisc.c optional acpi contrib/dev/acpica/components/events/evregion.c optional acpi contrib/dev/acpica/components/events/evrgnini.c optional acpi contrib/dev/acpica/components/events/evsci.c optional acpi contrib/dev/acpica/components/events/evxface.c optional acpi contrib/dev/acpica/components/events/evxfevnt.c optional acpi contrib/dev/acpica/components/events/evxfgpe.c optional acpi contrib/dev/acpica/components/events/evxfregn.c optional acpi contrib/dev/acpica/components/executer/exconcat.c optional acpi contrib/dev/acpica/components/executer/exconfig.c optional acpi contrib/dev/acpica/components/executer/exconvrt.c optional acpi contrib/dev/acpica/components/executer/excreate.c optional acpi contrib/dev/acpica/components/executer/exdebug.c optional acpi contrib/dev/acpica/components/executer/exdump.c optional acpi contrib/dev/acpica/components/executer/exfield.c optional acpi contrib/dev/acpica/components/executer/exfldio.c optional acpi contrib/dev/acpica/components/executer/exmisc.c optional acpi contrib/dev/acpica/components/executer/exmutex.c optional acpi contrib/dev/acpica/components/executer/exnames.c optional acpi contrib/dev/acpica/components/executer/exoparg1.c optional acpi contrib/dev/acpica/components/executer/exoparg2.c optional acpi contrib/dev/acpica/components/executer/exoparg3.c optional acpi contrib/dev/acpica/components/executer/exoparg6.c optional acpi contrib/dev/acpica/components/executer/exprep.c optional acpi contrib/dev/acpica/components/executer/exregion.c optional acpi contrib/dev/acpica/components/executer/exresnte.c optional acpi contrib/dev/acpica/components/executer/exresolv.c optional acpi contrib/dev/acpica/components/executer/exresop.c optional acpi contrib/dev/acpica/components/executer/exserial.c optional acpi contrib/dev/acpica/components/executer/exstore.c optional acpi contrib/dev/acpica/components/executer/exstoren.c optional acpi contrib/dev/acpica/components/executer/exstorob.c optional acpi contrib/dev/acpica/components/executer/exsystem.c optional acpi contrib/dev/acpica/components/executer/extrace.c optional acpi contrib/dev/acpica/components/executer/exutils.c optional acpi contrib/dev/acpica/components/hardware/hwacpi.c optional acpi contrib/dev/acpica/components/hardware/hwesleep.c optional acpi contrib/dev/acpica/components/hardware/hwgpe.c optional acpi contrib/dev/acpica/components/hardware/hwpci.c optional acpi contrib/dev/acpica/components/hardware/hwregs.c optional acpi contrib/dev/acpica/components/hardware/hwsleep.c optional acpi contrib/dev/acpica/components/hardware/hwtimer.c optional acpi contrib/dev/acpica/components/hardware/hwvalid.c optional acpi contrib/dev/acpica/components/hardware/hwxface.c optional acpi contrib/dev/acpica/components/hardware/hwxfsleep.c optional acpi contrib/dev/acpica/components/namespace/nsaccess.c optional acpi contrib/dev/acpica/components/namespace/nsalloc.c optional acpi contrib/dev/acpica/components/namespace/nsarguments.c optional acpi contrib/dev/acpica/components/namespace/nsconvert.c optional acpi contrib/dev/acpica/components/namespace/nsdump.c optional acpi contrib/dev/acpica/components/namespace/nseval.c optional acpi contrib/dev/acpica/components/namespace/nsinit.c optional acpi contrib/dev/acpica/components/namespace/nsload.c optional acpi contrib/dev/acpica/components/namespace/nsnames.c optional acpi contrib/dev/acpica/components/namespace/nsobject.c optional acpi contrib/dev/acpica/components/namespace/nsparse.c optional acpi contrib/dev/acpica/components/namespace/nspredef.c optional acpi contrib/dev/acpica/components/namespace/nsprepkg.c optional acpi contrib/dev/acpica/components/namespace/nsrepair.c optional acpi contrib/dev/acpica/components/namespace/nsrepair2.c optional acpi contrib/dev/acpica/components/namespace/nssearch.c optional acpi contrib/dev/acpica/components/namespace/nsutils.c optional acpi contrib/dev/acpica/components/namespace/nswalk.c optional acpi contrib/dev/acpica/components/namespace/nsxfeval.c optional acpi contrib/dev/acpica/components/namespace/nsxfname.c optional acpi contrib/dev/acpica/components/namespace/nsxfobj.c optional acpi contrib/dev/acpica/components/parser/psargs.c optional acpi contrib/dev/acpica/components/parser/psloop.c optional acpi contrib/dev/acpica/components/parser/psobject.c optional acpi contrib/dev/acpica/components/parser/psopcode.c optional acpi contrib/dev/acpica/components/parser/psopinfo.c optional acpi contrib/dev/acpica/components/parser/psparse.c optional acpi contrib/dev/acpica/components/parser/psscope.c optional acpi contrib/dev/acpica/components/parser/pstree.c optional acpi contrib/dev/acpica/components/parser/psutils.c optional acpi contrib/dev/acpica/components/parser/pswalk.c optional acpi contrib/dev/acpica/components/parser/psxface.c optional acpi contrib/dev/acpica/components/resources/rsaddr.c optional acpi contrib/dev/acpica/components/resources/rscalc.c optional acpi contrib/dev/acpica/components/resources/rscreate.c optional acpi contrib/dev/acpica/components/resources/rsdump.c optional acpi acpi_debug contrib/dev/acpica/components/resources/rsdumpinfo.c optional acpi contrib/dev/acpica/components/resources/rsinfo.c optional acpi contrib/dev/acpica/components/resources/rsio.c optional acpi contrib/dev/acpica/components/resources/rsirq.c optional acpi contrib/dev/acpica/components/resources/rslist.c optional acpi contrib/dev/acpica/components/resources/rsmemory.c optional acpi contrib/dev/acpica/components/resources/rsmisc.c optional acpi contrib/dev/acpica/components/resources/rsserial.c optional acpi contrib/dev/acpica/components/resources/rsutils.c optional acpi contrib/dev/acpica/components/resources/rsxface.c optional acpi contrib/dev/acpica/components/tables/tbdata.c optional acpi contrib/dev/acpica/components/tables/tbfadt.c optional acpi contrib/dev/acpica/components/tables/tbfind.c optional acpi contrib/dev/acpica/components/tables/tbinstal.c optional acpi contrib/dev/acpica/components/tables/tbprint.c optional acpi contrib/dev/acpica/components/tables/tbutils.c optional acpi contrib/dev/acpica/components/tables/tbxface.c optional acpi contrib/dev/acpica/components/tables/tbxfload.c optional acpi contrib/dev/acpica/components/tables/tbxfroot.c optional acpi contrib/dev/acpica/components/utilities/utaddress.c optional acpi contrib/dev/acpica/components/utilities/utalloc.c optional acpi contrib/dev/acpica/components/utilities/utascii.c optional acpi contrib/dev/acpica/components/utilities/utbuffer.c optional acpi contrib/dev/acpica/components/utilities/utcache.c optional acpi contrib/dev/acpica/components/utilities/utcopy.c optional acpi contrib/dev/acpica/components/utilities/utdebug.c optional acpi contrib/dev/acpica/components/utilities/utdecode.c optional acpi contrib/dev/acpica/components/utilities/utdelete.c optional acpi contrib/dev/acpica/components/utilities/uterror.c optional acpi contrib/dev/acpica/components/utilities/uteval.c optional acpi contrib/dev/acpica/components/utilities/utexcep.c optional acpi contrib/dev/acpica/components/utilities/utglobal.c optional acpi contrib/dev/acpica/components/utilities/uthex.c optional acpi contrib/dev/acpica/components/utilities/utids.c optional acpi contrib/dev/acpica/components/utilities/utinit.c optional acpi contrib/dev/acpica/components/utilities/utlock.c optional acpi contrib/dev/acpica/components/utilities/utmath.c optional acpi contrib/dev/acpica/components/utilities/utmisc.c optional acpi contrib/dev/acpica/components/utilities/utmutex.c optional acpi contrib/dev/acpica/components/utilities/utnonansi.c optional acpi contrib/dev/acpica/components/utilities/utobject.c optional acpi contrib/dev/acpica/components/utilities/utosi.c optional acpi contrib/dev/acpica/components/utilities/utownerid.c optional acpi contrib/dev/acpica/components/utilities/utpredef.c optional acpi contrib/dev/acpica/components/utilities/utresdecode.c optional acpi acpi_debug contrib/dev/acpica/components/utilities/utresrc.c optional acpi contrib/dev/acpica/components/utilities/utstate.c optional acpi contrib/dev/acpica/components/utilities/utstring.c optional acpi contrib/dev/acpica/components/utilities/utstrsuppt.c optional acpi contrib/dev/acpica/components/utilities/utstrtoul64.c optional acpi contrib/dev/acpica/components/utilities/utuuid.c optional acpi acpi_debug contrib/dev/acpica/components/utilities/utxface.c optional acpi contrib/dev/acpica/components/utilities/utxferror.c optional acpi contrib/dev/acpica/components/utilities/utxfinit.c optional acpi contrib/dev/acpica/os_specific/service_layers/osgendbg.c optional acpi acpi_debug contrib/ipfilter/netinet/fil.c optional ipfilter inet \ compile-with "${NORMAL_C} ${NO_WSELF_ASSIGN} -Wno-unused -I$S/contrib/ipfilter" contrib/ipfilter/netinet/ip_auth.c optional ipfilter inet \ compile-with "${NORMAL_C} -Wno-unused -I$S/contrib/ipfilter" contrib/ipfilter/netinet/ip_fil_freebsd.c optional ipfilter inet \ compile-with "${NORMAL_C} -Wno-unused -I$S/contrib/ipfilter" contrib/ipfilter/netinet/ip_frag.c optional ipfilter inet \ compile-with "${NORMAL_C} -Wno-unused -I$S/contrib/ipfilter" contrib/ipfilter/netinet/ip_log.c optional ipfilter inet \ compile-with "${NORMAL_C} -I$S/contrib/ipfilter" contrib/ipfilter/netinet/ip_nat.c optional ipfilter inet \ compile-with "${NORMAL_C} -Wno-unused -I$S/contrib/ipfilter" contrib/ipfilter/netinet/ip_proxy.c optional ipfilter inet \ compile-with "${NORMAL_C} ${NO_WSELF_ASSIGN} -Wno-unused -I$S/contrib/ipfilter" contrib/ipfilter/netinet/ip_state.c optional ipfilter inet \ compile-with "${NORMAL_C} -Wno-unused -I$S/contrib/ipfilter" contrib/ipfilter/netinet/ip_lookup.c optional ipfilter inet \ compile-with "${NORMAL_C} ${NO_WSELF_ASSIGN} -Wno-unused -Wno-error -I$S/contrib/ipfilter" contrib/ipfilter/netinet/ip_pool.c optional ipfilter inet \ compile-with "${NORMAL_C} -Wno-unused -I$S/contrib/ipfilter" contrib/ipfilter/netinet/ip_htable.c optional ipfilter inet \ compile-with "${NORMAL_C} -Wno-unused -I$S/contrib/ipfilter ${NO_WTAUTOLOGICAL_POINTER_COMPARE}" contrib/ipfilter/netinet/ip_sync.c optional ipfilter inet \ compile-with "${NORMAL_C} -Wno-unused -I$S/contrib/ipfilter" contrib/ipfilter/netinet/mlfk_ipl.c optional ipfilter inet \ compile-with "${NORMAL_C} -I$S/contrib/ipfilter" contrib/ipfilter/netinet/ip_nat6.c optional ipfilter inet \ compile-with "${NORMAL_C} -Wno-unused -I$S/contrib/ipfilter" contrib/ipfilter/netinet/ip_rules.c optional ipfilter inet \ compile-with "${NORMAL_C} -I$S/contrib/ipfilter" contrib/ipfilter/netinet/ip_scan.c optional ipfilter inet \ compile-with "${NORMAL_C} -Wno-unused -I$S/contrib/ipfilter" contrib/ipfilter/netinet/ip_dstlist.c optional ipfilter inet \ compile-with "${NORMAL_C} -Wno-unused -I$S/contrib/ipfilter" contrib/ipfilter/netinet/radix_ipf.c optional ipfilter inet \ compile-with "${NORMAL_C} -I$S/contrib/ipfilter" contrib/libfdt/fdt.c optional fdt contrib/libfdt/fdt_ro.c optional fdt contrib/libfdt/fdt_rw.c optional fdt contrib/libfdt/fdt_strerror.c optional fdt contrib/libfdt/fdt_sw.c optional fdt contrib/libfdt/fdt_wip.c optional fdt contrib/libnv/cnvlist.c standard contrib/libnv/dnvlist.c standard contrib/libnv/nvlist.c standard contrib/libnv/nvpair.c standard contrib/ngatm/netnatm/api/cc_conn.c optional ngatm_ccatm \ compile-with "${NORMAL_C_NOWERROR} -I$S/contrib/ngatm" contrib/ngatm/netnatm/api/cc_data.c optional ngatm_ccatm \ compile-with "${NORMAL_C} -I$S/contrib/ngatm" contrib/ngatm/netnatm/api/cc_dump.c optional ngatm_ccatm \ compile-with "${NORMAL_C} -I$S/contrib/ngatm" contrib/ngatm/netnatm/api/cc_port.c optional ngatm_ccatm \ compile-with "${NORMAL_C} -I$S/contrib/ngatm" contrib/ngatm/netnatm/api/cc_sig.c optional ngatm_ccatm \ compile-with "${NORMAL_C} -I$S/contrib/ngatm" contrib/ngatm/netnatm/api/cc_user.c optional ngatm_ccatm \ compile-with "${NORMAL_C} -I$S/contrib/ngatm" contrib/ngatm/netnatm/api/unisap.c optional ngatm_ccatm \ compile-with "${NORMAL_C} -I$S/contrib/ngatm" contrib/ngatm/netnatm/misc/straddr.c optional ngatm_atmbase \ compile-with "${NORMAL_C} -I$S/contrib/ngatm" contrib/ngatm/netnatm/misc/unimsg_common.c optional ngatm_atmbase \ compile-with "${NORMAL_C} -I$S/contrib/ngatm" contrib/ngatm/netnatm/msg/traffic.c optional ngatm_atmbase \ compile-with "${NORMAL_C} -I$S/contrib/ngatm" contrib/ngatm/netnatm/msg/uni_ie.c optional ngatm_atmbase \ compile-with "${NORMAL_C} -I$S/contrib/ngatm" contrib/ngatm/netnatm/msg/uni_msg.c optional ngatm_atmbase \ compile-with "${NORMAL_C} -I$S/contrib/ngatm" contrib/ngatm/netnatm/saal/saal_sscfu.c optional ngatm_sscfu \ compile-with "${NORMAL_C} -I$S/contrib/ngatm" contrib/ngatm/netnatm/saal/saal_sscop.c optional ngatm_sscop \ compile-with "${NORMAL_C} -I$S/contrib/ngatm" contrib/ngatm/netnatm/sig/sig_call.c optional ngatm_uni \ compile-with "${NORMAL_C} -I$S/contrib/ngatm" contrib/ngatm/netnatm/sig/sig_coord.c optional ngatm_uni \ compile-with "${NORMAL_C} -I$S/contrib/ngatm" contrib/ngatm/netnatm/sig/sig_party.c optional ngatm_uni \ compile-with "${NORMAL_C} -I$S/contrib/ngatm" contrib/ngatm/netnatm/sig/sig_print.c optional ngatm_uni \ compile-with "${NORMAL_C} -I$S/contrib/ngatm" contrib/ngatm/netnatm/sig/sig_reset.c optional ngatm_uni \ compile-with "${NORMAL_C} -I$S/contrib/ngatm" contrib/ngatm/netnatm/sig/sig_uni.c optional ngatm_uni \ compile-with "${NORMAL_C} -I$S/contrib/ngatm" contrib/ngatm/netnatm/sig/sig_unimsgcpy.c optional ngatm_uni \ compile-with "${NORMAL_C} -I$S/contrib/ngatm" contrib/ngatm/netnatm/sig/sig_verify.c optional ngatm_uni \ compile-with "${NORMAL_C} -I$S/contrib/ngatm" # xz dev/xz/xz_mod.c optional xz \ compile-with "${NORMAL_C} -I$S/contrib/xz-embedded/freebsd/ -I$S/contrib/xz-embedded/linux/lib/xz/ -I$S/contrib/xz-embedded/linux/include/linux/" contrib/xz-embedded/linux/lib/xz/xz_crc32.c optional xz \ compile-with "${NORMAL_C} -I$S/contrib/xz-embedded/freebsd/ -I$S/contrib/xz-embedded/linux/lib/xz/ -I$S/contrib/xz-embedded/linux/include/linux/" contrib/xz-embedded/linux/lib/xz/xz_dec_bcj.c optional xz \ compile-with "${NORMAL_C} -I$S/contrib/xz-embedded/freebsd/ -I$S/contrib/xz-embedded/linux/lib/xz/ -I$S/contrib/xz-embedded/linux/include/linux/" contrib/xz-embedded/linux/lib/xz/xz_dec_lzma2.c optional xz \ compile-with "${NORMAL_C} -I$S/contrib/xz-embedded/freebsd/ -I$S/contrib/xz-embedded/linux/lib/xz/ -I$S/contrib/xz-embedded/linux/include/linux/" contrib/xz-embedded/linux/lib/xz/xz_dec_stream.c optional xz \ compile-with "${NORMAL_C} -I$S/contrib/xz-embedded/freebsd/ -I$S/contrib/xz-embedded/linux/lib/xz/ -I$S/contrib/xz-embedded/linux/include/linux/" # Zstd contrib/zstd/lib/freebsd/zstd_kmalloc.c optional zstdio compile-with ${ZSTD_C} contrib/zstd/lib/common/zstd_common.c optional zstdio compile-with ${ZSTD_C} contrib/zstd/lib/common/fse_decompress.c optional zstdio compile-with ${ZSTD_C} contrib/zstd/lib/common/entropy_common.c optional zstdio compile-with ${ZSTD_C} contrib/zstd/lib/common/error_private.c optional zstdio compile-with ${ZSTD_C} contrib/zstd/lib/common/xxhash.c optional zstdio compile-with ${ZSTD_C} contrib/zstd/lib/compress/zstd_compress.c optional zstdio compile-with ${ZSTD_C} contrib/zstd/lib/compress/zstd_compress_literals.c optional zstdio compile-with ${ZSTD_C} contrib/zstd/lib/compress/zstd_compress_sequences.c optional zstdio compile-with ${ZSTD_C} contrib/zstd/lib/compress/fse_compress.c optional zstdio compile-with ${ZSTD_C} contrib/zstd/lib/compress/hist.c optional zstdio compile-with ${ZSTD_C} contrib/zstd/lib/compress/huf_compress.c optional zstdio compile-with ${ZSTD_C} contrib/zstd/lib/compress/zstd_double_fast.c optional zstdio compile-with ${ZSTD_C} contrib/zstd/lib/compress/zstd_fast.c optional zstdio compile-with ${ZSTD_C} contrib/zstd/lib/compress/zstd_lazy.c optional zstdio compile-with ${ZSTD_C} contrib/zstd/lib/compress/zstd_ldm.c optional zstdio compile-with ${ZSTD_C} contrib/zstd/lib/compress/zstd_opt.c optional zstdio compile-with ${ZSTD_C} contrib/zstd/lib/decompress/zstd_ddict.c optional zstdio compile-with ${ZSTD_C} contrib/zstd/lib/decompress/zstd_decompress.c optional zstdio compile-with ${ZSTD_C} # See comment in sys/conf/kern.pre.mk contrib/zstd/lib/decompress/zstd_decompress_block.c optional zstdio \ compile-with "${ZSTD_C} ${ZSTD_DECOMPRESS_BLOCK_FLAGS}" contrib/zstd/lib/decompress/huf_decompress.c optional zstdio compile-with ${ZSTD_C} # Blake 2 contrib/libb2/blake2b-ref.c optional crypto | ipsec | ipsec_support \ compile-with "${NORMAL_C} -I$S/crypto/blake2 -Wno-cast-qual -DSUFFIX=_ref -Wno-unused-function" contrib/libb2/blake2s-ref.c optional crypto | ipsec | ipsec_support \ compile-with "${NORMAL_C} -I$S/crypto/blake2 -Wno-cast-qual -DSUFFIX=_ref -Wno-unused-function" crypto/blake2/blake2-sw.c optional crypto | ipsec | ipsec_support \ compile-with "${NORMAL_C} -I$S/crypto/blake2 -Wno-cast-qual" crypto/blowfish/bf_ecb.c optional ipsec | ipsec_support crypto/blowfish/bf_skey.c optional crypto | ipsec | ipsec_support crypto/camellia/camellia.c optional crypto | ipsec | ipsec_support crypto/camellia/camellia-api.c optional crypto | ipsec | ipsec_support crypto/chacha20/chacha.c standard crypto/chacha20/chacha-sw.c optional crypto | ipsec | ipsec_support crypto/des/des_ecb.c optional crypto | ipsec | ipsec_support | netsmb crypto/des/des_setkey.c optional crypto | ipsec | ipsec_support | netsmb crypto/rc4/rc4.c optional netgraph_mppc_encryption | kgssapi crypto/rijndael/rijndael-alg-fst.c optional crypto | ekcd | geom_bde | \ ipsec | ipsec_support | !random_loadable | wlan_ccmp crypto/rijndael/rijndael-api-fst.c optional ekcd | geom_bde | !random_loadable crypto/rijndael/rijndael-api.c optional crypto | ipsec | ipsec_support | \ wlan_ccmp crypto/sha1.c optional carp | crypto | ether | ipsec | \ ipsec_support | netgraph_mppc_encryption | sctp crypto/sha2/sha256c.c optional crypto | ekcd | geom_bde | ipsec | \ ipsec_support | !random_loadable | sctp | zfs crypto/sha2/sha512c.c optional crypto | geom_bde | ipsec | \ ipsec_support | zfs crypto/skein/skein.c optional crypto | zfs crypto/skein/skein_block.c optional crypto | zfs crypto/siphash/siphash.c optional inet | inet6 crypto/siphash/siphash_test.c optional inet | inet6 ddb/db_access.c optional ddb ddb/db_break.c optional ddb ddb/db_capture.c optional ddb ddb/db_command.c optional ddb ddb/db_examine.c optional ddb ddb/db_expr.c optional ddb ddb/db_input.c optional ddb ddb/db_lex.c optional ddb ddb/db_main.c optional ddb ddb/db_output.c optional ddb ddb/db_print.c optional ddb ddb/db_ps.c optional ddb ddb/db_run.c optional ddb ddb/db_script.c optional ddb ddb/db_sym.c optional ddb ddb/db_thread.c optional ddb ddb/db_textdump.c optional ddb ddb/db_variables.c optional ddb ddb/db_watch.c optional ddb ddb/db_write_cmd.c optional ddb dev/aac/aac.c optional aac dev/aac/aac_cam.c optional aacp aac dev/aac/aac_debug.c optional aac dev/aac/aac_disk.c optional aac dev/aac/aac_linux.c optional aac compat_linux dev/aac/aac_pci.c optional aac pci dev/aacraid/aacraid.c optional aacraid dev/aacraid/aacraid_cam.c optional aacraid scbus dev/aacraid/aacraid_debug.c optional aacraid dev/aacraid/aacraid_linux.c optional aacraid compat_linux dev/aacraid/aacraid_pci.c optional aacraid pci dev/acpi_support/acpi_wmi.c optional acpi_wmi acpi dev/acpi_support/acpi_asus.c optional acpi_asus acpi dev/acpi_support/acpi_asus_wmi.c optional acpi_asus_wmi acpi dev/acpi_support/acpi_fujitsu.c optional acpi_fujitsu acpi dev/acpi_support/acpi_hp.c optional acpi_hp acpi dev/acpi_support/acpi_ibm.c optional acpi_ibm acpi dev/acpi_support/acpi_panasonic.c optional acpi_panasonic acpi dev/acpi_support/acpi_sony.c optional acpi_sony acpi dev/acpi_support/acpi_toshiba.c optional acpi_toshiba acpi dev/acpi_support/atk0110.c optional aibs acpi dev/acpica/Osd/OsdDebug.c optional acpi dev/acpica/Osd/OsdHardware.c optional acpi dev/acpica/Osd/OsdInterrupt.c optional acpi dev/acpica/Osd/OsdMemory.c optional acpi dev/acpica/Osd/OsdSchedule.c optional acpi dev/acpica/Osd/OsdStream.c optional acpi dev/acpica/Osd/OsdSynch.c optional acpi dev/acpica/Osd/OsdTable.c optional acpi dev/acpica/acpi.c optional acpi dev/acpica/acpi_acad.c optional acpi dev/acpica/acpi_battery.c optional acpi dev/acpica/acpi_button.c optional acpi dev/acpica/acpi_cmbat.c optional acpi dev/acpica/acpi_cpu.c optional acpi dev/acpica/acpi_ec.c optional acpi dev/acpica/acpi_isab.c optional acpi isa dev/acpica/acpi_lid.c optional acpi dev/acpica/acpi_package.c optional acpi dev/acpica/acpi_perf.c optional acpi dev/acpica/acpi_powerres.c optional acpi dev/acpica/acpi_quirk.c optional acpi dev/acpica/acpi_resource.c optional acpi dev/acpica/acpi_container.c optional acpi dev/acpica/acpi_smbat.c optional acpi dev/acpica/acpi_thermal.c optional acpi dev/acpica/acpi_throttle.c optional acpi dev/acpica/acpi_video.c optional acpi_video acpi dev/acpica/acpi_dock.c optional acpi_dock acpi dev/adlink/adlink.c optional adlink dev/ae/if_ae.c optional ae pci dev/age/if_age.c optional age pci dev/agp/agp.c optional agp pci dev/agp/agp_if.m optional agp pci dev/ahci/ahci.c optional ahci dev/ahci/ahciem.c optional ahci dev/ahci/ahci_pci.c optional ahci pci dev/aic7xxx/ahc_isa.c optional ahc isa dev/aic7xxx/ahc_pci.c optional ahc pci \ compile-with "${NORMAL_C} ${NO_WCONSTANT_CONVERSION}" dev/aic7xxx/ahd_pci.c optional ahd pci \ compile-with "${NORMAL_C} ${NO_WCONSTANT_CONVERSION}" dev/aic7xxx/aic7770.c optional ahc dev/aic7xxx/aic79xx.c optional ahd pci dev/aic7xxx/aic79xx_osm.c optional ahd pci dev/aic7xxx/aic79xx_pci.c optional ahd pci dev/aic7xxx/aic79xx_reg_print.c optional ahd pci ahd_reg_pretty_print dev/aic7xxx/aic7xxx.c optional ahc dev/aic7xxx/aic7xxx_93cx6.c optional ahc dev/aic7xxx/aic7xxx_osm.c optional ahc dev/aic7xxx/aic7xxx_pci.c optional ahc pci dev/aic7xxx/aic7xxx_reg_print.c optional ahc ahc_reg_pretty_print dev/al_eth/al_eth.c optional al_eth fdt \ no-depend \ compile-with "${CC} -c -o ${.TARGET} ${CFLAGS} -I$S/contrib/alpine-hal -I$S/contrib/alpine-hal/eth ${PROF} ${.IMPSRC}" dev/al_eth/al_init_eth_lm.c optional al_eth fdt \ no-depend \ compile-with "${CC} -c -o ${.TARGET} ${CFLAGS} -I$S/contrib/alpine-hal -I$S/contrib/alpine-hal/eth ${PROF} ${.IMPSRC}" dev/al_eth/al_init_eth_kr.c optional al_eth fdt \ no-depend \ compile-with "${CC} -c -o ${.TARGET} ${CFLAGS} -I$S/contrib/alpine-hal -I$S/contrib/alpine-hal/eth ${PROF} ${.IMPSRC}" contrib/alpine-hal/al_hal_iofic.c optional al_iofic \ no-depend \ compile-with "${CC} -c -o ${.TARGET} ${CFLAGS} -I$S/contrib/alpine-hal -I$S/contrib/alpine-hal/eth ${PROF} ${.IMPSRC}" contrib/alpine-hal/al_hal_serdes_25g.c optional al_serdes \ no-depend \ compile-with "${CC} -c -o ${.TARGET} ${CFLAGS} -I$S/contrib/alpine-hal -I$S/contrib/alpine-hal/eth ${PROF} ${.IMPSRC}" contrib/alpine-hal/al_hal_serdes_hssp.c optional al_serdes \ no-depend \ compile-with "${CC} -c -o ${.TARGET} ${CFLAGS} -I$S/contrib/alpine-hal -I$S/contrib/alpine-hal/eth ${PROF} ${.IMPSRC}" contrib/alpine-hal/al_hal_udma_config.c optional al_udma \ no-depend \ compile-with "${CC} -c -o ${.TARGET} ${CFLAGS} -I$S/contrib/alpine-hal -I$S/contrib/alpine-hal/eth ${PROF} ${.IMPSRC}" contrib/alpine-hal/al_hal_udma_debug.c optional al_udma \ no-depend \ compile-with "${CC} -c -o ${.TARGET} ${CFLAGS} -I$S/contrib/alpine-hal -I$S/contrib/alpine-hal/eth ${PROF} ${.IMPSRC}" contrib/alpine-hal/al_hal_udma_iofic.c optional al_udma \ no-depend \ compile-with "${CC} -c -o ${.TARGET} ${CFLAGS} -I$S/contrib/alpine-hal -I$S/contrib/alpine-hal/eth ${PROF} ${.IMPSRC}" contrib/alpine-hal/al_hal_udma_main.c optional al_udma \ no-depend \ compile-with "${CC} -c -o ${.TARGET} ${CFLAGS} -I$S/contrib/alpine-hal -I$S/contrib/alpine-hal/eth ${PROF} ${.IMPSRC}" contrib/alpine-hal/al_serdes.c optional al_serdes \ no-depend \ compile-with "${CC} -c -o ${.TARGET} ${CFLAGS} -I$S/contrib/alpine-hal -I$S/contrib/alpine-hal/eth ${PROF} ${.IMPSRC}" contrib/alpine-hal/eth/al_hal_eth_kr.c optional al_eth \ no-depend \ compile-with "${CC} -c -o ${.TARGET} ${CFLAGS} -I$S/contrib/alpine-hal -I$S/contrib/alpine-hal/eth ${PROF} ${.IMPSRC}" contrib/alpine-hal/eth/al_hal_eth_main.c optional al_eth \ no-depend \ compile-with "${CC} -c -o ${.TARGET} ${CFLAGS} -I$S/contrib/alpine-hal -I$S/contrib/alpine-hal/eth ${PROF} ${.IMPSRC}" dev/alc/if_alc.c optional alc pci dev/ale/if_ale.c optional ale pci dev/alpm/alpm.c optional alpm pci dev/altera/avgen/altera_avgen.c optional altera_avgen dev/altera/avgen/altera_avgen_fdt.c optional altera_avgen fdt dev/altera/avgen/altera_avgen_nexus.c optional altera_avgen dev/altera/msgdma/msgdma.c optional altera_msgdma xdma dev/altera/sdcard/altera_sdcard.c optional altera_sdcard dev/altera/sdcard/altera_sdcard_disk.c optional altera_sdcard dev/altera/sdcard/altera_sdcard_io.c optional altera_sdcard dev/altera/sdcard/altera_sdcard_fdt.c optional altera_sdcard fdt dev/altera/sdcard/altera_sdcard_nexus.c optional altera_sdcard dev/altera/softdma/softdma.c optional altera_softdma xdma fdt dev/altera/pio/pio.c optional altera_pio dev/altera/pio/pio_if.m optional altera_pio dev/amdpm/amdpm.c optional amdpm pci | nfpm pci dev/amdsmb/amdsmb.c optional amdsmb pci dev/amr/amr.c optional amr dev/amr/amr_cam.c optional amrp amr dev/amr/amr_disk.c optional amr dev/amr/amr_linux.c optional amr compat_linux dev/amr/amr_pci.c optional amr pci dev/an/if_an.c optional an dev/an/if_an_isa.c optional an isa dev/an/if_an_pccard.c optional an pccard dev/an/if_an_pci.c optional an pci # dev/ata/ata_if.m optional ata | atacore dev/ata/ata-all.c optional ata | atacore dev/ata/ata-dma.c optional ata | atacore dev/ata/ata-lowlevel.c optional ata | atacore dev/ata/ata-sata.c optional ata | atacore dev/ata/ata-card.c optional ata pccard | atapccard dev/ata/ata-isa.c optional ata isa | ataisa dev/ata/ata-pci.c optional ata pci | atapci dev/ata/chipsets/ata-acard.c optional ata pci | ataacard dev/ata/chipsets/ata-acerlabs.c optional ata pci | ataacerlabs dev/ata/chipsets/ata-amd.c optional ata pci | ataamd dev/ata/chipsets/ata-ati.c optional ata pci | ataati dev/ata/chipsets/ata-cenatek.c optional ata pci | atacenatek dev/ata/chipsets/ata-cypress.c optional ata pci | atacypress dev/ata/chipsets/ata-cyrix.c optional ata pci | atacyrix dev/ata/chipsets/ata-highpoint.c optional ata pci | atahighpoint dev/ata/chipsets/ata-intel.c optional ata pci | ataintel dev/ata/chipsets/ata-ite.c optional ata pci | ataite dev/ata/chipsets/ata-jmicron.c optional ata pci | atajmicron dev/ata/chipsets/ata-marvell.c optional ata pci | atamarvell dev/ata/chipsets/ata-micron.c optional ata pci | atamicron dev/ata/chipsets/ata-national.c optional ata pci | atanational dev/ata/chipsets/ata-netcell.c optional ata pci | atanetcell dev/ata/chipsets/ata-nvidia.c optional ata pci | atanvidia dev/ata/chipsets/ata-promise.c optional ata pci | atapromise dev/ata/chipsets/ata-serverworks.c optional ata pci | ataserverworks dev/ata/chipsets/ata-siliconimage.c optional ata pci | atasiliconimage | ataati dev/ata/chipsets/ata-sis.c optional ata pci | atasis dev/ata/chipsets/ata-via.c optional ata pci | atavia # dev/ath/if_ath_pci.c optional ath_pci pci \ compile-with "${NORMAL_C} -I$S/dev/ath" # dev/ath/if_ath_ahb.c optional ath_ahb \ compile-with "${NORMAL_C} -I$S/dev/ath" # dev/ath/if_ath.c optional ath \ compile-with "${NORMAL_C} -I$S/dev/ath" dev/ath/if_ath_alq.c optional ath \ compile-with "${NORMAL_C} -I$S/dev/ath" dev/ath/if_ath_beacon.c optional ath \ compile-with "${NORMAL_C} -I$S/dev/ath" dev/ath/if_ath_btcoex.c optional ath \ compile-with "${NORMAL_C} -I$S/dev/ath" dev/ath/if_ath_btcoex_mci.c optional ath \ compile-with "${NORMAL_C} -I$S/dev/ath" dev/ath/if_ath_debug.c optional ath \ compile-with "${NORMAL_C} -I$S/dev/ath" dev/ath/if_ath_descdma.c optional ath \ compile-with "${NORMAL_C} -I$S/dev/ath" dev/ath/if_ath_keycache.c optional ath \ compile-with "${NORMAL_C} -I$S/dev/ath" dev/ath/if_ath_ioctl.c optional ath \ compile-with "${NORMAL_C} -I$S/dev/ath" dev/ath/if_ath_led.c optional ath \ compile-with "${NORMAL_C} -I$S/dev/ath" dev/ath/if_ath_lna_div.c optional ath \ compile-with "${NORMAL_C} -I$S/dev/ath" dev/ath/if_ath_tx.c optional ath \ compile-with "${NORMAL_C} -I$S/dev/ath" dev/ath/if_ath_tx_edma.c optional ath \ compile-with "${NORMAL_C} -I$S/dev/ath" dev/ath/if_ath_tx_ht.c optional ath \ compile-with "${NORMAL_C} -I$S/dev/ath" dev/ath/if_ath_tdma.c optional ath \ compile-with "${NORMAL_C} -I$S/dev/ath" dev/ath/if_ath_sysctl.c optional ath \ compile-with "${NORMAL_C} -I$S/dev/ath" dev/ath/if_ath_rx.c optional ath \ compile-with "${NORMAL_C} -I$S/dev/ath" dev/ath/if_ath_rx_edma.c optional ath \ compile-with "${NORMAL_C} -I$S/dev/ath" dev/ath/if_ath_spectral.c optional ath \ compile-with "${NORMAL_C} -I$S/dev/ath" dev/ath/ah_osdep.c optional ath \ compile-with "${NORMAL_C} -I$S/dev/ath" # dev/ath/ath_hal/ah.c optional ath \ compile-with "${NORMAL_C} -I$S/dev/ath" dev/ath/ath_hal/ah_eeprom_v1.c optional ath_hal | ath_ar5210 \ compile-with "${NORMAL_C} -I$S/dev/ath" dev/ath/ath_hal/ah_eeprom_v3.c optional ath_hal | ath_ar5211 | ath_ar5212 \ compile-with "${NORMAL_C} -I$S/dev/ath" dev/ath/ath_hal/ah_eeprom_v14.c \ optional ath_hal | ath_ar5416 | ath_ar9160 | ath_ar9280 \ compile-with "${NORMAL_C} -I$S/dev/ath" dev/ath/ath_hal/ah_eeprom_v4k.c \ optional ath_hal | ath_ar9285 \ compile-with "${NORMAL_C} -I$S/dev/ath" dev/ath/ath_hal/ah_eeprom_9287.c \ optional ath_hal | ath_ar9287 \ compile-with "${NORMAL_C} -I$S/dev/ath" dev/ath/ath_hal/ah_regdomain.c optional ath \ compile-with "${NORMAL_C} ${NO_WSHIFT_COUNT_NEGATIVE} ${NO_WSHIFT_COUNT_OVERFLOW} -I$S/dev/ath" # ar5210 dev/ath/ath_hal/ar5210/ar5210_attach.c optional ath_hal | ath_ar5210 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5210/ar5210_beacon.c optional ath_hal | ath_ar5210 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5210/ar5210_interrupts.c optional ath_hal | ath_ar5210 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5210/ar5210_keycache.c optional ath_hal | ath_ar5210 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5210/ar5210_misc.c optional ath_hal | ath_ar5210 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5210/ar5210_phy.c optional ath_hal | ath_ar5210 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5210/ar5210_power.c optional ath_hal | ath_ar5210 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5210/ar5210_recv.c optional ath_hal | ath_ar5210 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5210/ar5210_reset.c optional ath_hal | ath_ar5210 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5210/ar5210_xmit.c optional ath_hal | ath_ar5210 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" # ar5211 dev/ath/ath_hal/ar5211/ar5211_attach.c optional ath_hal | ath_ar5211 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5211/ar5211_beacon.c optional ath_hal | ath_ar5211 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5211/ar5211_interrupts.c optional ath_hal | ath_ar5211 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5211/ar5211_keycache.c optional ath_hal | ath_ar5211 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5211/ar5211_misc.c optional ath_hal | ath_ar5211 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5211/ar5211_phy.c optional ath_hal | ath_ar5211 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5211/ar5211_power.c optional ath_hal | ath_ar5211 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5211/ar5211_recv.c optional ath_hal | ath_ar5211 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5211/ar5211_reset.c optional ath_hal | ath_ar5211 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5211/ar5211_xmit.c optional ath_hal | ath_ar5211 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" # ar5212 dev/ath/ath_hal/ar5212/ar5212_ani.c \ optional ath_hal | ath_ar5212 | ath_ar5416 | ath_ar9160 | ath_ar9280 | \ ath_ar9285 ath_ar9287 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5212/ar5212_attach.c \ optional ath_hal | ath_ar5212 | ath_ar5416 | ath_ar9160 | ath_ar9280 | \ ath_ar9285 ath_ar9287 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5212/ar5212_beacon.c \ optional ath_hal | ath_ar5212 | ath_ar5416 | ath_ar9160 | ath_ar9280 | \ ath_ar9285 ath_ar9287 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5212/ar5212_eeprom.c \ optional ath_hal | ath_ar5212 | ath_ar5416 | ath_ar9160 | ath_ar9280 | \ ath_ar9285 ath_ar9287 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5212/ar5212_gpio.c \ optional ath_hal | ath_ar5212 | ath_ar5416 | ath_ar9160 | ath_ar9280 | \ ath_ar9285 ath_ar9287 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5212/ar5212_interrupts.c \ optional ath_hal | ath_ar5212 | ath_ar5416 | ath_ar9160 | ath_ar9280 | \ ath_ar9285 ath_ar9287 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5212/ar5212_keycache.c \ optional ath_hal | ath_ar5212 | ath_ar5416 | ath_ar9160 | ath_ar9280 | \ ath_ar9285 ath_ar9287 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5212/ar5212_misc.c \ optional ath_hal | ath_ar5212 | ath_ar5416 | ath_ar9160 | ath_ar9280 | \ ath_ar9285 ath_ar9287 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5212/ar5212_phy.c \ optional ath_hal | ath_ar5212 | ath_ar5416 | ath_ar9160 | ath_ar9280 | \ ath_ar9285 ath_ar9287 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5212/ar5212_power.c \ optional ath_hal | ath_ar5212 | ath_ar5416 | ath_ar9160 | ath_ar9280 | \ ath_ar9285 ath_ar9287 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5212/ar5212_recv.c \ optional ath_hal | ath_ar5212 | ath_ar5416 | ath_ar9160 | ath_ar9280 | \ ath_ar9285 ath_ar9287 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5212/ar5212_reset.c \ optional ath_hal | ath_ar5212 | ath_ar5416 | ath_ar9160 | ath_ar9280 | \ ath_ar9285 ath_ar9287 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5212/ar5212_rfgain.c \ optional ath_hal | ath_ar5212 | ath_ar5416 | ath_ar9160 | ath_ar9280 | \ ath_ar9285 ath_ar9287 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5212/ar5212_xmit.c \ optional ath_hal | ath_ar5212 | ath_ar5416 | ath_ar9160 | ath_ar9280 | \ ath_ar9285 ath_ar9287 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" # ar5416 (depends on ar5212) dev/ath/ath_hal/ar5416/ar5416_ani.c \ optional ath_hal | ath_ar5416 | ath_ar9160 | ath_ar9280 | ath_ar9285 | \ ath_ar9287 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5416/ar5416_attach.c \ optional ath_hal | ath_ar5416 | ath_ar9160 | ath_ar9280 | ath_ar9285 | \ ath_ar9287 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5416/ar5416_beacon.c \ optional ath_hal | ath_ar5416 | ath_ar9160 | ath_ar9280 | ath_ar9285 | \ ath_ar9287 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5416/ar5416_btcoex.c \ optional ath_hal | ath_ar5416 | ath_ar9160 | ath_ar9280 | ath_ar9285 | \ ath_ar9287 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5416/ar5416_cal.c \ optional ath_hal | ath_ar5416 | ath_ar9160 | ath_ar9280 | ath_ar9285 | \ ath_ar9287 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5416/ar5416_cal_iq.c \ optional ath_hal | ath_ar5416 | ath_ar9160 | ath_ar9280 | ath_ar9285 | \ ath_ar9287 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5416/ar5416_cal_adcgain.c \ optional ath_hal | ath_ar5416 | ath_ar9160 | ath_ar9280 | ath_ar9285 | \ ath_ar9287 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5416/ar5416_cal_adcdc.c \ optional ath_hal | ath_ar5416 | ath_ar9160 | ath_ar9280 | ath_ar9285 | \ ath_ar9287 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5416/ar5416_eeprom.c \ optional ath_hal | ath_ar5416 | ath_ar9160 | ath_ar9280 | ath_ar9285 | \ ath_ar9287 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5416/ar5416_gpio.c \ optional ath_hal | ath_ar5416 | ath_ar9160 | ath_ar9280 | ath_ar9285 | \ ath_ar9287 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5416/ar5416_interrupts.c \ optional ath_hal | ath_ar5416 | ath_ar9160 | ath_ar9280 | ath_ar9285 | \ ath_ar9287 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5416/ar5416_keycache.c \ optional ath_hal | ath_ar5416 | ath_ar9160 | ath_ar9280 | ath_ar9285 | \ ath_ar9287 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5416/ar5416_misc.c \ optional ath_hal | ath_ar5416 | ath_ar9160 | ath_ar9280 | ath_ar9285 | \ ath_ar9287 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5416/ar5416_phy.c \ optional ath_hal | ath_ar5416 | ath_ar9160 | ath_ar9280 | ath_ar9285 | \ ath_ar9287 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5416/ar5416_power.c \ optional ath_hal | ath_ar5416 | ath_ar9160 | ath_ar9280 | ath_ar9285 | \ ath_ar9287 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5416/ar5416_radar.c \ optional ath_hal | ath_ar5416 | ath_ar9160 | ath_ar9280 | ath_ar9285 | \ ath_ar9287 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5416/ar5416_recv.c \ optional ath_hal | ath_ar5416 | ath_ar9160 | ath_ar9280 | ath_ar9285 | \ ath_ar9287 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5416/ar5416_reset.c \ optional ath_hal | ath_ar5416 | ath_ar9160 | ath_ar9280 | ath_ar9285 | \ ath_ar9287 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5416/ar5416_spectral.c \ optional ath_hal | ath_ar5416 | ath_ar9160 | ath_ar9280 | ath_ar9285 | \ ath_ar9287 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5416/ar5416_xmit.c \ optional ath_hal | ath_ar5416 | ath_ar9160 | ath_ar9280 | ath_ar9285 | \ ath_ar9287 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" # ar9130 (depends upon ar5416) - also requires AH_SUPPORT_AR9130 # # Since this is an embedded MAC SoC, there's no need to compile it into the # default HAL. dev/ath/ath_hal/ar9001/ar9130_attach.c optional ath_ar9130 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar9001/ar9130_phy.c optional ath_ar9130 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar9001/ar9130_eeprom.c optional ath_ar9130 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" # ar9160 (depends on ar5416) dev/ath/ath_hal/ar9001/ar9160_attach.c optional ath_hal | ath_ar9160 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" # ar9280 (depends on ar5416) dev/ath/ath_hal/ar9002/ar9280_attach.c optional ath_hal | ath_ar9280 | \ ath_ar9285 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar9002/ar9280_olc.c optional ath_hal | ath_ar9280 | \ ath_ar9285 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" # ar9285 (depends on ar5416 and ar9280) dev/ath/ath_hal/ar9002/ar9285_attach.c optional ath_hal | ath_ar9285 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar9002/ar9285_btcoex.c optional ath_hal | ath_ar9285 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar9002/ar9285_reset.c optional ath_hal | ath_ar9285 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar9002/ar9285_cal.c optional ath_hal | ath_ar9285 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar9002/ar9285_phy.c optional ath_hal | ath_ar9285 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar9002/ar9285_diversity.c optional ath_hal | ath_ar9285 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" # ar9287 (depends on ar5416) dev/ath/ath_hal/ar9002/ar9287_attach.c optional ath_hal | ath_ar9287 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar9002/ar9287_reset.c optional ath_hal | ath_ar9287 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar9002/ar9287_cal.c optional ath_hal | ath_ar9287 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar9002/ar9287_olc.c optional ath_hal | ath_ar9287 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" # ar9300 contrib/dev/ath/ath_hal/ar9300/ar9300_ani.c optional ath_hal | ath_ar9300 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal -I$S/contrib/dev/ath/ath_hal" contrib/dev/ath/ath_hal/ar9300/ar9300_attach.c optional ath_hal | ath_ar9300 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal -I$S/contrib/dev/ath/ath_hal" contrib/dev/ath/ath_hal/ar9300/ar9300_beacon.c optional ath_hal | ath_ar9300 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal -I$S/contrib/dev/ath/ath_hal" contrib/dev/ath/ath_hal/ar9300/ar9300_eeprom.c optional ath_hal | ath_ar9300 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal -I$S/contrib/dev/ath/ath_hal ${NO_WCONSTANT_CONVERSION}" contrib/dev/ath/ath_hal/ar9300/ar9300_freebsd.c optional ath_hal | ath_ar9300 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal -I$S/contrib/dev/ath/ath_hal" contrib/dev/ath/ath_hal/ar9300/ar9300_gpio.c optional ath_hal | ath_ar9300 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal -I$S/contrib/dev/ath/ath_hal" contrib/dev/ath/ath_hal/ar9300/ar9300_interrupts.c optional ath_hal | ath_ar9300 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal -I$S/contrib/dev/ath/ath_hal" contrib/dev/ath/ath_hal/ar9300/ar9300_keycache.c optional ath_hal | ath_ar9300 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal -I$S/contrib/dev/ath/ath_hal" contrib/dev/ath/ath_hal/ar9300/ar9300_mci.c optional ath_hal | ath_ar9300 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal -I$S/contrib/dev/ath/ath_hal" contrib/dev/ath/ath_hal/ar9300/ar9300_misc.c optional ath_hal | ath_ar9300 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal -I$S/contrib/dev/ath/ath_hal" contrib/dev/ath/ath_hal/ar9300/ar9300_paprd.c optional ath_hal | ath_ar9300 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal -I$S/contrib/dev/ath/ath_hal" contrib/dev/ath/ath_hal/ar9300/ar9300_phy.c optional ath_hal | ath_ar9300 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal -I$S/contrib/dev/ath/ath_hal" contrib/dev/ath/ath_hal/ar9300/ar9300_power.c optional ath_hal | ath_ar9300 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal -I$S/contrib/dev/ath/ath_hal" contrib/dev/ath/ath_hal/ar9300/ar9300_radar.c optional ath_hal | ath_ar9300 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal -I$S/contrib/dev/ath/ath_hal" contrib/dev/ath/ath_hal/ar9300/ar9300_radio.c optional ath_hal | ath_ar9300 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal -I$S/contrib/dev/ath/ath_hal" contrib/dev/ath/ath_hal/ar9300/ar9300_recv.c optional ath_hal | ath_ar9300 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal -I$S/contrib/dev/ath/ath_hal" contrib/dev/ath/ath_hal/ar9300/ar9300_recv_ds.c optional ath_hal | ath_ar9300 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal -I$S/contrib/dev/ath/ath_hal" contrib/dev/ath/ath_hal/ar9300/ar9300_reset.c optional ath_hal | ath_ar9300 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal -I$S/contrib/dev/ath/ath_hal ${NO_WSOMETIMES_UNINITIALIZED} -Wno-unused-function" contrib/dev/ath/ath_hal/ar9300/ar9300_stub.c optional ath_hal | ath_ar9300 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal -I$S/contrib/dev/ath/ath_hal" contrib/dev/ath/ath_hal/ar9300/ar9300_stub_funcs.c optional ath_hal | ath_ar9300 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal -I$S/contrib/dev/ath/ath_hal" contrib/dev/ath/ath_hal/ar9300/ar9300_spectral.c optional ath_hal | ath_ar9300 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal -I$S/contrib/dev/ath/ath_hal" contrib/dev/ath/ath_hal/ar9300/ar9300_timer.c optional ath_hal | ath_ar9300 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal -I$S/contrib/dev/ath/ath_hal" contrib/dev/ath/ath_hal/ar9300/ar9300_xmit.c optional ath_hal | ath_ar9300 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal -I$S/contrib/dev/ath/ath_hal" contrib/dev/ath/ath_hal/ar9300/ar9300_xmit_ds.c optional ath_hal | ath_ar9300 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal -I$S/contrib/dev/ath/ath_hal" # rf backends dev/ath/ath_hal/ar5212/ar2316.c optional ath_rf2316 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5212/ar2317.c optional ath_rf2317 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5212/ar2413.c optional ath_hal | ath_rf2413 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5212/ar2425.c optional ath_hal | ath_rf2425 | ath_rf2417 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5212/ar5111.c optional ath_hal | ath_rf5111 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5212/ar5112.c optional ath_hal | ath_rf5112 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5212/ar5413.c optional ath_hal | ath_rf5413 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar5416/ar2133.c optional ath_hal | ath_ar5416 | \ ath_ar9130 | ath_ar9160 | ath_ar9280 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar9002/ar9280.c optional ath_hal | ath_ar9280 | ath_ar9285 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar9002/ar9285.c optional ath_hal | ath_ar9285 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" dev/ath/ath_hal/ar9002/ar9287.c optional ath_hal | ath_ar9287 \ compile-with "${NORMAL_C} -I$S/dev/ath -I$S/dev/ath/ath_hal" # ath rate control algorithms dev/ath/ath_rate/amrr/amrr.c optional ath_rate_amrr \ compile-with "${NORMAL_C} -I$S/dev/ath" dev/ath/ath_rate/onoe/onoe.c optional ath_rate_onoe \ compile-with "${NORMAL_C} -I$S/dev/ath" dev/ath/ath_rate/sample/sample.c optional ath_rate_sample \ compile-with "${NORMAL_C} -I$S/dev/ath" # ath DFS modules dev/ath/ath_dfs/null/dfs_null.c optional ath \ compile-with "${NORMAL_C} -I$S/dev/ath" # dev/bce/if_bce.c optional bce dev/bfe/if_bfe.c optional bfe dev/bge/if_bge.c optional bge dev/bhnd/bhnd.c optional bhnd dev/bhnd/bhnd_erom.c optional bhnd dev/bhnd/bhnd_erom_if.m optional bhnd dev/bhnd/bhnd_subr.c optional bhnd dev/bhnd/bhnd_bus_if.m optional bhnd dev/bhnd/bhndb/bhnd_bhndb.c optional bhndb bhnd dev/bhnd/bhndb/bhndb.c optional bhndb bhnd dev/bhnd/bhndb/bhndb_bus_if.m optional bhndb bhnd dev/bhnd/bhndb/bhndb_hwdata.c optional bhndb bhnd dev/bhnd/bhndb/bhndb_if.m optional bhndb bhnd dev/bhnd/bhndb/bhndb_pci.c optional bhndb_pci bhndb bhnd pci dev/bhnd/bhndb/bhndb_pci_hwdata.c optional bhndb_pci bhndb bhnd pci dev/bhnd/bhndb/bhndb_pci_sprom.c optional bhndb_pci bhndb bhnd pci dev/bhnd/bhndb/bhndb_subr.c optional bhndb bhnd dev/bhnd/bcma/bcma.c optional bcma bhnd dev/bhnd/bcma/bcma_bhndb.c optional bcma bhnd bhndb dev/bhnd/bcma/bcma_erom.c optional bcma bhnd dev/bhnd/bcma/bcma_subr.c optional bcma bhnd dev/bhnd/cores/chipc/bhnd_chipc_if.m optional bhnd dev/bhnd/cores/chipc/bhnd_sprom_chipc.c optional bhnd dev/bhnd/cores/chipc/bhnd_pmu_chipc.c optional bhnd dev/bhnd/cores/chipc/chipc.c optional bhnd dev/bhnd/cores/chipc/chipc_cfi.c optional bhnd cfi dev/bhnd/cores/chipc/chipc_gpio.c optional bhnd gpio dev/bhnd/cores/chipc/chipc_slicer.c optional bhnd cfi | bhnd spibus dev/bhnd/cores/chipc/chipc_spi.c optional bhnd spibus dev/bhnd/cores/chipc/chipc_subr.c optional bhnd dev/bhnd/cores/chipc/pwrctl/bhnd_pwrctl.c optional bhnd dev/bhnd/cores/chipc/pwrctl/bhnd_pwrctl_if.m optional bhnd dev/bhnd/cores/chipc/pwrctl/bhnd_pwrctl_hostb_if.m optional bhnd dev/bhnd/cores/chipc/pwrctl/bhnd_pwrctl_subr.c optional bhnd dev/bhnd/cores/pci/bhnd_pci.c optional bhnd pci dev/bhnd/cores/pci/bhnd_pci_hostb.c optional bhndb bhnd pci dev/bhnd/cores/pci/bhnd_pcib.c optional bhnd_pcib bhnd pci dev/bhnd/cores/pcie2/bhnd_pcie2.c optional bhnd pci dev/bhnd/cores/pcie2/bhnd_pcie2_hostb.c optional bhndb bhnd pci dev/bhnd/cores/pcie2/bhnd_pcie2b.c optional bhnd_pcie2b bhnd pci dev/bhnd/cores/pmu/bhnd_pmu.c optional bhnd dev/bhnd/cores/pmu/bhnd_pmu_core.c optional bhnd dev/bhnd/cores/pmu/bhnd_pmu_if.m optional bhnd dev/bhnd/cores/pmu/bhnd_pmu_subr.c optional bhnd dev/bhnd/nvram/bhnd_nvram_data.c optional bhnd dev/bhnd/nvram/bhnd_nvram_data_bcm.c optional bhnd dev/bhnd/nvram/bhnd_nvram_data_bcmraw.c optional bhnd dev/bhnd/nvram/bhnd_nvram_data_btxt.c optional bhnd dev/bhnd/nvram/bhnd_nvram_data_sprom.c optional bhnd dev/bhnd/nvram/bhnd_nvram_data_sprom_subr.c optional bhnd dev/bhnd/nvram/bhnd_nvram_data_tlv.c optional bhnd dev/bhnd/nvram/bhnd_nvram_if.m optional bhnd dev/bhnd/nvram/bhnd_nvram_io.c optional bhnd dev/bhnd/nvram/bhnd_nvram_iobuf.c optional bhnd dev/bhnd/nvram/bhnd_nvram_ioptr.c optional bhnd dev/bhnd/nvram/bhnd_nvram_iores.c optional bhnd dev/bhnd/nvram/bhnd_nvram_plist.c optional bhnd dev/bhnd/nvram/bhnd_nvram_store.c optional bhnd dev/bhnd/nvram/bhnd_nvram_store_subr.c optional bhnd dev/bhnd/nvram/bhnd_nvram_subr.c optional bhnd dev/bhnd/nvram/bhnd_nvram_value.c optional bhnd dev/bhnd/nvram/bhnd_nvram_value_fmts.c optional bhnd dev/bhnd/nvram/bhnd_nvram_value_prf.c optional bhnd dev/bhnd/nvram/bhnd_nvram_value_subr.c optional bhnd dev/bhnd/nvram/bhnd_sprom.c optional bhnd dev/bhnd/siba/siba.c optional siba bhnd dev/bhnd/siba/siba_bhndb.c optional siba bhnd bhndb dev/bhnd/siba/siba_erom.c optional siba bhnd dev/bhnd/siba/siba_subr.c optional siba bhnd # dev/bnxt/bnxt_hwrm.c optional bnxt iflib pci dev/bnxt/bnxt_sysctl.c optional bnxt iflib pci dev/bnxt/bnxt_txrx.c optional bnxt iflib pci dev/bnxt/if_bnxt.c optional bnxt iflib pci dev/bwi/bwimac.c optional bwi dev/bwi/bwiphy.c optional bwi dev/bwi/bwirf.c optional bwi dev/bwi/if_bwi.c optional bwi dev/bwi/if_bwi_pci.c optional bwi pci dev/bwn/if_bwn.c optional bwn bhnd dev/bwn/if_bwn_pci.c optional bwn pci bhnd bhndb bhndb_pci dev/bwn/if_bwn_phy_common.c optional bwn bhnd dev/bwn/if_bwn_phy_g.c optional bwn bhnd dev/bwn/if_bwn_phy_lp.c optional bwn bhnd dev/bwn/if_bwn_phy_n.c optional bwn bhnd dev/bwn/if_bwn_util.c optional bwn bhnd dev/cardbus/cardbus.c optional cardbus dev/cardbus/cardbus_cis.c optional cardbus dev/cardbus/cardbus_device.c optional cardbus dev/cas/if_cas.c optional cas dev/cfi/cfi_bus_fdt.c optional cfi fdt dev/cfi/cfi_bus_nexus.c optional cfi dev/cfi/cfi_core.c optional cfi dev/cfi/cfi_dev.c optional cfi dev/cfi/cfi_disk.c optional cfid dev/chromebook_platform/chromebook_platform.c optional chromebook_platform dev/ciss/ciss.c optional ciss dev/cmx/cmx.c optional cmx dev/cmx/cmx_pccard.c optional cmx pccard dev/cpufreq/ichss.c optional cpufreq pci dev/cxgb/cxgb_main.c optional cxgb pci \ compile-with "${NORMAL_C} -I$S/dev/cxgb" dev/cxgb/cxgb_sge.c optional cxgb pci \ compile-with "${NORMAL_C} -I$S/dev/cxgb" dev/cxgb/common/cxgb_mc5.c optional cxgb pci \ compile-with "${NORMAL_C} -I$S/dev/cxgb" dev/cxgb/common/cxgb_vsc7323.c optional cxgb pci \ compile-with "${NORMAL_C} -I$S/dev/cxgb" dev/cxgb/common/cxgb_vsc8211.c optional cxgb pci \ compile-with "${NORMAL_C} -I$S/dev/cxgb" dev/cxgb/common/cxgb_ael1002.c optional cxgb pci \ compile-with "${NORMAL_C} -I$S/dev/cxgb" dev/cxgb/common/cxgb_aq100x.c optional cxgb pci \ compile-with "${NORMAL_C} -I$S/dev/cxgb" dev/cxgb/common/cxgb_mv88e1xxx.c optional cxgb pci \ compile-with "${NORMAL_C} -I$S/dev/cxgb" dev/cxgb/common/cxgb_xgmac.c optional cxgb pci \ compile-with "${NORMAL_C} -I$S/dev/cxgb" dev/cxgb/common/cxgb_t3_hw.c optional cxgb pci \ compile-with "${NORMAL_C} -I$S/dev/cxgb" dev/cxgb/common/cxgb_tn1010.c optional cxgb pci \ compile-with "${NORMAL_C} -I$S/dev/cxgb" dev/cxgb/sys/uipc_mvec.c optional cxgb pci \ compile-with "${NORMAL_C} -I$S/dev/cxgb" dev/cxgb/cxgb_t3fw.c optional cxgb cxgb_t3fw \ compile-with "${NORMAL_C} -I$S/dev/cxgb" dev/cxgbe/t4_clip.c optional cxgbe pci \ compile-with "${NORMAL_C} -I$S/dev/cxgbe" dev/cxgbe/t4_filter.c optional cxgbe pci \ compile-with "${NORMAL_C} -I$S/dev/cxgbe" dev/cxgbe/t4_if.m optional cxgbe pci dev/cxgbe/t4_iov.c optional cxgbe pci \ compile-with "${NORMAL_C} -I$S/dev/cxgbe" dev/cxgbe/t4_mp_ring.c optional cxgbe pci \ compile-with "${NORMAL_C} -I$S/dev/cxgbe" dev/cxgbe/t4_main.c optional cxgbe pci \ compile-with "${NORMAL_C} -I$S/dev/cxgbe" dev/cxgbe/t4_netmap.c optional cxgbe pci \ compile-with "${NORMAL_C} -I$S/dev/cxgbe" dev/cxgbe/t4_sched.c optional cxgbe pci \ compile-with "${NORMAL_C} -I$S/dev/cxgbe" dev/cxgbe/t4_sge.c optional cxgbe pci \ compile-with "${NORMAL_C} -I$S/dev/cxgbe" dev/cxgbe/t4_smt.c optional cxgbe pci \ compile-with "${NORMAL_C} -I$S/dev/cxgbe" dev/cxgbe/t4_l2t.c optional cxgbe pci \ compile-with "${NORMAL_C} -I$S/dev/cxgbe" dev/cxgbe/t4_tracer.c optional cxgbe pci \ compile-with "${NORMAL_C} -I$S/dev/cxgbe" dev/cxgbe/t4_vf.c optional cxgbev pci \ compile-with "${NORMAL_C} -I$S/dev/cxgbe" dev/cxgbe/common/t4_hw.c optional cxgbe pci \ compile-with "${NORMAL_C} -I$S/dev/cxgbe" dev/cxgbe/common/t4vf_hw.c optional cxgbev pci \ compile-with "${NORMAL_C} -I$S/dev/cxgbe" dev/cxgbe/crypto/t4_kern_tls.c optional cxgbe pci kern_tls \ compile-with "${NORMAL_C} -I$S/dev/cxgbe" dev/cxgbe/crypto/t4_keyctx.c optional cxgbe pci \ compile-with "${NORMAL_C} -I$S/dev/cxgbe" dev/cxgbe/cudbg/cudbg_common.c optional cxgbe \ compile-with "${NORMAL_C} -I$S/dev/cxgbe" dev/cxgbe/cudbg/cudbg_flash_utils.c optional cxgbe \ compile-with "${NORMAL_C} -I$S/dev/cxgbe" dev/cxgbe/cudbg/cudbg_lib.c optional cxgbe \ compile-with "${NORMAL_C} -I$S/dev/cxgbe" dev/cxgbe/cudbg/cudbg_wtp.c optional cxgbe \ compile-with "${NORMAL_C} -I$S/dev/cxgbe" dev/cxgbe/cudbg/fastlz.c optional cxgbe \ compile-with "${NORMAL_C} -I$S/dev/cxgbe" dev/cxgbe/cudbg/fastlz_api.c optional cxgbe \ compile-with "${NORMAL_C} -I$S/dev/cxgbe" t4fw_cfg.c optional cxgbe \ compile-with "${AWK} -f $S/tools/fw_stub.awk t4fw_cfg.fw:t4fw_cfg t4fw_cfg_uwire.fw:t4fw_cfg_uwire t4fw.fw:t4fw -mt4fw_cfg -c${.TARGET}" \ no-implicit-rule before-depend local \ clean "t4fw_cfg.c" t4fw_cfg.fwo optional cxgbe \ dependency "t4fw_cfg.fw" \ compile-with "${NORMAL_FWO}" \ no-implicit-rule \ clean "t4fw_cfg.fwo" t4fw_cfg.fw optional cxgbe \ dependency "$S/dev/cxgbe/firmware/t4fw_cfg.txt" \ compile-with "${CP} ${.ALLSRC} ${.TARGET}" \ no-obj no-implicit-rule \ clean "t4fw_cfg.fw" t4fw_cfg_uwire.fwo optional cxgbe \ dependency "t4fw_cfg_uwire.fw" \ compile-with "${NORMAL_FWO}" \ no-implicit-rule \ clean "t4fw_cfg_uwire.fwo" t4fw_cfg_uwire.fw optional cxgbe \ dependency "$S/dev/cxgbe/firmware/t4fw_cfg_uwire.txt" \ compile-with "${CP} ${.ALLSRC} ${.TARGET}" \ no-obj no-implicit-rule \ clean "t4fw_cfg_uwire.fw" t4fw.fwo optional cxgbe \ dependency "t4fw.fw" \ compile-with "${NORMAL_FWO}" \ no-implicit-rule \ clean "t4fw.fwo" t4fw.fw optional cxgbe \ dependency "$S/dev/cxgbe/firmware/t4fw-1.24.12.0.bin" \ compile-with "${CP} ${.ALLSRC} ${.TARGET}" \ no-obj no-implicit-rule \ clean "t4fw.fw" t5fw_cfg.c optional cxgbe \ compile-with "${AWK} -f $S/tools/fw_stub.awk t5fw_cfg.fw:t5fw_cfg t5fw_cfg_uwire.fw:t5fw_cfg_uwire t5fw.fw:t5fw -mt5fw_cfg -c${.TARGET}" \ no-implicit-rule before-depend local \ clean "t5fw_cfg.c" t5fw_cfg.fwo optional cxgbe \ dependency "t5fw_cfg.fw" \ compile-with "${NORMAL_FWO}" \ no-implicit-rule \ clean "t5fw_cfg.fwo" t5fw_cfg.fw optional cxgbe \ dependency "$S/dev/cxgbe/firmware/t5fw_cfg.txt" \ compile-with "${CP} ${.ALLSRC} ${.TARGET}" \ no-obj no-implicit-rule \ clean "t5fw_cfg.fw" t5fw_cfg_uwire.fwo optional cxgbe \ dependency "t5fw_cfg_uwire.fw" \ compile-with "${NORMAL_FWO}" \ no-implicit-rule \ clean "t5fw_cfg_uwire.fwo" t5fw_cfg_uwire.fw optional cxgbe \ dependency "$S/dev/cxgbe/firmware/t5fw_cfg_uwire.txt" \ compile-with "${CP} ${.ALLSRC} ${.TARGET}" \ no-obj no-implicit-rule \ clean "t5fw_cfg_uwire.fw" t5fw.fwo optional cxgbe \ dependency "t5fw.fw" \ compile-with "${NORMAL_FWO}" \ no-implicit-rule \ clean "t5fw.fwo" t5fw.fw optional cxgbe \ dependency "$S/dev/cxgbe/firmware/t5fw-1.24.12.0.bin" \ compile-with "${CP} ${.ALLSRC} ${.TARGET}" \ no-obj no-implicit-rule \ clean "t5fw.fw" t6fw_cfg.c optional cxgbe \ compile-with "${AWK} -f $S/tools/fw_stub.awk t6fw_cfg.fw:t6fw_cfg t6fw_cfg_uwire.fw:t6fw_cfg_uwire t6fw.fw:t6fw -mt6fw_cfg -c${.TARGET}" \ no-implicit-rule before-depend local \ clean "t6fw_cfg.c" t6fw_cfg.fwo optional cxgbe \ dependency "t6fw_cfg.fw" \ compile-with "${NORMAL_FWO}" \ no-implicit-rule \ clean "t6fw_cfg.fwo" t6fw_cfg.fw optional cxgbe \ dependency "$S/dev/cxgbe/firmware/t6fw_cfg.txt" \ compile-with "${CP} ${.ALLSRC} ${.TARGET}" \ no-obj no-implicit-rule \ clean "t6fw_cfg.fw" t6fw_cfg_uwire.fwo optional cxgbe \ dependency "t6fw_cfg_uwire.fw" \ compile-with "${NORMAL_FWO}" \ no-implicit-rule \ clean "t6fw_cfg_uwire.fwo" t6fw_cfg_uwire.fw optional cxgbe \ dependency "$S/dev/cxgbe/firmware/t6fw_cfg_uwire.txt" \ compile-with "${CP} ${.ALLSRC} ${.TARGET}" \ no-obj no-implicit-rule \ clean "t6fw_cfg_uwire.fw" t6fw.fwo optional cxgbe \ dependency "t6fw.fw" \ compile-with "${NORMAL_FWO}" \ no-implicit-rule \ clean "t6fw.fwo" t6fw.fw optional cxgbe \ dependency "$S/dev/cxgbe/firmware/t6fw-1.24.12.0.bin" \ compile-with "${CP} ${.ALLSRC} ${.TARGET}" \ no-obj no-implicit-rule \ clean "t6fw.fw" dev/cxgbe/crypto/t4_crypto.c optional ccr \ compile-with "${NORMAL_C} -I$S/dev/cxgbe" dev/cy/cy.c optional cy dev/cy/cy_isa.c optional cy isa dev/cy/cy_pci.c optional cy pci dev/cyapa/cyapa.c optional cyapa iicbus dev/dc/if_dc.c optional dc pci dev/dc/dcphy.c optional dc pci dev/dc/pnphy.c optional dc pci dev/dcons/dcons.c optional dcons dev/dcons/dcons_crom.c optional dcons_crom dev/dcons/dcons_os.c optional dcons dev/dme/if_dme.c optional dme dev/drm2/drm_agpsupport.c optional drm2 dev/drm2/drm_auth.c optional drm2 dev/drm2/drm_bufs.c optional drm2 dev/drm2/drm_buffer.c optional drm2 dev/drm2/drm_context.c optional drm2 dev/drm2/drm_crtc.c optional drm2 dev/drm2/drm_crtc_helper.c optional drm2 dev/drm2/drm_dma.c optional drm2 dev/drm2/drm_dp_helper.c optional drm2 dev/drm2/drm_dp_iic_helper.c optional drm2 dev/drm2/drm_drv.c optional drm2 dev/drm2/drm_edid.c optional drm2 dev/drm2/drm_fb_helper.c optional drm2 dev/drm2/drm_fops.c optional drm2 dev/drm2/drm_gem.c optional drm2 dev/drm2/drm_gem_names.c optional drm2 dev/drm2/drm_global.c optional drm2 dev/drm2/drm_hashtab.c optional drm2 dev/drm2/drm_ioctl.c optional drm2 dev/drm2/drm_irq.c optional drm2 dev/drm2/drm_linux_list_sort.c optional drm2 dev/drm2/drm_lock.c optional drm2 dev/drm2/drm_memory.c optional drm2 dev/drm2/drm_mm.c optional drm2 dev/drm2/drm_modes.c optional drm2 dev/drm2/drm_pci.c optional drm2 dev/drm2/drm_platform.c optional drm2 dev/drm2/drm_scatter.c optional drm2 dev/drm2/drm_stub.c optional drm2 dev/drm2/drm_sysctl.c optional drm2 dev/drm2/drm_vm.c optional drm2 dev/drm2/drm_os_freebsd.c optional drm2 dev/drm2/ttm/ttm_agp_backend.c optional drm2 dev/drm2/ttm/ttm_lock.c optional drm2 dev/drm2/ttm/ttm_object.c optional drm2 dev/drm2/ttm/ttm_tt.c optional drm2 dev/drm2/ttm/ttm_bo_util.c optional drm2 dev/drm2/ttm/ttm_bo.c optional drm2 dev/drm2/ttm/ttm_bo_manager.c optional drm2 dev/drm2/ttm/ttm_execbuf_util.c optional drm2 dev/drm2/ttm/ttm_memory.c optional drm2 dev/drm2/ttm/ttm_page_alloc.c optional drm2 dev/drm2/ttm/ttm_bo_vm.c optional drm2 dev/efidev/efidev.c optional efirt dev/efidev/efirt.c optional efirt dev/efidev/efirtc.c optional efirt dev/e1000/if_em.c optional em \ compile-with "${NORMAL_C} -I$S/dev/e1000" dev/e1000/em_txrx.c optional em \ compile-with "${NORMAL_C} -I$S/dev/e1000" dev/e1000/igb_txrx.c optional em \ compile-with "${NORMAL_C} -I$S/dev/e1000" dev/e1000/e1000_80003es2lan.c optional em \ compile-with "${NORMAL_C} -I$S/dev/e1000" dev/e1000/e1000_82540.c optional em \ compile-with "${NORMAL_C} -I$S/dev/e1000" dev/e1000/e1000_82541.c optional em \ compile-with "${NORMAL_C} -I$S/dev/e1000" dev/e1000/e1000_82542.c optional em \ compile-with "${NORMAL_C} -I$S/dev/e1000" dev/e1000/e1000_82543.c optional em \ compile-with "${NORMAL_C} -I$S/dev/e1000" dev/e1000/e1000_82571.c optional em \ compile-with "${NORMAL_C} -I$S/dev/e1000" dev/e1000/e1000_82575.c optional em \ compile-with "${NORMAL_C} -I$S/dev/e1000" dev/e1000/e1000_ich8lan.c optional em \ compile-with "${NORMAL_C} -I$S/dev/e1000" dev/e1000/e1000_i210.c optional em \ compile-with "${NORMAL_C} -I$S/dev/e1000" dev/e1000/e1000_api.c optional em \ compile-with "${NORMAL_C} -I$S/dev/e1000" dev/e1000/e1000_mac.c optional em \ compile-with "${NORMAL_C} -I$S/dev/e1000" dev/e1000/e1000_manage.c optional em \ compile-with "${NORMAL_C} -I$S/dev/e1000" dev/e1000/e1000_nvm.c optional em \ compile-with "${NORMAL_C} -I$S/dev/e1000" dev/e1000/e1000_phy.c optional em \ compile-with "${NORMAL_C} -I$S/dev/e1000" dev/e1000/e1000_vf.c optional em \ compile-with "${NORMAL_C} -I$S/dev/e1000" dev/e1000/e1000_mbx.c optional em \ compile-with "${NORMAL_C} -I$S/dev/e1000" dev/e1000/e1000_osdep.c optional em \ compile-with "${NORMAL_C} -I$S/dev/e1000" dev/et/if_et.c optional et dev/ena/ena.c optional ena \ compile-with "${NORMAL_C} -I$S/contrib" dev/ena/ena_sysctl.c optional ena \ compile-with "${NORMAL_C} -I$S/contrib" contrib/ena-com/ena_com.c optional ena contrib/ena-com/ena_eth_com.c optional ena dev/esp/esp_pci.c optional esp pci dev/esp/ncr53c9x.c optional esp dev/etherswitch/arswitch/arswitch.c optional arswitch dev/etherswitch/arswitch/arswitch_reg.c optional arswitch dev/etherswitch/arswitch/arswitch_phy.c optional arswitch dev/etherswitch/arswitch/arswitch_8216.c optional arswitch dev/etherswitch/arswitch/arswitch_8226.c optional arswitch dev/etherswitch/arswitch/arswitch_8316.c optional arswitch dev/etherswitch/arswitch/arswitch_8327.c optional arswitch dev/etherswitch/arswitch/arswitch_7240.c optional arswitch dev/etherswitch/arswitch/arswitch_9340.c optional arswitch dev/etherswitch/arswitch/arswitch_vlans.c optional arswitch dev/etherswitch/etherswitch.c optional etherswitch dev/etherswitch/etherswitch_if.m optional etherswitch dev/etherswitch/ip17x/ip17x.c optional ip17x dev/etherswitch/ip17x/ip175c.c optional ip17x dev/etherswitch/ip17x/ip175d.c optional ip17x dev/etherswitch/ip17x/ip17x_phy.c optional ip17x dev/etherswitch/ip17x/ip17x_vlans.c optional ip17x dev/etherswitch/miiproxy.c optional miiproxy dev/etherswitch/rtl8366/rtl8366rb.c optional rtl8366rb dev/etherswitch/e6000sw/e6000sw.c optional e6000sw dev/etherswitch/e6000sw/e6060sw.c optional e6060sw dev/etherswitch/infineon/adm6996fc.c optional adm6996fc dev/etherswitch/micrel/ksz8995ma.c optional ksz8995ma dev/etherswitch/ukswitch/ukswitch.c optional ukswitch dev/evdev/cdev.c optional evdev dev/evdev/evdev.c optional evdev dev/evdev/evdev_mt.c optional evdev dev/evdev/evdev_utils.c optional evdev dev/evdev/uinput.c optional evdev uinput dev/exca/exca.c optional cbb dev/extres/clk/clk.c optional ext_resources clk fdt dev/extres/clk/clkdev_if.m optional ext_resources clk fdt dev/extres/clk/clknode_if.m optional ext_resources clk fdt dev/extres/clk/clk_bus.c optional ext_resources clk fdt dev/extres/clk/clk_div.c optional ext_resources clk fdt dev/extres/clk/clk_fixed.c optional ext_resources clk fdt dev/extres/clk/clk_gate.c optional ext_resources clk fdt dev/extres/clk/clk_link.c optional ext_resources clk fdt dev/extres/clk/clk_mux.c optional ext_resources clk fdt dev/extres/phy/phy.c optional ext_resources phy fdt dev/extres/phy/phydev_if.m optional ext_resources phy fdt dev/extres/phy/phynode_if.m optional ext_resources phy fdt dev/extres/phy/phy_usb.c optional ext_resources phy fdt dev/extres/phy/phynode_usb_if.m optional ext_resources phy fdt dev/extres/hwreset/hwreset.c optional ext_resources hwreset fdt dev/extres/hwreset/hwreset_if.m optional ext_resources hwreset fdt dev/extres/nvmem/nvmem.c optional ext_resources nvmem fdt dev/extres/nvmem/nvmem_if.m optional ext_resources nvmem fdt dev/extres/regulator/regdev_if.m optional ext_resources regulator fdt dev/extres/regulator/regnode_if.m optional ext_resources regulator fdt dev/extres/regulator/regulator.c optional ext_resources regulator fdt dev/extres/regulator/regulator_bus.c optional ext_resources regulator fdt dev/extres/regulator/regulator_fixed.c optional ext_resources regulator fdt dev/extres/syscon/syscon.c optional ext_resources syscon dev/extres/syscon/syscon_generic.c optional ext_resources syscon fdt dev/extres/syscon/syscon_if.m optional ext_resources syscon dev/fb/fbd.c optional fbd | vt dev/fb/fb_if.m standard dev/fb/splash.c optional sc splash dev/fdt/fdt_clock.c optional fdt fdt_clock dev/fdt/fdt_clock_if.m optional fdt fdt_clock dev/fdt/fdt_common.c optional fdt dev/fdt/fdt_pinctrl.c optional fdt fdt_pinctrl dev/fdt/fdt_pinctrl_if.m optional fdt fdt_pinctrl dev/fdt/fdt_slicer.c optional fdt cfi | fdt mx25l | fdt n25q | fdt at45d dev/fdt/fdt_static_dtb.S optional fdt fdt_dtb_static \ dependency "${FDT_DTS_FILE:T:R}.dtb" dev/fdt/simplebus.c optional fdt dev/fdt/simple_mfd.c optional syscon fdt dev/filemon/filemon.c optional filemon dev/firewire/firewire.c optional firewire dev/firewire/fwcrom.c optional firewire dev/firewire/fwdev.c optional firewire dev/firewire/fwdma.c optional firewire dev/firewire/fwmem.c optional firewire dev/firewire/fwohci.c optional firewire dev/firewire/fwohci_pci.c optional firewire pci dev/firewire/if_fwe.c optional fwe dev/firewire/if_fwip.c optional fwip dev/firewire/sbp.c optional sbp dev/firewire/sbp_targ.c optional sbp_targ dev/flash/at45d.c optional at45d dev/flash/cqspi.c optional cqspi fdt xdma dev/flash/mx25l.c optional mx25l dev/flash/n25q.c optional n25q fdt dev/flash/qspi_if.m optional cqspi fdt | n25q fdt dev/fxp/if_fxp.c optional fxp dev/fxp/inphy.c optional fxp dev/gem/if_gem.c optional gem dev/gem/if_gem_pci.c optional gem pci dev/gpio/dwgpio/dwgpio.c optional gpio dwgpio fdt dev/gpio/dwgpio/dwgpio_bus.c optional gpio dwgpio fdt dev/gpio/dwgpio/dwgpio_if.m optional gpio dwgpio fdt dev/gpio/gpiobacklight.c optional gpiobacklight fdt dev/gpio/gpiokeys.c optional gpiokeys fdt dev/gpio/gpiokeys_codes.c optional gpiokeys fdt dev/gpio/gpiobus.c optional gpio \ dependency "gpiobus_if.h" dev/gpio/gpioc.c optional gpio \ dependency "gpio_if.h" dev/gpio/gpioiic.c optional gpioiic dev/gpio/gpioled.c optional gpioled !fdt dev/gpio/gpioled_fdt.c optional gpioled fdt dev/gpio/gpiomdio.c optional gpiomdio mii_bitbang dev/gpio/gpiopower.c optional gpiopower fdt dev/gpio/gpioregulator.c optional gpioregulator fdt ext_resources dev/gpio/gpiospi.c optional gpiospi dev/gpio/gpioths.c optional gpioths dev/gpio/gpio_if.m optional gpio dev/gpio/gpiobus_if.m optional gpio dev/gpio/gpiopps.c optional gpiopps fdt dev/gpio/ofw_gpiobus.c optional fdt gpio dev/hifn/hifn7751.c optional hifn dev/hme/if_hme.c optional hme dev/hme/if_hme_pci.c optional hme pci dev/hptiop/hptiop.c optional hptiop scbus dev/hwpmc/hwpmc_logging.c optional hwpmc dev/hwpmc/hwpmc_mod.c optional hwpmc dev/hwpmc/hwpmc_soft.c optional hwpmc dev/ichiic/ig4_acpi.c optional ig4 acpi iicbus dev/ichiic/ig4_iic.c optional ig4 iicbus dev/ichiic/ig4_pci.c optional ig4 pci iicbus dev/ichsmb/ichsmb.c optional ichsmb dev/ichsmb/ichsmb_pci.c optional ichsmb pci dev/ida/ida.c optional ida dev/ida/ida_disk.c optional ida dev/ida/ida_pci.c optional ida pci dev/iicbus/ad7418.c optional ad7418 dev/iicbus/ads111x.c optional ads111x dev/iicbus/ds1307.c optional ds1307 dev/iicbus/ds13rtc.c optional ds13rtc | ds133x | ds1374 dev/iicbus/ds1672.c optional ds1672 dev/iicbus/ds3231.c optional ds3231 dev/iicbus/syr827.c optional syr827 ext_resources fdt dev/iicbus/icee.c optional icee dev/iicbus/if_ic.c optional ic dev/iicbus/iic.c optional iic dev/iicbus/iic_recover_bus.c optional iicbus dev/iicbus/iicbb.c optional iicbb dev/iicbus/iicbb_if.m optional iicbb dev/iicbus/iicbus.c optional iicbus dev/iicbus/iicbus_if.m optional iicbus dev/iicbus/iiconf.c optional iicbus dev/iicbus/iicsmb.c optional iicsmb \ dependency "iicbus_if.h" dev/iicbus/iicoc.c optional iicoc dev/iicbus/iicoc_fdt.c optional iicoc ext_resources fdt dev/iicbus/iicoc_pci.c optional iicoc pci dev/iicbus/isl12xx.c optional isl12xx dev/iicbus/lm75.c optional lm75 dev/iicbus/mux/iicmux.c optional iicmux dev/iicbus/mux/iicmux_if.m optional iicmux dev/iicbus/mux/iic_gpiomux.c optional iic_gpiomux fdt dev/iicbus/mux/ltc430x.c optional ltc430x dev/iicbus/nxprtc.c optional nxprtc | pcf8563 dev/iicbus/ofw_iicbus.c optional fdt iicbus dev/iicbus/rtc8583.c optional rtc8583 dev/iicbus/s35390a.c optional s35390a dev/iicbus/sy8106a.c optional sy8106a ext_resources fdt dev/iir/iir.c optional iir dev/iir/iir_ctrl.c optional iir dev/iir/iir_pci.c optional iir pci dev/intpm/intpm.c optional intpm pci # XXX Work around clang warning, until maintainer approves fix. dev/ips/ips.c optional ips \ compile-with "${NORMAL_C} ${NO_WSOMETIMES_UNINITIALIZED}" dev/ips/ips_commands.c optional ips dev/ips/ips_disk.c optional ips dev/ips/ips_ioctl.c optional ips dev/ips/ips_pci.c optional ips pci dev/ipw/if_ipw.c optional ipw ipwbssfw.c optional ipwbssfw | ipwfw \ compile-with "${AWK} -f $S/tools/fw_stub.awk ipw_bss.fw:ipw_bss:130 -lintel_ipw -mipw_bss -c${.TARGET}" \ no-implicit-rule before-depend local \ clean "ipwbssfw.c" ipw_bss.fwo optional ipwbssfw | ipwfw \ dependency "ipw_bss.fw" \ compile-with "${NORMAL_FWO}" \ no-implicit-rule \ clean "ipw_bss.fwo" ipw_bss.fw optional ipwbssfw | ipwfw \ dependency "$S/contrib/dev/ipw/ipw2100-1.3.fw.uu" \ compile-with "${NORMAL_FW}" \ no-obj no-implicit-rule \ clean "ipw_bss.fw" ipwibssfw.c optional ipwibssfw | ipwfw \ compile-with "${AWK} -f $S/tools/fw_stub.awk ipw_ibss.fw:ipw_ibss:130 -lintel_ipw -mipw_ibss -c${.TARGET}" \ no-implicit-rule before-depend local \ clean "ipwibssfw.c" ipw_ibss.fwo optional ipwibssfw | ipwfw \ dependency "ipw_ibss.fw" \ compile-with "${NORMAL_FWO}" \ no-implicit-rule \ clean "ipw_ibss.fwo" ipw_ibss.fw optional ipwibssfw | ipwfw \ dependency "$S/contrib/dev/ipw/ipw2100-1.3-i.fw.uu" \ compile-with "${NORMAL_FW}" \ no-obj no-implicit-rule \ clean "ipw_ibss.fw" ipwmonitorfw.c optional ipwmonitorfw | ipwfw \ compile-with "${AWK} -f $S/tools/fw_stub.awk ipw_monitor.fw:ipw_monitor:130 -lintel_ipw -mipw_monitor -c${.TARGET}" \ no-implicit-rule before-depend local \ clean "ipwmonitorfw.c" ipw_monitor.fwo optional ipwmonitorfw | ipwfw \ dependency "ipw_monitor.fw" \ compile-with "${NORMAL_FWO}" \ no-implicit-rule \ clean "ipw_monitor.fwo" ipw_monitor.fw optional ipwmonitorfw | ipwfw \ dependency "$S/contrib/dev/ipw/ipw2100-1.3-p.fw.uu" \ compile-with "${NORMAL_FW}" \ no-obj no-implicit-rule \ clean "ipw_monitor.fw" dev/iscsi/icl.c optional iscsi dev/iscsi/icl_conn_if.m optional cfiscsi | iscsi dev/iscsi/icl_soft.c optional iscsi dev/iscsi/icl_soft_proxy.c optional iscsi dev/iscsi/iscsi.c optional iscsi scbus dev/iscsi_initiator/iscsi.c optional iscsi_initiator scbus dev/iscsi_initiator/iscsi_subr.c optional iscsi_initiator scbus dev/iscsi_initiator/isc_cam.c optional iscsi_initiator scbus dev/iscsi_initiator/isc_soc.c optional iscsi_initiator scbus dev/iscsi_initiator/isc_sm.c optional iscsi_initiator scbus dev/iscsi_initiator/isc_subr.c optional iscsi_initiator scbus dev/ismt/ismt.c optional ismt dev/isl/isl.c optional isl iicbus dev/isp/isp.c optional isp dev/isp/isp_freebsd.c optional isp dev/isp/isp_library.c optional isp dev/isp/isp_pci.c optional isp pci dev/isp/isp_target.c optional isp dev/ispfw/ispfw.c optional ispfw dev/iwi/if_iwi.c optional iwi iwibssfw.c optional iwibssfw | iwifw \ compile-with "${AWK} -f $S/tools/fw_stub.awk iwi_bss.fw:iwi_bss:300 -lintel_iwi -miwi_bss -c${.TARGET}" \ no-implicit-rule before-depend local \ clean "iwibssfw.c" iwi_bss.fwo optional iwibssfw | iwifw \ dependency "iwi_bss.fw" \ compile-with "${NORMAL_FWO}" \ no-implicit-rule \ clean "iwi_bss.fwo" iwi_bss.fw optional iwibssfw | iwifw \ dependency "$S/contrib/dev/iwi/ipw2200-bss.fw.uu" \ compile-with "${NORMAL_FW}" \ no-obj no-implicit-rule \ clean "iwi_bss.fw" iwiibssfw.c optional iwiibssfw | iwifw \ compile-with "${AWK} -f $S/tools/fw_stub.awk iwi_ibss.fw:iwi_ibss:300 -lintel_iwi -miwi_ibss -c${.TARGET}" \ no-implicit-rule before-depend local \ clean "iwiibssfw.c" iwi_ibss.fwo optional iwiibssfw | iwifw \ dependency "iwi_ibss.fw" \ compile-with "${NORMAL_FWO}" \ no-implicit-rule \ clean "iwi_ibss.fwo" iwi_ibss.fw optional iwiibssfw | iwifw \ dependency "$S/contrib/dev/iwi/ipw2200-ibss.fw.uu" \ compile-with "${NORMAL_FW}" \ no-obj no-implicit-rule \ clean "iwi_ibss.fw" iwimonitorfw.c optional iwimonitorfw | iwifw \ compile-with "${AWK} -f $S/tools/fw_stub.awk iwi_monitor.fw:iwi_monitor:300 -lintel_iwi -miwi_monitor -c${.TARGET}" \ no-implicit-rule before-depend local \ clean "iwimonitorfw.c" iwi_monitor.fwo optional iwimonitorfw | iwifw \ dependency "iwi_monitor.fw" \ compile-with "${NORMAL_FWO}" \ no-implicit-rule \ clean "iwi_monitor.fwo" iwi_monitor.fw optional iwimonitorfw | iwifw \ dependency "$S/contrib/dev/iwi/ipw2200-sniffer.fw.uu" \ compile-with "${NORMAL_FW}" \ no-obj no-implicit-rule \ clean "iwi_monitor.fw" dev/iwm/if_iwm.c optional iwm dev/iwm/if_iwm_7000.c optional iwm dev/iwm/if_iwm_8000.c optional iwm dev/iwm/if_iwm_9000.c optional iwm dev/iwm/if_iwm_9260.c optional iwm dev/iwm/if_iwm_binding.c optional iwm dev/iwm/if_iwm_fw.c optional iwm dev/iwm/if_iwm_led.c optional iwm dev/iwm/if_iwm_mac_ctxt.c optional iwm dev/iwm/if_iwm_notif_wait.c optional iwm dev/iwm/if_iwm_pcie_trans.c optional iwm dev/iwm/if_iwm_phy_ctxt.c optional iwm dev/iwm/if_iwm_phy_db.c optional iwm dev/iwm/if_iwm_power.c optional iwm dev/iwm/if_iwm_scan.c optional iwm dev/iwm/if_iwm_sf.c optional iwm dev/iwm/if_iwm_sta.c optional iwm dev/iwm/if_iwm_time_event.c optional iwm dev/iwm/if_iwm_util.c optional iwm iwm3160fw.c optional iwm3160fw | iwmfw \ compile-with "${AWK} -f $S/tools/fw_stub.awk iwm3160.fw:iwm3160fw -miwm3160fw -c${.TARGET}" \ no-implicit-rule before-depend local \ clean "iwm3160fw.c" iwm3160fw.fwo optional iwm3160fw | iwmfw \ dependency "iwm3160.fw" \ compile-with "${NORMAL_FWO}" \ no-implicit-rule \ clean "iwm3160fw.fwo" iwm3160.fw optional iwm3160fw | iwmfw \ dependency "$S/contrib/dev/iwm/iwm-3160-17.fw.uu" \ compile-with "${NORMAL_FW}" \ no-obj no-implicit-rule \ clean "iwm3160.fw" iwm3168fw.c optional iwm3168fw | iwmfw \ compile-with "${AWK} -f $S/tools/fw_stub.awk iwm3168.fw:iwm3168fw -miwm3168fw -c${.TARGET}" \ no-implicit-rule before-depend local \ clean "iwm3168fw.c" iwm3168fw.fwo optional iwm3168fw | iwmfw \ dependency "iwm3168.fw" \ compile-with "${NORMAL_FWO}" \ no-implicit-rule \ clean "iwm3168fw.fwo" iwm3168.fw optional iwm3168fw | iwmfw \ dependency "$S/contrib/dev/iwm/iwm-3168-22.fw.uu" \ compile-with "${NORMAL_FW}" \ no-obj no-implicit-rule \ clean "iwm3168.fw" iwm7260fw.c optional iwm7260fw | iwmfw \ compile-with "${AWK} -f $S/tools/fw_stub.awk iwm7260.fw:iwm7260fw -miwm7260fw -c${.TARGET}" \ no-implicit-rule before-depend local \ clean "iwm7260fw.c" iwm7260fw.fwo optional iwm7260fw | iwmfw \ dependency "iwm7260.fw" \ compile-with "${NORMAL_FWO}" \ no-implicit-rule \ clean "iwm7260fw.fwo" iwm7260.fw optional iwm7260fw | iwmfw \ dependency "$S/contrib/dev/iwm/iwm-7260-17.fw.uu" \ compile-with "${NORMAL_FW}" \ no-obj no-implicit-rule \ clean "iwm7260.fw" iwm7265fw.c optional iwm7265fw | iwmfw \ compile-with "${AWK} -f $S/tools/fw_stub.awk iwm7265.fw:iwm7265fw -miwm7265fw -c${.TARGET}" \ no-implicit-rule before-depend local \ clean "iwm7265fw.c" iwm7265fw.fwo optional iwm7265fw | iwmfw \ dependency "iwm7265.fw" \ compile-with "${NORMAL_FWO}" \ no-implicit-rule \ clean "iwm7265fw.fwo" iwm7265.fw optional iwm7265fw | iwmfw \ dependency "$S/contrib/dev/iwm/iwm-7265-17.fw.uu" \ compile-with "${NORMAL_FW}" \ no-obj no-implicit-rule \ clean "iwm7265.fw" iwm7265Dfw.c optional iwm7265Dfw | iwmfw \ compile-with "${AWK} -f $S/tools/fw_stub.awk iwm7265D.fw:iwm7265Dfw -miwm7265Dfw -c${.TARGET}" \ no-implicit-rule before-depend local \ clean "iwm7265Dfw.c" iwm7265Dfw.fwo optional iwm7265Dfw | iwmfw \ dependency "iwm7265D.fw" \ compile-with "${NORMAL_FWO}" \ no-implicit-rule \ clean "iwm7265Dfw.fwo" iwm7265D.fw optional iwm7265Dfw | iwmfw \ dependency "$S/contrib/dev/iwm/iwm-7265D-17.fw.uu" \ compile-with "${NORMAL_FW}" \ no-obj no-implicit-rule \ clean "iwm7265D.fw" iwm8000Cfw.c optional iwm8000Cfw | iwmfw \ compile-with "${AWK} -f $S/tools/fw_stub.awk iwm8000C.fw:iwm8000Cfw -miwm8000Cfw -c${.TARGET}" \ no-implicit-rule before-depend local \ clean "iwm8000Cfw.c" iwm8000Cfw.fwo optional iwm8000Cfw | iwmfw \ dependency "iwm8000C.fw" \ compile-with "${NORMAL_FWO}" \ no-implicit-rule \ clean "iwm8000Cfw.fwo" iwm8000C.fw optional iwm8000Cfw | iwmfw \ dependency "$S/contrib/dev/iwm/iwm-8000C-16.fw.uu" \ compile-with "${NORMAL_FW}" \ no-obj no-implicit-rule \ clean "iwm8000C.fw" iwm8265.fw optional iwm8265fw | iwmfw \ dependency "$S/contrib/dev/iwm/iwm-8265-22.fw.uu" \ compile-with "${NORMAL_FW}" \ no-obj no-implicit-rule \ clean "iwm8265.fw" iwm8265fw.c optional iwm8265fw | iwmfw \ compile-with "${AWK} -f $S/tools/fw_stub.awk iwm8265.fw:iwm8265fw -miwm8265fw -c${.TARGET}" \ no-implicit-rule before-depend local \ clean "iwm8265fw.c" iwm8265fw.fwo optional iwm8265fw | iwmfw \ dependency "iwm8265.fw" \ compile-with "${NORMAL_FWO}" \ no-implicit-rule \ clean "iwm8265fw.fwo" dev/iwn/if_iwn.c optional iwn iwn1000fw.c optional iwn1000fw | iwnfw \ compile-with "${AWK} -f $S/tools/fw_stub.awk iwn1000.fw:iwn1000fw -miwn1000fw -c${.TARGET}" \ no-implicit-rule before-depend local \ clean "iwn1000fw.c" iwn1000fw.fwo optional iwn1000fw | iwnfw \ dependency "iwn1000.fw" \ compile-with "${NORMAL_FWO}" \ no-implicit-rule \ clean "iwn1000fw.fwo" iwn1000.fw optional iwn1000fw | iwnfw \ dependency "$S/contrib/dev/iwn/iwlwifi-1000-39.31.5.1.fw.uu" \ compile-with "${NORMAL_FW}" \ no-obj no-implicit-rule \ clean "iwn1000.fw" iwn100fw.c optional iwn100fw | iwnfw \ compile-with "${AWK} -f $S/tools/fw_stub.awk iwn100.fw:iwn100fw -miwn100fw -c${.TARGET}" \ no-implicit-rule before-depend local \ clean "iwn100fw.c" iwn100fw.fwo optional iwn100fw | iwnfw \ dependency "iwn100.fw" \ compile-with "${NORMAL_FWO}" \ no-implicit-rule \ clean "iwn100fw.fwo" iwn100.fw optional iwn100fw | iwnfw \ dependency "$S/contrib/dev/iwn/iwlwifi-100-39.31.5.1.fw.uu" \ compile-with "${NORMAL_FW}" \ no-obj no-implicit-rule \ clean "iwn100.fw" iwn105fw.c optional iwn105fw | iwnfw \ compile-with "${AWK} -f $S/tools/fw_stub.awk iwn105.fw:iwn105fw -miwn105fw -c${.TARGET}" \ no-implicit-rule before-depend local \ clean "iwn105fw.c" iwn105fw.fwo optional iwn105fw | iwnfw \ dependency "iwn105.fw" \ compile-with "${NORMAL_FWO}" \ no-implicit-rule \ clean "iwn105fw.fwo" iwn105.fw optional iwn105fw | iwnfw \ dependency "$S/contrib/dev/iwn/iwlwifi-105-6-18.168.6.1.fw.uu" \ compile-with "${NORMAL_FW}" \ no-obj no-implicit-rule \ clean "iwn105.fw" iwn135fw.c optional iwn135fw | iwnfw \ compile-with "${AWK} -f $S/tools/fw_stub.awk iwn135.fw:iwn135fw -miwn135fw -c${.TARGET}" \ no-implicit-rule before-depend local \ clean "iwn135fw.c" iwn135fw.fwo optional iwn135fw | iwnfw \ dependency "iwn135.fw" \ compile-with "${NORMAL_FWO}" \ no-implicit-rule \ clean "iwn135fw.fwo" iwn135.fw optional iwn135fw | iwnfw \ dependency "$S/contrib/dev/iwn/iwlwifi-135-6-18.168.6.1.fw.uu" \ compile-with "${NORMAL_FW}" \ no-obj no-implicit-rule \ clean "iwn135.fw" iwn2000fw.c optional iwn2000fw | iwnfw \ compile-with "${AWK} -f $S/tools/fw_stub.awk iwn2000.fw:iwn2000fw -miwn2000fw -c${.TARGET}" \ no-implicit-rule before-depend local \ clean "iwn2000fw.c" iwn2000fw.fwo optional iwn2000fw | iwnfw \ dependency "iwn2000.fw" \ compile-with "${NORMAL_FWO}" \ no-implicit-rule \ clean "iwn2000fw.fwo" iwn2000.fw optional iwn2000fw | iwnfw \ dependency "$S/contrib/dev/iwn/iwlwifi-2000-18.168.6.1.fw.uu" \ compile-with "${NORMAL_FW}" \ no-obj no-implicit-rule \ clean "iwn2000.fw" iwn2030fw.c optional iwn2030fw | iwnfw \ compile-with "${AWK} -f $S/tools/fw_stub.awk iwn2030.fw:iwn2030fw -miwn2030fw -c${.TARGET}" \ no-implicit-rule before-depend local \ clean "iwn2030fw.c" iwn2030fw.fwo optional iwn2030fw | iwnfw \ dependency "iwn2030.fw" \ compile-with "${NORMAL_FWO}" \ no-implicit-rule \ clean "iwn2030fw.fwo" iwn2030.fw optional iwn2030fw | iwnfw \ dependency "$S/contrib/dev/iwn/iwnwifi-2030-18.168.6.1.fw.uu" \ compile-with "${NORMAL_FW}" \ no-obj no-implicit-rule \ clean "iwn2030.fw" iwn4965fw.c optional iwn4965fw | iwnfw \ compile-with "${AWK} -f $S/tools/fw_stub.awk iwn4965.fw:iwn4965fw -miwn4965fw -c${.TARGET}" \ no-implicit-rule before-depend local \ clean "iwn4965fw.c" iwn4965fw.fwo optional iwn4965fw | iwnfw \ dependency "iwn4965.fw" \ compile-with "${NORMAL_FWO}" \ no-implicit-rule \ clean "iwn4965fw.fwo" iwn4965.fw optional iwn4965fw | iwnfw \ dependency "$S/contrib/dev/iwn/iwlwifi-4965-228.61.2.24.fw.uu" \ compile-with "${NORMAL_FW}" \ no-obj no-implicit-rule \ clean "iwn4965.fw" iwn5000fw.c optional iwn5000fw | iwnfw \ compile-with "${AWK} -f $S/tools/fw_stub.awk iwn5000.fw:iwn5000fw -miwn5000fw -c${.TARGET}" \ no-implicit-rule before-depend local \ clean "iwn5000fw.c" iwn5000fw.fwo optional iwn5000fw | iwnfw \ dependency "iwn5000.fw" \ compile-with "${NORMAL_FWO}" \ no-implicit-rule \ clean "iwn5000fw.fwo" iwn5000.fw optional iwn5000fw | iwnfw \ dependency "$S/contrib/dev/iwn/iwlwifi-5000-8.83.5.1.fw.uu" \ compile-with "${NORMAL_FW}" \ no-obj no-implicit-rule \ clean "iwn5000.fw" iwn5150fw.c optional iwn5150fw | iwnfw \ compile-with "${AWK} -f $S/tools/fw_stub.awk iwn5150.fw:iwn5150fw -miwn5150fw -c${.TARGET}" \ no-implicit-rule before-depend local \ clean "iwn5150fw.c" iwn5150fw.fwo optional iwn5150fw | iwnfw \ dependency "iwn5150.fw" \ compile-with "${NORMAL_FWO}" \ no-implicit-rule \ clean "iwn5150fw.fwo" iwn5150.fw optional iwn5150fw | iwnfw \ dependency "$S/contrib/dev/iwn/iwlwifi-5150-8.24.2.2.fw.uu"\ compile-with "${NORMAL_FW}" \ no-obj no-implicit-rule \ clean "iwn5150.fw" iwn6000fw.c optional iwn6000fw | iwnfw \ compile-with "${AWK} -f $S/tools/fw_stub.awk iwn6000.fw:iwn6000fw -miwn6000fw -c${.TARGET}" \ no-implicit-rule before-depend local \ clean "iwn6000fw.c" iwn6000fw.fwo optional iwn6000fw | iwnfw \ dependency "iwn6000.fw" \ compile-with "${NORMAL_FWO}" \ no-implicit-rule \ clean "iwn6000fw.fwo" iwn6000.fw optional iwn6000fw | iwnfw \ dependency "$S/contrib/dev/iwn/iwlwifi-6000-9.221.4.1.fw.uu" \ compile-with "${NORMAL_FW}" \ no-obj no-implicit-rule \ clean "iwn6000.fw" iwn6000g2afw.c optional iwn6000g2afw | iwnfw \ compile-with "${AWK} -f $S/tools/fw_stub.awk iwn6000g2a.fw:iwn6000g2afw -miwn6000g2afw -c${.TARGET}" \ no-implicit-rule before-depend local \ clean "iwn6000g2afw.c" iwn6000g2afw.fwo optional iwn6000g2afw | iwnfw \ dependency "iwn6000g2a.fw" \ compile-with "${NORMAL_FWO}" \ no-implicit-rule \ clean "iwn6000g2afw.fwo" iwn6000g2a.fw optional iwn6000g2afw | iwnfw \ dependency "$S/contrib/dev/iwn/iwlwifi-6000g2a-18.168.6.1.fw.uu" \ compile-with "${NORMAL_FW}" \ no-obj no-implicit-rule \ clean "iwn6000g2a.fw" iwn6000g2bfw.c optional iwn6000g2bfw | iwnfw \ compile-with "${AWK} -f $S/tools/fw_stub.awk iwn6000g2b.fw:iwn6000g2bfw -miwn6000g2bfw -c${.TARGET}" \ no-implicit-rule before-depend local \ clean "iwn6000g2bfw.c" iwn6000g2bfw.fwo optional iwn6000g2bfw | iwnfw \ dependency "iwn6000g2b.fw" \ compile-with "${NORMAL_FWO}" \ no-implicit-rule \ clean "iwn6000g2bfw.fwo" iwn6000g2b.fw optional iwn6000g2bfw | iwnfw \ dependency "$S/contrib/dev/iwn/iwlwifi-6000g2b-18.168.6.1.fw.uu" \ compile-with "${NORMAL_FW}" \ no-obj no-implicit-rule \ clean "iwn6000g2b.fw" iwn6050fw.c optional iwn6050fw | iwnfw \ compile-with "${AWK} -f $S/tools/fw_stub.awk iwn6050.fw:iwn6050fw -miwn6050fw -c${.TARGET}" \ no-implicit-rule before-depend local \ clean "iwn6050fw.c" iwn6050fw.fwo optional iwn6050fw | iwnfw \ dependency "iwn6050.fw" \ compile-with "${NORMAL_FWO}" \ no-implicit-rule \ clean "iwn6050fw.fwo" iwn6050.fw optional iwn6050fw | iwnfw \ dependency "$S/contrib/dev/iwn/iwlwifi-6050-41.28.5.1.fw.uu" \ compile-with "${NORMAL_FW}" \ no-obj no-implicit-rule \ clean "iwn6050.fw" dev/ixgbe/if_ix.c optional ix inet \ compile-with "${NORMAL_C} -I$S/dev/ixgbe -DSMP" dev/ixgbe/if_ixv.c optional ixv inet \ compile-with "${NORMAL_C} -I$S/dev/ixgbe -DSMP" dev/ixgbe/if_bypass.c optional ix inet \ compile-with "${NORMAL_C} -I$S/dev/ixgbe" dev/ixgbe/if_fdir.c optional ix inet | ixv inet \ compile-with "${NORMAL_C} -I$S/dev/ixgbe" dev/ixgbe/if_sriov.c optional ix inet \ compile-with "${NORMAL_C} -I$S/dev/ixgbe" dev/ixgbe/ix_txrx.c optional ix inet | ixv inet \ compile-with "${NORMAL_C} -I$S/dev/ixgbe" dev/ixgbe/ixgbe_osdep.c optional ix inet | ixv inet \ compile-with "${NORMAL_C} -I$S/dev/ixgbe" dev/ixgbe/ixgbe_phy.c optional ix inet | ixv inet \ compile-with "${NORMAL_C} -I$S/dev/ixgbe" dev/ixgbe/ixgbe_api.c optional ix inet | ixv inet \ compile-with "${NORMAL_C} -I$S/dev/ixgbe" dev/ixgbe/ixgbe_common.c optional ix inet | ixv inet \ compile-with "${NORMAL_C} -I$S/dev/ixgbe" dev/ixgbe/ixgbe_mbx.c optional ix inet | ixv inet \ compile-with "${NORMAL_C} -I$S/dev/ixgbe" dev/ixgbe/ixgbe_vf.c optional ix inet | ixv inet \ compile-with "${NORMAL_C} -I$S/dev/ixgbe" dev/ixgbe/ixgbe_82598.c optional ix inet | ixv inet \ compile-with "${NORMAL_C} -I$S/dev/ixgbe" dev/ixgbe/ixgbe_82599.c optional ix inet | ixv inet \ compile-with "${NORMAL_C} -I$S/dev/ixgbe" dev/ixgbe/ixgbe_x540.c optional ix inet | ixv inet \ compile-with "${NORMAL_C} -I$S/dev/ixgbe" dev/ixgbe/ixgbe_x550.c optional ix inet | ixv inet \ compile-with "${NORMAL_C} -I$S/dev/ixgbe" dev/ixgbe/ixgbe_dcb.c optional ix inet | ixv inet \ compile-with "${NORMAL_C} -I$S/dev/ixgbe" dev/ixgbe/ixgbe_dcb_82598.c optional ix inet | ixv inet \ compile-with "${NORMAL_C} -I$S/dev/ixgbe" dev/ixgbe/ixgbe_dcb_82599.c optional ix inet | ixv inet \ compile-with "${NORMAL_C} -I$S/dev/ixgbe" dev/jedec_dimm/jedec_dimm.c optional jedec_dimm smbus dev/jme/if_jme.c optional jme pci dev/kbd/kbd.c optional atkbd | pckbd | sc | ukbd | vt dev/kbdmux/kbdmux.c optional kbdmux dev/ksyms/ksyms.c optional ksyms dev/le/am7990.c optional le dev/le/am79900.c optional le dev/le/if_le_pci.c optional le pci dev/le/lance.c optional le dev/led/led.c standard dev/lge/if_lge.c optional lge dev/liquidio/base/cn23xx_pf_device.c optional lio \ compile-with "${NORMAL_C} \ -I$S/dev/liquidio -I$S/dev/liquidio/base -DSMP" dev/liquidio/base/lio_console.c optional lio \ compile-with "${NORMAL_C} \ -I$S/dev/liquidio -I$S/dev/liquidio/base -DSMP" dev/liquidio/base/lio_ctrl.c optional lio \ compile-with "${NORMAL_C} \ -I$S/dev/liquidio -I$S/dev/liquidio/base -DSMP" dev/liquidio/base/lio_device.c optional lio \ compile-with "${NORMAL_C} \ -I$S/dev/liquidio -I$S/dev/liquidio/base -DSMP" dev/liquidio/base/lio_droq.c optional lio \ compile-with "${NORMAL_C} \ -I$S/dev/liquidio -I$S/dev/liquidio/base -DSMP" dev/liquidio/base/lio_mem_ops.c optional lio \ compile-with "${NORMAL_C} \ -I$S/dev/liquidio -I$S/dev/liquidio/base -DSMP" dev/liquidio/base/lio_request_manager.c optional lio \ compile-with "${NORMAL_C} \ -I$S/dev/liquidio -I$S/dev/liquidio/base -DSMP" dev/liquidio/base/lio_response_manager.c optional lio \ compile-with "${NORMAL_C} \ -I$S/dev/liquidio -I$S/dev/liquidio/base -DSMP" dev/liquidio/lio_core.c optional lio \ compile-with "${NORMAL_C} \ -I$S/dev/liquidio -I$S/dev/liquidio/base -DSMP" dev/liquidio/lio_ioctl.c optional lio \ compile-with "${NORMAL_C} \ -I$S/dev/liquidio -I$S/dev/liquidio/base -DSMP" dev/liquidio/lio_main.c optional lio \ compile-with "${NORMAL_C} \ -I$S/dev/liquidio -I$S/dev/liquidio/base -DSMP" dev/liquidio/lio_rss.c optional lio \ compile-with "${NORMAL_C} \ -I$S/dev/liquidio -I$S/dev/liquidio/base -DSMP" dev/liquidio/lio_rxtx.c optional lio \ compile-with "${NORMAL_C} \ -I$S/dev/liquidio -I$S/dev/liquidio/base -DSMP" dev/liquidio/lio_sysctl.c optional lio \ compile-with "${NORMAL_C} \ -I$S/dev/liquidio -I$S/dev/liquidio/base -DSMP" lio.c optional lio \ compile-with "${AWK} -f $S/tools/fw_stub.awk lio_23xx_nic.bin.fw:lio_23xx_nic.bin -mlio_23xx_nic.bin -c${.TARGET}" \ no-implicit-rule before-depend local \ clean "lio.c" lio_23xx_nic.bin.fw.fwo optional lio \ dependency "lio_23xx_nic.bin.fw" \ compile-with "${NORMAL_FWO}" \ no-implicit-rule \ clean "lio_23xx_nic.bin.fw.fwo" lio_23xx_nic.bin.fw optional lio \ dependency "$S/contrib/dev/liquidio/lio_23xx_nic.bin.uu" \ compile-with "${NORMAL_FW}" \ no-obj no-implicit-rule \ clean "lio_23xx_nic.bin.fw" dev/malo/if_malo.c optional malo dev/malo/if_malohal.c optional malo dev/malo/if_malo_pci.c optional malo pci dev/mc146818/mc146818.c optional mc146818 dev/md/md.c optional md dev/mdio/mdio_if.m optional miiproxy | mdio dev/mdio/mdio.c optional miiproxy | mdio dev/mem/memdev.c optional mem dev/mem/memutil.c optional mem dev/mfi/mfi.c optional mfi dev/mfi/mfi_debug.c optional mfi dev/mfi/mfi_pci.c optional mfi pci dev/mfi/mfi_disk.c optional mfi dev/mfi/mfi_syspd.c optional mfi dev/mfi/mfi_tbolt.c optional mfi dev/mfi/mfi_linux.c optional mfi compat_linux dev/mfi/mfi_cam.c optional mfip scbus dev/mii/acphy.c optional miibus | acphy dev/mii/amphy.c optional miibus | amphy dev/mii/atphy.c optional miibus | atphy dev/mii/axphy.c optional miibus | axphy dev/mii/bmtphy.c optional miibus | bmtphy dev/mii/brgphy.c optional miibus | brgphy dev/mii/ciphy.c optional miibus | ciphy dev/mii/e1000phy.c optional miibus | e1000phy dev/mii/gentbi.c optional miibus | gentbi dev/mii/icsphy.c optional miibus | icsphy dev/mii/ip1000phy.c optional miibus | ip1000phy dev/mii/jmphy.c optional miibus | jmphy dev/mii/lxtphy.c optional miibus | lxtphy dev/mii/micphy.c optional miibus fdt | micphy fdt dev/mii/mii.c optional miibus | mii dev/mii/mii_bitbang.c optional miibus | mii_bitbang dev/mii/mii_physubr.c optional miibus | mii dev/mii/mii_fdt.c optional miibus fdt | mii fdt dev/mii/miibus_if.m optional miibus | mii dev/mii/mlphy.c optional miibus | mlphy dev/mii/nsgphy.c optional miibus | nsgphy dev/mii/nsphy.c optional miibus | nsphy dev/mii/nsphyter.c optional miibus | nsphyter dev/mii/pnaphy.c optional miibus | pnaphy dev/mii/qsphy.c optional miibus | qsphy dev/mii/rdcphy.c optional miibus | rdcphy dev/mii/rgephy.c optional miibus | rgephy dev/mii/rlphy.c optional miibus | rlphy dev/mii/rlswitch.c optional rlswitch dev/mii/smcphy.c optional miibus | smcphy dev/mii/smscphy.c optional miibus | smscphy dev/mii/tdkphy.c optional miibus | tdkphy dev/mii/tlphy.c optional miibus | tlphy dev/mii/truephy.c optional miibus | truephy dev/mii/ukphy.c optional miibus | mii dev/mii/ukphy_subr.c optional miibus | mii dev/mii/vscphy.c optional miibus | vscphy dev/mii/xmphy.c optional miibus | xmphy dev/mk48txx/mk48txx.c optional mk48txx dev/mlxfw/mlxfw_fsm.c optional mlxfw \ compile-with "${MLXFW_C}" dev/mlxfw/mlxfw_mfa2.c optional mlxfw \ compile-with "${MLXFW_C}" dev/mlxfw/mlxfw_mfa2_tlv_multi.c optional mlxfw \ compile-with "${MLXFW_C}" dev/mlx/mlx.c optional mlx dev/mlx/mlx_disk.c optional mlx dev/mlx/mlx_pci.c optional mlx pci dev/mly/mly.c optional mly dev/mmc/mmc_subr.c optional mmc | mmcsd !mmccam dev/mmc/mmc.c optional mmc !mmccam dev/mmc/mmcbr_if.m standard dev/mmc/mmcbus_if.m standard dev/mmc/mmcsd.c optional mmcsd !mmccam dev/mmcnull/mmcnull.c optional mmcnull dev/mn/if_mn.c optional mn pci dev/mpr/mpr.c optional mpr dev/mpr/mpr_config.c optional mpr # XXX Work around clang warning, until maintainer approves fix. dev/mpr/mpr_mapping.c optional mpr \ compile-with "${NORMAL_C} ${NO_WSOMETIMES_UNINITIALIZED}" dev/mpr/mpr_pci.c optional mpr pci dev/mpr/mpr_sas.c optional mpr \ compile-with "${NORMAL_C} ${NO_WUNNEEDED_INTERNAL_DECL}" dev/mpr/mpr_sas_lsi.c optional mpr dev/mpr/mpr_table.c optional mpr dev/mpr/mpr_user.c optional mpr dev/mps/mps.c optional mps dev/mps/mps_config.c optional mps # XXX Work around clang warning, until maintainer approves fix. dev/mps/mps_mapping.c optional mps \ compile-with "${NORMAL_C} ${NO_WSOMETIMES_UNINITIALIZED}" dev/mps/mps_pci.c optional mps pci dev/mps/mps_sas.c optional mps \ compile-with "${NORMAL_C} ${NO_WUNNEEDED_INTERNAL_DECL}" dev/mps/mps_sas_lsi.c optional mps dev/mps/mps_table.c optional mps dev/mps/mps_user.c optional mps dev/mpt/mpt.c optional mpt dev/mpt/mpt_cam.c optional mpt dev/mpt/mpt_debug.c optional mpt dev/mpt/mpt_pci.c optional mpt pci dev/mpt/mpt_raid.c optional mpt dev/mpt/mpt_user.c optional mpt dev/mrsas/mrsas.c optional mrsas dev/mrsas/mrsas_cam.c optional mrsas dev/mrsas/mrsas_ioctl.c optional mrsas dev/mrsas/mrsas_fp.c optional mrsas dev/msk/if_msk.c optional msk dev/mvs/mvs.c optional mvs dev/mvs/mvs_if.m optional mvs dev/mvs/mvs_pci.c optional mvs pci dev/mwl/if_mwl.c optional mwl dev/mwl/if_mwl_pci.c optional mwl pci dev/mwl/mwlhal.c optional mwl mwlfw.c optional mwlfw \ compile-with "${AWK} -f $S/tools/fw_stub.awk mw88W8363.fw:mw88W8363fw mwlboot.fw:mwlboot -mmwl -c${.TARGET}" \ no-implicit-rule before-depend local \ clean "mwlfw.c" mw88W8363.fwo optional mwlfw \ dependency "mw88W8363.fw" \ compile-with "${NORMAL_FWO}" \ no-implicit-rule \ clean "mw88W8363.fwo" mw88W8363.fw optional mwlfw \ dependency "$S/contrib/dev/mwl/mw88W8363.fw.uu" \ compile-with "${NORMAL_FW}" \ no-obj no-implicit-rule \ clean "mw88W8363.fw" mwlboot.fwo optional mwlfw \ dependency "mwlboot.fw" \ compile-with "${NORMAL_FWO}" \ no-implicit-rule \ clean "mwlboot.fwo" mwlboot.fw optional mwlfw \ dependency "$S/contrib/dev/mwl/mwlboot.fw.uu" \ compile-with "${NORMAL_FW}" \ no-obj no-implicit-rule \ clean "mwlboot.fw" dev/mxge/if_mxge.c optional mxge pci dev/mxge/mxge_eth_z8e.c optional mxge pci dev/mxge/mxge_ethp_z8e.c optional mxge pci dev/mxge/mxge_rss_eth_z8e.c optional mxge pci dev/mxge/mxge_rss_ethp_z8e.c optional mxge pci dev/my/if_my.c optional my dev/netmap/if_ptnet.c optional netmap inet dev/netmap/netmap.c optional netmap dev/netmap/netmap_bdg.c optional netmap dev/netmap/netmap_freebsd.c optional netmap dev/netmap/netmap_generic.c optional netmap dev/netmap/netmap_kloop.c optional netmap dev/netmap/netmap_legacy.c optional netmap dev/netmap/netmap_mbq.c optional netmap dev/netmap/netmap_mem2.c optional netmap dev/netmap/netmap_monitor.c optional netmap dev/netmap/netmap_null.c optional netmap dev/netmap/netmap_offloadings.c optional netmap dev/netmap/netmap_pipe.c optional netmap dev/netmap/netmap_vale.c optional netmap # compile-with "${NORMAL_C} -Wconversion -Wextra" dev/nfsmb/nfsmb.c optional nfsmb pci dev/nge/if_nge.c optional nge dev/nmdm/nmdm.c optional nmdm dev/null/null.c standard dev/nvd/nvd.c optional nvd nvme dev/nvme/nvme.c optional nvme dev/nvme/nvme_ahci.c optional nvme ahci dev/nvme/nvme_ctrlr.c optional nvme dev/nvme/nvme_ctrlr_cmd.c optional nvme dev/nvme/nvme_ns.c optional nvme dev/nvme/nvme_ns_cmd.c optional nvme dev/nvme/nvme_pci.c optional nvme pci dev/nvme/nvme_qpair.c optional nvme dev/nvme/nvme_sim.c optional nvme scbus dev/nvme/nvme_sysctl.c optional nvme dev/nvme/nvme_test.c optional nvme dev/nvme/nvme_util.c optional nvme dev/oce/oce_hw.c optional oce pci dev/oce/oce_if.c optional oce pci dev/oce/oce_mbox.c optional oce pci dev/oce/oce_queue.c optional oce pci dev/oce/oce_sysctl.c optional oce pci dev/oce/oce_util.c optional oce pci dev/ocs_fc/ocs_pci.c optional ocs_fc pci dev/ocs_fc/ocs_ioctl.c optional ocs_fc pci dev/ocs_fc/ocs_os.c optional ocs_fc pci dev/ocs_fc/ocs_utils.c optional ocs_fc pci dev/ocs_fc/ocs_hw.c optional ocs_fc pci dev/ocs_fc/ocs_hw_queues.c optional ocs_fc pci dev/ocs_fc/sli4.c optional ocs_fc pci dev/ocs_fc/ocs_sm.c optional ocs_fc pci dev/ocs_fc/ocs_device.c optional ocs_fc pci dev/ocs_fc/ocs_xport.c optional ocs_fc pci dev/ocs_fc/ocs_domain.c optional ocs_fc pci dev/ocs_fc/ocs_sport.c optional ocs_fc pci dev/ocs_fc/ocs_els.c optional ocs_fc pci dev/ocs_fc/ocs_fabric.c optional ocs_fc pci dev/ocs_fc/ocs_io.c optional ocs_fc pci dev/ocs_fc/ocs_node.c optional ocs_fc pci dev/ocs_fc/ocs_scsi.c optional ocs_fc pci dev/ocs_fc/ocs_unsol.c optional ocs_fc pci dev/ocs_fc/ocs_ddump.c optional ocs_fc pci dev/ocs_fc/ocs_mgmt.c optional ocs_fc pci dev/ocs_fc/ocs_cam.c optional ocs_fc pci dev/ofw/ofw_bus_if.m optional fdt dev/ofw/ofw_bus_subr.c optional fdt dev/ofw/ofw_cpu.c optional fdt dev/ofw/ofw_fdt.c optional fdt dev/ofw/ofw_if.m optional fdt dev/ofw/ofw_graph.c optional fdt dev/ofw/ofw_subr.c optional fdt dev/ofw/ofwbus.c optional fdt dev/ofw/openfirm.c optional fdt dev/ofw/openfirmio.c optional fdt dev/ow/ow.c optional ow \ dependency "owll_if.h" \ dependency "own_if.h" dev/ow/owll_if.m optional ow dev/ow/own_if.m optional ow dev/ow/ow_temp.c optional ow_temp dev/ow/owc_gpiobus.c optional owc gpio dev/pbio/pbio.c optional pbio isa dev/pccard/card_if.m standard dev/pccard/pccard.c optional pccard dev/pccard/pccard_cis.c optional pccard dev/pccard/pccard_cis_quirks.c optional pccard dev/pccard/pccard_device.c optional pccard dev/pccard/power_if.m standard dev/pccbb/pccbb.c optional cbb dev/pccbb/pccbb_pci.c optional cbb pci dev/pcf/pcf.c optional pcf dev/pci/fixup_pci.c optional pci dev/pci/hostb_pci.c optional pci dev/pci/ignore_pci.c optional pci dev/pci/isa_pci.c optional pci isa dev/pci/pci.c optional pci dev/pci/pci_if.m standard dev/pci/pci_iov.c optional pci pci_iov dev/pci/pci_iov_if.m standard dev/pci/pci_iov_schema.c optional pci pci_iov dev/pci/pci_pci.c optional pci dev/pci/pci_subr.c optional pci dev/pci/pci_user.c optional pci dev/pci/pcib_if.m standard dev/pci/pcib_support.c standard dev/pci/vga_pci.c optional pci dev/pms/freebsd/driver/ini/src/agtiapi.c optional pmspcv \ compile-with "${NORMAL_C} -Wunused-variable -Woverflow -Wparentheses -w" dev/pms/RefTisa/sallsdk/spc/sadisc.c optional pmspcv \ compile-with "${NORMAL_C} -Wunused-variable -Woverflow -Wparentheses -w" dev/pms/RefTisa/sallsdk/spc/mpi.c optional pmspcv \ compile-with "${NORMAL_C} -Wunused-variable -Woverflow -Wparentheses -w" dev/pms/RefTisa/sallsdk/spc/saframe.c optional pmspcv \ compile-with "${NORMAL_C} -Wunused-variable -Woverflow -Wparentheses -w" dev/pms/RefTisa/sallsdk/spc/sahw.c optional pmspcv \ compile-with "${NORMAL_C} -Wunused-variable -Woverflow -Wparentheses -w" dev/pms/RefTisa/sallsdk/spc/sainit.c optional pmspcv \ compile-with "${NORMAL_C} -Wunused-variable -Woverflow -Wparentheses -w" dev/pms/RefTisa/sallsdk/spc/saint.c optional pmspcv \ compile-with "${NORMAL_C} -Wunused-variable -Woverflow -Wparentheses -w" dev/pms/RefTisa/sallsdk/spc/sampicmd.c optional pmspcv \ compile-with "${NORMAL_C} -Wunused-variable -Woverflow -Wparentheses -w" dev/pms/RefTisa/sallsdk/spc/sampirsp.c optional pmspcv \ compile-with "${NORMAL_C} -Wunused-variable -Woverflow -Wparentheses -w" dev/pms/RefTisa/sallsdk/spc/saphy.c optional pmspcv \ compile-with "${NORMAL_C} -Wunused-variable -Woverflow -Wparentheses -w" dev/pms/RefTisa/sallsdk/spc/saport.c optional pmspcv \ compile-with "${NORMAL_C} -Wunused-variable -Woverflow -Wparentheses -w" dev/pms/RefTisa/sallsdk/spc/sasata.c optional pmspcv \ compile-with "${NORMAL_C} -Wunused-variable -Woverflow -Wparentheses -w" dev/pms/RefTisa/sallsdk/spc/sasmp.c optional pmspcv \ compile-with "${NORMAL_C} -Wunused-variable -Woverflow -Wparentheses -w" dev/pms/RefTisa/sallsdk/spc/sassp.c optional pmspcv \ compile-with "${NORMAL_C} -Wunused-variable -Woverflow -Wparentheses -w" dev/pms/RefTisa/sallsdk/spc/satimer.c optional pmspcv \ compile-with "${NORMAL_C} -Wunused-variable -Woverflow -Wparentheses -w" dev/pms/RefTisa/sallsdk/spc/sautil.c optional pmspcv \ compile-with "${NORMAL_C} -Wunused-variable -Woverflow -Wparentheses -w" dev/pms/RefTisa/sallsdk/spc/saioctlcmd.c optional pmspcv \ compile-with "${NORMAL_C} -Wunused-variable -Woverflow -Wparentheses -w" dev/pms/RefTisa/sallsdk/spc/mpidebug.c optional pmspcv \ compile-with "${NORMAL_C} -Wunused-variable -Woverflow -Wparentheses -w" dev/pms/RefTisa/discovery/dm/dminit.c optional pmspcv \ compile-with "${NORMAL_C} -Wunused-variable -Woverflow -Wparentheses -w" dev/pms/RefTisa/discovery/dm/dmsmp.c optional pmspcv \ compile-with "${NORMAL_C} -Wunused-variable -Woverflow -Wparentheses -w" dev/pms/RefTisa/discovery/dm/dmdisc.c optional pmspcv \ compile-with "${NORMAL_C} -Wunused-variable -Woverflow -Wparentheses -w" dev/pms/RefTisa/discovery/dm/dmport.c optional pmspcv \ compile-with "${NORMAL_C} -Wunused-variable -Woverflow -Wparentheses -w" dev/pms/RefTisa/discovery/dm/dmtimer.c optional pmspcv \ compile-with "${NORMAL_C} -Wunused-variable -Woverflow -Wparentheses -w" dev/pms/RefTisa/discovery/dm/dmmisc.c optional pmspcv \ compile-with "${NORMAL_C} -Wunused-variable -Woverflow -Wparentheses -w" dev/pms/RefTisa/sat/src/sminit.c optional pmspcv \ compile-with "${NORMAL_C} -Wunused-variable -Woverflow -Wparentheses -w" dev/pms/RefTisa/sat/src/smmisc.c optional pmspcv \ compile-with "${NORMAL_C} -Wunused-variable -Woverflow -Wparentheses -w" dev/pms/RefTisa/sat/src/smsat.c optional pmspcv \ compile-with "${NORMAL_C} -Wunused-variable -Woverflow -Wparentheses -w" dev/pms/RefTisa/sat/src/smsatcb.c optional pmspcv \ compile-with "${NORMAL_C} -Wunused-variable -Woverflow -Wparentheses -w" dev/pms/RefTisa/sat/src/smsathw.c optional pmspcv \ compile-with "${NORMAL_C} -Wunused-variable -Woverflow -Wparentheses -w" dev/pms/RefTisa/sat/src/smtimer.c optional pmspcv \ compile-with "${NORMAL_C} -Wunused-variable -Woverflow -Wparentheses -w" dev/pms/RefTisa/tisa/sassata/common/tdinit.c optional pmspcv \ compile-with "${NORMAL_C} -Wunused-variable -Woverflow -Wparentheses -w" dev/pms/RefTisa/tisa/sassata/common/tdmisc.c optional pmspcv \ compile-with "${NORMAL_C} -Wunused-variable -Woverflow -Wparentheses -w" dev/pms/RefTisa/tisa/sassata/common/tdesgl.c optional pmspcv \ compile-with "${NORMAL_C} -Wunused-variable -Woverflow -Wparentheses -w" dev/pms/RefTisa/tisa/sassata/common/tdport.c optional pmspcv \ compile-with "${NORMAL_C} -Wunused-variable -Woverflow -Wparentheses -w" dev/pms/RefTisa/tisa/sassata/common/tdint.c optional pmspcv \ compile-with "${NORMAL_C} -Wunused-variable -Woverflow -Wparentheses -w" dev/pms/RefTisa/tisa/sassata/common/tdioctl.c optional pmspcv \ compile-with "${NORMAL_C} -Wunused-variable -Woverflow -Wparentheses -w" dev/pms/RefTisa/tisa/sassata/common/tdhw.c optional pmspcv \ compile-with "${NORMAL_C} -Wunused-variable -Woverflow -Wparentheses -w" dev/pms/RefTisa/tisa/sassata/common/ossacmnapi.c optional pmspcv \ compile-with "${NORMAL_C} -Wunused-variable -Woverflow -Wparentheses -w" dev/pms/RefTisa/tisa/sassata/common/tddmcmnapi.c optional pmspcv \ compile-with "${NORMAL_C} -Wunused-variable -Woverflow -Wparentheses -w" dev/pms/RefTisa/tisa/sassata/common/tdsmcmnapi.c optional pmspcv \ compile-with "${NORMAL_C} -Wunused-variable -Woverflow -Wparentheses -w" dev/pms/RefTisa/tisa/sassata/common/tdtimers.c optional pmspcv \ compile-with "${NORMAL_C} -Wunused-variable -Woverflow -Wparentheses -w" dev/pms/RefTisa/tisa/sassata/sas/ini/itdio.c optional pmspcv \ compile-with "${NORMAL_C} -Wunused-variable -Woverflow -Wparentheses -w" dev/pms/RefTisa/tisa/sassata/sas/ini/itdcb.c optional pmspcv \ compile-with "${NORMAL_C} -Wunused-variable -Woverflow -Wparentheses -w" dev/pms/RefTisa/tisa/sassata/sas/ini/itdinit.c optional pmspcv \ compile-with "${NORMAL_C} -Wunused-variable -Woverflow -Wparentheses -w" dev/pms/RefTisa/tisa/sassata/sas/ini/itddisc.c optional pmspcv \ compile-with "${NORMAL_C} -Wunused-variable -Woverflow -Wparentheses -w" dev/pms/RefTisa/tisa/sassata/sata/host/sat.c optional pmspcv \ compile-with "${NORMAL_C} -Wunused-variable -Woverflow -Wparentheses -w" dev/pms/RefTisa/tisa/sassata/sata/host/ossasat.c optional pmspcv \ compile-with "${NORMAL_C} -Wunused-variable -Woverflow -Wparentheses -w" dev/pms/RefTisa/tisa/sassata/sata/host/sathw.c optional pmspcv \ compile-with "${NORMAL_C} -Wunused-variable -Woverflow -Wparentheses -w" dev/ppbus/if_plip.c optional plip dev/ppbus/lpbb.c optional lpbb dev/ppbus/lpt.c optional lpt dev/ppbus/pcfclock.c optional pcfclock dev/ppbus/ppb_1284.c optional ppbus dev/ppbus/ppb_base.c optional ppbus dev/ppbus/ppb_msq.c optional ppbus dev/ppbus/ppbconf.c optional ppbus dev/ppbus/ppbus_if.m optional ppbus dev/ppbus/ppi.c optional ppi dev/ppbus/pps.c optional pps dev/ppc/ppc.c optional ppc dev/ppc/ppc_acpi.c optional ppc acpi dev/ppc/ppc_isa.c optional ppc isa dev/ppc/ppc_pci.c optional ppc pci dev/ppc/ppc_puc.c optional ppc puc dev/proto/proto_bus_isa.c optional proto acpi | proto isa dev/proto/proto_bus_pci.c optional proto pci dev/proto/proto_busdma.c optional proto dev/proto/proto_core.c optional proto dev/pst/pst-iop.c optional pst dev/pst/pst-pci.c optional pst pci dev/pst/pst-raid.c optional pst dev/pty/pty.c optional pty dev/puc/puc.c optional puc dev/puc/puc_cfg.c optional puc dev/puc/puc_pccard.c optional puc pccard dev/puc/puc_pci.c optional puc pci dev/pwm/pwmc.c optional pwm | pwmc dev/pwm/pwmbus.c optional pwm | pwmbus dev/pwm/pwmbus_if.m optional pwm | pwmbus dev/pwm/ofw_pwm.c optional pwm fdt | pwmbus fdt dev/pwm/ofw_pwmbus.c optional pwm fdt | pwmbus fdt dev/quicc/quicc_core.c optional quicc dev/ral/rt2560.c optional ral dev/ral/rt2661.c optional ral dev/ral/rt2860.c optional ral dev/ral/if_ral_pci.c optional ral pci rt2561fw.c optional rt2561fw | ralfw \ compile-with "${AWK} -f $S/tools/fw_stub.awk rt2561.fw:rt2561fw -mrt2561 -c${.TARGET}" \ no-implicit-rule before-depend local \ clean "rt2561fw.c" rt2561fw.fwo optional rt2561fw | ralfw \ dependency "rt2561.fw" \ compile-with "${NORMAL_FWO}" \ no-implicit-rule \ clean "rt2561fw.fwo" rt2561.fw optional rt2561fw | ralfw \ dependency "$S/contrib/dev/ral/rt2561.fw.uu" \ compile-with "${NORMAL_FW}" \ no-obj no-implicit-rule \ clean "rt2561.fw" rt2561sfw.c optional rt2561sfw | ralfw \ compile-with "${AWK} -f $S/tools/fw_stub.awk rt2561s.fw:rt2561sfw -mrt2561s -c${.TARGET}" \ no-implicit-rule before-depend local \ clean "rt2561sfw.c" rt2561sfw.fwo optional rt2561sfw | ralfw \ dependency "rt2561s.fw" \ compile-with "${NORMAL_FWO}" \ no-implicit-rule \ clean "rt2561sfw.fwo" rt2561s.fw optional rt2561sfw | ralfw \ dependency "$S/contrib/dev/ral/rt2561s.fw.uu" \ compile-with "${NORMAL_FW}" \ no-obj no-implicit-rule \ clean "rt2561s.fw" rt2661fw.c optional rt2661fw | ralfw \ compile-with "${AWK} -f $S/tools/fw_stub.awk rt2661.fw:rt2661fw -mrt2661 -c${.TARGET}" \ no-implicit-rule before-depend local \ clean "rt2661fw.c" rt2661fw.fwo optional rt2661fw | ralfw \ dependency "rt2661.fw" \ compile-with "${NORMAL_FWO}" \ no-implicit-rule \ clean "rt2661fw.fwo" rt2661.fw optional rt2661fw | ralfw \ dependency "$S/contrib/dev/ral/rt2661.fw.uu" \ compile-with "${NORMAL_FW}" \ no-obj no-implicit-rule \ clean "rt2661.fw" rt2860fw.c optional rt2860fw | ralfw \ compile-with "${AWK} -f $S/tools/fw_stub.awk rt2860.fw:rt2860fw -mrt2860 -c${.TARGET}" \ no-implicit-rule before-depend local \ clean "rt2860fw.c" rt2860fw.fwo optional rt2860fw | ralfw \ dependency "rt2860.fw" \ compile-with "${NORMAL_FWO}" \ no-implicit-rule \ clean "rt2860fw.fwo" rt2860.fw optional rt2860fw | ralfw \ dependency "$S/contrib/dev/ral/rt2860.fw.uu" \ compile-with "${NORMAL_FW}" \ no-obj no-implicit-rule \ clean "rt2860.fw" dev/random/random_infra.c standard dev/random/random_harvestq.c standard dev/random/randomdev.c optional !random_loadable dev/random/fortuna.c optional !random_loadable dev/random/hash.c optional !random_loadable dev/rc/rc.c optional rc dev/rccgpio/rccgpio.c optional rccgpio gpio dev/re/if_re.c optional re dev/rl/if_rl.c optional rl pci dev/rndtest/rndtest.c optional rndtest dev/rp/rp.c optional rp dev/rp/rp_isa.c optional rp isa dev/rp/rp_pci.c optional rp pci # dev/rtwn/if_rtwn.c optional rtwn dev/rtwn/if_rtwn_beacon.c optional rtwn dev/rtwn/if_rtwn_calib.c optional rtwn dev/rtwn/if_rtwn_cam.c optional rtwn dev/rtwn/if_rtwn_efuse.c optional rtwn dev/rtwn/if_rtwn_fw.c optional rtwn dev/rtwn/if_rtwn_rx.c optional rtwn dev/rtwn/if_rtwn_task.c optional rtwn dev/rtwn/if_rtwn_tx.c optional rtwn # dev/rtwn/pci/rtwn_pci_attach.c optional rtwn_pci pci dev/rtwn/pci/rtwn_pci_reg.c optional rtwn_pci pci dev/rtwn/pci/rtwn_pci_rx.c optional rtwn_pci pci dev/rtwn/pci/rtwn_pci_tx.c optional rtwn_pci pci # dev/rtwn/usb/rtwn_usb_attach.c optional rtwn_usb dev/rtwn/usb/rtwn_usb_ep.c optional rtwn_usb dev/rtwn/usb/rtwn_usb_reg.c optional rtwn_usb dev/rtwn/usb/rtwn_usb_rx.c optional rtwn_usb dev/rtwn/usb/rtwn_usb_tx.c optional rtwn_usb # RTL8188E dev/rtwn/rtl8188e/r88e_beacon.c optional rtwn dev/rtwn/rtl8188e/r88e_calib.c optional rtwn dev/rtwn/rtl8188e/r88e_chan.c optional rtwn dev/rtwn/rtl8188e/r88e_fw.c optional rtwn dev/rtwn/rtl8188e/r88e_init.c optional rtwn dev/rtwn/rtl8188e/r88e_led.c optional rtwn dev/rtwn/rtl8188e/r88e_tx.c optional rtwn dev/rtwn/rtl8188e/r88e_rf.c optional rtwn dev/rtwn/rtl8188e/r88e_rom.c optional rtwn dev/rtwn/rtl8188e/r88e_rx.c optional rtwn dev/rtwn/rtl8188e/pci/r88ee_attach.c optional rtwn_pci pci dev/rtwn/rtl8188e/pci/r88ee_init.c optional rtwn_pci pci dev/rtwn/rtl8188e/pci/r88ee_rx.c optional rtwn_pci pci dev/rtwn/rtl8188e/usb/r88eu_attach.c optional rtwn_usb dev/rtwn/rtl8188e/usb/r88eu_init.c optional rtwn_usb # RTL8192C dev/rtwn/rtl8192c/r92c_attach.c optional rtwn dev/rtwn/rtl8192c/r92c_beacon.c optional rtwn dev/rtwn/rtl8192c/r92c_calib.c optional rtwn dev/rtwn/rtl8192c/r92c_chan.c optional rtwn dev/rtwn/rtl8192c/r92c_fw.c optional rtwn dev/rtwn/rtl8192c/r92c_init.c optional rtwn dev/rtwn/rtl8192c/r92c_llt.c optional rtwn dev/rtwn/rtl8192c/r92c_rf.c optional rtwn dev/rtwn/rtl8192c/r92c_rom.c optional rtwn dev/rtwn/rtl8192c/r92c_rx.c optional rtwn dev/rtwn/rtl8192c/r92c_tx.c optional rtwn dev/rtwn/rtl8192c/pci/r92ce_attach.c optional rtwn_pci pci dev/rtwn/rtl8192c/pci/r92ce_calib.c optional rtwn_pci pci dev/rtwn/rtl8192c/pci/r92ce_fw.c optional rtwn_pci pci dev/rtwn/rtl8192c/pci/r92ce_init.c optional rtwn_pci pci dev/rtwn/rtl8192c/pci/r92ce_led.c optional rtwn_pci pci dev/rtwn/rtl8192c/pci/r92ce_rx.c optional rtwn_pci pci dev/rtwn/rtl8192c/pci/r92ce_tx.c optional rtwn_pci pci dev/rtwn/rtl8192c/usb/r92cu_attach.c optional rtwn_usb dev/rtwn/rtl8192c/usb/r92cu_init.c optional rtwn_usb dev/rtwn/rtl8192c/usb/r92cu_led.c optional rtwn_usb dev/rtwn/rtl8192c/usb/r92cu_rx.c optional rtwn_usb dev/rtwn/rtl8192c/usb/r92cu_tx.c optional rtwn_usb # RTL8192E dev/rtwn/rtl8192e/r92e_chan.c optional rtwn dev/rtwn/rtl8192e/r92e_fw.c optional rtwn dev/rtwn/rtl8192e/r92e_init.c optional rtwn dev/rtwn/rtl8192e/r92e_led.c optional rtwn dev/rtwn/rtl8192e/r92e_rf.c optional rtwn dev/rtwn/rtl8192e/r92e_rom.c optional rtwn dev/rtwn/rtl8192e/r92e_rx.c optional rtwn dev/rtwn/rtl8192e/usb/r92eu_attach.c optional rtwn_usb dev/rtwn/rtl8192e/usb/r92eu_init.c optional rtwn_usb # RTL8812A dev/rtwn/rtl8812a/r12a_beacon.c optional rtwn dev/rtwn/rtl8812a/r12a_calib.c optional rtwn dev/rtwn/rtl8812a/r12a_caps.c optional rtwn dev/rtwn/rtl8812a/r12a_chan.c optional rtwn dev/rtwn/rtl8812a/r12a_fw.c optional rtwn dev/rtwn/rtl8812a/r12a_init.c optional rtwn dev/rtwn/rtl8812a/r12a_led.c optional rtwn dev/rtwn/rtl8812a/r12a_rf.c optional rtwn dev/rtwn/rtl8812a/r12a_rom.c optional rtwn dev/rtwn/rtl8812a/r12a_rx.c optional rtwn dev/rtwn/rtl8812a/r12a_tx.c optional rtwn dev/rtwn/rtl8812a/usb/r12au_attach.c optional rtwn_usb dev/rtwn/rtl8812a/usb/r12au_init.c optional rtwn_usb dev/rtwn/rtl8812a/usb/r12au_rx.c optional rtwn_usb dev/rtwn/rtl8812a/usb/r12au_tx.c optional rtwn_usb # RTL8821A dev/rtwn/rtl8821a/r21a_beacon.c optional rtwn dev/rtwn/rtl8821a/r21a_calib.c optional rtwn dev/rtwn/rtl8821a/r21a_chan.c optional rtwn dev/rtwn/rtl8821a/r21a_fw.c optional rtwn dev/rtwn/rtl8821a/r21a_init.c optional rtwn dev/rtwn/rtl8821a/r21a_led.c optional rtwn dev/rtwn/rtl8821a/r21a_rom.c optional rtwn dev/rtwn/rtl8821a/r21a_rx.c optional rtwn dev/rtwn/rtl8821a/usb/r21au_attach.c optional rtwn_usb dev/rtwn/rtl8821a/usb/r21au_dfs.c optional rtwn_usb dev/rtwn/rtl8821a/usb/r21au_init.c optional rtwn_usb rtwn-rtl8188eefw.c optional rtwn-rtl8188eefw | rtwnfw \ compile-with "${AWK} -f $S/tools/fw_stub.awk rtwn-rtl8188eefw.fw:rtwn-rtl8188eefw:111 -mrtwn-rtl8188eefw -c${.TARGET}" \ no-implicit-rule before-depend local \ clean "rtwn-rtl8188eefw.c" rtwn-rtl8188eefw.fwo optional rtwn-rtl8188eefw | rtwnfw \ dependency "rtwn-rtl8188eefw.fw" \ compile-with "${NORMAL_FWO}" \ no-implicit-rule \ clean "rtwn-rtl8188eefw.fwo" rtwn-rtl8188eefw.fw optional rtwn-rtl8188eefw | rtwnfw \ dependency "$S/contrib/dev/rtwn/rtwn-rtl8188eefw.fw.uu" \ compile-with "${NORMAL_FW}" \ no-obj no-implicit-rule \ clean "rtwn-rtl8188eefw.fw" rtwn-rtl8188eufw.c optional rtwn-rtl8188eufw | rtwnfw \ compile-with "${AWK} -f $S/tools/fw_stub.awk rtwn-rtl8188eufw.fw:rtwn-rtl8188eufw:111 -mrtwn-rtl8188eufw -c${.TARGET}" \ no-implicit-rule before-depend local \ clean "rtwn-rtl8188eufw.c" rtwn-rtl8188eufw.fwo optional rtwn-rtl8188eufw | rtwnfw \ dependency "rtwn-rtl8188eufw.fw" \ compile-with "${NORMAL_FWO}" \ no-implicit-rule \ clean "rtwn-rtl8188eufw.fwo" rtwn-rtl8188eufw.fw optional rtwn-rtl8188eufw | rtwnfw \ dependency "$S/contrib/dev/rtwn/rtwn-rtl8188eufw.fw.uu" \ compile-with "${NORMAL_FW}" \ no-obj no-implicit-rule \ clean "rtwn-rtl8188eufw.fw" rtwn-rtl8192cfwE.c optional rtwn-rtl8192cfwE | rtwnfw \ compile-with "${AWK} -f $S/tools/fw_stub.awk rtwn-rtl8192cfwE.fw:rtwn-rtl8192cfwE:111 -mrtwn-rtl8192cfwE -c${.TARGET}" \ no-implicit-rule before-depend local \ clean "rtwn-rtl8192cfwE.c" rtwn-rtl8192cfwE.fwo optional rtwn-rtl8192cfwE | rtwnfw \ dependency "rtwn-rtl8192cfwE.fw" \ compile-with "${NORMAL_FWO}" \ no-implicit-rule \ clean "rtwn-rtl8192cfwE.fwo" rtwn-rtl8192cfwE.fw optional rtwn-rtl8192cfwE | rtwnfw \ dependency "$S/contrib/dev/rtwn/rtwn-rtl8192cfwE.fw.uu" \ compile-with "${NORMAL_FW}" \ no-obj no-implicit-rule \ clean "rtwn-rtl8192cfwE.fw" rtwn-rtl8192cfwE_B.c optional rtwn-rtl8192cfwE_B | rtwnfw \ compile-with "${AWK} -f $S/tools/fw_stub.awk rtwn-rtl8192cfwE_B.fw:rtwn-rtl8192cfwE_B:111 -mrtwn-rtl8192cfwE_B -c${.TARGET}" \ no-implicit-rule before-depend local \ clean "rtwn-rtl8192cfwE_B.c" rtwn-rtl8192cfwE_B.fwo optional rtwn-rtl8192cfwE_B | rtwnfw \ dependency "rtwn-rtl8192cfwE_B.fw" \ compile-with "${NORMAL_FWO}" \ no-implicit-rule \ clean "rtwn-rtl8192cfwE_B.fwo" rtwn-rtl8192cfwE_B.fw optional rtwn-rtl8192cfwE_B | rtwnfw \ dependency "$S/contrib/dev/rtwn/rtwn-rtl8192cfwE_B.fw.uu" \ compile-with "${NORMAL_FW}" \ no-obj no-implicit-rule \ clean "rtwn-rtl8192cfwE_B.fw" rtwn-rtl8192cfwT.c optional rtwn-rtl8192cfwT | rtwnfw \ compile-with "${AWK} -f $S/tools/fw_stub.awk rtwn-rtl8192cfwT.fw:rtwn-rtl8192cfwT:111 -mrtwn-rtl8192cfwT -c${.TARGET}" \ no-implicit-rule before-depend local \ clean "rtwn-rtl8192cfwT.c" rtwn-rtl8192cfwT.fwo optional rtwn-rtl8192cfwT | rtwnfw \ dependency "rtwn-rtl8192cfwT.fw" \ compile-with "${NORMAL_FWO}" \ no-implicit-rule \ clean "rtwn-rtl8192cfwT.fwo" rtwn-rtl8192cfwT.fw optional rtwn-rtl8192cfwT | rtwnfw \ dependency "$S/contrib/dev/rtwn/rtwn-rtl8192cfwT.fw.uu" \ compile-with "${NORMAL_FW}" \ no-obj no-implicit-rule \ clean "rtwn-rtl8192cfwT.fw" rtwn-rtl8192cfwU.c optional rtwn-rtl8192cfwU | rtwnfw \ compile-with "${AWK} -f $S/tools/fw_stub.awk rtwn-rtl8192cfwU.fw:rtwn-rtl8192cfwU:111 -mrtwn-rtl8192cfwU -c${.TARGET}" \ no-implicit-rule before-depend local \ clean "rtwn-rtl8192cfwU.c" rtwn-rtl8192cfwU.fwo optional rtwn-rtl8192cfwU | rtwnfw \ dependency "rtwn-rtl8192cfwU.fw" \ compile-with "${NORMAL_FWO}" \ no-implicit-rule \ clean "rtwn-rtl8192cfwU.fwo" rtwn-rtl8192cfwU.fw optional rtwn-rtl8192cfwU | rtwnfw \ dependency "$S/contrib/dev/rtwn/rtwn-rtl8192cfwU.fw.uu" \ compile-with "${NORMAL_FW}" \ no-obj no-implicit-rule \ clean "rtwn-rtl8192cfwU.fw" rtwn-rtl8192eufw.c optional rtwn-rtl8192eufw | rtwnfw \ compile-with "${AWK} -f $S/tools/fw_stub.awk rtwn-rtl8192eufw.fw:rtwn-rtl8192eufw:111 -mrtwn-rtl8192eufw -c${.TARGET}" \ no-implicit-rule before-depend local \ clean "rtwn-rtl8192eufw.c" rtwn-rtl8192eufw.fwo optional rtwn-rtl8192eufw | rtwnfw \ dependency "rtwn-rtl8192eufw.fw" \ compile-with "${NORMAL_FWO}" \ no-implicit-rule \ clean "rtwn-rtl8192eufw.fwo" rtwn-rtl8192eufw.fw optional rtwn-rtl8192eufw | rtwnfw \ dependency "$S/contrib/dev/rtwn/rtwn-rtl8192eufw.fw.uu" \ compile-with "${NORMAL_FW}" \ no-obj no-implicit-rule \ clean "rtwn-rtl8192eufw.fw" rtwn-rtl8812aufw.c optional rtwn-rtl8812aufw | rtwnfw \ compile-with "${AWK} -f $S/tools/fw_stub.awk rtwn-rtl8812aufw.fw:rtwn-rtl8812aufw:111 -mrtwn-rtl8812aufw -c${.TARGET}" \ no-implicit-rule before-depend local \ clean "rtwn-rtl8812aufw.c" rtwn-rtl8812aufw.fwo optional rtwn-rtl8812aufw | rtwnfw \ dependency "rtwn-rtl8812aufw.fw" \ compile-with "${NORMAL_FWO}" \ no-implicit-rule \ clean "rtwn-rtl8812aufw.fwo" rtwn-rtl8812aufw.fw optional rtwn-rtl8812aufw | rtwnfw \ dependency "$S/contrib/dev/rtwn/rtwn-rtl8812aufw.fw.uu" \ compile-with "${NORMAL_FW}" \ no-obj no-implicit-rule \ clean "rtwn-rtl8812aufw.fw" rtwn-rtl8821aufw.c optional rtwn-rtl8821aufw | rtwnfw \ compile-with "${AWK} -f $S/tools/fw_stub.awk rtwn-rtl8821aufw.fw:rtwn-rtl8821aufw:111 -mrtwn-rtl8821aufw -c${.TARGET}" \ no-implicit-rule before-depend local \ clean "rtwn-rtl8821aufw.c" rtwn-rtl8821aufw.fwo optional rtwn-rtl8821aufw | rtwnfw \ dependency "rtwn-rtl8821aufw.fw" \ compile-with "${NORMAL_FWO}" \ no-implicit-rule \ clean "rtwn-rtl8821aufw.fwo" rtwn-rtl8821aufw.fw optional rtwn-rtl8821aufw | rtwnfw \ dependency "$S/contrib/dev/rtwn/rtwn-rtl8821aufw.fw.uu" \ compile-with "${NORMAL_FW}" \ no-obj no-implicit-rule \ clean "rtwn-rtl8821aufw.fw" dev/safe/safe.c optional safe dev/scc/scc_if.m optional scc dev/scc/scc_bfe_quicc.c optional scc quicc dev/scc/scc_core.c optional scc dev/scc/scc_dev_quicc.c optional scc quicc dev/scc/scc_dev_sab82532.c optional scc dev/scc/scc_dev_z8530.c optional scc dev/sdhci/sdhci.c optional sdhci dev/sdhci/sdhci_fdt.c optional sdhci fdt dev/sdhci/sdhci_fdt_gpio.c optional sdhci fdt gpio dev/sdhci/sdhci_if.m optional sdhci dev/sdhci/sdhci_acpi.c optional sdhci acpi dev/sdhci/sdhci_pci.c optional sdhci pci dev/sdio/sdio_if.m optional mmccam dev/sdio/sdio_subr.c optional mmccam dev/sdio/sdiob.c optional mmccam dev/sge/if_sge.c optional sge pci dev/siis/siis.c optional siis pci dev/sis/if_sis.c optional sis pci dev/sk/if_sk.c optional sk pci dev/smbus/smb.c optional smb dev/smbus/smbconf.c optional smbus dev/smbus/smbus.c optional smbus dev/smbus/smbus_if.m optional smbus dev/smc/if_smc.c optional smc dev/smc/if_smc_fdt.c optional smc fdt dev/snp/snp.c optional snp dev/sound/clone.c optional sound dev/sound/unit.c optional sound dev/sound/isa/ad1816.c optional snd_ad1816 isa dev/sound/isa/ess.c optional snd_ess isa dev/sound/isa/gusc.c optional snd_gusc isa dev/sound/isa/mss.c optional snd_mss isa dev/sound/isa/sb16.c optional snd_sb16 isa dev/sound/isa/sb8.c optional snd_sb8 isa dev/sound/isa/sbc.c optional snd_sbc isa dev/sound/isa/sndbuf_dma.c optional sound isa dev/sound/pci/als4000.c optional snd_als4000 pci dev/sound/pci/atiixp.c optional snd_atiixp pci dev/sound/pci/cmi.c optional snd_cmi pci dev/sound/pci/cs4281.c optional snd_cs4281 pci dev/sound/pci/csa.c optional snd_csa pci dev/sound/pci/csapcm.c optional snd_csa pci dev/sound/pci/ds1.c optional snd_ds1 pci dev/sound/pci/emu10k1.c optional snd_emu10k1 pci dev/sound/pci/emu10kx.c optional snd_emu10kx pci dev/sound/pci/emu10kx-pcm.c optional snd_emu10kx pci dev/sound/pci/emu10kx-midi.c optional snd_emu10kx pci dev/sound/pci/envy24.c optional snd_envy24 pci dev/sound/pci/envy24ht.c optional snd_envy24ht pci dev/sound/pci/es137x.c optional snd_es137x pci dev/sound/pci/fm801.c optional snd_fm801 pci dev/sound/pci/ich.c optional snd_ich pci dev/sound/pci/maestro.c optional snd_maestro pci dev/sound/pci/maestro3.c optional snd_maestro3 pci dev/sound/pci/neomagic.c optional snd_neomagic pci dev/sound/pci/solo.c optional snd_solo pci dev/sound/pci/spicds.c optional snd_spicds pci dev/sound/pci/t4dwave.c optional snd_t4dwave pci dev/sound/pci/via8233.c optional snd_via8233 pci dev/sound/pci/via82c686.c optional snd_via82c686 pci dev/sound/pci/vibes.c optional snd_vibes pci dev/sound/pci/hda/hdaa.c optional snd_hda pci dev/sound/pci/hda/hdaa_patches.c optional snd_hda pci dev/sound/pci/hda/hdac.c optional snd_hda pci dev/sound/pci/hda/hdac_if.m optional snd_hda pci dev/sound/pci/hda/hdacc.c optional snd_hda pci dev/sound/pci/hdspe.c optional snd_hdspe pci dev/sound/pci/hdspe-pcm.c optional snd_hdspe pci dev/sound/pcm/ac97.c optional sound dev/sound/pcm/ac97_if.m optional sound dev/sound/pcm/ac97_patch.c optional sound dev/sound/pcm/buffer.c optional sound \ dependency "snd_fxdiv_gen.h" dev/sound/pcm/channel.c optional sound dev/sound/pcm/channel_if.m optional sound dev/sound/pcm/dsp.c optional sound dev/sound/pcm/feeder.c optional sound dev/sound/pcm/feeder_chain.c optional sound dev/sound/pcm/feeder_eq.c optional sound \ dependency "feeder_eq_gen.h" \ dependency "snd_fxdiv_gen.h" dev/sound/pcm/feeder_if.m optional sound dev/sound/pcm/feeder_format.c optional sound \ dependency "snd_fxdiv_gen.h" dev/sound/pcm/feeder_matrix.c optional sound \ dependency "snd_fxdiv_gen.h" dev/sound/pcm/feeder_mixer.c optional sound \ dependency "snd_fxdiv_gen.h" dev/sound/pcm/feeder_rate.c optional sound \ dependency "feeder_rate_gen.h" \ dependency "snd_fxdiv_gen.h" dev/sound/pcm/feeder_volume.c optional sound \ dependency "snd_fxdiv_gen.h" dev/sound/pcm/mixer.c optional sound dev/sound/pcm/mixer_if.m optional sound dev/sound/pcm/sndstat.c optional sound dev/sound/pcm/sound.c optional sound dev/sound/pcm/vchan.c optional sound dev/sound/usb/uaudio.c optional snd_uaudio usb dev/sound/usb/uaudio_pcm.c optional snd_uaudio usb dev/sound/midi/midi.c optional sound dev/sound/midi/mpu401.c optional sound dev/sound/midi/mpu_if.m optional sound dev/sound/midi/mpufoi_if.m optional sound dev/sound/midi/sequencer.c optional sound dev/sound/midi/synth_if.m optional sound dev/spibus/ofw_spibus.c optional fdt spibus dev/spibus/spibus.c optional spibus \ dependency "spibus_if.h" dev/spibus/spigen.c optional spigen dev/spibus/spibus_if.m optional spibus dev/ste/if_ste.c optional ste pci dev/stge/if_stge.c optional stge dev/sym/sym_hipd.c optional sym \ dependency "$S/dev/sym/sym_{conf,defs}.h" dev/syscons/blank/blank_saver.c optional blank_saver dev/syscons/daemon/daemon_saver.c optional daemon_saver dev/syscons/dragon/dragon_saver.c optional dragon_saver dev/syscons/fade/fade_saver.c optional fade_saver dev/syscons/fire/fire_saver.c optional fire_saver dev/syscons/green/green_saver.c optional green_saver dev/syscons/logo/logo.c optional logo_saver dev/syscons/logo/logo_saver.c optional logo_saver dev/syscons/rain/rain_saver.c optional rain_saver dev/syscons/schistory.c optional sc dev/syscons/scmouse.c optional sc dev/syscons/scterm.c optional sc dev/syscons/scterm-dumb.c optional sc !SC_NO_TERM_DUMB dev/syscons/scterm-sc.c optional sc !SC_NO_TERM_SC dev/syscons/scterm-teken.c optional sc !SC_NO_TERM_TEKEN dev/syscons/scvidctl.c optional sc dev/syscons/scvtb.c optional sc dev/syscons/snake/snake_saver.c optional snake_saver dev/syscons/star/star_saver.c optional star_saver dev/syscons/syscons.c optional sc dev/syscons/sysmouse.c optional sc dev/syscons/warp/warp_saver.c optional warp_saver dev/tcp_log/tcp_log_dev.c optional tcp_blackbox inet | tcp_blackbox inet6 dev/tdfx/tdfx_linux.c optional tdfx_linux tdfx compat_linux dev/tdfx/tdfx_pci.c optional tdfx pci dev/ti/if_ti.c optional ti pci dev/twa/tw_cl_init.c optional twa \ compile-with "${NORMAL_C} -I$S/dev/twa" dev/twa/tw_cl_intr.c optional twa \ compile-with "${NORMAL_C} -I$S/dev/twa" dev/twa/tw_cl_io.c optional twa \ compile-with "${NORMAL_C} -I$S/dev/twa" dev/twa/tw_cl_misc.c optional twa \ compile-with "${NORMAL_C} -I$S/dev/twa" dev/twa/tw_osl_cam.c optional twa \ compile-with "${NORMAL_C} -I$S/dev/twa" dev/twa/tw_osl_freebsd.c optional twa \ compile-with "${NORMAL_C} -I$S/dev/twa" dev/twe/twe.c optional twe dev/twe/twe_freebsd.c optional twe dev/tws/tws.c optional tws dev/tws/tws_cam.c optional tws dev/tws/tws_hdm.c optional tws dev/tws/tws_services.c optional tws dev/tws/tws_user.c optional tws dev/uart/uart_bus_acpi.c optional uart acpi dev/uart/uart_bus_fdt.c optional uart fdt dev/uart/uart_bus_isa.c optional uart isa dev/uart/uart_bus_pccard.c optional uart pccard dev/uart/uart_bus_pci.c optional uart pci dev/uart/uart_bus_puc.c optional uart puc dev/uart/uart_bus_scc.c optional uart scc dev/uart/uart_core.c optional uart dev/uart/uart_cpu_acpi.c optional uart acpi dev/uart/uart_dbg.c optional uart gdb dev/uart/uart_dev_msm.c optional uart uart_msm fdt dev/uart/uart_dev_mvebu.c optional uart uart_mvebu dev/uart/uart_dev_ns8250.c optional uart uart_ns8250 | uart uart_snps dev/uart/uart_dev_pl011.c optional uart pl011 dev/uart/uart_dev_quicc.c optional uart quicc dev/uart/uart_dev_sab82532.c optional uart uart_sab82532 | uart scc dev/uart/uart_dev_snps.c optional uart uart_snps fdt dev/uart/uart_dev_z8530.c optional uart uart_z8530 | uart scc dev/uart/uart_if.m optional uart dev/uart/uart_subr.c optional uart dev/uart/uart_tty.c optional uart dev/ubsec/ubsec.c optional ubsec # # USB controller drivers # dev/usb/controller/musb_otg.c optional musb dev/usb/controller/dwc_otg.c optional dwcotg dev/usb/controller/dwc_otg_fdt.c optional dwcotg fdt dev/usb/controller/ehci.c optional ehci dev/usb/controller/ehci_msm.c optional ehci_msm fdt dev/usb/controller/ehci_pci.c optional ehci pci dev/usb/controller/ohci.c optional ohci dev/usb/controller/ohci_pci.c optional ohci pci dev/usb/controller/uhci.c optional uhci dev/usb/controller/uhci_pci.c optional uhci pci dev/usb/controller/xhci.c optional xhci dev/usb/controller/xhci_pci.c optional xhci pci dev/usb/controller/saf1761_otg.c optional saf1761otg dev/usb/controller/saf1761_otg_fdt.c optional saf1761otg fdt dev/usb/controller/uss820dci.c optional uss820dci dev/usb/controller/usb_controller.c optional usb # # USB storage drivers # dev/usb/storage/cfumass.c optional cfumass ctl dev/usb/storage/umass.c optional umass dev/usb/storage/urio.c optional urio dev/usb/storage/ustorage_fs.c optional usfs # # USB core # dev/usb/usb_busdma.c optional usb dev/usb/usb_core.c optional usb dev/usb/usb_debug.c optional usb dev/usb/usb_dev.c optional usb dev/usb/usb_device.c optional usb dev/usb/usb_dynamic.c optional usb dev/usb/usb_error.c optional usb dev/usb/usb_fdt_support.c optional usb fdt dev/usb/usb_generic.c optional usb dev/usb/usb_handle_request.c optional usb dev/usb/usb_hid.c optional usb dev/usb/usb_hub.c optional usb dev/usb/usb_hub_acpi.c optional uacpi acpi dev/usb/usb_if.m optional usb dev/usb/usb_lookup.c optional usb dev/usb/usb_mbuf.c optional usb dev/usb/usb_msctest.c optional usb dev/usb/usb_parse.c optional usb dev/usb/usb_pf.c optional usb dev/usb/usb_process.c optional usb dev/usb/usb_request.c optional usb dev/usb/usb_transfer.c optional usb dev/usb/usb_util.c optional usb # # USB network drivers # dev/usb/net/if_aue.c optional aue dev/usb/net/if_axe.c optional axe dev/usb/net/if_axge.c optional axge dev/usb/net/if_cdce.c optional cdce dev/usb/net/if_cdceem.c optional cdceem dev/usb/net/if_cue.c optional cue dev/usb/net/if_ipheth.c optional ipheth dev/usb/net/if_kue.c optional kue dev/usb/net/if_mos.c optional mos dev/usb/net/if_muge.c optional muge dev/usb/net/if_rue.c optional rue dev/usb/net/if_smsc.c optional smsc dev/usb/net/if_udav.c optional udav dev/usb/net/if_ure.c optional ure dev/usb/net/if_usie.c optional usie dev/usb/net/if_urndis.c optional urndis dev/usb/net/ruephy.c optional rue dev/usb/net/usb_ethernet.c optional uether | aue | axe | axge | cdce | \ cdceem | cue | ipheth | kue | mos | \ rue | smsc | udav | ure | urndis | muge dev/usb/net/uhso.c optional uhso # # USB WLAN drivers # dev/usb/wlan/if_rsu.c optional rsu rsu-rtl8712fw.c optional rsu-rtl8712fw | rsufw \ compile-with "${AWK} -f $S/tools/fw_stub.awk rsu-rtl8712fw.fw:rsu-rtl8712fw:120 -mrsu-rtl8712fw -c${.TARGET}" \ no-implicit-rule before-depend local \ clean "rsu-rtl8712fw.c" rsu-rtl8712fw.fwo optional rsu-rtl8712fw | rsufw \ dependency "rsu-rtl8712fw.fw" \ compile-with "${NORMAL_FWO}" \ no-implicit-rule \ clean "rsu-rtl8712fw.fwo" rsu-rtl8712fw.fw optional rsu-rtl8712.fw | rsufw \ dependency "$S/contrib/dev/rsu/rsu-rtl8712fw.fw.uu" \ compile-with "${NORMAL_FW}" \ no-obj no-implicit-rule \ clean "rsu-rtl8712fw.fw" dev/usb/wlan/if_rum.c optional rum dev/usb/wlan/if_run.c optional run runfw.c optional runfw \ compile-with "${AWK} -f $S/tools/fw_stub.awk run.fw:runfw -mrunfw -c${.TARGET}" \ no-implicit-rule before-depend local \ clean "runfw.c" runfw.fwo optional runfw \ dependency "run.fw" \ compile-with "${NORMAL_FWO}" \ no-implicit-rule \ clean "runfw.fwo" run.fw optional runfw \ dependency "$S/contrib/dev/run/rt2870.fw.uu" \ compile-with "${NORMAL_FW}" \ no-obj no-implicit-rule \ clean "run.fw" dev/usb/wlan/if_uath.c optional uath dev/usb/wlan/if_upgt.c optional upgt dev/usb/wlan/if_ural.c optional ural dev/usb/wlan/if_urtw.c optional urtw dev/usb/wlan/if_zyd.c optional zyd # # USB serial and parallel port drivers # dev/usb/serial/u3g.c optional u3g dev/usb/serial/uark.c optional uark dev/usb/serial/ubsa.c optional ubsa dev/usb/serial/ubser.c optional ubser dev/usb/serial/uchcom.c optional uchcom dev/usb/serial/ucycom.c optional ucycom dev/usb/serial/ufoma.c optional ufoma dev/usb/serial/uftdi.c optional uftdi dev/usb/serial/ugensa.c optional ugensa dev/usb/serial/uipaq.c optional uipaq dev/usb/serial/ulpt.c optional ulpt dev/usb/serial/umcs.c optional umcs dev/usb/serial/umct.c optional umct dev/usb/serial/umodem.c optional umodem dev/usb/serial/umoscom.c optional umoscom dev/usb/serial/uplcom.c optional uplcom dev/usb/serial/uslcom.c optional uslcom dev/usb/serial/uvisor.c optional uvisor dev/usb/serial/uvscom.c optional uvscom dev/usb/serial/usb_serial.c optional ucom | u3g | uark | ubsa | ubser | \ uchcom | ucycom | ufoma | uftdi | \ ugensa | uipaq | umcs | umct | \ umodem | umoscom | uplcom | usie | \ uslcom | uvisor | uvscom # # USB misc drivers # dev/usb/misc/ufm.c optional ufm dev/usb/misc/udbp.c optional udbp dev/usb/misc/ugold.c optional ugold dev/usb/misc/uled.c optional uled # # USB input drivers # dev/usb/input/atp.c optional atp dev/usb/input/uep.c optional uep dev/usb/input/uhid.c optional uhid dev/usb/input/uhid_snes.c optional uhid_snes dev/usb/input/ukbd.c optional ukbd dev/usb/input/ums.c optional ums dev/usb/input/wmt.c optional wmt dev/usb/input/wsp.c optional wsp # # USB quirks # dev/usb/quirk/usb_quirk.c optional usb # # USB templates # dev/usb/template/usb_template.c optional usb_template dev/usb/template/usb_template_audio.c optional usb_template dev/usb/template/usb_template_cdce.c optional usb_template dev/usb/template/usb_template_kbd.c optional usb_template dev/usb/template/usb_template_modem.c optional usb_template dev/usb/template/usb_template_mouse.c optional usb_template dev/usb/template/usb_template_msc.c optional usb_template dev/usb/template/usb_template_mtp.c optional usb_template dev/usb/template/usb_template_phone.c optional usb_template dev/usb/template/usb_template_serialnet.c optional usb_template dev/usb/template/usb_template_midi.c optional usb_template dev/usb/template/usb_template_multi.c optional usb_template dev/usb/template/usb_template_cdceem.c optional usb_template # # USB video drivers # dev/usb/video/udl.c optional udl # # USB END # dev/videomode/videomode.c optional videomode dev/videomode/edid.c optional videomode dev/videomode/pickmode.c optional videomode dev/videomode/vesagtf.c optional videomode dev/veriexec/verified_exec.c optional veriexec mac_veriexec dev/vge/if_vge.c optional vge dev/viapm/viapm.c optional viapm pci dev/virtio/virtio.c optional virtio dev/virtio/virtqueue.c optional virtio dev/virtio/virtio_bus_if.m optional virtio dev/virtio/virtio_if.m optional virtio dev/virtio/pci/virtio_pci.c optional virtio_pci dev/virtio/mmio/virtio_mmio.c optional virtio_mmio dev/virtio/mmio/virtio_mmio_acpi.c optional virtio_mmio acpi dev/virtio/mmio/virtio_mmio_fdt.c optional virtio_mmio fdt dev/virtio/mmio/virtio_mmio_if.m optional virtio_mmio dev/virtio/network/if_vtnet.c optional vtnet dev/virtio/block/virtio_blk.c optional virtio_blk dev/virtio/balloon/virtio_balloon.c optional virtio_balloon dev/virtio/scsi/virtio_scsi.c optional virtio_scsi dev/virtio/random/virtio_random.c optional virtio_random dev/virtio/console/virtio_console.c optional virtio_console dev/vkbd/vkbd.c optional vkbd dev/vmgenc/vmgenc_acpi.c optional acpi dev/vr/if_vr.c optional vr pci dev/vt/colors/vt_termcolors.c optional vt dev/vt/font/vt_font_default.c optional vt dev/vt/font/vt_mouse_cursor.c optional vt dev/vt/hw/efifb/efifb.c optional vt_efifb dev/vt/hw/fb/vt_fb.c optional vt dev/vt/hw/vga/vt_vga.c optional vt vt_vga dev/vt/logo/logo_freebsd.c optional vt splash dev/vt/logo/logo_beastie.c optional vt splash dev/vt/vt_buf.c optional vt dev/vt/vt_consolectl.c optional vt dev/vt/vt_core.c optional vt dev/vt/vt_cpulogos.c optional vt splash dev/vt/vt_font.c optional vt dev/vt/vt_sysmouse.c optional vt dev/vte/if_vte.c optional vte pci dev/watchdog/watchdog.c standard dev/wi/if_wi.c optional wi dev/wi/if_wi_pccard.c optional wi pccard dev/wi/if_wi_pci.c optional wi pci dev/wpi/if_wpi.c optional wpi pci wpifw.c optional wpifw \ compile-with "${AWK} -f $S/tools/fw_stub.awk wpi.fw:wpifw:153229 -mwpi -c${.TARGET}" \ no-implicit-rule before-depend local \ clean "wpifw.c" wpifw.fwo optional wpifw \ dependency "wpi.fw" \ compile-with "${NORMAL_FWO}" \ no-implicit-rule \ clean "wpifw.fwo" wpi.fw optional wpifw \ dependency "$S/contrib/dev/wpi/iwlwifi-3945-15.32.2.9.fw.uu" \ compile-with "${NORMAL_FW}" \ no-obj no-implicit-rule \ clean "wpi.fw" dev/xdma/controller/pl330.c optional xdma pl330 dev/xdma/xdma.c optional xdma dev/xdma/xdma_bank.c optional xdma dev/xdma/xdma_bio.c optional xdma dev/xdma/xdma_fdt_test.c optional xdma xdma_test fdt dev/xdma/xdma_if.m optional xdma dev/xdma/xdma_iommu.c optional xdma dev/xdma/xdma_mbuf.c optional xdma dev/xdma/xdma_queue.c optional xdma dev/xdma/xdma_sg.c optional xdma dev/xdma/xdma_sglist.c optional xdma dev/xen/balloon/balloon.c optional xenhvm dev/xen/blkfront/blkfront.c optional xenhvm dev/xen/blkback/blkback.c optional xenhvm dev/xen/console/xen_console.c optional xenhvm dev/xen/control/control.c optional xenhvm dev/xen/grant_table/grant_table.c optional xenhvm dev/xen/netback/netback.c optional xenhvm dev/xen/netfront/netfront.c optional xenhvm dev/xen/xenpci/xenpci.c optional xenpci dev/xen/timer/timer.c optional xenhvm dev/xen/pvcpu/pvcpu.c optional xenhvm dev/xen/xenstore/xenstore.c optional xenhvm dev/xen/xenstore/xenstore_dev.c optional xenhvm dev/xen/xenstore/xenstored_dev.c optional xenhvm dev/xen/evtchn/evtchn_dev.c optional xenhvm dev/xen/privcmd/privcmd.c optional xenhvm dev/xen/gntdev/gntdev.c optional xenhvm dev/xen/debug/debug.c optional xenhvm dev/xl/if_xl.c optional xl pci dev/xl/xlphy.c optional xl pci fs/autofs/autofs.c optional autofs fs/autofs/autofs_vfsops.c optional autofs fs/autofs/autofs_vnops.c optional autofs fs/deadfs/dead_vnops.c standard fs/devfs/devfs_devs.c standard fs/devfs/devfs_dir.c standard fs/devfs/devfs_rule.c standard fs/devfs/devfs_vfsops.c standard fs/devfs/devfs_vnops.c standard fs/fdescfs/fdesc_vfsops.c optional fdescfs fs/fdescfs/fdesc_vnops.c optional fdescfs fs/fifofs/fifo_vnops.c standard fs/cuse/cuse.c optional cuse fs/fuse/fuse_device.c optional fusefs fs/fuse/fuse_file.c optional fusefs fs/fuse/fuse_internal.c optional fusefs fs/fuse/fuse_io.c optional fusefs fs/fuse/fuse_ipc.c optional fusefs fs/fuse/fuse_main.c optional fusefs fs/fuse/fuse_node.c optional fusefs fs/fuse/fuse_vfsops.c optional fusefs fs/fuse/fuse_vnops.c optional fusefs +fs/mntfs/mntfs_vnops.c standard fs/msdosfs/msdosfs_conv.c optional msdosfs fs/msdosfs/msdosfs_denode.c optional msdosfs fs/msdosfs/msdosfs_fat.c optional msdosfs fs/msdosfs/msdosfs_iconv.c optional msdosfs_iconv fs/msdosfs/msdosfs_lookup.c optional msdosfs fs/msdosfs/msdosfs_vfsops.c optional msdosfs fs/msdosfs/msdosfs_vnops.c optional msdosfs fs/nfs/nfs_commonkrpc.c optional nfscl | nfsd fs/nfs/nfs_commonsubs.c optional nfscl | nfsd fs/nfs/nfs_commonport.c optional nfscl | nfsd fs/nfs/nfs_commonacl.c optional nfscl | nfsd fs/nfsclient/nfs_clcomsubs.c optional nfscl fs/nfsclient/nfs_clsubs.c optional nfscl fs/nfsclient/nfs_clstate.c optional nfscl fs/nfsclient/nfs_clkrpc.c optional nfscl fs/nfsclient/nfs_clrpcops.c optional nfscl fs/nfsclient/nfs_clvnops.c optional nfscl fs/nfsclient/nfs_clnode.c optional nfscl fs/nfsclient/nfs_clvfsops.c optional nfscl fs/nfsclient/nfs_clport.c optional nfscl fs/nfsclient/nfs_clbio.c optional nfscl fs/nfsclient/nfs_clnfsiod.c optional nfscl fs/nfsserver/nfs_fha_new.c optional nfsd inet fs/nfsserver/nfs_nfsdsocket.c optional nfsd inet fs/nfsserver/nfs_nfsdsubs.c optional nfsd inet fs/nfsserver/nfs_nfsdstate.c optional nfsd inet fs/nfsserver/nfs_nfsdkrpc.c optional nfsd inet fs/nfsserver/nfs_nfsdserv.c optional nfsd inet fs/nfsserver/nfs_nfsdport.c optional nfsd inet fs/nfsserver/nfs_nfsdcache.c optional nfsd inet fs/nullfs/null_subr.c optional nullfs fs/nullfs/null_vfsops.c optional nullfs fs/nullfs/null_vnops.c optional nullfs fs/procfs/procfs.c optional procfs fs/procfs/procfs_dbregs.c optional procfs fs/procfs/procfs_fpregs.c optional procfs fs/procfs/procfs_ioctl.c optional procfs fs/procfs/procfs_map.c optional procfs fs/procfs/procfs_mem.c optional procfs fs/procfs/procfs_note.c optional procfs fs/procfs/procfs_osrel.c optional procfs fs/procfs/procfs_regs.c optional procfs fs/procfs/procfs_rlimit.c optional procfs fs/procfs/procfs_status.c optional procfs fs/procfs/procfs_type.c optional procfs fs/pseudofs/pseudofs.c optional pseudofs fs/pseudofs/pseudofs_fileno.c optional pseudofs fs/pseudofs/pseudofs_vncache.c optional pseudofs fs/pseudofs/pseudofs_vnops.c optional pseudofs fs/smbfs/smbfs_io.c optional smbfs fs/smbfs/smbfs_node.c optional smbfs fs/smbfs/smbfs_smb.c optional smbfs fs/smbfs/smbfs_subr.c optional smbfs fs/smbfs/smbfs_vfsops.c optional smbfs fs/smbfs/smbfs_vnops.c optional smbfs fs/udf/osta.c optional udf fs/udf/udf_iconv.c optional udf_iconv fs/udf/udf_vfsops.c optional udf fs/udf/udf_vnops.c optional udf fs/unionfs/union_subr.c optional unionfs fs/unionfs/union_vfsops.c optional unionfs fs/unionfs/union_vnops.c optional unionfs fs/tmpfs/tmpfs_vnops.c optional tmpfs fs/tmpfs/tmpfs_fifoops.c optional tmpfs fs/tmpfs/tmpfs_vfsops.c optional tmpfs fs/tmpfs/tmpfs_subr.c optional tmpfs gdb/gdb_cons.c optional gdb gdb/gdb_main.c optional gdb gdb/gdb_packet.c optional gdb gdb/netgdb.c optional ddb debugnet gdb netgdb inet geom/bde/g_bde.c optional geom_bde geom/bde/g_bde_crypt.c optional geom_bde geom/bde/g_bde_lock.c optional geom_bde geom/bde/g_bde_work.c optional geom_bde geom/cache/g_cache.c optional geom_cache geom/concat/g_concat.c optional geom_concat geom/eli/g_eli.c optional geom_eli geom/eli/g_eli_crypto.c optional geom_eli geom/eli/g_eli_ctl.c optional geom_eli geom/eli/g_eli_hmac.c optional geom_eli geom/eli/g_eli_integrity.c optional geom_eli geom/eli/g_eli_key.c optional geom_eli geom/eli/g_eli_key_cache.c optional geom_eli geom/eli/g_eli_privacy.c optional geom_eli geom/eli/pkcs5v2.c optional geom_eli geom/gate/g_gate.c optional geom_gate geom/geom_bsd_enc.c optional geom_part_bsd geom/geom_ccd.c optional ccd | geom_ccd geom/geom_ctl.c standard geom/geom_dev.c standard geom/geom_disk.c standard geom/geom_dump.c standard geom/geom_event.c standard geom/geom_flashmap.c optional fdt cfi | fdt mx25l | mmcsd | fdt n25q | fdt at45d geom/geom_io.c standard geom/geom_kern.c standard geom/geom_map.c optional geom_map geom/geom_redboot.c optional geom_redboot geom/geom_slice.c standard geom/geom_subr.c standard geom/geom_vfs.c standard geom/journal/g_journal.c optional geom_journal geom/journal/g_journal_ufs.c optional geom_journal geom/label/g_label.c optional geom_label | geom_label_gpt geom/label/g_label_ext2fs.c optional geom_label geom/label/g_label_flashmap.c optional geom_label geom/label/g_label_iso9660.c optional geom_label geom/label/g_label_msdosfs.c optional geom_label geom/label/g_label_ntfs.c optional geom_label geom/label/g_label_reiserfs.c optional geom_label geom/label/g_label_ufs.c optional geom_label geom/label/g_label_gpt.c optional geom_label | geom_label_gpt geom/label/g_label_disk_ident.c optional geom_label geom/linux_lvm/g_linux_lvm.c optional geom_linux_lvm geom/mirror/g_mirror.c optional geom_mirror geom/mirror/g_mirror_ctl.c optional geom_mirror geom/mountver/g_mountver.c optional geom_mountver geom/multipath/g_multipath.c optional geom_multipath geom/nop/g_nop.c optional geom_nop geom/part/g_part.c standard geom/part/g_part_if.m standard geom/part/g_part_apm.c optional geom_part_apm geom/part/g_part_bsd.c optional geom_part_bsd geom/part/g_part_bsd64.c optional geom_part_bsd64 geom/part/g_part_ebr.c optional geom_part_ebr geom/part/g_part_gpt.c optional geom_part_gpt geom/part/g_part_ldm.c optional geom_part_ldm geom/part/g_part_mbr.c optional geom_part_mbr geom/part/g_part_vtoc8.c optional geom_part_vtoc8 geom/raid/g_raid.c optional geom_raid geom/raid/g_raid_ctl.c optional geom_raid geom/raid/g_raid_md_if.m optional geom_raid geom/raid/g_raid_tr_if.m optional geom_raid geom/raid/md_ddf.c optional geom_raid geom/raid/md_intel.c optional geom_raid geom/raid/md_jmicron.c optional geom_raid geom/raid/md_nvidia.c optional geom_raid geom/raid/md_promise.c optional geom_raid geom/raid/md_sii.c optional geom_raid geom/raid/tr_concat.c optional geom_raid geom/raid/tr_raid0.c optional geom_raid geom/raid/tr_raid1.c optional geom_raid geom/raid/tr_raid1e.c optional geom_raid geom/raid/tr_raid5.c optional geom_raid geom/raid3/g_raid3.c optional geom_raid3 geom/raid3/g_raid3_ctl.c optional geom_raid3 geom/shsec/g_shsec.c optional geom_shsec geom/stripe/g_stripe.c optional geom_stripe geom/uzip/g_uzip.c optional geom_uzip geom/uzip/g_uzip_lzma.c optional geom_uzip geom/uzip/g_uzip_wrkthr.c optional geom_uzip geom/uzip/g_uzip_zlib.c optional geom_uzip geom/uzip/g_uzip_zstd.c optional geom_uzip zstdio \ compile-with "${NORMAL_C} -I$S/contrib/zstd/lib/freebsd" geom/vinum/geom_vinum.c optional geom_vinum geom/vinum/geom_vinum_create.c optional geom_vinum geom/vinum/geom_vinum_drive.c optional geom_vinum geom/vinum/geom_vinum_plex.c optional geom_vinum geom/vinum/geom_vinum_volume.c optional geom_vinum geom/vinum/geom_vinum_subr.c optional geom_vinum geom/vinum/geom_vinum_raid5.c optional geom_vinum geom/vinum/geom_vinum_share.c optional geom_vinum geom/vinum/geom_vinum_list.c optional geom_vinum geom/vinum/geom_vinum_rm.c optional geom_vinum geom/vinum/geom_vinum_init.c optional geom_vinum geom/vinum/geom_vinum_state.c optional geom_vinum geom/vinum/geom_vinum_rename.c optional geom_vinum geom/vinum/geom_vinum_move.c optional geom_vinum geom/vinum/geom_vinum_events.c optional geom_vinum geom/virstor/binstream.c optional geom_virstor geom/virstor/g_virstor.c optional geom_virstor geom/virstor/g_virstor_md.c optional geom_virstor geom/zero/g_zero.c optional geom_zero fs/ext2fs/ext2_acl.c optional ext2fs fs/ext2fs/ext2_alloc.c optional ext2fs fs/ext2fs/ext2_balloc.c optional ext2fs fs/ext2fs/ext2_bmap.c optional ext2fs fs/ext2fs/ext2_csum.c optional ext2fs fs/ext2fs/ext2_extattr.c optional ext2fs fs/ext2fs/ext2_extents.c optional ext2fs fs/ext2fs/ext2_inode.c optional ext2fs fs/ext2fs/ext2_inode_cnv.c optional ext2fs fs/ext2fs/ext2_hash.c optional ext2fs fs/ext2fs/ext2_htree.c optional ext2fs fs/ext2fs/ext2_lookup.c optional ext2fs fs/ext2fs/ext2_subr.c optional ext2fs fs/ext2fs/ext2_vfsops.c optional ext2fs fs/ext2fs/ext2_vnops.c optional ext2fs # isa/isa_if.m standard isa/isa_common.c optional isa isa/isahint.c optional isa isa/pnp.c optional isa isapnp isa/pnpparse.c optional isa isapnp fs/cd9660/cd9660_bmap.c optional cd9660 fs/cd9660/cd9660_lookup.c optional cd9660 fs/cd9660/cd9660_node.c optional cd9660 fs/cd9660/cd9660_rrip.c optional cd9660 fs/cd9660/cd9660_util.c optional cd9660 fs/cd9660/cd9660_vfsops.c optional cd9660 fs/cd9660/cd9660_vnops.c optional cd9660 fs/cd9660/cd9660_iconv.c optional cd9660_iconv gnu/gcov/gcc_4_7.c optional gcov \ warning "kernel contains GPL licensed gcov support" gnu/gcov/gcov_fs.c optional gcov lindebugfs \ compile-with "${LINUXKPI_C}" gnu/gcov/gcov_subr.c optional gcov kern/bus_if.m standard kern/clock_if.m standard kern/cpufreq_if.m standard kern/device_if.m standard kern/imgact_binmisc.c optional imagact_binmisc kern/imgact_elf.c standard kern/imgact_elf32.c optional compat_freebsd32 kern/imgact_shell.c standard kern/init_main.c standard kern/init_sysent.c standard kern/ksched.c optional _kposix_priority_scheduling kern/kern_acct.c standard kern/kern_alq.c optional alq kern/kern_clock.c standard kern/kern_condvar.c standard kern/kern_conf.c standard kern/kern_cons.c standard kern/kern_cpu.c standard kern/kern_cpuset.c standard kern/kern_context.c standard kern/kern_descrip.c standard kern/kern_dtrace.c optional kdtrace_hooks kern/kern_dump.c standard kern/kern_environment.c standard kern/kern_et.c standard kern/kern_event.c standard kern/kern_exec.c standard kern/kern_exit.c standard kern/kern_fail.c standard kern/kern_ffclock.c standard kern/kern_fork.c standard kern/kern_hhook.c standard kern/kern_idle.c standard kern/kern_intr.c standard kern/kern_jail.c standard kern/kern_kcov.c optional kcov \ compile-with "${NORMAL_C:N-fsanitize*}" kern/kern_khelp.c standard kern/kern_kthread.c standard kern/kern_ktr.c optional ktr kern/kern_ktrace.c standard kern/kern_linker.c standard kern/kern_lock.c standard kern/kern_lockf.c standard kern/kern_lockstat.c optional kdtrace_hooks kern/kern_loginclass.c standard kern/kern_malloc.c standard kern/kern_mbuf.c standard kern/kern_mib.c standard kern/kern_module.c standard kern/kern_mtxpool.c standard kern/kern_mutex.c standard kern/kern_ntptime.c standard kern/kern_osd.c standard kern/kern_physio.c standard kern/kern_pmc.c standard kern/kern_poll.c optional device_polling kern/kern_priv.c standard kern/kern_proc.c standard kern/kern_procctl.c standard kern/kern_prot.c standard kern/kern_racct.c standard kern/kern_rangelock.c standard kern/kern_rctl.c standard kern/kern_resource.c standard kern/kern_rmlock.c standard kern/kern_rwlock.c standard kern/kern_sdt.c optional kdtrace_hooks kern/kern_sema.c standard kern/kern_sendfile.c standard kern/kern_sharedpage.c standard kern/kern_shutdown.c standard kern/kern_sig.c standard kern/kern_switch.c standard kern/kern_sx.c standard kern/kern_synch.c standard kern/kern_syscalls.c standard kern/kern_sysctl.c standard kern/kern_tc.c standard kern/kern_thr.c standard kern/kern_thread.c standard kern/kern_time.c standard kern/kern_timeout.c standard kern/kern_tslog.c optional tslog kern/kern_ubsan.c optional kubsan kern/kern_umtx.c standard kern/kern_uuid.c standard kern/kern_xxx.c standard kern/link_elf.c standard kern/linker_if.m standard kern/md4c.c optional netsmb kern/md5c.c standard kern/p1003_1b.c standard kern/posix4_mib.c standard kern/sched_4bsd.c optional sched_4bsd kern/sched_ule.c optional sched_ule kern/serdev_if.m standard kern/stack_protector.c standard \ compile-with "${NORMAL_C:N-fstack-protector*}" kern/subr_acl_nfs4.c optional ufs_acl | zfs kern/subr_acl_posix1e.c optional ufs_acl kern/subr_autoconf.c standard kern/subr_blist.c standard kern/subr_boot.c standard kern/subr_bus.c standard kern/subr_bus_dma.c standard kern/subr_bufring.c standard kern/subr_capability.c standard kern/subr_clock.c standard kern/subr_compressor.c standard \ compile-with "${NORMAL_C} -I$S/contrib/zstd/lib/freebsd" kern/subr_coverage.c optional coverage \ compile-with "${NORMAL_C:N-fsanitize*}" kern/subr_counter.c standard kern/subr_csan.c optional kcsan \ compile-with "${NORMAL_C:N-fsanitize*}" kern/subr_devstat.c standard kern/subr_disk.c standard kern/subr_early.c standard kern/subr_epoch.c standard kern/subr_eventhandler.c standard kern/subr_fattime.c standard kern/subr_firmware.c optional firmware kern/subr_filter.c standard kern/subr_gtaskqueue.c standard kern/subr_hash.c standard kern/subr_hints.c standard kern/subr_kdb.c standard kern/subr_kobj.c standard kern/subr_lock.c standard kern/subr_log.c standard kern/subr_mchain.c optional libmchain kern/subr_module.c standard kern/subr_msgbuf.c standard kern/subr_param.c standard kern/subr_pcpu.c standard kern/subr_pctrie.c standard kern/subr_pidctrl.c standard kern/subr_power.c standard kern/subr_prf.c standard kern/subr_prof.c standard kern/subr_rangeset.c standard kern/subr_rman.c standard kern/subr_rtc.c standard kern/subr_sbuf.c standard kern/subr_scanf.c standard kern/subr_sglist.c standard kern/subr_sleepqueue.c standard kern/subr_smp.c standard kern/subr_smr.c standard kern/subr_stack.c optional ddb | stack | ktr kern/subr_stats.c optional stats kern/subr_taskqueue.c standard kern/subr_terminal.c optional vt kern/subr_trap.c standard kern/subr_turnstile.c standard kern/subr_uio.c standard kern/subr_unit.c standard kern/subr_vmem.c standard kern/subr_witness.c optional witness kern/sys_capability.c standard kern/sys_generic.c standard kern/sys_getrandom.c standard kern/sys_pipe.c standard kern/sys_procdesc.c standard kern/sys_process.c standard kern/sys_socket.c standard kern/syscalls.c standard kern/sysv_ipc.c standard kern/sysv_msg.c optional sysvmsg kern/sysv_sem.c optional sysvsem kern/sysv_shm.c optional sysvshm kern/tty.c standard kern/tty_compat.c optional compat_43tty kern/tty_info.c standard kern/tty_inq.c standard kern/tty_outq.c standard kern/tty_pts.c standard kern/tty_tty.c standard kern/tty_ttydisc.c standard kern/uipc_accf.c standard kern/uipc_debug.c optional ddb kern/uipc_domain.c standard kern/uipc_ktls.c optional kern_tls kern/uipc_mbuf.c standard kern/uipc_mbuf2.c standard kern/uipc_mbufhash.c standard kern/uipc_mqueue.c optional p1003_1b_mqueue kern/uipc_sem.c optional p1003_1b_semaphores kern/uipc_shm.c standard kern/uipc_sockbuf.c standard kern/uipc_socket.c standard kern/uipc_syscalls.c standard kern/uipc_usrreq.c standard kern/vfs_acl.c standard kern/vfs_aio.c standard kern/vfs_bio.c standard kern/vfs_cache.c standard kern/vfs_cluster.c standard kern/vfs_default.c standard kern/vfs_export.c standard kern/vfs_extattr.c standard kern/vfs_hash.c standard kern/vfs_init.c standard kern/vfs_lookup.c standard kern/vfs_mount.c standard kern/vfs_mountroot.c standard kern/vfs_subr.c standard kern/vfs_syscalls.c standard kern/vfs_vnops.c standard # # Kernel GSS-API # gssd.h optional kgssapi \ dependency "$S/kgssapi/gssd.x" \ compile-with "RPCGEN_CPP='${CPP}' rpcgen -hM $S/kgssapi/gssd.x | grep -v pthread.h > gssd.h" \ no-obj no-implicit-rule before-depend local \ clean "gssd.h" gssd_xdr.c optional kgssapi \ dependency "$S/kgssapi/gssd.x gssd.h" \ compile-with "RPCGEN_CPP='${CPP}' rpcgen -c $S/kgssapi/gssd.x -o gssd_xdr.c" \ no-implicit-rule before-depend local \ clean "gssd_xdr.c" gssd_clnt.c optional kgssapi \ dependency "$S/kgssapi/gssd.x gssd.h" \ compile-with "RPCGEN_CPP='${CPP}' rpcgen -lM $S/kgssapi/gssd.x | grep -v string.h > gssd_clnt.c" \ no-implicit-rule before-depend local \ clean "gssd_clnt.c" kgssapi/gss_accept_sec_context.c optional kgssapi kgssapi/gss_add_oid_set_member.c optional kgssapi kgssapi/gss_acquire_cred.c optional kgssapi kgssapi/gss_canonicalize_name.c optional kgssapi kgssapi/gss_create_empty_oid_set.c optional kgssapi kgssapi/gss_delete_sec_context.c optional kgssapi kgssapi/gss_display_status.c optional kgssapi kgssapi/gss_export_name.c optional kgssapi kgssapi/gss_get_mic.c optional kgssapi kgssapi/gss_init_sec_context.c optional kgssapi kgssapi/gss_impl.c optional kgssapi kgssapi/gss_import_name.c optional kgssapi kgssapi/gss_names.c optional kgssapi kgssapi/gss_pname_to_uid.c optional kgssapi kgssapi/gss_release_buffer.c optional kgssapi kgssapi/gss_release_cred.c optional kgssapi kgssapi/gss_release_name.c optional kgssapi kgssapi/gss_release_oid_set.c optional kgssapi kgssapi/gss_set_cred_option.c optional kgssapi kgssapi/gss_test_oid_set_member.c optional kgssapi kgssapi/gss_unwrap.c optional kgssapi kgssapi/gss_verify_mic.c optional kgssapi kgssapi/gss_wrap.c optional kgssapi kgssapi/gss_wrap_size_limit.c optional kgssapi kgssapi/gssd_prot.c optional kgssapi kgssapi/krb5/krb5_mech.c optional kgssapi kgssapi/krb5/kcrypto.c optional kgssapi kgssapi/krb5/kcrypto_aes.c optional kgssapi kgssapi/krb5/kcrypto_arcfour.c optional kgssapi kgssapi/krb5/kcrypto_des.c optional kgssapi kgssapi/krb5/kcrypto_des3.c optional kgssapi kgssapi/kgss_if.m optional kgssapi kgssapi/gsstest.c optional kgssapi_debug # These files in libkern/ are those needed by all architectures. Some # of the files in libkern/ are only needed on some architectures, e.g., # libkern/divdi3.c is needed by i386 but not alpha. Also, some of these # routines may be optimized for a particular platform. In either case, # the file should be moved to conf/files. from here. # libkern/arc4random.c standard libkern/asprintf.c standard libkern/bcd.c standard libkern/bsearch.c standard libkern/explicit_bzero.c standard libkern/fnmatch.c standard libkern/gsb_crc32.c standard libkern/iconv.c optional libiconv libkern/iconv_converter_if.m optional libiconv libkern/iconv_ucs.c optional libiconv libkern/iconv_xlat.c optional libiconv libkern/iconv_xlat16.c optional libiconv libkern/inet_aton.c standard libkern/inet_ntoa.c standard libkern/inet_ntop.c standard libkern/inet_pton.c standard libkern/jenkins_hash.c standard libkern/murmur3_32.c standard libkern/mcount.c optional profiling-routine libkern/memcchr.c standard libkern/memchr.c standard libkern/memmem.c optional gdb libkern/qsort.c standard libkern/qsort_r.c standard libkern/random.c standard libkern/scanc.c standard libkern/strcasecmp.c standard libkern/strcat.c standard libkern/strchr.c standard libkern/strchrnul.c optional gdb libkern/strcmp.c standard libkern/strcpy.c standard libkern/strcspn.c standard libkern/strdup.c standard libkern/strndup.c standard libkern/strlcat.c standard libkern/strlcpy.c standard libkern/strlen.c standard libkern/strncat.c standard libkern/strncmp.c standard libkern/strncpy.c standard libkern/strnlen.c standard libkern/strrchr.c standard libkern/strsep.c standard libkern/strspn.c standard libkern/strstr.c standard libkern/strtol.c standard libkern/strtoq.c standard libkern/strtoul.c standard libkern/strtouq.c standard libkern/strvalid.c standard libkern/timingsafe_bcmp.c standard contrib/zlib/adler32.c optional crypto | geom_uzip | ipsec | \ ipsec_support | mxge | ddb_ctf | gzio | zfs | zlib contrib/zlib/compress.c optional crypto | geom_uzip | ipsec | \ ipsec_support | mxge | ddb_ctf | gzio | zfs | zlib \ compile-with "${NORMAL_C} -Wno-cast-qual" contrib/zlib/crc32.c optional crypto | geom_uzip | ipsec | \ ipsec_support | mxge | ddb_ctf | gzio | zfs | zlib contrib/zlib/deflate.c optional crypto | geom_uzip | ipsec | \ ipsec_support | mxge | ddb_ctf | gzio | zfs | zlib \ compile-with "${NORMAL_C} -Wno-cast-qual" contrib/zlib/inffast.c optional crypto | geom_uzip | ipsec | \ ipsec_support | mxge | ddb_ctf | gzio | zfs | zlib contrib/zlib/inflate.c optional crypto | geom_uzip | ipsec | \ ipsec_support | mxge | ddb_ctf | gzio | zfs | zlib contrib/zlib/inftrees.c optional crypto | geom_uzip | ipsec | \ ipsec_support | mxge | ddb_ctf | gzio | zfs | zlib contrib/zlib/trees.c optional crypto | geom_uzip | ipsec | \ ipsec_support | mxge | ddb_ctf | gzio | zfs | zlib contrib/zlib/uncompr.c optional crypto | geom_uzip | ipsec | \ ipsec_support | mxge | ddb_ctf | gzio | zfs | zlib \ compile-with "${NORMAL_C} -Wno-cast-qual" contrib/zlib/zutil.c optional crypto | geom_uzip | ipsec | \ ipsec_support | mxge | ddb_ctf | gzio | zfs | zlib dev/zlib/zlib_mod.c optional crypto | geom_uzip | ipsec | \ ipsec_support | mxge | ddb_ctf | gzio | zfs | zlib dev/zlib/zcalloc.c optional crypto | geom_uzip | ipsec | \ ipsec_support | mxge | ddb_ctf | gzio | zfs | zlib net/altq/altq_cbq.c optional altq net/altq/altq_codel.c optional altq net/altq/altq_hfsc.c optional altq net/altq/altq_fairq.c optional altq net/altq/altq_priq.c optional altq net/altq/altq_red.c optional altq net/altq/altq_rio.c optional altq net/altq/altq_rmclass.c optional altq net/altq/altq_subr.c optional altq net/bpf.c standard net/bpf_buffer.c optional bpf net/bpf_jitter.c optional bpf_jitter net/bpf_filter.c optional bpf | netgraph_bpf net/bpf_zerocopy.c optional bpf net/bridgestp.c optional bridge | if_bridge net/ieee8023ad_lacp.c optional lagg net/if.c standard net/if_bridge.c optional bridge inet | if_bridge inet net/if_clone.c standard net/if_dead.c standard net/if_debug.c optional ddb net/if_disc.c optional disc net/if_edsc.c optional edsc net/if_enc.c optional enc inet | enc inet6 net/if_epair.c optional epair net/if_ethersubr.c optional ether net/if_fwsubr.c optional fwip net/if_gif.c optional gif inet | gif inet6 | \ netgraph_gif inet | netgraph_gif inet6 net/if_gre.c optional gre inet | gre inet6 net/if_ipsec.c optional inet ipsec | inet6 ipsec net/if_lagg.c optional lagg net/if_loop.c optional loop net/if_llatbl.c standard net/if_me.c optional me inet net/if_media.c standard net/if_mib.c standard net/if_spppfr.c optional sppp | netgraph_sppp net/if_spppsubr.c optional sppp | netgraph_sppp net/if_stf.c optional stf inet inet6 net/if_tuntap.c optional tuntap net/if_vlan.c optional vlan net/if_vxlan.c optional vxlan inet | vxlan inet6 net/ifdi_if.m optional ether pci iflib net/iflib.c optional ether pci iflib net/iflib_clone.c optional ether pci iflib net/mp_ring.c optional ether iflib net/mppcc.c optional netgraph_mppc_compression net/mppcd.c optional netgraph_mppc_compression net/netisr.c standard net/debugnet.c optional inet debugnet net/debugnet_inet.c optional inet debugnet net/pfil.c optional ether | inet net/radix.c standard net/radix_mpath.c standard net/raw_cb.c standard net/raw_usrreq.c standard net/route.c standard net/route_temporal.c standard net/rss_config.c optional inet rss | inet6 rss net/rtsock.c standard net/slcompress.c optional netgraph_vjc | sppp | \ netgraph_sppp net/toeplitz.c optional inet rss | inet6 rss net/vnet.c optional vimage net80211/ieee80211.c optional wlan net80211/ieee80211_acl.c optional wlan wlan_acl net80211/ieee80211_action.c optional wlan net80211/ieee80211_adhoc.c optional wlan \ compile-with "${NORMAL_C} -Wno-unused-function" net80211/ieee80211_ageq.c optional wlan net80211/ieee80211_amrr.c optional wlan | wlan_amrr net80211/ieee80211_crypto.c optional wlan \ compile-with "${NORMAL_C} -Wno-unused-function" net80211/ieee80211_crypto_ccmp.c optional wlan wlan_ccmp net80211/ieee80211_crypto_none.c optional wlan net80211/ieee80211_crypto_tkip.c optional wlan wlan_tkip net80211/ieee80211_crypto_wep.c optional wlan wlan_wep net80211/ieee80211_ddb.c optional wlan ddb net80211/ieee80211_dfs.c optional wlan net80211/ieee80211_freebsd.c optional wlan net80211/ieee80211_hostap.c optional wlan \ compile-with "${NORMAL_C} -Wno-unused-function" net80211/ieee80211_ht.c optional wlan net80211/ieee80211_hwmp.c optional wlan ieee80211_support_mesh net80211/ieee80211_input.c optional wlan net80211/ieee80211_ioctl.c optional wlan net80211/ieee80211_mesh.c optional wlan ieee80211_support_mesh \ compile-with "${NORMAL_C} -Wno-unused-function" net80211/ieee80211_monitor.c optional wlan net80211/ieee80211_node.c optional wlan net80211/ieee80211_output.c optional wlan net80211/ieee80211_phy.c optional wlan net80211/ieee80211_power.c optional wlan net80211/ieee80211_proto.c optional wlan net80211/ieee80211_radiotap.c optional wlan net80211/ieee80211_ratectl.c optional wlan net80211/ieee80211_ratectl_none.c optional wlan net80211/ieee80211_regdomain.c optional wlan net80211/ieee80211_rssadapt.c optional wlan wlan_rssadapt net80211/ieee80211_scan.c optional wlan net80211/ieee80211_scan_sta.c optional wlan net80211/ieee80211_sta.c optional wlan \ compile-with "${NORMAL_C} -Wno-unused-function" net80211/ieee80211_superg.c optional wlan ieee80211_support_superg net80211/ieee80211_scan_sw.c optional wlan net80211/ieee80211_tdma.c optional wlan ieee80211_support_tdma net80211/ieee80211_vht.c optional wlan net80211/ieee80211_wds.c optional wlan net80211/ieee80211_xauth.c optional wlan wlan_xauth net80211/ieee80211_alq.c optional wlan ieee80211_alq netgraph/atm/ccatm/ng_ccatm.c optional ngatm_ccatm \ compile-with "${NORMAL_C} -I$S/contrib/ngatm" netgraph/atm/ngatmbase.c optional ngatm_atmbase \ compile-with "${NORMAL_C} -I$S/contrib/ngatm" netgraph/atm/sscfu/ng_sscfu.c optional ngatm_sscfu \ compile-with "${NORMAL_C} -I$S/contrib/ngatm" netgraph/atm/sscop/ng_sscop.c optional ngatm_sscop \ compile-with "${NORMAL_C} -I$S/contrib/ngatm" netgraph/atm/uni/ng_uni.c optional ngatm_uni \ compile-with "${NORMAL_C} -I$S/contrib/ngatm" netgraph/bluetooth/common/ng_bluetooth.c optional netgraph_bluetooth netgraph/bluetooth/drivers/bt3c/ng_bt3c_pccard.c optional netgraph_bluetooth_bt3c netgraph/bluetooth/drivers/h4/ng_h4.c optional netgraph_bluetooth_h4 netgraph/bluetooth/drivers/ubt/ng_ubt.c optional netgraph_bluetooth_ubt usb netgraph/bluetooth/drivers/ubt/ng_ubt_intel.c optional netgraph_bluetooth_ubt usb netgraph/bluetooth/drivers/ubtbcmfw/ubtbcmfw.c optional netgraph_bluetooth_ubtbcmfw usb netgraph/bluetooth/hci/ng_hci_cmds.c optional netgraph_bluetooth_hci netgraph/bluetooth/hci/ng_hci_evnt.c optional netgraph_bluetooth_hci netgraph/bluetooth/hci/ng_hci_main.c optional netgraph_bluetooth_hci netgraph/bluetooth/hci/ng_hci_misc.c optional netgraph_bluetooth_hci netgraph/bluetooth/hci/ng_hci_ulpi.c optional netgraph_bluetooth_hci netgraph/bluetooth/l2cap/ng_l2cap_cmds.c optional netgraph_bluetooth_l2cap netgraph/bluetooth/l2cap/ng_l2cap_evnt.c optional netgraph_bluetooth_l2cap netgraph/bluetooth/l2cap/ng_l2cap_llpi.c optional netgraph_bluetooth_l2cap netgraph/bluetooth/l2cap/ng_l2cap_main.c optional netgraph_bluetooth_l2cap netgraph/bluetooth/l2cap/ng_l2cap_misc.c optional netgraph_bluetooth_l2cap netgraph/bluetooth/l2cap/ng_l2cap_ulpi.c optional netgraph_bluetooth_l2cap netgraph/bluetooth/socket/ng_btsocket.c optional netgraph_bluetooth_socket netgraph/bluetooth/socket/ng_btsocket_hci_raw.c optional netgraph_bluetooth_socket netgraph/bluetooth/socket/ng_btsocket_l2cap.c optional netgraph_bluetooth_socket netgraph/bluetooth/socket/ng_btsocket_l2cap_raw.c optional netgraph_bluetooth_socket netgraph/bluetooth/socket/ng_btsocket_rfcomm.c optional netgraph_bluetooth_socket netgraph/bluetooth/socket/ng_btsocket_sco.c optional netgraph_bluetooth_socket netgraph/netflow/netflow.c optional netgraph_netflow netgraph/netflow/netflow_v9.c optional netgraph_netflow netgraph/netflow/ng_netflow.c optional netgraph_netflow netgraph/ng_UI.c optional netgraph_UI netgraph/ng_async.c optional netgraph_async netgraph/ng_atmllc.c optional netgraph_atmllc netgraph/ng_base.c optional netgraph netgraph/ng_bpf.c optional netgraph_bpf netgraph/ng_bridge.c optional netgraph_bridge netgraph/ng_car.c optional netgraph_car netgraph/ng_checksum.c optional netgraph_checksum netgraph/ng_cisco.c optional netgraph_cisco netgraph/ng_deflate.c optional netgraph_deflate netgraph/ng_device.c optional netgraph_device netgraph/ng_echo.c optional netgraph_echo netgraph/ng_eiface.c optional netgraph_eiface netgraph/ng_ether.c optional netgraph_ether netgraph/ng_ether_echo.c optional netgraph_ether_echo netgraph/ng_frame_relay.c optional netgraph_frame_relay netgraph/ng_gif.c optional netgraph_gif inet6 | netgraph_gif inet netgraph/ng_gif_demux.c optional netgraph_gif_demux netgraph/ng_hole.c optional netgraph_hole netgraph/ng_iface.c optional netgraph_iface netgraph/ng_ip_input.c optional netgraph_ip_input netgraph/ng_ipfw.c optional netgraph_ipfw inet ipfirewall netgraph/ng_ksocket.c optional netgraph_ksocket netgraph/ng_l2tp.c optional netgraph_l2tp netgraph/ng_lmi.c optional netgraph_lmi netgraph/ng_mppc.c optional netgraph_mppc_compression | \ netgraph_mppc_encryption netgraph/ng_nat.c optional netgraph_nat inet libalias netgraph/ng_one2many.c optional netgraph_one2many netgraph/ng_parse.c optional netgraph netgraph/ng_patch.c optional netgraph_patch netgraph/ng_pipe.c optional netgraph_pipe netgraph/ng_ppp.c optional netgraph_ppp netgraph/ng_pppoe.c optional netgraph_pppoe netgraph/ng_pptpgre.c optional netgraph_pptpgre netgraph/ng_pred1.c optional netgraph_pred1 netgraph/ng_rfc1490.c optional netgraph_rfc1490 netgraph/ng_socket.c optional netgraph_socket netgraph/ng_split.c optional netgraph_split netgraph/ng_sppp.c optional netgraph_sppp netgraph/ng_tag.c optional netgraph_tag netgraph/ng_tcpmss.c optional netgraph_tcpmss netgraph/ng_tee.c optional netgraph_tee netgraph/ng_tty.c optional netgraph_tty netgraph/ng_vjc.c optional netgraph_vjc netgraph/ng_vlan.c optional netgraph_vlan netinet/accf_data.c optional accept_filter_data inet netinet/accf_dns.c optional accept_filter_dns inet netinet/accf_http.c optional accept_filter_http inet netinet/if_ether.c optional inet ether netinet/igmp.c optional inet netinet/in.c optional inet netinet/in_debug.c optional inet ddb netinet/in_kdtrace.c optional inet | inet6 netinet/ip_carp.c optional inet carp | inet6 carp netinet/in_fib.c optional inet netinet/in_gif.c optional gif inet | netgraph_gif inet netinet/ip_gre.c optional gre inet netinet/ip_id.c optional inet netinet/in_jail.c optional inet netinet/in_mcast.c optional inet netinet/in_pcb.c optional inet | inet6 netinet/in_pcbgroup.c optional inet pcbgroup | inet6 pcbgroup netinet/in_prot.c optional inet | inet6 netinet/in_proto.c optional inet | inet6 netinet/in_rmx.c optional inet netinet/in_rss.c optional inet rss netinet/ip_divert.c optional inet ipdivert ipfirewall netinet/ip_ecn.c optional inet | inet6 netinet/ip_encap.c optional inet | inet6 netinet/ip_fastfwd.c optional inet netinet/ip_icmp.c optional inet | inet6 netinet/ip_input.c optional inet netinet/ip_mroute.c optional mrouting inet netinet/ip_options.c optional inet netinet/ip_output.c optional inet netinet/ip_reass.c optional inet netinet/raw_ip.c optional inet | inet6 netinet/cc/cc.c optional inet | inet6 netinet/cc/cc_newreno.c optional inet | inet6 netinet/sctp_asconf.c optional inet sctp | inet6 sctp netinet/sctp_auth.c optional inet sctp | inet6 sctp netinet/sctp_bsd_addr.c optional inet sctp | inet6 sctp netinet/sctp_cc_functions.c optional inet sctp | inet6 sctp netinet/sctp_crc32.c optional inet | inet6 netinet/sctp_indata.c optional inet sctp | inet6 sctp netinet/sctp_input.c optional inet sctp | inet6 sctp netinet/sctp_kdtrace.c optional inet sctp | inet6 sctp netinet/sctp_output.c optional inet sctp | inet6 sctp netinet/sctp_pcb.c optional inet sctp | inet6 sctp netinet/sctp_peeloff.c optional inet sctp | inet6 sctp netinet/sctp_ss_functions.c optional inet sctp | inet6 sctp netinet/sctp_syscalls.c optional inet sctp | inet6 sctp netinet/sctp_sysctl.c optional inet sctp | inet6 sctp netinet/sctp_timer.c optional inet sctp | inet6 sctp netinet/sctp_usrreq.c optional inet sctp | inet6 sctp netinet/sctputil.c optional inet sctp | inet6 sctp netinet/siftr.c optional inet siftr alq | inet6 siftr alq netinet/tcp_debug.c optional tcpdebug netinet/tcp_fastopen.c optional inet tcp_rfc7413 | inet6 tcp_rfc7413 netinet/tcp_hostcache.c optional inet | inet6 netinet/tcp_input.c optional inet | inet6 netinet/tcp_log_buf.c optional tcp_blackbox inet | tcp_blackbox inet6 netinet/tcp_lro.c optional inet | inet6 netinet/tcp_output.c optional inet | inet6 netinet/tcp_offload.c optional tcp_offload inet | tcp_offload inet6 netinet/tcp_hpts.c optional tcphpts inet | tcphpts inet6 netinet/tcp_ratelimit.c optional ratelimit inet | ratelimit inet6 netinet/tcp_pcap.c optional inet tcppcap | inet6 tcppcap \ compile-with "${NORMAL_C} ${NO_WNONNULL}" netinet/tcp_reass.c optional inet | inet6 netinet/tcp_sack.c optional inet | inet6 netinet/tcp_stats.c optional stats inet | stats inet6 netinet/tcp_subr.c optional inet | inet6 netinet/tcp_syncache.c optional inet | inet6 netinet/tcp_timer.c optional inet | inet6 netinet/tcp_timewait.c optional inet | inet6 netinet/tcp_usrreq.c optional inet | inet6 netinet/udp_usrreq.c optional inet | inet6 netinet/libalias/alias.c optional libalias inet | netgraph_nat inet netinet/libalias/alias_db.c optional libalias inet | netgraph_nat inet netinet/libalias/alias_mod.c optional libalias | netgraph_nat netinet/libalias/alias_proxy.c optional libalias inet | netgraph_nat inet netinet/libalias/alias_util.c optional libalias inet | netgraph_nat inet netinet/libalias/alias_sctp.c optional libalias inet | netgraph_nat inet netinet/netdump/netdump_client.c optional inet debugnet netdump netinet6/dest6.c optional inet6 netinet6/frag6.c optional inet6 netinet6/icmp6.c optional inet6 netinet6/in6.c optional inet6 netinet6/in6_cksum.c optional inet6 netinet6/in6_fib.c optional inet6 netinet6/in6_gif.c optional gif inet6 | netgraph_gif inet6 netinet6/in6_ifattach.c optional inet6 netinet6/in6_jail.c optional inet6 netinet6/in6_mcast.c optional inet6 netinet6/in6_pcb.c optional inet6 netinet6/in6_pcbgroup.c optional inet6 pcbgroup netinet6/in6_proto.c optional inet6 netinet6/in6_rmx.c optional inet6 netinet6/in6_rss.c optional inet6 rss netinet6/in6_src.c optional inet6 netinet6/ip6_fastfwd.c optional inet6 netinet6/ip6_forward.c optional inet6 netinet6/ip6_gre.c optional gre inet6 netinet6/ip6_id.c optional inet6 netinet6/ip6_input.c optional inet6 netinet6/ip6_mroute.c optional mrouting inet6 netinet6/ip6_output.c optional inet6 netinet6/mld6.c optional inet6 netinet6/nd6.c optional inet6 netinet6/nd6_nbr.c optional inet6 netinet6/nd6_rtr.c optional inet6 netinet6/raw_ip6.c optional inet6 netinet6/route6.c optional inet6 netinet6/scope6.c optional inet6 netinet6/sctp6_usrreq.c optional inet6 sctp netinet6/udp6_usrreq.c optional inet6 netipsec/ipsec.c optional ipsec inet | ipsec inet6 netipsec/ipsec_input.c optional ipsec inet | ipsec inet6 netipsec/ipsec_mbuf.c optional ipsec inet | ipsec inet6 netipsec/ipsec_mod.c optional ipsec inet | ipsec inet6 netipsec/ipsec_output.c optional ipsec inet | ipsec inet6 netipsec/ipsec_pcb.c optional ipsec inet | ipsec inet6 | \ ipsec_support inet | ipsec_support inet6 netipsec/key.c optional ipsec inet | ipsec inet6 | \ ipsec_support inet | ipsec_support inet6 netipsec/key_debug.c optional ipsec inet | ipsec inet6 | \ ipsec_support inet | ipsec_support inet6 netipsec/keysock.c optional ipsec inet | ipsec inet6 | \ ipsec_support inet | ipsec_support inet6 netipsec/subr_ipsec.c optional ipsec inet | ipsec inet6 | \ ipsec_support inet | ipsec_support inet6 netipsec/udpencap.c optional ipsec inet netipsec/xform_ah.c optional ipsec inet | ipsec inet6 netipsec/xform_esp.c optional ipsec inet | ipsec inet6 netipsec/xform_ipcomp.c optional ipsec inet | ipsec inet6 netipsec/xform_tcp.c optional ipsec inet tcp_signature | \ ipsec inet6 tcp_signature | ipsec_support inet tcp_signature | \ ipsec_support inet6 tcp_signature netpfil/ipfw/dn_aqm_codel.c optional inet dummynet netpfil/ipfw/dn_aqm_pie.c optional inet dummynet netpfil/ipfw/dn_heap.c optional inet dummynet netpfil/ipfw/dn_sched_fifo.c optional inet dummynet netpfil/ipfw/dn_sched_fq_codel.c optional inet dummynet netpfil/ipfw/dn_sched_fq_pie.c optional inet dummynet netpfil/ipfw/dn_sched_prio.c optional inet dummynet netpfil/ipfw/dn_sched_qfq.c optional inet dummynet netpfil/ipfw/dn_sched_rr.c optional inet dummynet netpfil/ipfw/dn_sched_wf2q.c optional inet dummynet netpfil/ipfw/ip_dummynet.c optional inet dummynet netpfil/ipfw/ip_dn_io.c optional inet dummynet netpfil/ipfw/ip_dn_glue.c optional inet dummynet netpfil/ipfw/ip_fw2.c optional inet ipfirewall netpfil/ipfw/ip_fw_bpf.c optional inet ipfirewall netpfil/ipfw/ip_fw_dynamic.c optional inet ipfirewall \ compile-with "${NORMAL_C} -I$S/contrib/ck/include" netpfil/ipfw/ip_fw_eaction.c optional inet ipfirewall netpfil/ipfw/ip_fw_log.c optional inet ipfirewall netpfil/ipfw/ip_fw_pfil.c optional inet ipfirewall netpfil/ipfw/ip_fw_sockopt.c optional inet ipfirewall netpfil/ipfw/ip_fw_table.c optional inet ipfirewall netpfil/ipfw/ip_fw_table_algo.c optional inet ipfirewall netpfil/ipfw/ip_fw_table_value.c optional inet ipfirewall netpfil/ipfw/ip_fw_iface.c optional inet ipfirewall netpfil/ipfw/ip_fw_nat.c optional inet ipfirewall_nat netpfil/ipfw/nat64/ip_fw_nat64.c optional inet inet6 ipfirewall \ ipfirewall_nat64 netpfil/ipfw/nat64/nat64clat.c optional inet inet6 ipfirewall \ ipfirewall_nat64 netpfil/ipfw/nat64/nat64clat_control.c optional inet inet6 ipfirewall \ ipfirewall_nat64 netpfil/ipfw/nat64/nat64lsn.c optional inet inet6 ipfirewall \ ipfirewall_nat64 compile-with "${NORMAL_C} -I$S/contrib/ck/include" netpfil/ipfw/nat64/nat64lsn_control.c optional inet inet6 ipfirewall \ ipfirewall_nat64 compile-with "${NORMAL_C} -I$S/contrib/ck/include" netpfil/ipfw/nat64/nat64stl.c optional inet inet6 ipfirewall \ ipfirewall_nat64 netpfil/ipfw/nat64/nat64stl_control.c optional inet inet6 ipfirewall \ ipfirewall_nat64 netpfil/ipfw/nat64/nat64_translate.c optional inet inet6 ipfirewall \ ipfirewall_nat64 netpfil/ipfw/nptv6/ip_fw_nptv6.c optional inet inet6 ipfirewall \ ipfirewall_nptv6 netpfil/ipfw/nptv6/nptv6.c optional inet inet6 ipfirewall \ ipfirewall_nptv6 netpfil/ipfw/pmod/ip_fw_pmod.c optional inet ipfirewall_pmod netpfil/ipfw/pmod/tcpmod.c optional inet ipfirewall_pmod netpfil/pf/if_pflog.c optional pflog pf inet netpfil/pf/if_pfsync.c optional pfsync pf inet netpfil/pf/pf.c optional pf inet netpfil/pf/pf_if.c optional pf inet netpfil/pf/pf_ioctl.c optional pf inet netpfil/pf/pf_lb.c optional pf inet netpfil/pf/pf_norm.c optional pf inet netpfil/pf/pf_osfp.c optional pf inet netpfil/pf/pf_ruleset.c optional pf inet netpfil/pf/pf_table.c optional pf inet netpfil/pf/in4_cksum.c optional pf inet netsmb/smb_conn.c optional netsmb netsmb/smb_crypt.c optional netsmb netsmb/smb_dev.c optional netsmb netsmb/smb_iod.c optional netsmb netsmb/smb_rq.c optional netsmb netsmb/smb_smb.c optional netsmb netsmb/smb_subr.c optional netsmb netsmb/smb_trantcp.c optional netsmb netsmb/smb_usr.c optional netsmb nfs/bootp_subr.c optional bootp nfscl nfs/krpc_subr.c optional bootp nfscl nfs/nfs_diskless.c optional nfscl nfs_root nfs/nfs_fha.c optional nfsd nfs/nfs_lock.c optional nfscl | nfslockd | nfsd nfs/nfs_nfssvc.c optional nfscl | nfsd nlm/nlm_advlock.c optional nfslockd | nfsd nlm/nlm_prot_clnt.c optional nfslockd | nfsd nlm/nlm_prot_impl.c optional nfslockd | nfsd nlm/nlm_prot_server.c optional nfslockd | nfsd nlm/nlm_prot_svc.c optional nfslockd | nfsd nlm/nlm_prot_xdr.c optional nfslockd | nfsd nlm/sm_inter_xdr.c optional nfslockd | nfsd # Linux Kernel Programming Interface compat/linuxkpi/common/src/linux_kmod.c optional compat_linuxkpi \ compile-with "${LINUXKPI_C}" compat/linuxkpi/common/src/linux_compat.c optional compat_linuxkpi \ compile-with "${LINUXKPI_C}" compat/linuxkpi/common/src/linux_current.c optional compat_linuxkpi \ compile-with "${LINUXKPI_C}" compat/linuxkpi/common/src/linux_hrtimer.c optional compat_linuxkpi \ compile-with "${LINUXKPI_C}" compat/linuxkpi/common/src/linux_kthread.c optional compat_linuxkpi \ compile-with "${LINUXKPI_C}" compat/linuxkpi/common/src/linux_lock.c optional compat_linuxkpi \ compile-with "${LINUXKPI_C}" compat/linuxkpi/common/src/linux_page.c optional compat_linuxkpi \ compile-with "${LINUXKPI_C}" compat/linuxkpi/common/src/linux_pci.c optional compat_linuxkpi pci \ compile-with "${LINUXKPI_C}" compat/linuxkpi/common/src/linux_tasklet.c optional compat_linuxkpi \ compile-with "${LINUXKPI_C}" compat/linuxkpi/common/src/linux_idr.c optional compat_linuxkpi \ compile-with "${LINUXKPI_C}" compat/linuxkpi/common/src/linux_radix.c optional compat_linuxkpi \ compile-with "${LINUXKPI_C}" compat/linuxkpi/common/src/linux_rcu.c optional compat_linuxkpi \ compile-with "${LINUXKPI_C} -I$S/contrib/ck/include" compat/linuxkpi/common/src/linux_schedule.c optional compat_linuxkpi \ compile-with "${LINUXKPI_C}" compat/linuxkpi/common/src/linux_shmemfs.c optional compat_linuxkpi \ compile-with "${LINUXKPI_C}" compat/linuxkpi/common/src/linux_slab.c optional compat_linuxkpi \ compile-with "${LINUXKPI_C}" compat/linuxkpi/common/src/linux_usb.c optional compat_linuxkpi usb \ compile-with "${LINUXKPI_C}" compat/linuxkpi/common/src/linux_work.c optional compat_linuxkpi \ compile-with "${LINUXKPI_C}" compat/linuxkpi/common/src/linux_seq_file.c optional compat_linuxkpi | lindebugfs \ compile-with "${LINUXKPI_C}" compat/lindebugfs/lindebugfs.c optional lindebugfs \ compile-with "${LINUXKPI_C}" # OpenFabrics Enterprise Distribution (Infiniband) ofed/drivers/infiniband/core/ib_addr.c optional ofed \ compile-with "${OFED_C}" ofed/drivers/infiniband/core/ib_agent.c optional ofed \ compile-with "${OFED_C}" ofed/drivers/infiniband/core/ib_cache.c optional ofed \ compile-with "${OFED_C}" ofed/drivers/infiniband/core/ib_cm.c optional ofed \ compile-with "${OFED_C}" ofed/drivers/infiniband/core/ib_cma.c optional ofed \ compile-with "${OFED_C}" ofed/drivers/infiniband/core/ib_cq.c optional ofed \ compile-with "${OFED_C}" ofed/drivers/infiniband/core/ib_device.c optional ofed \ compile-with "${OFED_C}" ofed/drivers/infiniband/core/ib_fmr_pool.c optional ofed \ compile-with "${OFED_C}" ofed/drivers/infiniband/core/ib_iwcm.c optional ofed \ compile-with "${OFED_C}" ofed/drivers/infiniband/core/ib_iwpm_msg.c optional ofed \ compile-with "${OFED_C}" ofed/drivers/infiniband/core/ib_iwpm_util.c optional ofed \ compile-with "${OFED_C}" ofed/drivers/infiniband/core/ib_mad.c optional ofed \ compile-with "${OFED_C}" ofed/drivers/infiniband/core/ib_mad_rmpp.c optional ofed \ compile-with "${OFED_C}" ofed/drivers/infiniband/core/ib_multicast.c optional ofed \ compile-with "${OFED_C}" ofed/drivers/infiniband/core/ib_packer.c optional ofed \ compile-with "${OFED_C}" ofed/drivers/infiniband/core/ib_roce_gid_mgmt.c optional ofed \ compile-with "${OFED_C}" ofed/drivers/infiniband/core/ib_sa_query.c optional ofed \ compile-with "${OFED_C}" ofed/drivers/infiniband/core/ib_smi.c optional ofed \ compile-with "${OFED_C}" ofed/drivers/infiniband/core/ib_sysfs.c optional ofed \ compile-with "${OFED_C}" ofed/drivers/infiniband/core/ib_ucm.c optional ofed \ compile-with "${OFED_C}" ofed/drivers/infiniband/core/ib_ucma.c optional ofed \ compile-with "${OFED_C}" ofed/drivers/infiniband/core/ib_ud_header.c optional ofed \ compile-with "${OFED_C}" ofed/drivers/infiniband/core/ib_umem.c optional ofed \ compile-with "${OFED_C}" ofed/drivers/infiniband/core/ib_user_mad.c optional ofed \ compile-with "${OFED_C}" ofed/drivers/infiniband/core/ib_uverbs_cmd.c optional ofed \ compile-with "${OFED_C}" ofed/drivers/infiniband/core/ib_uverbs_main.c optional ofed \ compile-with "${OFED_C}" ofed/drivers/infiniband/core/ib_uverbs_marshall.c optional ofed \ compile-with "${OFED_C}" ofed/drivers/infiniband/core/ib_verbs.c optional ofed \ compile-with "${OFED_C}" ofed/drivers/infiniband/ulp/ipoib/ipoib_cm.c optional ipoib \ compile-with "${OFED_C} -I$S/ofed/drivers/infiniband/ulp/ipoib/" #ofed/drivers/infiniband/ulp/ipoib/ipoib_fs.c optional ipoib \ # compile-with "${OFED_C} -I$S/ofed/drivers/infiniband/ulp/ipoib/" ofed/drivers/infiniband/ulp/ipoib/ipoib_ib.c optional ipoib \ compile-with "${OFED_C} -I$S/ofed/drivers/infiniband/ulp/ipoib/" ofed/drivers/infiniband/ulp/ipoib/ipoib_main.c optional ipoib \ compile-with "${OFED_C} -I$S/ofed/drivers/infiniband/ulp/ipoib/" ofed/drivers/infiniband/ulp/ipoib/ipoib_multicast.c optional ipoib \ compile-with "${OFED_C} -I$S/ofed/drivers/infiniband/ulp/ipoib/" ofed/drivers/infiniband/ulp/ipoib/ipoib_verbs.c optional ipoib \ compile-with "${OFED_C} -I$S/ofed/drivers/infiniband/ulp/ipoib/" #ofed/drivers/infiniband/ulp/ipoib/ipoib_vlan.c optional ipoib \ # compile-with "${OFED_C} -I$S/ofed/drivers/infiniband/ulp/ipoib/" ofed/drivers/infiniband/ulp/sdp/sdp_bcopy.c optional sdp inet \ compile-with "${OFED_C} -I$S/ofed/drivers/infiniband/ulp/sdp/" ofed/drivers/infiniband/ulp/sdp/sdp_main.c optional sdp inet \ compile-with "${OFED_C} -I$S/ofed/drivers/infiniband/ulp/sdp/" ofed/drivers/infiniband/ulp/sdp/sdp_rx.c optional sdp inet \ compile-with "${OFED_C} -I$S/ofed/drivers/infiniband/ulp/sdp/" ofed/drivers/infiniband/ulp/sdp/sdp_cma.c optional sdp inet \ compile-with "${OFED_C} -I$S/ofed/drivers/infiniband/ulp/sdp/" ofed/drivers/infiniband/ulp/sdp/sdp_tx.c optional sdp inet \ compile-with "${OFED_C} -I$S/ofed/drivers/infiniband/ulp/sdp/" dev/mthca/mthca_allocator.c optional mthca pci ofed \ compile-with "${OFED_C}" dev/mthca/mthca_av.c optional mthca pci ofed \ compile-with "${OFED_C}" dev/mthca/mthca_catas.c optional mthca pci ofed \ compile-with "${OFED_C}" dev/mthca/mthca_cmd.c optional mthca pci ofed \ compile-with "${OFED_C}" dev/mthca/mthca_cq.c optional mthca pci ofed \ compile-with "${OFED_C}" dev/mthca/mthca_eq.c optional mthca pci ofed \ compile-with "${OFED_C}" dev/mthca/mthca_mad.c optional mthca pci ofed \ compile-with "${OFED_C}" dev/mthca/mthca_main.c optional mthca pci ofed \ compile-with "${OFED_C}" dev/mthca/mthca_mcg.c optional mthca pci ofed \ compile-with "${OFED_C}" dev/mthca/mthca_memfree.c optional mthca pci ofed \ compile-with "${OFED_C}" dev/mthca/mthca_mr.c optional mthca pci ofed \ compile-with "${OFED_C}" dev/mthca/mthca_pd.c optional mthca pci ofed \ compile-with "${OFED_C}" dev/mthca/mthca_profile.c optional mthca pci ofed \ compile-with "${OFED_C}" dev/mthca/mthca_provider.c optional mthca pci ofed \ compile-with "${OFED_C}" dev/mthca/mthca_qp.c optional mthca pci ofed \ compile-with "${OFED_C}" dev/mthca/mthca_reset.c optional mthca pci ofed \ compile-with "${OFED_C}" dev/mthca/mthca_srq.c optional mthca pci ofed \ compile-with "${OFED_C}" dev/mthca/mthca_uar.c optional mthca pci ofed \ compile-with "${OFED_C}" dev/mlx4/mlx4_ib/mlx4_ib_alias_GUID.c optional mlx4ib pci ofed \ compile-with "${OFED_C}" dev/mlx4/mlx4_ib/mlx4_ib_mcg.c optional mlx4ib pci ofed \ compile-with "${OFED_C}" dev/mlx4/mlx4_ib/mlx4_ib_sysfs.c optional mlx4ib pci ofed \ compile-with "${OFED_C}" dev/mlx4/mlx4_ib/mlx4_ib_cm.c optional mlx4ib pci ofed \ compile-with "${OFED_C}" dev/mlx4/mlx4_ib/mlx4_ib_ah.c optional mlx4ib pci ofed \ compile-with "${OFED_C}" dev/mlx4/mlx4_ib/mlx4_ib_cq.c optional mlx4ib pci ofed \ compile-with "${OFED_C}" dev/mlx4/mlx4_ib/mlx4_ib_doorbell.c optional mlx4ib pci ofed \ compile-with "${OFED_C}" dev/mlx4/mlx4_ib/mlx4_ib_mad.c optional mlx4ib pci ofed \ compile-with "${OFED_C}" dev/mlx4/mlx4_ib/mlx4_ib_main.c optional mlx4ib pci ofed \ compile-with "${OFED_C}" dev/mlx4/mlx4_ib/mlx4_ib_mr.c optional mlx4ib pci ofed \ compile-with "${OFED_C}" dev/mlx4/mlx4_ib/mlx4_ib_qp.c optional mlx4ib pci ofed \ compile-with "${OFED_C}" dev/mlx4/mlx4_ib/mlx4_ib_srq.c optional mlx4ib pci ofed \ compile-with "${OFED_C}" dev/mlx4/mlx4_ib/mlx4_ib_wc.c optional mlx4ib pci ofed \ compile-with "${OFED_C}" dev/mlx4/mlx4_core/mlx4_alloc.c optional mlx4 pci \ compile-with "${OFED_C}" dev/mlx4/mlx4_core/mlx4_catas.c optional mlx4 pci \ compile-with "${OFED_C}" dev/mlx4/mlx4_core/mlx4_cmd.c optional mlx4 pci \ compile-with "${OFED_C}" dev/mlx4/mlx4_core/mlx4_cq.c optional mlx4 pci \ compile-with "${OFED_C}" dev/mlx4/mlx4_core/mlx4_eq.c optional mlx4 pci \ compile-with "${OFED_C}" dev/mlx4/mlx4_core/mlx4_fw.c optional mlx4 pci \ compile-with "${OFED_C}" dev/mlx4/mlx4_core/mlx4_fw_qos.c optional mlx4 pci \ compile-with "${OFED_C}" dev/mlx4/mlx4_core/mlx4_icm.c optional mlx4 pci \ compile-with "${OFED_C}" dev/mlx4/mlx4_core/mlx4_intf.c optional mlx4 pci \ compile-with "${OFED_C}" dev/mlx4/mlx4_core/mlx4_main.c optional mlx4 pci \ compile-with "${OFED_C}" dev/mlx4/mlx4_core/mlx4_mcg.c optional mlx4 pci \ compile-with "${OFED_C}" dev/mlx4/mlx4_core/mlx4_mr.c optional mlx4 pci \ compile-with "${OFED_C}" dev/mlx4/mlx4_core/mlx4_pd.c optional mlx4 pci \ compile-with "${OFED_C}" dev/mlx4/mlx4_core/mlx4_port.c optional mlx4 pci \ compile-with "${OFED_C}" dev/mlx4/mlx4_core/mlx4_profile.c optional mlx4 pci \ compile-with "${OFED_C}" dev/mlx4/mlx4_core/mlx4_qp.c optional mlx4 pci \ compile-with "${OFED_C}" dev/mlx4/mlx4_core/mlx4_reset.c optional mlx4 pci \ compile-with "${OFED_C}" dev/mlx4/mlx4_core/mlx4_sense.c optional mlx4 pci \ compile-with "${OFED_C}" dev/mlx4/mlx4_core/mlx4_srq.c optional mlx4 pci \ compile-with "${OFED_C}" dev/mlx4/mlx4_core/mlx4_resource_tracker.c optional mlx4 pci \ compile-with "${OFED_C}" dev/mlx4/mlx4_en/mlx4_en_cq.c optional mlx4en pci inet inet6 \ compile-with "${OFED_C}" dev/mlx4/mlx4_en/mlx4_en_main.c optional mlx4en pci inet inet6 \ compile-with "${OFED_C}" dev/mlx4/mlx4_en/mlx4_en_netdev.c optional mlx4en pci inet inet6 \ compile-with "${OFED_C}" dev/mlx4/mlx4_en/mlx4_en_port.c optional mlx4en pci inet inet6 \ compile-with "${OFED_C}" dev/mlx4/mlx4_en/mlx4_en_resources.c optional mlx4en pci inet inet6 \ compile-with "${OFED_C}" dev/mlx4/mlx4_en/mlx4_en_rx.c optional mlx4en pci inet inet6 \ compile-with "${OFED_C}" dev/mlx4/mlx4_en/mlx4_en_tx.c optional mlx4en pci inet inet6 \ compile-with "${OFED_C}" dev/mlx5/mlx5_ib/mlx5_ib_ah.c optional mlx5ib pci ofed \ compile-with "${OFED_C}" dev/mlx5/mlx5_ib/mlx5_ib_cong.c optional mlx5ib pci ofed \ compile-with "${OFED_C}" dev/mlx5/mlx5_ib/mlx5_ib_cq.c optional mlx5ib pci ofed \ compile-with "${OFED_C}" dev/mlx5/mlx5_ib/mlx5_ib_doorbell.c optional mlx5ib pci ofed \ compile-with "${OFED_C}" dev/mlx5/mlx5_ib/mlx5_ib_gsi.c optional mlx5ib pci ofed \ compile-with "${OFED_C}" dev/mlx5/mlx5_ib/mlx5_ib_mad.c optional mlx5ib pci ofed \ compile-with "${OFED_C}" dev/mlx5/mlx5_ib/mlx5_ib_main.c optional mlx5ib pci ofed \ compile-with "${OFED_C}" dev/mlx5/mlx5_ib/mlx5_ib_mem.c optional mlx5ib pci ofed \ compile-with "${OFED_C}" dev/mlx5/mlx5_ib/mlx5_ib_mr.c optional mlx5ib pci ofed \ compile-with "${OFED_C}" dev/mlx5/mlx5_ib/mlx5_ib_qp.c optional mlx5ib pci ofed \ compile-with "${OFED_C}" dev/mlx5/mlx5_ib/mlx5_ib_srq.c optional mlx5ib pci ofed \ compile-with "${OFED_C}" dev/mlx5/mlx5_ib/mlx5_ib_virt.c optional mlx5ib pci ofed \ compile-with "${OFED_C}" dev/mlx5/mlx5_core/mlx5_alloc.c optional mlx5 pci \ compile-with "${OFED_C}" dev/mlx5/mlx5_core/mlx5_cmd.c optional mlx5 pci \ compile-with "${OFED_C}" dev/mlx5/mlx5_core/mlx5_cq.c optional mlx5 pci \ compile-with "${OFED_C}" dev/mlx5/mlx5_core/mlx5_diagnostics.c optional mlx5 pci \ compile-with "${OFED_C}" dev/mlx5/mlx5_core/mlx5_eq.c optional mlx5 pci \ compile-with "${OFED_C}" dev/mlx5/mlx5_core/mlx5_fs_cmd.c optional mlx5 pci \ compile-with "${OFED_C}" dev/mlx5/mlx5_core/mlx5_fs_tree.c optional mlx5 pci \ compile-with "${OFED_C}" dev/mlx5/mlx5_core/mlx5_fw.c optional mlx5 pci \ compile-with "${OFED_C}" dev/mlx5/mlx5_core/mlx5_fwdump.c optional mlx5 pci \ compile-with "${OFED_C}" dev/mlx5/mlx5_core/mlx5_health.c optional mlx5 pci \ compile-with "${OFED_C}" dev/mlx5/mlx5_core/mlx5_mad.c optional mlx5 pci \ compile-with "${OFED_C}" dev/mlx5/mlx5_core/mlx5_main.c optional mlx5 pci \ compile-with "${OFED_C}" dev/mlx5/mlx5_core/mlx5_mcg.c optional mlx5 pci \ compile-with "${OFED_C}" dev/mlx5/mlx5_core/mlx5_mpfs.c optional mlx5 pci \ compile-with "${OFED_C}" dev/mlx5/mlx5_core/mlx5_mr.c optional mlx5 pci \ compile-with "${OFED_C}" dev/mlx5/mlx5_core/mlx5_pagealloc.c optional mlx5 pci \ compile-with "${OFED_C}" dev/mlx5/mlx5_core/mlx5_pd.c optional mlx5 pci \ compile-with "${OFED_C}" dev/mlx5/mlx5_core/mlx5_port.c optional mlx5 pci \ compile-with "${OFED_C}" dev/mlx5/mlx5_core/mlx5_qp.c optional mlx5 pci \ compile-with "${OFED_C}" dev/mlx5/mlx5_core/mlx5_rl.c optional mlx5 pci \ compile-with "${OFED_C}" dev/mlx5/mlx5_core/mlx5_srq.c optional mlx5 pci \ compile-with "${OFED_C}" dev/mlx5/mlx5_core/mlx5_tls.c optional mlx5 pci \ compile-with "${OFED_C}" dev/mlx5/mlx5_core/mlx5_transobj.c optional mlx5 pci \ compile-with "${OFED_C}" dev/mlx5/mlx5_core/mlx5_uar.c optional mlx5 pci \ compile-with "${OFED_C}" dev/mlx5/mlx5_core/mlx5_vport.c optional mlx5 pci \ compile-with "${OFED_C}" dev/mlx5/mlx5_core/mlx5_vsc.c optional mlx5 pci \ compile-with "${OFED_C}" dev/mlx5/mlx5_core/mlx5_wq.c optional mlx5 pci \ compile-with "${OFED_C}" dev/mlx5/mlx5_lib/mlx5_gid.c optional mlx5 pci \ compile-with "${OFED_C}" dev/mlx5/mlx5_en/mlx5_en_dim.c optional mlx5en pci inet inet6 \ compile-with "${OFED_C}" dev/mlx5/mlx5_en/mlx5_en_ethtool.c optional mlx5en pci inet inet6 \ compile-with "${OFED_C}" dev/mlx5/mlx5_en/mlx5_en_main.c optional mlx5en pci inet inet6 \ compile-with "${OFED_C}" dev/mlx5/mlx5_en/mlx5_en_tx.c optional mlx5en pci inet inet6 \ compile-with "${OFED_C}" dev/mlx5/mlx5_en/mlx5_en_flow_table.c optional mlx5en pci inet inet6 \ compile-with "${OFED_C}" dev/mlx5/mlx5_en/mlx5_en_hw_tls.c optional mlx5en pci inet inet6 \ compile-with "${OFED_C}" dev/mlx5/mlx5_en/mlx5_en_rx.c optional mlx5en pci inet inet6 \ compile-with "${OFED_C}" dev/mlx5/mlx5_en/mlx5_en_rl.c optional mlx5en pci inet inet6 \ compile-with "${OFED_C}" dev/mlx5/mlx5_en/mlx5_en_txrx.c optional mlx5en pci inet inet6 \ compile-with "${OFED_C}" dev/mlx5/mlx5_en/mlx5_en_port_buffer.c optional mlx5en pci inet inet6 \ compile-with "${OFED_C}" # crypto support opencrypto/cast.c optional crypto | ipsec | ipsec_support opencrypto/criov.c optional crypto | ipsec | ipsec_support opencrypto/crypto.c optional crypto | ipsec | ipsec_support opencrypto/cryptodev.c optional cryptodev opencrypto/cryptodev_if.m optional crypto | ipsec | ipsec_support opencrypto/cryptosoft.c optional crypto | ipsec | ipsec_support opencrypto/cryptodeflate.c optional crypto | ipsec | ipsec_support opencrypto/gmac.c optional crypto | ipsec | ipsec_support opencrypto/gfmult.c optional crypto | ipsec | ipsec_support opencrypto/rmd160.c optional crypto | ipsec | ipsec_support opencrypto/skipjack.c optional crypto | ipsec | ipsec_support opencrypto/xform.c optional crypto | ipsec | ipsec_support opencrypto/xform_poly1305.c optional crypto \ compile-with "${NORMAL_C} -I$S/contrib/libsodium/src/libsodium/include -I$S/crypto/libsodium" contrib/libsodium/src/libsodium/crypto_onetimeauth/poly1305/onetimeauth_poly1305.c \ optional crypto \ compile-with "${NORMAL_C} -I$S/contrib/libsodium/src/libsodium/include/sodium -I$S/crypto/libsodium" contrib/libsodium/src/libsodium/crypto_onetimeauth/poly1305/donna/poly1305_donna.c \ optional crypto \ compile-with "${NORMAL_C} -I$S/contrib/libsodium/src/libsodium/include/sodium -I$S/crypto/libsodium" contrib/libsodium/src/libsodium/crypto_verify/sodium/verify.c \ optional crypto \ compile-with "${NORMAL_C} -I$S/contrib/libsodium/src/libsodium/include/sodium -I$S/crypto/libsodium" crypto/libsodium/randombytes.c optional crypto \ compile-with "${NORMAL_C} -I$S/contrib/libsodium/src/libsodium/include -I$S/crypto/libsodium" crypto/libsodium/utils.c optional crypto \ compile-with "${NORMAL_C} -I$S/contrib/libsodium/src/libsodium/include -I$S/crypto/libsodium" contrib/libsodium/src/libsodium/crypto_core/hchacha20/core_hchacha20.c \ optional crypto \ compile-with "${NORMAL_C} -I$S/contrib/libsodium/src/libsodium/include/sodium -I$S/crypto/libsodium" contrib/libsodium/src/libsodium/crypto_stream/chacha20/stream_chacha20.c \ optional crypto \ compile-with "${NORMAL_C} -I$S/contrib/libsodium/src/libsodium/include/sodium -I$S/crypto/libsodium" contrib/libsodium/src/libsodium/crypto_stream/chacha20/ref/chacha20_ref.c \ optional crypto \ compile-with "${NORMAL_C} -I$S/contrib/libsodium/src/libsodium/include/sodium -I$S/crypto/libsodium" contrib/libsodium/src/libsodium/crypto_aead/chacha20poly1305/sodium/aead_chacha20poly1305.c \ optional crypto \ compile-with "${NORMAL_C} -I$S/contrib/libsodium/src/libsodium/include/sodium -I$S/crypto/libsodium" contrib/libsodium/src/libsodium/crypto_aead/xchacha20poly1305/sodium/aead_xchacha20poly1305.c \ optional crypto \ compile-with "${NORMAL_C} -I$S/contrib/libsodium/src/libsodium/include/sodium -I$S/crypto/libsodium" opencrypto/cbc_mac.c optional crypto opencrypto/xform_cbc_mac.c optional crypto rpc/auth_none.c optional krpc | nfslockd | nfscl | nfsd rpc/auth_unix.c optional krpc | nfslockd | nfscl | nfsd rpc/authunix_prot.c optional krpc | nfslockd | nfscl | nfsd rpc/clnt_bck.c optional krpc | nfslockd | nfscl | nfsd rpc/clnt_dg.c optional krpc | nfslockd | nfscl | nfsd rpc/clnt_rc.c optional krpc | nfslockd | nfscl | nfsd rpc/clnt_vc.c optional krpc | nfslockd | nfscl | nfsd rpc/getnetconfig.c optional krpc | nfslockd | nfscl | nfsd rpc/replay.c optional krpc | nfslockd | nfscl | nfsd rpc/rpc_callmsg.c optional krpc | nfslockd | nfscl | nfsd rpc/rpc_generic.c optional krpc | nfslockd | nfscl | nfsd rpc/rpc_prot.c optional krpc | nfslockd | nfscl | nfsd rpc/rpcb_clnt.c optional krpc | nfslockd | nfscl | nfsd rpc/rpcb_prot.c optional krpc | nfslockd | nfscl | nfsd rpc/svc.c optional krpc | nfslockd | nfscl | nfsd rpc/svc_auth.c optional krpc | nfslockd | nfscl | nfsd rpc/svc_auth_unix.c optional krpc | nfslockd | nfscl | nfsd rpc/svc_dg.c optional krpc | nfslockd | nfscl | nfsd rpc/svc_generic.c optional krpc | nfslockd | nfscl | nfsd rpc/svc_vc.c optional krpc | nfslockd | nfscl | nfsd rpc/rpcsec_gss/rpcsec_gss.c optional krpc kgssapi | nfslockd kgssapi | nfscl kgssapi | nfsd kgssapi rpc/rpcsec_gss/rpcsec_gss_conf.c optional krpc kgssapi | nfslockd kgssapi | nfscl kgssapi | nfsd kgssapi rpc/rpcsec_gss/rpcsec_gss_misc.c optional krpc kgssapi | nfslockd kgssapi | nfscl kgssapi | nfsd kgssapi rpc/rpcsec_gss/rpcsec_gss_prot.c optional krpc kgssapi | nfslockd kgssapi | nfscl kgssapi | nfsd kgssapi rpc/rpcsec_gss/svc_rpcsec_gss.c optional krpc kgssapi | nfslockd kgssapi | nfscl kgssapi | nfsd kgssapi security/audit/audit.c optional audit security/audit/audit_arg.c optional audit security/audit/audit_bsm.c optional audit security/audit/audit_bsm_db.c optional audit security/audit/audit_bsm_klib.c optional audit security/audit/audit_dtrace.c optional dtaudit audit | dtraceall audit compile-with "${CDDL_C}" security/audit/audit_pipe.c optional audit security/audit/audit_syscalls.c standard security/audit/audit_trigger.c optional audit security/audit/audit_worker.c optional audit security/audit/bsm_domain.c optional audit security/audit/bsm_errno.c optional audit security/audit/bsm_fcntl.c optional audit security/audit/bsm_socket_type.c optional audit security/audit/bsm_token.c optional audit security/mac/mac_audit.c optional mac audit security/mac/mac_cred.c optional mac security/mac/mac_framework.c optional mac security/mac/mac_inet.c optional mac inet | mac inet6 security/mac/mac_inet6.c optional mac inet6 security/mac/mac_label.c optional mac security/mac/mac_net.c optional mac security/mac/mac_pipe.c optional mac security/mac/mac_posix_sem.c optional mac security/mac/mac_posix_shm.c optional mac security/mac/mac_priv.c optional mac security/mac/mac_process.c optional mac security/mac/mac_socket.c optional mac security/mac/mac_syscalls.c standard security/mac/mac_system.c optional mac security/mac/mac_sysv_msg.c optional mac security/mac/mac_sysv_sem.c optional mac security/mac/mac_sysv_shm.c optional mac security/mac/mac_vfs.c optional mac security/mac_biba/mac_biba.c optional mac_biba security/mac_bsdextended/mac_bsdextended.c optional mac_bsdextended security/mac_bsdextended/ugidfw_system.c optional mac_bsdextended security/mac_bsdextended/ugidfw_vnode.c optional mac_bsdextended security/mac_ifoff/mac_ifoff.c optional mac_ifoff security/mac_lomac/mac_lomac.c optional mac_lomac security/mac_mls/mac_mls.c optional mac_mls security/mac_none/mac_none.c optional mac_none security/mac_ntpd/mac_ntpd.c optional mac_ntpd security/mac_partition/mac_partition.c optional mac_partition security/mac_portacl/mac_portacl.c optional mac_portacl security/mac_seeotheruids/mac_seeotheruids.c optional mac_seeotheruids security/mac_stub/mac_stub.c optional mac_stub security/mac_test/mac_test.c optional mac_test security/mac_veriexec/mac_veriexec.c optional mac_veriexec security/mac_veriexec/veriexec_fingerprint.c optional mac_veriexec security/mac_veriexec/veriexec_metadata.c optional mac_veriexec security/mac_veriexec_parser/mac_veriexec_parser.c optional mac_veriexec mac_veriexec_parser security/mac_veriexec/mac_veriexec_rmd160.c optional mac_veriexec_rmd160 security/mac_veriexec/mac_veriexec_sha1.c optional mac_veriexec_sha1 security/mac_veriexec/mac_veriexec_sha256.c optional mac_veriexec_sha256 security/mac_veriexec/mac_veriexec_sha384.c optional mac_veriexec_sha384 security/mac_veriexec/mac_veriexec_sha512.c optional mac_veriexec_sha512 teken/teken.c optional sc !SC_NO_TERM_TEKEN | vt ufs/ffs/ffs_alloc.c optional ffs ufs/ffs/ffs_balloc.c optional ffs ufs/ffs/ffs_inode.c optional ffs ufs/ffs/ffs_snapshot.c optional ffs ufs/ffs/ffs_softdep.c optional ffs ufs/ffs/ffs_subr.c optional ffs | geom_label ufs/ffs/ffs_tables.c optional ffs | geom_label ufs/ffs/ffs_vfsops.c optional ffs ufs/ffs/ffs_vnops.c optional ffs ufs/ffs/ffs_rawread.c optional ffs directio ufs/ffs/ffs_suspend.c optional ffs ufs/ufs/ufs_acl.c optional ffs ufs/ufs/ufs_bmap.c optional ffs ufs/ufs/ufs_dirhash.c optional ffs ufs/ufs/ufs_extattr.c optional ffs ufs/ufs/ufs_gjournal.c optional ffs UFS_GJOURNAL ufs/ufs/ufs_inode.c optional ffs ufs/ufs/ufs_lookup.c optional ffs ufs/ufs/ufs_quota.c optional ffs ufs/ufs/ufs_vfsops.c optional ffs ufs/ufs/ufs_vnops.c optional ffs vm/default_pager.c standard vm/device_pager.c standard vm/phys_pager.c standard vm/redzone.c optional DEBUG_REDZONE vm/sg_pager.c standard vm/swap_pager.c standard vm/uma_core.c standard vm/uma_dbg.c standard vm/memguard.c optional DEBUG_MEMGUARD vm/vm_domainset.c standard vm/vm_fault.c standard vm/vm_glue.c standard vm/vm_init.c standard vm/vm_kern.c standard vm/vm_map.c standard vm/vm_meter.c standard vm/vm_mmap.c standard vm/vm_object.c standard vm/vm_page.c standard vm/vm_pageout.c standard vm/vm_pager.c standard vm/vm_phys.c standard vm/vm_radix.c standard vm/vm_reserv.c standard vm/vm_swapout.c optional !NO_SWAPPING vm/vm_swapout_dummy.c optional NO_SWAPPING vm/vm_unix.c standard vm/vnode_pager.c standard xen/features.c optional xenhvm xen/xenbus/xenbus_if.m optional xenhvm xen/xenbus/xenbus.c optional xenhvm xen/xenbus/xenbusb_if.m optional xenhvm xen/xenbus/xenbusb.c optional xenhvm xen/xenbus/xenbusb_front.c optional xenhvm xen/xenbus/xenbusb_back.c optional xenhvm xen/xenmem/xenmem_if.m optional xenhvm xdr/xdr.c optional krpc | nfslockd | nfscl | nfsd xdr/xdr_array.c optional krpc | nfslockd | nfscl | nfsd xdr/xdr_mbuf.c optional krpc | nfslockd | nfscl | nfsd xdr/xdr_mem.c optional krpc | nfslockd | nfscl | nfsd xdr/xdr_reference.c optional krpc | nfslockd | nfscl | nfsd xdr/xdr_sizeof.c optional krpc | nfslockd | nfscl | nfsd diff --git a/sys/dev/cxgbe/tom/t4_tom.c b/sys/dev/cxgbe/tom/t4_tom.c index 7967cb185163..46de597c28d0 100644 --- a/sys/dev/cxgbe/tom/t4_tom.c +++ b/sys/dev/cxgbe/tom/t4_tom.c @@ -1,1933 +1,1937 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2012 Chelsio Communications, Inc. * All rights reserved. * Written by: Navdeep Parhar * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include "opt_inet.h" #include "opt_inet6.h" #include "opt_kern_tls.h" #include "opt_ratelimit.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #define TCPSTATES #include #include #include #include #include #ifdef TCP_OFFLOAD #include "common/common.h" #include "common/t4_msg.h" #include "common/t4_regs.h" #include "common/t4_regs_values.h" #include "common/t4_tcb.h" #include "t4_clip.h" #include "tom/t4_tom_l2t.h" #include "tom/t4_tom.h" #include "tom/t4_tls.h" static struct protosw toe_protosw; static struct pr_usrreqs toe_usrreqs; static struct protosw toe6_protosw; static struct pr_usrreqs toe6_usrreqs; /* Module ops */ static int t4_tom_mod_load(void); static int t4_tom_mod_unload(void); static int t4_tom_modevent(module_t, int, void *); /* ULD ops and helpers */ static int t4_tom_activate(struct adapter *); static int t4_tom_deactivate(struct adapter *); static struct uld_info tom_uld_info = { .uld_id = ULD_TOM, .activate = t4_tom_activate, .deactivate = t4_tom_deactivate, }; static void release_offload_resources(struct toepcb *); static int alloc_tid_tabs(struct tid_info *); static void free_tid_tabs(struct tid_info *); static void free_tom_data(struct adapter *, struct tom_data *); static void reclaim_wr_resources(void *, int); struct toepcb * alloc_toepcb(struct vi_info *vi, int flags) { struct port_info *pi = vi->pi; struct adapter *sc = pi->adapter; struct toepcb *toep; int tx_credits, txsd_total, len; /* * The firmware counts tx work request credits in units of 16 bytes * each. Reserve room for an ABORT_REQ so the driver never has to worry * about tx credits if it wants to abort a connection. */ tx_credits = sc->params.ofldq_wr_cred; tx_credits -= howmany(sizeof(struct cpl_abort_req), 16); /* * Shortest possible tx work request is a fw_ofld_tx_data_wr + 1 byte * immediate payload, and firmware counts tx work request credits in * units of 16 byte. Calculate the maximum work requests possible. */ txsd_total = tx_credits / howmany(sizeof(struct fw_ofld_tx_data_wr) + 1, 16); len = offsetof(struct toepcb, txsd) + txsd_total * sizeof(struct ofld_tx_sdesc); toep = malloc(len, M_CXGBE, M_ZERO | flags); if (toep == NULL) return (NULL); refcount_init(&toep->refcount, 1); toep->td = sc->tom_softc; toep->vi = vi; toep->tid = -1; toep->tx_total = tx_credits; toep->tx_credits = tx_credits; mbufq_init(&toep->ulp_pduq, INT_MAX); mbufq_init(&toep->ulp_pdu_reclaimq, INT_MAX); toep->txsd_total = txsd_total; toep->txsd_avail = txsd_total; toep->txsd_pidx = 0; toep->txsd_cidx = 0; aiotx_init_toep(toep); return (toep); } /* * Initialize a toepcb after its params have been filled out. */ int init_toepcb(struct vi_info *vi, struct toepcb *toep) { struct conn_params *cp = &toep->params; struct port_info *pi = vi->pi; struct adapter *sc = pi->adapter; struct tx_cl_rl_params *tc; if (cp->tc_idx >= 0 && cp->tc_idx < sc->chip_params->nsched_cls) { tc = &pi->sched_params->cl_rl[cp->tc_idx]; mtx_lock(&sc->tc_lock); if (tc->flags & CLRL_ERR) { log(LOG_ERR, "%s: failed to associate traffic class %u with tid %u\n", device_get_nameunit(vi->dev), cp->tc_idx, toep->tid); cp->tc_idx = -1; } else { tc->refcount++; } mtx_unlock(&sc->tc_lock); } toep->ofld_txq = &sc->sge.ofld_txq[cp->txq_idx]; toep->ofld_rxq = &sc->sge.ofld_rxq[cp->rxq_idx]; toep->ctrlq = &sc->sge.ctrlq[pi->port_id]; tls_init_toep(toep); if (ulp_mode(toep) == ULP_MODE_TCPDDP) ddp_init_toep(toep); + toep->flags |= TPF_INITIALIZED; + return (0); } struct toepcb * hold_toepcb(struct toepcb *toep) { refcount_acquire(&toep->refcount); return (toep); } void free_toepcb(struct toepcb *toep) { if (refcount_release(&toep->refcount) == 0) return; KASSERT(!(toep->flags & TPF_ATTACHED), ("%s: attached to an inpcb", __func__)); KASSERT(!(toep->flags & TPF_CPL_PENDING), ("%s: CPL pending", __func__)); - if (ulp_mode(toep) == ULP_MODE_TCPDDP) - ddp_uninit_toep(toep); - tls_uninit_toep(toep); + if (toep->flags & TPF_INITIALIZED) { + if (ulp_mode(toep) == ULP_MODE_TCPDDP) + ddp_uninit_toep(toep); + tls_uninit_toep(toep); + } free(toep, M_CXGBE); } /* * Set up the socket for TCP offload. */ void offload_socket(struct socket *so, struct toepcb *toep) { struct tom_data *td = toep->td; struct inpcb *inp = sotoinpcb(so); struct tcpcb *tp = intotcpcb(inp); struct sockbuf *sb; INP_WLOCK_ASSERT(inp); /* Update socket */ sb = &so->so_snd; SOCKBUF_LOCK(sb); sb->sb_flags |= SB_NOCOALESCE; SOCKBUF_UNLOCK(sb); sb = &so->so_rcv; SOCKBUF_LOCK(sb); sb->sb_flags |= SB_NOCOALESCE; if (inp->inp_vflag & INP_IPV6) so->so_proto = &toe6_protosw; else so->so_proto = &toe_protosw; SOCKBUF_UNLOCK(sb); /* Update TCP PCB */ tp->tod = &td->tod; tp->t_toe = toep; tp->t_flags |= TF_TOE; /* Install an extra hold on inp */ toep->inp = inp; toep->flags |= TPF_ATTACHED; in_pcbref(inp); /* Add the TOE PCB to the active list */ mtx_lock(&td->toep_list_lock); TAILQ_INSERT_HEAD(&td->toep_list, toep, link); mtx_unlock(&td->toep_list_lock); } /* This is _not_ the normal way to "unoffload" a socket. */ void undo_offload_socket(struct socket *so) { struct inpcb *inp = sotoinpcb(so); struct tcpcb *tp = intotcpcb(inp); struct toepcb *toep = tp->t_toe; struct tom_data *td = toep->td; struct sockbuf *sb; INP_WLOCK_ASSERT(inp); sb = &so->so_snd; SOCKBUF_LOCK(sb); sb->sb_flags &= ~SB_NOCOALESCE; SOCKBUF_UNLOCK(sb); sb = &so->so_rcv; SOCKBUF_LOCK(sb); sb->sb_flags &= ~SB_NOCOALESCE; SOCKBUF_UNLOCK(sb); tp->tod = NULL; tp->t_toe = NULL; tp->t_flags &= ~TF_TOE; toep->inp = NULL; toep->flags &= ~TPF_ATTACHED; if (in_pcbrele_wlocked(inp)) panic("%s: inp freed.", __func__); mtx_lock(&td->toep_list_lock); TAILQ_REMOVE(&td->toep_list, toep, link); mtx_unlock(&td->toep_list_lock); } static void release_offload_resources(struct toepcb *toep) { struct tom_data *td = toep->td; struct adapter *sc = td_adapter(td); int tid = toep->tid; KASSERT(!(toep->flags & TPF_CPL_PENDING), ("%s: %p has CPL pending.", __func__, toep)); KASSERT(!(toep->flags & TPF_ATTACHED), ("%s: %p is still attached.", __func__, toep)); CTR5(KTR_CXGBE, "%s: toep %p (tid %d, l2te %p, ce %p)", __func__, toep, tid, toep->l2te, toep->ce); /* * These queues should have been emptied at approximately the same time * that a normal connection's socket's so_snd would have been purged or * drained. Do _not_ clean up here. */ MPASS(mbufq_len(&toep->ulp_pduq) == 0); MPASS(mbufq_len(&toep->ulp_pdu_reclaimq) == 0); #ifdef INVARIANTS if (ulp_mode(toep) == ULP_MODE_TCPDDP) ddp_assert_empty(toep); #endif MPASS(TAILQ_EMPTY(&toep->aiotx_jobq)); if (toep->l2te) t4_l2t_release(toep->l2te); if (tid >= 0) { remove_tid(sc, tid, toep->ce ? 2 : 1); release_tid(sc, tid, toep->ctrlq); } if (toep->ce) t4_release_lip(sc, toep->ce); if (toep->params.tc_idx != -1) t4_release_cl_rl(sc, toep->vi->pi->port_id, toep->params.tc_idx); mtx_lock(&td->toep_list_lock); TAILQ_REMOVE(&td->toep_list, toep, link); mtx_unlock(&td->toep_list_lock); free_toepcb(toep); } /* * The kernel is done with the TCP PCB and this is our opportunity to unhook the * toepcb hanging off of it. If the TOE driver is also done with the toepcb (no * pending CPL) then it is time to release all resources tied to the toepcb. * * Also gets called when an offloaded active open fails and the TOM wants the * kernel to take the TCP PCB back. */ static void t4_pcb_detach(struct toedev *tod __unused, struct tcpcb *tp) { #if defined(KTR) || defined(INVARIANTS) struct inpcb *inp = tp->t_inpcb; #endif struct toepcb *toep = tp->t_toe; INP_WLOCK_ASSERT(inp); KASSERT(toep != NULL, ("%s: toep is NULL", __func__)); KASSERT(toep->flags & TPF_ATTACHED, ("%s: not attached", __func__)); #ifdef KTR if (tp->t_state == TCPS_SYN_SENT) { CTR6(KTR_CXGBE, "%s: atid %d, toep %p (0x%x), inp %p (0x%x)", __func__, toep->tid, toep, toep->flags, inp, inp->inp_flags); } else { CTR6(KTR_CXGBE, "t4_pcb_detach: tid %d (%s), toep %p (0x%x), inp %p (0x%x)", toep->tid, tcpstates[tp->t_state], toep, toep->flags, inp, inp->inp_flags); } #endif tp->t_toe = NULL; tp->t_flags &= ~TF_TOE; toep->flags &= ~TPF_ATTACHED; if (!(toep->flags & TPF_CPL_PENDING)) release_offload_resources(toep); } /* * setsockopt handler. */ static void t4_ctloutput(struct toedev *tod, struct tcpcb *tp, int dir, int name) { struct adapter *sc = tod->tod_softc; struct toepcb *toep = tp->t_toe; if (dir == SOPT_GET) return; CTR4(KTR_CXGBE, "%s: tp %p, dir %u, name %u", __func__, tp, dir, name); switch (name) { case TCP_NODELAY: if (tp->t_state != TCPS_ESTABLISHED) break; toep->params.nagle = tp->t_flags & TF_NODELAY ? 0 : 1; t4_set_tcb_field(sc, toep->ctrlq, toep, W_TCB_T_FLAGS, V_TF_NAGLE(1), V_TF_NAGLE(toep->params.nagle), 0, 0); break; default: break; } } static inline uint64_t get_tcb_tflags(const uint64_t *tcb) { return ((be64toh(tcb[14]) << 32) | (be64toh(tcb[15]) >> 32)); } static inline uint32_t get_tcb_field(const uint64_t *tcb, u_int word, uint32_t mask, u_int shift) { #define LAST_WORD ((TCB_SIZE / 4) - 1) uint64_t t1, t2; int flit_idx; MPASS(mask != 0); MPASS(word <= LAST_WORD); MPASS(shift < 32); flit_idx = (LAST_WORD - word) / 2; if (word & 0x1) shift += 32; t1 = be64toh(tcb[flit_idx]) >> shift; t2 = 0; if (fls(mask) > 64 - shift) { /* * Will spill over into the next logical flit, which is the flit * before this one. The flit_idx before this one must be valid. */ MPASS(flit_idx > 0); t2 = be64toh(tcb[flit_idx - 1]) << (64 - shift); } return ((t2 | t1) & mask); #undef LAST_WORD } #define GET_TCB_FIELD(tcb, F) \ get_tcb_field(tcb, W_TCB_##F, M_TCB_##F, S_TCB_##F) /* * Issues a CPL_GET_TCB to read the entire TCB for the tid. */ static int send_get_tcb(struct adapter *sc, u_int tid) { struct cpl_get_tcb *cpl; struct wrq_cookie cookie; MPASS(tid < sc->tids.ntids); cpl = start_wrq_wr(&sc->sge.ctrlq[0], howmany(sizeof(*cpl), 16), &cookie); if (__predict_false(cpl == NULL)) return (ENOMEM); bzero(cpl, sizeof(*cpl)); INIT_TP_WR(cpl, tid); OPCODE_TID(cpl) = htobe32(MK_OPCODE_TID(CPL_GET_TCB, tid)); cpl->reply_ctrl = htobe16(V_REPLY_CHAN(0) | V_QUEUENO(sc->sge.ofld_rxq[0].iq.cntxt_id)); cpl->cookie = 0xff; commit_wrq_wr(&sc->sge.ctrlq[0], cpl, &cookie); return (0); } static struct tcb_histent * alloc_tcb_histent(struct adapter *sc, u_int tid, int flags) { struct tcb_histent *te; MPASS(flags == M_NOWAIT || flags == M_WAITOK); te = malloc(sizeof(*te), M_CXGBE, M_ZERO | flags); if (te == NULL) return (NULL); mtx_init(&te->te_lock, "TCB entry", NULL, MTX_DEF); callout_init_mtx(&te->te_callout, &te->te_lock, 0); te->te_adapter = sc; te->te_tid = tid; return (te); } static void free_tcb_histent(struct tcb_histent *te) { mtx_destroy(&te->te_lock); free(te, M_CXGBE); } /* * Start tracking the tid in the TCB history. */ int add_tid_to_history(struct adapter *sc, u_int tid) { struct tcb_histent *te = NULL; struct tom_data *td = sc->tom_softc; int rc; MPASS(tid < sc->tids.ntids); if (td->tcb_history == NULL) return (ENXIO); rw_wlock(&td->tcb_history_lock); if (td->tcb_history[tid] != NULL) { rc = EEXIST; goto done; } te = alloc_tcb_histent(sc, tid, M_NOWAIT); if (te == NULL) { rc = ENOMEM; goto done; } mtx_lock(&te->te_lock); rc = send_get_tcb(sc, tid); if (rc == 0) { te->te_flags |= TE_RPL_PENDING; td->tcb_history[tid] = te; } else { free(te, M_CXGBE); } mtx_unlock(&te->te_lock); done: rw_wunlock(&td->tcb_history_lock); return (rc); } static void remove_tcb_histent(struct tcb_histent *te) { struct adapter *sc = te->te_adapter; struct tom_data *td = sc->tom_softc; rw_assert(&td->tcb_history_lock, RA_WLOCKED); mtx_assert(&te->te_lock, MA_OWNED); MPASS(td->tcb_history[te->te_tid] == te); td->tcb_history[te->te_tid] = NULL; free_tcb_histent(te); rw_wunlock(&td->tcb_history_lock); } static inline struct tcb_histent * lookup_tcb_histent(struct adapter *sc, u_int tid, bool addrem) { struct tcb_histent *te; struct tom_data *td = sc->tom_softc; MPASS(tid < sc->tids.ntids); if (td->tcb_history == NULL) return (NULL); if (addrem) rw_wlock(&td->tcb_history_lock); else rw_rlock(&td->tcb_history_lock); te = td->tcb_history[tid]; if (te != NULL) { mtx_lock(&te->te_lock); return (te); /* with both locks held */ } if (addrem) rw_wunlock(&td->tcb_history_lock); else rw_runlock(&td->tcb_history_lock); return (te); } static inline void release_tcb_histent(struct tcb_histent *te) { struct adapter *sc = te->te_adapter; struct tom_data *td = sc->tom_softc; mtx_assert(&te->te_lock, MA_OWNED); mtx_unlock(&te->te_lock); rw_assert(&td->tcb_history_lock, RA_RLOCKED); rw_runlock(&td->tcb_history_lock); } static void request_tcb(void *arg) { struct tcb_histent *te = arg; mtx_assert(&te->te_lock, MA_OWNED); /* Noone else is supposed to update the histent. */ MPASS(!(te->te_flags & TE_RPL_PENDING)); if (send_get_tcb(te->te_adapter, te->te_tid) == 0) te->te_flags |= TE_RPL_PENDING; else callout_schedule(&te->te_callout, hz / 100); } static void update_tcb_histent(struct tcb_histent *te, const uint64_t *tcb) { struct tom_data *td = te->te_adapter->tom_softc; uint64_t tflags = get_tcb_tflags(tcb); uint8_t sample = 0; if (GET_TCB_FIELD(tcb, SND_MAX_RAW) != GET_TCB_FIELD(tcb, SND_UNA_RAW)) { if (GET_TCB_FIELD(tcb, T_RXTSHIFT) != 0) sample |= TS_RTO; if (GET_TCB_FIELD(tcb, T_DUPACKS) != 0) sample |= TS_DUPACKS; if (GET_TCB_FIELD(tcb, T_DUPACKS) >= td->dupack_threshold) sample |= TS_FASTREXMT; } if (GET_TCB_FIELD(tcb, SND_MAX_RAW) != 0) { uint32_t snd_wnd; sample |= TS_SND_BACKLOGGED; /* for whatever reason. */ snd_wnd = GET_TCB_FIELD(tcb, RCV_ADV); if (tflags & V_TF_RECV_SCALE(1)) snd_wnd <<= GET_TCB_FIELD(tcb, RCV_SCALE); if (GET_TCB_FIELD(tcb, SND_CWND) < snd_wnd) sample |= TS_CWND_LIMITED; /* maybe due to CWND */ } if (tflags & V_TF_CCTRL_ECN(1)) { /* * CE marker on incoming IP hdr, echoing ECE back in the TCP * hdr. Indicates congestion somewhere on the way from the peer * to this node. */ if (tflags & V_TF_CCTRL_ECE(1)) sample |= TS_ECN_ECE; /* * ECE seen and CWR sent (or about to be sent). Might indicate * congestion on the way to the peer. This node is reducing its * congestion window in response. */ if (tflags & (V_TF_CCTRL_CWR(1) | V_TF_CCTRL_RFR(1))) sample |= TS_ECN_CWR; } te->te_sample[te->te_pidx] = sample; if (++te->te_pidx == nitems(te->te_sample)) te->te_pidx = 0; memcpy(te->te_tcb, tcb, TCB_SIZE); te->te_flags |= TE_ACTIVE; } static int do_get_tcb_rpl(struct sge_iq *iq, const struct rss_header *rss, struct mbuf *m) { struct adapter *sc = iq->adapter; const struct cpl_get_tcb_rpl *cpl = mtod(m, const void *); const uint64_t *tcb = (const uint64_t *)(const void *)(cpl + 1); struct tcb_histent *te; const u_int tid = GET_TID(cpl); bool remove; remove = GET_TCB_FIELD(tcb, T_STATE) == TCPS_CLOSED; te = lookup_tcb_histent(sc, tid, remove); if (te == NULL) { /* Not in the history. Who issued the GET_TCB for this? */ device_printf(sc->dev, "tcb %u: flags 0x%016jx, state %u, " "srtt %u, sscale %u, rscale %u, cookie 0x%x\n", tid, (uintmax_t)get_tcb_tflags(tcb), GET_TCB_FIELD(tcb, T_STATE), GET_TCB_FIELD(tcb, T_SRTT), GET_TCB_FIELD(tcb, SND_SCALE), GET_TCB_FIELD(tcb, RCV_SCALE), cpl->cookie); goto done; } MPASS(te->te_flags & TE_RPL_PENDING); te->te_flags &= ~TE_RPL_PENDING; if (remove) { remove_tcb_histent(te); } else { update_tcb_histent(te, tcb); callout_reset(&te->te_callout, hz / 10, request_tcb, te); release_tcb_histent(te); } done: m_freem(m); return (0); } static void fill_tcp_info_from_tcb(struct adapter *sc, uint64_t *tcb, struct tcp_info *ti) { uint32_t v; ti->tcpi_state = GET_TCB_FIELD(tcb, T_STATE); v = GET_TCB_FIELD(tcb, T_SRTT); ti->tcpi_rtt = tcp_ticks_to_us(sc, v); v = GET_TCB_FIELD(tcb, T_RTTVAR); ti->tcpi_rttvar = tcp_ticks_to_us(sc, v); ti->tcpi_snd_ssthresh = GET_TCB_FIELD(tcb, SND_SSTHRESH); ti->tcpi_snd_cwnd = GET_TCB_FIELD(tcb, SND_CWND); ti->tcpi_rcv_nxt = GET_TCB_FIELD(tcb, RCV_NXT); v = GET_TCB_FIELD(tcb, TX_MAX); ti->tcpi_snd_nxt = v - GET_TCB_FIELD(tcb, SND_NXT_RAW); /* Receive window being advertised by us. */ ti->tcpi_rcv_wscale = GET_TCB_FIELD(tcb, SND_SCALE); /* Yes, SND. */ ti->tcpi_rcv_space = GET_TCB_FIELD(tcb, RCV_WND); /* Send window */ ti->tcpi_snd_wscale = GET_TCB_FIELD(tcb, RCV_SCALE); /* Yes, RCV. */ ti->tcpi_snd_wnd = GET_TCB_FIELD(tcb, RCV_ADV); if (get_tcb_tflags(tcb) & V_TF_RECV_SCALE(1)) ti->tcpi_snd_wnd <<= ti->tcpi_snd_wscale; else ti->tcpi_snd_wscale = 0; } static void fill_tcp_info_from_history(struct adapter *sc, struct tcb_histent *te, struct tcp_info *ti) { fill_tcp_info_from_tcb(sc, te->te_tcb, ti); } /* * Reads the TCB for the given tid using a memory window and copies it to 'buf' * in the same format as CPL_GET_TCB_RPL. */ static void read_tcb_using_memwin(struct adapter *sc, u_int tid, uint64_t *buf) { int i, j, k, rc; uint32_t addr; u_char *tcb, tmp; MPASS(tid < sc->tids.ntids); addr = t4_read_reg(sc, A_TP_CMM_TCB_BASE) + tid * TCB_SIZE; rc = read_via_memwin(sc, 2, addr, (uint32_t *)buf, TCB_SIZE); if (rc != 0) return; tcb = (u_char *)buf; for (i = 0, j = TCB_SIZE - 16; i < j; i += 16, j -= 16) { for (k = 0; k < 16; k++) { tmp = tcb[i + k]; tcb[i + k] = tcb[j + k]; tcb[j + k] = tmp; } } } static void fill_tcp_info(struct adapter *sc, u_int tid, struct tcp_info *ti) { uint64_t tcb[TCB_SIZE / sizeof(uint64_t)]; struct tcb_histent *te; ti->tcpi_toe_tid = tid; te = lookup_tcb_histent(sc, tid, false); if (te != NULL) { fill_tcp_info_from_history(sc, te, ti); release_tcb_histent(te); } else { if (!(sc->debug_flags & DF_DISABLE_TCB_CACHE)) { /* XXX: tell firmware to flush TCB cache. */ } read_tcb_using_memwin(sc, tid, tcb); fill_tcp_info_from_tcb(sc, tcb, ti); } } /* * Called by the kernel to allow the TOE driver to "refine" values filled up in * the tcp_info for an offloaded connection. */ static void t4_tcp_info(struct toedev *tod, struct tcpcb *tp, struct tcp_info *ti) { struct adapter *sc = tod->tod_softc; struct toepcb *toep = tp->t_toe; INP_WLOCK_ASSERT(tp->t_inpcb); MPASS(ti != NULL); fill_tcp_info(sc, toep->tid, ti); } #ifdef KERN_TLS static int t4_alloc_tls_session(struct toedev *tod, struct tcpcb *tp, struct ktls_session *tls) { struct toepcb *toep = tp->t_toe; INP_WLOCK_ASSERT(tp->t_inpcb); MPASS(tls != NULL); return (tls_alloc_ktls(toep, tls)); } #endif /* * The TOE driver will not receive any more CPLs for the tid associated with the * toepcb; release the hold on the inpcb. */ void final_cpl_received(struct toepcb *toep) { struct inpcb *inp = toep->inp; KASSERT(inp != NULL, ("%s: inp is NULL", __func__)); INP_WLOCK_ASSERT(inp); KASSERT(toep->flags & TPF_CPL_PENDING, ("%s: CPL not pending already?", __func__)); CTR6(KTR_CXGBE, "%s: tid %d, toep %p (0x%x), inp %p (0x%x)", __func__, toep->tid, toep, toep->flags, inp, inp->inp_flags); if (ulp_mode(toep) == ULP_MODE_TCPDDP) release_ddp_resources(toep); toep->inp = NULL; toep->flags &= ~TPF_CPL_PENDING; mbufq_drain(&toep->ulp_pdu_reclaimq); if (!(toep->flags & TPF_ATTACHED)) release_offload_resources(toep); if (!in_pcbrele_wlocked(inp)) INP_WUNLOCK(inp); } void insert_tid(struct adapter *sc, int tid, void *ctx, int ntids) { struct tid_info *t = &sc->tids; MPASS(tid >= t->tid_base); MPASS(tid - t->tid_base < t->ntids); t->tid_tab[tid - t->tid_base] = ctx; atomic_add_int(&t->tids_in_use, ntids); } void * lookup_tid(struct adapter *sc, int tid) { struct tid_info *t = &sc->tids; return (t->tid_tab[tid - t->tid_base]); } void update_tid(struct adapter *sc, int tid, void *ctx) { struct tid_info *t = &sc->tids; t->tid_tab[tid - t->tid_base] = ctx; } void remove_tid(struct adapter *sc, int tid, int ntids) { struct tid_info *t = &sc->tids; t->tid_tab[tid - t->tid_base] = NULL; atomic_subtract_int(&t->tids_in_use, ntids); } /* * What mtu_idx to use, given a 4-tuple. Note that both s->mss and tcp_mssopt * have the MSS that we should advertise in our SYN. Advertised MSS doesn't * account for any TCP options so the effective MSS (only payload, no headers or * options) could be different. */ static int find_best_mtu_idx(struct adapter *sc, struct in_conninfo *inc, struct offload_settings *s) { unsigned short *mtus = &sc->params.mtus[0]; int i, mss, mtu; MPASS(inc != NULL); mss = s->mss > 0 ? s->mss : tcp_mssopt(inc); if (inc->inc_flags & INC_ISIPV6) mtu = mss + sizeof(struct ip6_hdr) + sizeof(struct tcphdr); else mtu = mss + sizeof(struct ip) + sizeof(struct tcphdr); for (i = 0; i < NMTUS - 1 && mtus[i + 1] <= mtu; i++) continue; return (i); } /* * Determine the receive window size for a socket. */ u_long select_rcv_wnd(struct socket *so) { unsigned long wnd; SOCKBUF_LOCK_ASSERT(&so->so_rcv); wnd = sbspace(&so->so_rcv); if (wnd < MIN_RCV_WND) wnd = MIN_RCV_WND; return min(wnd, MAX_RCV_WND); } int select_rcv_wscale(void) { int wscale = 0; unsigned long space = sb_max; if (space > MAX_RCV_WND) space = MAX_RCV_WND; while (wscale < TCP_MAX_WINSHIFT && (TCP_MAXWIN << wscale) < space) wscale++; return (wscale); } __be64 calc_options0(struct vi_info *vi, struct conn_params *cp) { uint64_t opt0 = 0; opt0 |= F_TCAM_BYPASS; MPASS(cp->wscale >= 0 && cp->wscale <= M_WND_SCALE); opt0 |= V_WND_SCALE(cp->wscale); MPASS(cp->mtu_idx >= 0 && cp->mtu_idx < NMTUS); opt0 |= V_MSS_IDX(cp->mtu_idx); MPASS(cp->ulp_mode >= 0 && cp->ulp_mode <= M_ULP_MODE); opt0 |= V_ULP_MODE(cp->ulp_mode); MPASS(cp->opt0_bufsize >= 0 && cp->opt0_bufsize <= M_RCV_BUFSIZ); opt0 |= V_RCV_BUFSIZ(cp->opt0_bufsize); MPASS(cp->l2t_idx >= 0 && cp->l2t_idx < vi->pi->adapter->vres.l2t.size); opt0 |= V_L2T_IDX(cp->l2t_idx); opt0 |= V_SMAC_SEL(vi->smt_idx); opt0 |= V_TX_CHAN(vi->pi->tx_chan); MPASS(cp->keepalive == 0 || cp->keepalive == 1); opt0 |= V_KEEP_ALIVE(cp->keepalive); MPASS(cp->nagle == 0 || cp->nagle == 1); opt0 |= V_NAGLE(cp->nagle); return (htobe64(opt0)); } __be32 calc_options2(struct vi_info *vi, struct conn_params *cp) { uint32_t opt2 = 0; struct port_info *pi = vi->pi; struct adapter *sc = pi->adapter; /* * rx flow control, rx coalesce, congestion control, and tx pace are all * explicitly set by the driver. On T5+ the ISS is also set by the * driver to the value picked by the kernel. */ if (is_t4(sc)) { opt2 |= F_RX_FC_VALID | F_RX_COALESCE_VALID; opt2 |= F_CONG_CNTRL_VALID | F_PACE_VALID; } else { opt2 |= F_T5_OPT_2_VALID; /* all 4 valid */ opt2 |= F_T5_ISS; /* ISS provided in CPL */ } MPASS(cp->sack == 0 || cp->sack == 1); opt2 |= V_SACK_EN(cp->sack); MPASS(cp->tstamp == 0 || cp->tstamp == 1); opt2 |= V_TSTAMPS_EN(cp->tstamp); if (cp->wscale > 0) opt2 |= F_WND_SCALE_EN; MPASS(cp->ecn == 0 || cp->ecn == 1); opt2 |= V_CCTRL_ECN(cp->ecn); /* XXX: F_RX_CHANNEL for multiple rx c-chan support goes here. */ opt2 |= V_TX_QUEUE(sc->params.tp.tx_modq[pi->tx_chan]); opt2 |= V_PACE(0); opt2 |= F_RSS_QUEUE_VALID; opt2 |= V_RSS_QUEUE(sc->sge.ofld_rxq[cp->rxq_idx].iq.abs_id); MPASS(cp->cong_algo >= 0 && cp->cong_algo <= M_CONG_CNTRL); opt2 |= V_CONG_CNTRL(cp->cong_algo); MPASS(cp->rx_coalesce == 0 || cp->rx_coalesce == 1); if (cp->rx_coalesce == 1) opt2 |= V_RX_COALESCE(M_RX_COALESCE); opt2 |= V_RX_FC_DDP(0) | V_RX_FC_DISABLE(0); #ifdef USE_DDP_RX_FLOW_CONTROL if (cp->ulp_mode == ULP_MODE_TCPDDP) opt2 |= F_RX_FC_DDP; #endif if (cp->ulp_mode == ULP_MODE_TLS) opt2 |= F_RX_FC_DISABLE; return (htobe32(opt2)); } uint64_t select_ntuple(struct vi_info *vi, struct l2t_entry *e) { struct adapter *sc = vi->pi->adapter; struct tp_params *tp = &sc->params.tp; uint64_t ntuple = 0; /* * Initialize each of the fields which we care about which are present * in the Compressed Filter Tuple. */ if (tp->vlan_shift >= 0 && EVL_VLANOFTAG(e->vlan) != CPL_L2T_VLAN_NONE) ntuple |= (uint64_t)(F_FT_VLAN_VLD | e->vlan) << tp->vlan_shift; if (tp->port_shift >= 0) ntuple |= (uint64_t)e->lport << tp->port_shift; if (tp->protocol_shift >= 0) ntuple |= (uint64_t)IPPROTO_TCP << tp->protocol_shift; if (tp->vnic_shift >= 0 && tp->ingress_config & F_VNIC) { ntuple |= (uint64_t)(V_FT_VNID_ID_VF(vi->vin) | V_FT_VNID_ID_PF(sc->pf) | V_FT_VNID_ID_VLD(vi->vfvld)) << tp->vnic_shift; } if (is_t4(sc)) return (htobe32((uint32_t)ntuple)); else return (htobe64(V_FILTER_TUPLE(ntuple))); } static int is_tls_sock(struct socket *so, struct adapter *sc) { struct inpcb *inp = sotoinpcb(so); int i, rc; /* XXX: Eventually add a SO_WANT_TLS socket option perhaps? */ rc = 0; ADAPTER_LOCK(sc); for (i = 0; i < sc->tt.num_tls_rx_ports; i++) { if (inp->inp_lport == htons(sc->tt.tls_rx_ports[i]) || inp->inp_fport == htons(sc->tt.tls_rx_ports[i])) { rc = 1; break; } } ADAPTER_UNLOCK(sc); return (rc); } /* * Initialize various connection parameters. */ void init_conn_params(struct vi_info *vi , struct offload_settings *s, struct in_conninfo *inc, struct socket *so, const struct tcp_options *tcpopt, int16_t l2t_idx, struct conn_params *cp) { struct port_info *pi = vi->pi; struct adapter *sc = pi->adapter; struct tom_tunables *tt = &sc->tt; struct inpcb *inp = sotoinpcb(so); struct tcpcb *tp = intotcpcb(inp); u_long wnd; MPASS(s->offload != 0); /* Congestion control algorithm */ if (s->cong_algo >= 0) cp->cong_algo = s->cong_algo & M_CONG_CNTRL; else if (sc->tt.cong_algorithm >= 0) cp->cong_algo = tt->cong_algorithm & M_CONG_CNTRL; else { struct cc_algo *cc = CC_ALGO(tp); if (strcasecmp(cc->name, "reno") == 0) cp->cong_algo = CONG_ALG_RENO; else if (strcasecmp(cc->name, "tahoe") == 0) cp->cong_algo = CONG_ALG_TAHOE; if (strcasecmp(cc->name, "newreno") == 0) cp->cong_algo = CONG_ALG_NEWRENO; if (strcasecmp(cc->name, "highspeed") == 0) cp->cong_algo = CONG_ALG_HIGHSPEED; else { /* * Use newreno in case the algorithm selected by the * host stack is not supported by the hardware. */ cp->cong_algo = CONG_ALG_NEWRENO; } } /* Tx traffic scheduling class. */ if (s->sched_class >= 0 && s->sched_class < sc->chip_params->nsched_cls) { cp->tc_idx = s->sched_class; } else cp->tc_idx = -1; /* Nagle's algorithm. */ if (s->nagle >= 0) cp->nagle = s->nagle > 0 ? 1 : 0; else cp->nagle = tp->t_flags & TF_NODELAY ? 0 : 1; /* TCP Keepalive. */ if (V_tcp_always_keepalive || so_options_get(so) & SO_KEEPALIVE) cp->keepalive = 1; else cp->keepalive = 0; /* Optimization that's specific to T5 @ 40G. */ if (tt->tx_align >= 0) cp->tx_align = tt->tx_align > 0 ? 1 : 0; else if (chip_id(sc) == CHELSIO_T5 && (port_top_speed(pi) > 10 || sc->params.nports > 2)) cp->tx_align = 1; else cp->tx_align = 0; /* ULP mode. */ if (can_tls_offload(sc) && (s->tls > 0 || (s->tls < 0 && is_tls_sock(so, sc)))) cp->ulp_mode = ULP_MODE_TLS; else if (s->ddp > 0 || (s->ddp < 0 && sc->tt.ddp && (so_options_get(so) & SO_NO_DDP) == 0)) cp->ulp_mode = ULP_MODE_TCPDDP; else cp->ulp_mode = ULP_MODE_NONE; /* Rx coalescing. */ if (s->rx_coalesce >= 0) cp->rx_coalesce = s->rx_coalesce > 0 ? 1 : 0; else if (cp->ulp_mode == ULP_MODE_TLS) cp->rx_coalesce = 0; else if (tt->rx_coalesce >= 0) cp->rx_coalesce = tt->rx_coalesce > 0 ? 1 : 0; else cp->rx_coalesce = 1; /* default */ /* * Index in the PMTU table. This controls the MSS that we announce in * our SYN initially, but after ESTABLISHED it controls the MSS that we * use to send data. */ cp->mtu_idx = find_best_mtu_idx(sc, inc, s); /* Tx queue for this connection. */ if (s->txq >= 0 && s->txq < vi->nofldtxq) cp->txq_idx = s->txq; else cp->txq_idx = arc4random() % vi->nofldtxq; cp->txq_idx += vi->first_ofld_txq; /* Rx queue for this connection. */ if (s->rxq >= 0 && s->rxq < vi->nofldrxq) cp->rxq_idx = s->rxq; else cp->rxq_idx = arc4random() % vi->nofldrxq; cp->rxq_idx += vi->first_ofld_rxq; if (SOLISTENING(so)) { /* Passive open */ MPASS(tcpopt != NULL); /* TCP timestamp option */ if (tcpopt->tstamp && (s->tstamp > 0 || (s->tstamp < 0 && V_tcp_do_rfc1323))) cp->tstamp = 1; else cp->tstamp = 0; /* SACK */ if (tcpopt->sack && (s->sack > 0 || (s->sack < 0 && V_tcp_do_sack))) cp->sack = 1; else cp->sack = 0; /* Receive window scaling. */ if (tcpopt->wsf > 0 && tcpopt->wsf < 15 && V_tcp_do_rfc1323) cp->wscale = select_rcv_wscale(); else cp->wscale = 0; /* ECN */ if (tcpopt->ecn && /* XXX: review. */ (s->ecn > 0 || (s->ecn < 0 && V_tcp_do_ecn))) cp->ecn = 1; else cp->ecn = 0; wnd = max(so->sol_sbrcv_hiwat, MIN_RCV_WND); cp->opt0_bufsize = min(wnd >> 10, M_RCV_BUFSIZ); if (tt->sndbuf > 0) cp->sndbuf = tt->sndbuf; else if (so->sol_sbsnd_flags & SB_AUTOSIZE && V_tcp_do_autosndbuf) cp->sndbuf = 256 * 1024; else cp->sndbuf = so->sol_sbsnd_hiwat; } else { /* Active open */ /* TCP timestamp option */ if (s->tstamp > 0 || (s->tstamp < 0 && (tp->t_flags & TF_REQ_TSTMP))) cp->tstamp = 1; else cp->tstamp = 0; /* SACK */ if (s->sack > 0 || (s->sack < 0 && (tp->t_flags & TF_SACK_PERMIT))) cp->sack = 1; else cp->sack = 0; /* Receive window scaling */ if (tp->t_flags & TF_REQ_SCALE) cp->wscale = select_rcv_wscale(); else cp->wscale = 0; /* ECN */ if (s->ecn > 0 || (s->ecn < 0 && V_tcp_do_ecn == 1)) cp->ecn = 1; else cp->ecn = 0; SOCKBUF_LOCK(&so->so_rcv); wnd = max(select_rcv_wnd(so), MIN_RCV_WND); SOCKBUF_UNLOCK(&so->so_rcv); cp->opt0_bufsize = min(wnd >> 10, M_RCV_BUFSIZ); if (tt->sndbuf > 0) cp->sndbuf = tt->sndbuf; else { SOCKBUF_LOCK(&so->so_snd); if (so->so_snd.sb_flags & SB_AUTOSIZE && V_tcp_do_autosndbuf) cp->sndbuf = 256 * 1024; else cp->sndbuf = so->so_snd.sb_hiwat; SOCKBUF_UNLOCK(&so->so_snd); } } cp->l2t_idx = l2t_idx; /* This will be initialized on ESTABLISHED. */ cp->emss = 0; } int negative_advice(int status) { return (status == CPL_ERR_RTX_NEG_ADVICE || status == CPL_ERR_PERSIST_NEG_ADVICE || status == CPL_ERR_KEEPALV_NEG_ADVICE); } static int alloc_tid_tab(struct tid_info *t, int flags) { MPASS(t->ntids > 0); MPASS(t->tid_tab == NULL); t->tid_tab = malloc(t->ntids * sizeof(*t->tid_tab), M_CXGBE, M_ZERO | flags); if (t->tid_tab == NULL) return (ENOMEM); atomic_store_rel_int(&t->tids_in_use, 0); return (0); } static void free_tid_tab(struct tid_info *t) { KASSERT(t->tids_in_use == 0, ("%s: %d tids still in use.", __func__, t->tids_in_use)); free(t->tid_tab, M_CXGBE); t->tid_tab = NULL; } static int alloc_stid_tab(struct tid_info *t, int flags) { MPASS(t->nstids > 0); MPASS(t->stid_tab == NULL); t->stid_tab = malloc(t->nstids * sizeof(*t->stid_tab), M_CXGBE, M_ZERO | flags); if (t->stid_tab == NULL) return (ENOMEM); mtx_init(&t->stid_lock, "stid lock", NULL, MTX_DEF); t->stids_in_use = 0; TAILQ_INIT(&t->stids); t->nstids_free_head = t->nstids; return (0); } static void free_stid_tab(struct tid_info *t) { KASSERT(t->stids_in_use == 0, ("%s: %d tids still in use.", __func__, t->stids_in_use)); if (mtx_initialized(&t->stid_lock)) mtx_destroy(&t->stid_lock); free(t->stid_tab, M_CXGBE); t->stid_tab = NULL; } static void free_tid_tabs(struct tid_info *t) { free_tid_tab(t); free_stid_tab(t); } static int alloc_tid_tabs(struct tid_info *t) { int rc; rc = alloc_tid_tab(t, M_NOWAIT); if (rc != 0) goto failed; rc = alloc_stid_tab(t, M_NOWAIT); if (rc != 0) goto failed; return (0); failed: free_tid_tabs(t); return (rc); } static inline void alloc_tcb_history(struct adapter *sc, struct tom_data *td) { if (sc->tids.ntids == 0 || sc->tids.ntids > 1024) return; rw_init(&td->tcb_history_lock, "TCB history"); td->tcb_history = malloc(sc->tids.ntids * sizeof(*td->tcb_history), M_CXGBE, M_ZERO | M_NOWAIT); td->dupack_threshold = G_DUPACKTHRESH(t4_read_reg(sc, A_TP_PARA_REG0)); } static inline void free_tcb_history(struct adapter *sc, struct tom_data *td) { #ifdef INVARIANTS int i; if (td->tcb_history != NULL) { for (i = 0; i < sc->tids.ntids; i++) { MPASS(td->tcb_history[i] == NULL); } } #endif free(td->tcb_history, M_CXGBE); if (rw_initialized(&td->tcb_history_lock)) rw_destroy(&td->tcb_history_lock); } static void free_tom_data(struct adapter *sc, struct tom_data *td) { ASSERT_SYNCHRONIZED_OP(sc); KASSERT(TAILQ_EMPTY(&td->toep_list), ("%s: TOE PCB list is not empty.", __func__)); KASSERT(td->lctx_count == 0, ("%s: lctx hash table is not empty.", __func__)); t4_free_ppod_region(&td->pr); if (td->listen_mask != 0) hashdestroy(td->listen_hash, M_CXGBE, td->listen_mask); if (mtx_initialized(&td->unsent_wr_lock)) mtx_destroy(&td->unsent_wr_lock); if (mtx_initialized(&td->lctx_hash_lock)) mtx_destroy(&td->lctx_hash_lock); if (mtx_initialized(&td->toep_list_lock)) mtx_destroy(&td->toep_list_lock); free_tcb_history(sc, td); free_tid_tabs(&sc->tids); free(td, M_CXGBE); } static char * prepare_pkt(int open_type, uint16_t vtag, struct inpcb *inp, int *pktlen, int *buflen) { char *pkt; struct tcphdr *th; int ipv6, len; const int maxlen = max(sizeof(struct ether_header), sizeof(struct ether_vlan_header)) + max(sizeof(struct ip), sizeof(struct ip6_hdr)) + sizeof(struct tcphdr); MPASS(open_type == OPEN_TYPE_ACTIVE || open_type == OPEN_TYPE_LISTEN); pkt = malloc(maxlen, M_CXGBE, M_ZERO | M_NOWAIT); if (pkt == NULL) return (NULL); ipv6 = inp->inp_vflag & INP_IPV6; len = 0; if (EVL_VLANOFTAG(vtag) == 0xfff) { struct ether_header *eh = (void *)pkt; if (ipv6) eh->ether_type = htons(ETHERTYPE_IPV6); else eh->ether_type = htons(ETHERTYPE_IP); len += sizeof(*eh); } else { struct ether_vlan_header *evh = (void *)pkt; evh->evl_encap_proto = htons(ETHERTYPE_VLAN); evh->evl_tag = htons(vtag); if (ipv6) evh->evl_proto = htons(ETHERTYPE_IPV6); else evh->evl_proto = htons(ETHERTYPE_IP); len += sizeof(*evh); } if (ipv6) { struct ip6_hdr *ip6 = (void *)&pkt[len]; ip6->ip6_vfc = IPV6_VERSION; ip6->ip6_plen = htons(sizeof(struct tcphdr)); ip6->ip6_nxt = IPPROTO_TCP; if (open_type == OPEN_TYPE_ACTIVE) { ip6->ip6_src = inp->in6p_laddr; ip6->ip6_dst = inp->in6p_faddr; } else if (open_type == OPEN_TYPE_LISTEN) { ip6->ip6_src = inp->in6p_laddr; ip6->ip6_dst = ip6->ip6_src; } len += sizeof(*ip6); } else { struct ip *ip = (void *)&pkt[len]; ip->ip_v = IPVERSION; ip->ip_hl = sizeof(*ip) >> 2; ip->ip_tos = inp->inp_ip_tos; ip->ip_len = htons(sizeof(struct ip) + sizeof(struct tcphdr)); ip->ip_ttl = inp->inp_ip_ttl; ip->ip_p = IPPROTO_TCP; if (open_type == OPEN_TYPE_ACTIVE) { ip->ip_src = inp->inp_laddr; ip->ip_dst = inp->inp_faddr; } else if (open_type == OPEN_TYPE_LISTEN) { ip->ip_src = inp->inp_laddr; ip->ip_dst = ip->ip_src; } len += sizeof(*ip); } th = (void *)&pkt[len]; if (open_type == OPEN_TYPE_ACTIVE) { th->th_sport = inp->inp_lport; /* network byte order already */ th->th_dport = inp->inp_fport; /* ditto */ } else if (open_type == OPEN_TYPE_LISTEN) { th->th_sport = inp->inp_lport; /* network byte order already */ th->th_dport = th->th_sport; } len += sizeof(th); *pktlen = *buflen = len; return (pkt); } const struct offload_settings * lookup_offload_policy(struct adapter *sc, int open_type, struct mbuf *m, uint16_t vtag, struct inpcb *inp) { const struct t4_offload_policy *op; char *pkt; struct offload_rule *r; int i, matched, pktlen, buflen; static const struct offload_settings allow_offloading_settings = { .offload = 1, .rx_coalesce = -1, .cong_algo = -1, .sched_class = -1, .tstamp = -1, .sack = -1, .nagle = -1, .ecn = -1, .ddp = -1, .tls = -1, .txq = -1, .rxq = -1, .mss = -1, }; static const struct offload_settings disallow_offloading_settings = { .offload = 0, /* rest is irrelevant when offload is off. */ }; rw_assert(&sc->policy_lock, RA_LOCKED); /* * If there's no Connection Offloading Policy attached to the device * then we need to return a default static policy. If * "cop_managed_offloading" is true, then we need to disallow * offloading until a COP is attached to the device. Otherwise we * allow offloading ... */ op = sc->policy; if (op == NULL) { if (sc->tt.cop_managed_offloading) return (&disallow_offloading_settings); else return (&allow_offloading_settings); } switch (open_type) { case OPEN_TYPE_ACTIVE: case OPEN_TYPE_LISTEN: pkt = prepare_pkt(open_type, vtag, inp, &pktlen, &buflen); break; case OPEN_TYPE_PASSIVE: MPASS(m != NULL); pkt = mtod(m, char *); MPASS(*pkt == CPL_PASS_ACCEPT_REQ); pkt += sizeof(struct cpl_pass_accept_req); pktlen = m->m_pkthdr.len - sizeof(struct cpl_pass_accept_req); buflen = m->m_len - sizeof(struct cpl_pass_accept_req); break; default: MPASS(0); return (&disallow_offloading_settings); } if (pkt == NULL || pktlen == 0 || buflen == 0) return (&disallow_offloading_settings); matched = 0; r = &op->rule[0]; for (i = 0; i < op->nrules; i++, r++) { if (r->open_type != open_type && r->open_type != OPEN_TYPE_DONTCARE) { continue; } matched = bpf_filter(r->bpf_prog.bf_insns, pkt, pktlen, buflen); if (matched) break; } if (open_type == OPEN_TYPE_ACTIVE || open_type == OPEN_TYPE_LISTEN) free(pkt, M_CXGBE); return (matched ? &r->settings : &disallow_offloading_settings); } static void reclaim_wr_resources(void *arg, int count) { struct tom_data *td = arg; STAILQ_HEAD(, wrqe) twr_list = STAILQ_HEAD_INITIALIZER(twr_list); struct cpl_act_open_req *cpl; u_int opcode, atid, tid; struct wrqe *wr; struct adapter *sc = td_adapter(td); mtx_lock(&td->unsent_wr_lock); STAILQ_SWAP(&td->unsent_wr_list, &twr_list, wrqe); mtx_unlock(&td->unsent_wr_lock); while ((wr = STAILQ_FIRST(&twr_list)) != NULL) { STAILQ_REMOVE_HEAD(&twr_list, link); cpl = wrtod(wr); opcode = GET_OPCODE(cpl); switch (opcode) { case CPL_ACT_OPEN_REQ: case CPL_ACT_OPEN_REQ6: atid = G_TID_TID(be32toh(OPCODE_TID(cpl))); CTR2(KTR_CXGBE, "%s: atid %u ", __func__, atid); act_open_failure_cleanup(sc, atid, EHOSTUNREACH); free(wr, M_CXGBE); break; case CPL_PASS_ACCEPT_RPL: tid = GET_TID(cpl); CTR2(KTR_CXGBE, "%s: tid %u ", __func__, tid); synack_failure_cleanup(sc, tid); free(wr, M_CXGBE); break; default: log(LOG_ERR, "%s: leaked work request %p, wr_len %d, " "opcode %x\n", __func__, wr, wr->wr_len, opcode); /* WR not freed here; go look at it with a debugger. */ } } } /* * Ground control to Major TOM * Commencing countdown, engines on */ static int t4_tom_activate(struct adapter *sc) { struct tom_data *td; struct toedev *tod; struct vi_info *vi; int i, rc, v; ASSERT_SYNCHRONIZED_OP(sc); /* per-adapter softc for TOM */ td = malloc(sizeof(*td), M_CXGBE, M_ZERO | M_NOWAIT); if (td == NULL) return (ENOMEM); /* List of TOE PCBs and associated lock */ mtx_init(&td->toep_list_lock, "PCB list lock", NULL, MTX_DEF); TAILQ_INIT(&td->toep_list); /* Listen context */ mtx_init(&td->lctx_hash_lock, "lctx hash lock", NULL, MTX_DEF); td->listen_hash = hashinit_flags(LISTEN_HASH_SIZE, M_CXGBE, &td->listen_mask, HASH_NOWAIT); /* List of WRs for which L2 resolution failed */ mtx_init(&td->unsent_wr_lock, "Unsent WR list lock", NULL, MTX_DEF); STAILQ_INIT(&td->unsent_wr_list); TASK_INIT(&td->reclaim_wr_resources, 0, reclaim_wr_resources, td); /* TID tables */ rc = alloc_tid_tabs(&sc->tids); if (rc != 0) goto done; rc = t4_init_ppod_region(&td->pr, &sc->vres.ddp, t4_read_reg(sc, A_ULP_RX_TDDP_PSZ), "TDDP page pods"); if (rc != 0) goto done; t4_set_reg_field(sc, A_ULP_RX_TDDP_TAGMASK, V_TDDPTAGMASK(M_TDDPTAGMASK), td->pr.pr_tag_mask); alloc_tcb_history(sc, td); /* toedev ops */ tod = &td->tod; init_toedev(tod); tod->tod_softc = sc; tod->tod_connect = t4_connect; tod->tod_listen_start = t4_listen_start; tod->tod_listen_stop = t4_listen_stop; tod->tod_rcvd = t4_rcvd; tod->tod_output = t4_tod_output; tod->tod_send_rst = t4_send_rst; tod->tod_send_fin = t4_send_fin; tod->tod_pcb_detach = t4_pcb_detach; tod->tod_l2_update = t4_l2_update; tod->tod_syncache_added = t4_syncache_added; tod->tod_syncache_removed = t4_syncache_removed; tod->tod_syncache_respond = t4_syncache_respond; tod->tod_offload_socket = t4_offload_socket; tod->tod_ctloutput = t4_ctloutput; tod->tod_tcp_info = t4_tcp_info; #ifdef KERN_TLS tod->tod_alloc_tls_session = t4_alloc_tls_session; #endif for_each_port(sc, i) { for_each_vi(sc->port[i], v, vi) { TOEDEV(vi->ifp) = &td->tod; } } sc->tom_softc = td; register_toedev(sc->tom_softc); done: if (rc != 0) free_tom_data(sc, td); return (rc); } static int t4_tom_deactivate(struct adapter *sc) { int rc = 0; struct tom_data *td = sc->tom_softc; ASSERT_SYNCHRONIZED_OP(sc); if (td == NULL) return (0); /* XXX. KASSERT? */ if (sc->offload_map != 0) return (EBUSY); /* at least one port has IFCAP_TOE enabled */ if (uld_active(sc, ULD_IWARP) || uld_active(sc, ULD_ISCSI)) return (EBUSY); /* both iWARP and iSCSI rely on the TOE. */ mtx_lock(&td->toep_list_lock); if (!TAILQ_EMPTY(&td->toep_list)) rc = EBUSY; mtx_unlock(&td->toep_list_lock); mtx_lock(&td->lctx_hash_lock); if (td->lctx_count > 0) rc = EBUSY; mtx_unlock(&td->lctx_hash_lock); taskqueue_drain(taskqueue_thread, &td->reclaim_wr_resources); mtx_lock(&td->unsent_wr_lock); if (!STAILQ_EMPTY(&td->unsent_wr_list)) rc = EBUSY; mtx_unlock(&td->unsent_wr_lock); if (rc == 0) { unregister_toedev(sc->tom_softc); free_tom_data(sc, td); sc->tom_softc = NULL; } return (rc); } static int t4_aio_queue_tom(struct socket *so, struct kaiocb *job) { struct tcpcb *tp = so_sototcpcb(so); struct toepcb *toep = tp->t_toe; int error; if (ulp_mode(toep) == ULP_MODE_TCPDDP) { error = t4_aio_queue_ddp(so, job); if (error != EOPNOTSUPP) return (error); } return (t4_aio_queue_aiotx(so, job)); } static int t4_ctloutput_tom(struct socket *so, struct sockopt *sopt) { if (sopt->sopt_level != IPPROTO_TCP) return (tcp_ctloutput(so, sopt)); switch (sopt->sopt_name) { case TCP_TLSOM_SET_TLS_CONTEXT: case TCP_TLSOM_GET_TLS_TOM: case TCP_TLSOM_CLR_TLS_TOM: case TCP_TLSOM_CLR_QUIES: return (t4_ctloutput_tls(so, sopt)); default: return (tcp_ctloutput(so, sopt)); } } static int t4_tom_mod_load(void) { struct protosw *tcp_protosw, *tcp6_protosw; /* CPL handlers */ t4_register_cpl_handler(CPL_GET_TCB_RPL, do_get_tcb_rpl); t4_register_shared_cpl_handler(CPL_L2T_WRITE_RPL, do_l2t_write_rpl2, CPL_COOKIE_TOM); t4_init_connect_cpl_handlers(); t4_init_listen_cpl_handlers(); t4_init_cpl_io_handlers(); t4_ddp_mod_load(); t4_tls_mod_load(); tcp_protosw = pffindproto(PF_INET, IPPROTO_TCP, SOCK_STREAM); if (tcp_protosw == NULL) return (ENOPROTOOPT); bcopy(tcp_protosw, &toe_protosw, sizeof(toe_protosw)); bcopy(tcp_protosw->pr_usrreqs, &toe_usrreqs, sizeof(toe_usrreqs)); toe_usrreqs.pru_aio_queue = t4_aio_queue_tom; toe_protosw.pr_ctloutput = t4_ctloutput_tom; toe_protosw.pr_usrreqs = &toe_usrreqs; tcp6_protosw = pffindproto(PF_INET6, IPPROTO_TCP, SOCK_STREAM); if (tcp6_protosw == NULL) return (ENOPROTOOPT); bcopy(tcp6_protosw, &toe6_protosw, sizeof(toe6_protosw)); bcopy(tcp6_protosw->pr_usrreqs, &toe6_usrreqs, sizeof(toe6_usrreqs)); toe6_usrreqs.pru_aio_queue = t4_aio_queue_tom; toe6_protosw.pr_ctloutput = t4_ctloutput_tom; toe6_protosw.pr_usrreqs = &toe6_usrreqs; return (t4_register_uld(&tom_uld_info)); } static void tom_uninit(struct adapter *sc, void *arg __unused) { if (begin_synchronized_op(sc, NULL, SLEEP_OK | INTR_OK, "t4tomun")) return; /* Try to free resources (works only if no port has IFCAP_TOE) */ if (uld_active(sc, ULD_TOM)) t4_deactivate_uld(sc, ULD_TOM); end_synchronized_op(sc, 0); } static int t4_tom_mod_unload(void) { t4_iterate(tom_uninit, NULL); if (t4_unregister_uld(&tom_uld_info) == EBUSY) return (EBUSY); t4_tls_mod_unload(); t4_ddp_mod_unload(); t4_uninit_connect_cpl_handlers(); t4_uninit_listen_cpl_handlers(); t4_uninit_cpl_io_handlers(); t4_register_shared_cpl_handler(CPL_L2T_WRITE_RPL, NULL, CPL_COOKIE_TOM); return (0); } #endif /* TCP_OFFLOAD */ static int t4_tom_modevent(module_t mod, int cmd, void *arg) { int rc = 0; #ifdef TCP_OFFLOAD switch (cmd) { case MOD_LOAD: rc = t4_tom_mod_load(); break; case MOD_UNLOAD: rc = t4_tom_mod_unload(); break; default: rc = EINVAL; } #else printf("t4_tom: compiled without TCP_OFFLOAD support.\n"); rc = EOPNOTSUPP; #endif return (rc); } static moduledata_t t4_tom_moddata= { "t4_tom", t4_tom_modevent, 0 }; MODULE_VERSION(t4_tom, 1); MODULE_DEPEND(t4_tom, toecore, 1, 1, 1); MODULE_DEPEND(t4_tom, t4nex, 1, 1, 1); DECLARE_MODULE(t4_tom, t4_tom_moddata, SI_SUB_EXEC, SI_ORDER_ANY); diff --git a/sys/dev/cxgbe/tom/t4_tom.h b/sys/dev/cxgbe/tom/t4_tom.h index 3da4189014c8..604d9786e34d 100644 --- a/sys/dev/cxgbe/tom/t4_tom.h +++ b/sys/dev/cxgbe/tom/t4_tom.h @@ -1,455 +1,456 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2012, 2015 Chelsio Communications, Inc. * All rights reserved. * Written by: Navdeep Parhar * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ * */ #ifndef __T4_TOM_H__ #define __T4_TOM_H__ #include #include "common/t4_hw.h" #include "common/t4_msg.h" #include "tom/t4_tls.h" #define LISTEN_HASH_SIZE 32 /* * Min receive window. We want it to be large enough to accommodate receive * coalescing, handle jumbo frames, and not trigger sender SWS avoidance. */ #define MIN_RCV_WND (24 * 1024U) /* * Max receive window supported by HW in bytes. Only a small part of it can * be set through option0, the rest needs to be set through RX_DATA_ACK. */ #define MAX_RCV_WND ((1U << 27) - 1) #define DDP_RSVD_WIN (16 * 1024U) #define SB_DDP_INDICATE SB_IN_TOE /* soreceive must respond to indicate */ #define USE_DDP_RX_FLOW_CONTROL #define PPOD_SZ(n) ((n) * sizeof(struct pagepod)) #define PPOD_SIZE (PPOD_SZ(1)) /* TOE PCB flags */ enum { TPF_ATTACHED = (1 << 0), /* a tcpcb refers to this toepcb */ TPF_FLOWC_WR_SENT = (1 << 1), /* firmware flow context WR sent */ TPF_TX_DATA_SENT = (1 << 2), /* some data sent */ TPF_TX_SUSPENDED = (1 << 3), /* tx suspended for lack of resources */ TPF_SEND_FIN = (1 << 4), /* send FIN after all pending data */ TPF_FIN_SENT = (1 << 5), /* FIN has been sent */ TPF_ABORT_SHUTDOWN = (1 << 6), /* connection abort is in progress */ TPF_CPL_PENDING = (1 << 7), /* haven't received the last CPL */ TPF_SYNQE = (1 << 8), /* synq_entry, not really a toepcb */ TPF_SYNQE_EXPANDED = (1 << 9), /* toepcb ready, tid context updated */ TPF_FORCE_CREDITS = (1 << 10), /* always send credits */ TPF_KTLS = (1 << 11), /* send TLS records from KTLS */ + TPF_INITIALIZED = (1 << 12), /* init_toepcb has been called */ }; enum { DDP_OK = (1 << 0), /* OK to turn on DDP */ DDP_SC_REQ = (1 << 1), /* state change (on/off) requested */ DDP_ON = (1 << 2), /* DDP is turned on */ DDP_BUF0_ACTIVE = (1 << 3), /* buffer 0 in use (not invalidated) */ DDP_BUF1_ACTIVE = (1 << 4), /* buffer 1 in use (not invalidated) */ DDP_TASK_ACTIVE = (1 << 5), /* requeue task is queued / running */ DDP_DEAD = (1 << 6), /* toepcb is shutting down */ }; struct sockopt; struct offload_settings; /* * Connection parameters for an offloaded connection. These are mostly (but not * all) hardware TOE parameters. */ struct conn_params { int8_t rx_coalesce; int8_t cong_algo; int8_t tc_idx; int8_t tstamp; int8_t sack; int8_t nagle; int8_t keepalive; int8_t wscale; int8_t ecn; int8_t mtu_idx; int8_t ulp_mode; int8_t tx_align; int16_t txq_idx; /* ofld_txq = &sc->sge.ofld_txq[txq_idx] */ int16_t rxq_idx; /* ofld_rxq = &sc->sge.ofld_rxq[rxq_idx] */ int16_t l2t_idx; uint16_t emss; uint16_t opt0_bufsize; u_int sndbuf; /* controls TP tx pages */ }; struct ofld_tx_sdesc { uint32_t plen; /* payload length */ uint8_t tx_credits; /* firmware tx credits (unit is 16B) */ void *iv_buffer; /* optional buffer holding IVs for TLS */ }; struct ppod_region { u_int pr_start; u_int pr_len; u_int pr_page_shift[4]; uint32_t pr_tag_mask; /* hardware tagmask for this region. */ uint32_t pr_invalid_bit; /* OR with this to invalidate tag. */ uint32_t pr_alias_mask; /* AND with tag to get alias bits. */ u_int pr_alias_shift; /* shift this much for first alias bit. */ vmem_t *pr_arena; }; struct ppod_reservation { struct ppod_region *prsv_pr; uint32_t prsv_tag; /* Full tag: pgsz, alias, tag, color */ u_int prsv_nppods; }; struct pageset { TAILQ_ENTRY(pageset) link; vm_page_t *pages; int npages; int flags; int offset; /* offset in first page */ int len; struct ppod_reservation prsv; struct vmspace *vm; vm_offset_t start; u_int vm_timestamp; }; TAILQ_HEAD(pagesetq, pageset); #define PS_PPODS_WRITTEN 0x0001 /* Page pods written to the card. */ struct ddp_buffer { struct pageset *ps; struct kaiocb *job; int cancel_pending; }; struct ddp_pcb { u_int flags; struct ddp_buffer db[2]; TAILQ_HEAD(, pageset) cached_pagesets; TAILQ_HEAD(, kaiocb) aiojobq; u_int waiting_count; u_int active_count; u_int cached_count; int active_id; /* the currently active DDP buffer */ struct task requeue_task; struct kaiocb *queueing; struct mtx lock; }; struct toepcb { TAILQ_ENTRY(toepcb) link; /* toep_list */ u_int flags; /* miscellaneous flags */ int refcount; struct tom_data *td; struct inpcb *inp; /* backpointer to host stack's PCB */ struct vnet *vnet; struct vi_info *vi; /* virtual interface */ struct sge_wrq *ofld_txq; struct sge_ofld_rxq *ofld_rxq; struct sge_wrq *ctrlq; struct l2t_entry *l2te; /* L2 table entry used by this connection */ struct clip_entry *ce; /* CLIP table entry used by this tid */ int tid; /* Connection identifier */ /* tx credit handling */ u_int tx_total; /* total tx WR credits (in 16B units) */ u_int tx_credits; /* tx WR credits (in 16B units) available */ u_int tx_nocompl; /* tx WR credits since last compl request */ u_int plen_nocompl; /* payload since last compl request */ struct conn_params params; void *ulpcb; void *ulpcb2; struct mbufq ulp_pduq; /* PDUs waiting to be sent out. */ struct mbufq ulp_pdu_reclaimq; struct ddp_pcb ddp; struct tls_ofld_info tls; TAILQ_HEAD(, kaiocb) aiotx_jobq; struct task aiotx_task; struct socket *aiotx_so; /* Tx software descriptor */ uint8_t txsd_total; uint8_t txsd_pidx; uint8_t txsd_cidx; uint8_t txsd_avail; struct ofld_tx_sdesc txsd[]; }; static inline int ulp_mode(struct toepcb *toep) { return (toep->params.ulp_mode); } #define DDP_LOCK(toep) mtx_lock(&(toep)->ddp.lock) #define DDP_UNLOCK(toep) mtx_unlock(&(toep)->ddp.lock) #define DDP_ASSERT_LOCKED(toep) mtx_assert(&(toep)->ddp.lock, MA_OWNED) /* * Compressed state for embryonic connections for a listener. */ struct synq_entry { struct listen_ctx *lctx; /* backpointer to listen ctx */ struct mbuf *syn; int flags; /* same as toepcb's tp_flags */ volatile int ok_to_respond; volatile u_int refcnt; int tid; uint32_t iss; uint32_t irs; uint32_t ts; __be16 tcp_opt; /* from cpl_pass_establish */ struct toepcb *toep; struct conn_params params; }; /* listen_ctx flags */ #define LCTX_RPL_PENDING 1 /* waiting for a CPL_PASS_OPEN_RPL */ struct listen_ctx { LIST_ENTRY(listen_ctx) link; /* listen hash linkage */ volatile int refcount; int stid; struct stid_region stid_region; int flags; struct inpcb *inp; /* listening socket's inp */ struct vnet *vnet; struct sge_wrq *ctrlq; struct sge_ofld_rxq *ofld_rxq; struct clip_entry *ce; }; /* tcb_histent flags */ #define TE_RPL_PENDING 1 #define TE_ACTIVE 2 /* bits in one 8b tcb_histent sample. */ #define TS_RTO (1 << 0) #define TS_DUPACKS (1 << 1) #define TS_FASTREXMT (1 << 2) #define TS_SND_BACKLOGGED (1 << 3) #define TS_CWND_LIMITED (1 << 4) #define TS_ECN_ECE (1 << 5) #define TS_ECN_CWR (1 << 6) #define TS_RESERVED (1 << 7) /* Unused. */ struct tcb_histent { struct mtx te_lock; struct callout te_callout; uint64_t te_tcb[TCB_SIZE / sizeof(uint64_t)]; struct adapter *te_adapter; u_int te_flags; u_int te_tid; uint8_t te_pidx; uint8_t te_sample[100]; }; struct tom_data { struct toedev tod; /* toepcb's associated with this TOE device */ struct mtx toep_list_lock; TAILQ_HEAD(, toepcb) toep_list; struct mtx lctx_hash_lock; LIST_HEAD(, listen_ctx) *listen_hash; u_long listen_mask; int lctx_count; /* # of lctx in the hash table */ struct ppod_region pr; struct rwlock tcb_history_lock __aligned(CACHE_LINE_SIZE); struct tcb_histent **tcb_history; int dupack_threshold; /* WRs that will not be sent to the chip because L2 resolution failed */ struct mtx unsent_wr_lock; STAILQ_HEAD(, wrqe) unsent_wr_list; struct task reclaim_wr_resources; }; static inline struct tom_data * tod_td(struct toedev *tod) { return (__containerof(tod, struct tom_data, tod)); } static inline struct adapter * td_adapter(struct tom_data *td) { return (td->tod.tod_softc); } static inline void set_mbuf_ulp_submode(struct mbuf *m, uint8_t ulp_submode) { M_ASSERTPKTHDR(m); m->m_pkthdr.PH_per.eight[0] = ulp_submode; } static inline uint8_t mbuf_ulp_submode(struct mbuf *m) { M_ASSERTPKTHDR(m); return (m->m_pkthdr.PH_per.eight[0]); } /* t4_tom.c */ struct toepcb *alloc_toepcb(struct vi_info *, int); int init_toepcb(struct vi_info *, struct toepcb *); struct toepcb *hold_toepcb(struct toepcb *); void free_toepcb(struct toepcb *); void offload_socket(struct socket *, struct toepcb *); void undo_offload_socket(struct socket *); void final_cpl_received(struct toepcb *); void insert_tid(struct adapter *, int, void *, int); void *lookup_tid(struct adapter *, int); void update_tid(struct adapter *, int, void *); void remove_tid(struct adapter *, int, int); u_long select_rcv_wnd(struct socket *); int select_rcv_wscale(void); void init_conn_params(struct vi_info *, struct offload_settings *, struct in_conninfo *, struct socket *, const struct tcp_options *, int16_t, struct conn_params *cp); __be64 calc_options0(struct vi_info *, struct conn_params *); __be32 calc_options2(struct vi_info *, struct conn_params *); uint64_t select_ntuple(struct vi_info *, struct l2t_entry *); int negative_advice(int); int add_tid_to_history(struct adapter *, u_int); /* t4_connect.c */ void t4_init_connect_cpl_handlers(void); void t4_uninit_connect_cpl_handlers(void); int t4_connect(struct toedev *, struct socket *, struct rtentry *, struct sockaddr *); void act_open_failure_cleanup(struct adapter *, u_int, u_int); /* t4_listen.c */ void t4_init_listen_cpl_handlers(void); void t4_uninit_listen_cpl_handlers(void); int t4_listen_start(struct toedev *, struct tcpcb *); int t4_listen_stop(struct toedev *, struct tcpcb *); void t4_syncache_added(struct toedev *, void *); void t4_syncache_removed(struct toedev *, void *); int t4_syncache_respond(struct toedev *, void *, struct mbuf *); int do_abort_req_synqe(struct sge_iq *, const struct rss_header *, struct mbuf *); int do_abort_rpl_synqe(struct sge_iq *, const struct rss_header *, struct mbuf *); void t4_offload_socket(struct toedev *, void *, struct socket *); void synack_failure_cleanup(struct adapter *, int); /* t4_cpl_io.c */ void aiotx_init_toep(struct toepcb *); int t4_aio_queue_aiotx(struct socket *, struct kaiocb *); void t4_init_cpl_io_handlers(void); void t4_uninit_cpl_io_handlers(void); void send_abort_rpl(struct adapter *, struct sge_wrq *, int , int); void send_flowc_wr(struct toepcb *, struct tcpcb *); void send_reset(struct adapter *, struct toepcb *, uint32_t); int send_rx_credits(struct adapter *, struct toepcb *, int); void send_rx_modulate(struct adapter *, struct toepcb *); void make_established(struct toepcb *, uint32_t, uint32_t, uint16_t); int t4_close_conn(struct adapter *, struct toepcb *); void t4_rcvd(struct toedev *, struct tcpcb *); void t4_rcvd_locked(struct toedev *, struct tcpcb *); int t4_tod_output(struct toedev *, struct tcpcb *); int t4_send_fin(struct toedev *, struct tcpcb *); int t4_send_rst(struct toedev *, struct tcpcb *); void t4_set_tcb_field(struct adapter *, struct sge_wrq *, struct toepcb *, uint16_t, uint64_t, uint64_t, int, int); void t4_push_frames(struct adapter *, struct toepcb *, int); void t4_push_pdus(struct adapter *, struct toepcb *, int); /* t4_ddp.c */ int t4_init_ppod_region(struct ppod_region *, struct t4_range *, u_int, const char *); void t4_free_ppod_region(struct ppod_region *); int t4_alloc_page_pods_for_ps(struct ppod_region *, struct pageset *); int t4_alloc_page_pods_for_buf(struct ppod_region *, vm_offset_t, int, struct ppod_reservation *); int t4_write_page_pods_for_ps(struct adapter *, struct sge_wrq *, int, struct pageset *); int t4_write_page_pods_for_buf(struct adapter *, struct sge_wrq *, int tid, struct ppod_reservation *, vm_offset_t, int); void t4_free_page_pods(struct ppod_reservation *); int t4_soreceive_ddp(struct socket *, struct sockaddr **, struct uio *, struct mbuf **, struct mbuf **, int *); int t4_aio_queue_ddp(struct socket *, struct kaiocb *); void t4_ddp_mod_load(void); void t4_ddp_mod_unload(void); void ddp_assert_empty(struct toepcb *); void ddp_init_toep(struct toepcb *); void ddp_uninit_toep(struct toepcb *); void ddp_queue_toep(struct toepcb *); void release_ddp_resources(struct toepcb *toep); void handle_ddp_close(struct toepcb *, struct tcpcb *, uint32_t); void handle_ddp_indicate(struct toepcb *); void insert_ddp_data(struct toepcb *, uint32_t); const struct offload_settings *lookup_offload_policy(struct adapter *, int, struct mbuf *, uint16_t, struct inpcb *); /* t4_tls.c */ bool can_tls_offload(struct adapter *); int t4_ctloutput_tls(struct socket *, struct sockopt *); void t4_push_tls_records(struct adapter *, struct toepcb *, int); void t4_push_ktls(struct adapter *, struct toepcb *, int); void t4_tls_mod_load(void); void t4_tls_mod_unload(void); void tls_establish(struct toepcb *); void tls_init_toep(struct toepcb *); int tls_rx_key(struct toepcb *); void tls_stop_handshake_timer(struct toepcb *); int tls_tx_key(struct toepcb *); void tls_uninit_toep(struct toepcb *); int tls_alloc_ktls(struct toepcb *, struct ktls_session *); #endif diff --git a/sys/dev/iicbus/ad7417.c b/sys/dev/iicbus/ad7417.c index 96178e52a368..43170ce2e81f 100644 --- a/sys/dev/iicbus/ad7417.c +++ b/sys/dev/iicbus/ad7417.c @@ -1,629 +1,642 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2010 Andreas Tobler - * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* CPU A/B sensors, temp and adc: AD7417. */ #define AD7417_TEMP 0x00 #define AD7417_CONFIG 0x01 #define AD7417_ADC 0x04 #define AD7417_CONFIG2 0x05 #define AD7417_CONFMASK 0xe0 uint8_t adc741x_config; struct ad7417_sensor { struct pmac_therm therm; device_t dev; int id; enum { ADC7417_TEMP_SENSOR, ADC7417_ADC_SENSOR } type; }; struct write_data { uint8_t reg; uint8_t val; }; struct read_data { uint8_t reg; uint16_t val; }; /* Regular bus attachment functions */ static int ad7417_probe(device_t); static int ad7417_attach(device_t); /* Utility functions */ static int ad7417_sensor_sysctl(SYSCTL_HANDLER_ARGS); static int ad7417_write(device_t dev, uint32_t addr, uint8_t reg, uint8_t *buf, int len); static int ad7417_read_1(device_t dev, uint32_t addr, uint8_t reg, uint8_t *data); static int ad7417_read_2(device_t dev, uint32_t addr, uint8_t reg, uint16_t *data); static int ad7417_write_read(device_t dev, uint32_t addr, struct write_data out, struct read_data *in); static int ad7417_diode_read(struct ad7417_sensor *sens); static int ad7417_adc_read(struct ad7417_sensor *sens); static int ad7417_sensor_read(struct ad7417_sensor *sens); struct ad7417_softc { device_t sc_dev; uint32_t sc_addr; struct ad7417_sensor *sc_sensors; int sc_nsensors; int init_done; }; static device_method_t ad7417_methods[] = { /* Device interface */ DEVMETHOD(device_probe, ad7417_probe), DEVMETHOD(device_attach, ad7417_attach), { 0, 0 }, }; static driver_t ad7417_driver = { "ad7417", ad7417_methods, sizeof(struct ad7417_softc) }; static devclass_t ad7417_devclass; DRIVER_MODULE(ad7417, iicbus, ad7417_driver, ad7417_devclass, 0, 0); static MALLOC_DEFINE(M_AD7417, "ad7417", "Supply-Monitor AD7417"); static int ad7417_write(device_t dev, uint32_t addr, uint8_t reg, uint8_t *buff, int len) { unsigned char buf[4]; int try = 0; struct iic_msg msg[] = { { addr, IIC_M_WR, 0, buf } }; msg[0].len = len + 1; buf[0] = reg; memcpy(buf + 1, buff, len); for (;;) { - if (iicbus_transfer(dev, msg, 1) == 0) + if (iicbus_transfer(dev, msg, nitems(msg)) == 0) return (0); if (++try > 5) { device_printf(dev, "iicbus write failed\n"); return (-1); } pause("ad7417_write", hz); } } static int ad7417_read_1(device_t dev, uint32_t addr, uint8_t reg, uint8_t *data) { uint8_t buf[4]; int err, try = 0; struct iic_msg msg[2] = { { addr, IIC_M_WR | IIC_M_NOSTOP, 1, ® }, { addr, IIC_M_RD, 1, buf }, }; for (;;) { - err = iicbus_transfer(dev, msg, 2); + err = iicbus_transfer(dev, msg, nitems(msg)); if (err != 0) goto retry; *data = *((uint8_t*)buf); return (0); retry: if (++try > 5) { device_printf(dev, "iicbus read failed\n"); return (-1); } pause("ad7417_read_1", hz); } } static int ad7417_read_2(device_t dev, uint32_t addr, uint8_t reg, uint16_t *data) { uint8_t buf[4]; int err, try = 0; struct iic_msg msg[2] = { { addr, IIC_M_WR | IIC_M_NOSTOP, 1, ® }, { addr, IIC_M_RD, 2, buf }, }; for (;;) { - err = iicbus_transfer(dev, msg, 2); + err = iicbus_transfer(dev, msg, nitems(msg)); if (err != 0) goto retry; *data = *((uint16_t*)buf); return (0); retry: if (++try > 5) { device_printf(dev, "iicbus read failed\n"); return (-1); } pause("ad7417_read_2", hz); } } static int ad7417_write_read(device_t dev, uint32_t addr, struct write_data out, struct read_data *in) { uint8_t buf[4]; int err, try = 0; /* Do a combined write/read. */ struct iic_msg msg[3] = { { addr, IIC_M_WR, 2, buf }, { addr, IIC_M_WR | IIC_M_NOSTOP, 1, &in->reg }, { addr, IIC_M_RD, 2, buf }, }; /* Prepare the write msg. */ buf[0] = out.reg; buf[1] = out.val & 0xff; for (;;) { - err = iicbus_transfer(dev, msg, 3); + err = iicbus_transfer(dev, msg, nitems(msg)); if (err != 0) goto retry; in->val = *((uint16_t*)buf); return (0); retry: if (++try > 5) { device_printf(dev, "iicbus write/read failed\n"); return (-1); } pause("ad7417_write_read", hz); } } static int ad7417_init_adc(device_t dev, uint32_t addr) { uint8_t buf; int err; struct ad7417_softc *sc; sc = device_get_softc(dev); adc741x_config = 0; /* Clear Config2 */ buf = 0; - err = ad7417_write(dev, addr, AD7417_CONFIG2, &buf, 1); + err = ad7417_write(dev, addr, AD7417_CONFIG2, &buf, sizeof(buf)); /* Read & cache Config1 */ buf = 0; - err = ad7417_write(dev, addr, AD7417_CONFIG, &buf, 1); + err = ad7417_write(dev, addr, AD7417_CONFIG, &buf, sizeof(buf)); err = ad7417_read_1(dev, addr, AD7417_CONFIG, &buf); adc741x_config = (uint8_t)buf; /* Disable shutdown mode */ adc741x_config &= 0xfe; buf = adc741x_config; - err = ad7417_write(dev, addr, AD7417_CONFIG, &buf, 1); + err = ad7417_write(dev, addr, AD7417_CONFIG, &buf, sizeof(buf)); if (err < 0) return (-1); sc->init_done = 1; return (0); } static int ad7417_probe(device_t dev) { const char *name, *compatible; struct ad7417_softc *sc; name = ofw_bus_get_name(dev); compatible = ofw_bus_get_compat(dev); if (!name) return (ENXIO); if (strcmp(name, "supply-monitor") != 0 || strcmp(compatible, "ad7417") != 0) return (ENXIO); sc = device_get_softc(dev); sc->sc_dev = dev; sc->sc_addr = iicbus_get_addr(dev); device_set_desc(dev, "Supply-Monitor AD7417"); return (0); } /* * This function returns the number of sensors. If we call it the second time * and we have allocated memory for sc->sc_sensors, we fill in the properties. */ static int ad7417_fill_sensor_prop(device_t dev) { - phandle_t child; + phandle_t child, node; struct ad7417_softc *sc; u_int id[10]; char location[96]; char type[32]; int i = 0, j, len = 0, prop_len, prev_len = 0; sc = device_get_softc(dev); child = ofw_bus_get_node(dev); /* Fill the sensor location property. */ prop_len = OF_getprop(child, "hwsensor-location", location, sizeof(location)); while (len < prop_len) { if (sc->sc_sensors != NULL) strcpy(sc->sc_sensors[i].therm.name, location + len); prev_len = strlen(location + len) + 1; len += prev_len; i++; } if (sc->sc_sensors == NULL) return (i); /* Fill the sensor type property. */ len = 0; i = 0; prev_len = 0; prop_len = OF_getprop(child, "hwsensor-type", type, sizeof(type)); while (len < prop_len) { if (strcmp(type + len, "temperature") == 0) sc->sc_sensors[i].type = ADC7417_TEMP_SENSOR; else sc->sc_sensors[i].type = ADC7417_ADC_SENSOR; prev_len = strlen(type + len) + 1; len += prev_len; i++; } /* Fill the sensor id property. Taken from OF. */ prop_len = OF_getprop(child, "hwsensor-id", id, sizeof(id)); for (j = 0; j < i; j++) sc->sc_sensors[j].id = id[j]; /* Fill the sensor zone property. Taken from OF. */ prop_len = OF_getprop(child, "hwsensor-zone", id, sizeof(id)); for (j = 0; j < i; j++) sc->sc_sensors[j].therm.zone = id[j]; + /* Some PowerMac's have the real location of the sensors on + child nodes of the hwsensor-location node. Check for and + fix the name if needed. + This is needed to apply the below HACK with the diode. + */ + j = 0; + for (node = OF_child(child); node != 0; node = OF_peer(node)) { + + OF_getprop(node, "location", location, sizeof(location)); + strcpy(sc->sc_sensors[i].therm.name, location); + j++; + } + /* Finish setting up sensor properties */ for (j = 0; j < i; j++) { sc->sc_sensors[j].dev = dev; /* HACK: Apple wired a random diode to the ADC line */ - if (strstr(sc->sc_sensors[j].therm.name, "DIODE TEMP") - != NULL) { + if ((strstr(sc->sc_sensors[j].therm.name, "DIODE TEMP") + != NULL) + || (strstr(sc->sc_sensors[j].therm.name, "AD1") != NULL)) { sc->sc_sensors[j].type = ADC7417_TEMP_SENSOR; sc->sc_sensors[j].therm.read = (int (*)(struct pmac_therm *))(ad7417_diode_read); } else { sc->sc_sensors[j].therm.read = (int (*)(struct pmac_therm *))(ad7417_sensor_read); } if (sc->sc_sensors[j].type != ADC7417_TEMP_SENSOR) continue; /* Make up some ranges */ sc->sc_sensors[j].therm.target_temp = 500 + ZERO_C_TO_K; sc->sc_sensors[j].therm.max_temp = 900 + ZERO_C_TO_K; pmac_thermal_sensor_register(&sc->sc_sensors[j].therm); } return (i); } static int ad7417_attach(device_t dev) { struct ad7417_softc *sc; struct sysctl_oid *oid, *sensroot_oid; struct sysctl_ctx_list *ctx; char sysctl_name[32]; int i, j; const char *unit; const char *desc; sc = device_get_softc(dev); sc->sc_nsensors = 0; /* Count the actual number of sensors. */ sc->sc_nsensors = ad7417_fill_sensor_prop(dev); device_printf(dev, "%d sensors detected.\n", sc->sc_nsensors); if (sc->sc_nsensors == 0) device_printf(dev, "WARNING: No AD7417 sensors detected!\n"); sc->sc_sensors = malloc (sc->sc_nsensors * sizeof(struct ad7417_sensor), M_AD7417, M_WAITOK | M_ZERO); ctx = device_get_sysctl_ctx(dev); sensroot_oid = SYSCTL_ADD_NODE(ctx, SYSCTL_CHILDREN(device_get_sysctl_tree(dev)), OID_AUTO, "sensor", CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "AD7417 Sensor Information"); /* Now we can fill the properties into the allocated struct. */ sc->sc_nsensors = ad7417_fill_sensor_prop(dev); /* Add sysctls for the sensors. */ for (i = 0; i < sc->sc_nsensors; i++) { for (j = 0; j < strlen(sc->sc_sensors[i].therm.name); j++) { sysctl_name[j] = tolower(sc->sc_sensors[i].therm.name[j]); if (isspace(sysctl_name[j])) sysctl_name[j] = '_'; } sysctl_name[j] = 0; oid = SYSCTL_ADD_NODE(ctx, SYSCTL_CHILDREN(sensroot_oid), OID_AUTO, sysctl_name, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "Sensor Information"); if (sc->sc_sensors[i].type == ADC7417_TEMP_SENSOR) { unit = "temp"; desc = "sensor unit (C)"; } else { unit = "volt"; desc = "sensor unit (mV)"; } /* I use i to pass the sensor id. */ SYSCTL_ADD_PROC(ctx, SYSCTL_CHILDREN(oid), OID_AUTO, - unit, CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_NEEDGIANT, dev, - i, ad7417_sensor_sysctl, - sc->sc_sensors[i].type == ADC7417_TEMP_SENSOR ? "IK" : "I", - desc); + unit, CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, + dev, i, ad7417_sensor_sysctl, + sc->sc_sensors[i].type == ADC7417_TEMP_SENSOR ? + "IK" : "I", desc); } /* Dump sensor location, ID & type. */ if (bootverbose) { device_printf(dev, "Sensors\n"); for (i = 0; i < sc->sc_nsensors; i++) { device_printf(dev, "Location: %s ID: %d type: %d\n", sc->sc_sensors[i].therm.name, sc->sc_sensors[i].id, sc->sc_sensors[i].type); } } return (0); } static int ad7417_get_temp(device_t dev, uint32_t addr, int *temp) { uint16_t buf[2]; uint16_t read; int err; err = ad7417_read_2(dev, addr, AD7417_TEMP, buf); if (err < 0) return (-1); read = *((int16_t*)buf); /* The ADC is 10 bit, the resolution is 0.25 C. The temperature is in tenth kelvin. */ *temp = (((int16_t)(read & 0xffc0)) >> 6) * 25 / 10; return (0); } static int ad7417_get_adc(device_t dev, uint32_t addr, unsigned int *value, uint8_t chan) { uint8_t tmp; int err; struct write_data config; struct read_data data; tmp = chan << 5; config.reg = AD7417_CONFIG; data.reg = AD7417_ADC; data.val = 0; err = ad7417_read_1(dev, addr, AD7417_CONFIG, &config.val); config.val = (config.val & ~AD7417_CONFMASK) | (tmp & AD7417_CONFMASK); err = ad7417_write_read(dev, addr, config, &data); if (err < 0) return (-1); *value = ((uint32_t)data.val) >> 6; return (0); } static int ad7417_diode_read(struct ad7417_sensor *sens) { static int eeprom_read = 0; static cell_t eeprom[2][40]; phandle_t eeprom_node; int rawval, diode_slope, diode_offset; int temp; if (!eeprom_read) { eeprom_node = OF_finddevice("/u3/i2c/cpuid@a0"); OF_getprop(eeprom_node, "cpuid", eeprom[0], sizeof(eeprom[0])); eeprom_node = OF_finddevice("/u3/i2c/cpuid@a2"); OF_getprop(eeprom_node, "cpuid", eeprom[1], sizeof(eeprom[1])); eeprom_read = 1; } rawval = ad7417_adc_read(sens); if (rawval < 0) return (-1); if (strstr(sens->therm.name, "CPU B") != NULL) { diode_slope = eeprom[1][0x11] >> 16; diode_offset = (int16_t)(eeprom[1][0x11] & 0xffff) << 12; } else { diode_slope = eeprom[0][0x11] >> 16; diode_offset = (int16_t)(eeprom[0][0x11] & 0xffff) << 12; } temp = (rawval*diode_slope + diode_offset) >> 2; temp = (10*(temp >> 16)) + ((10*(temp & 0xffff)) >> 16); return (temp + ZERO_C_TO_K); } static int ad7417_adc_read(struct ad7417_sensor *sens) { struct ad7417_softc *sc; uint8_t chan; int temp; sc = device_get_softc(sens->dev); switch (sens->id) { case 11: case 16: chan = 1; break; case 12: case 17: chan = 2; break; case 13: case 18: chan = 3; break; case 14: case 19: chan = 4; break; default: chan = 1; } if (ad7417_get_adc(sc->sc_dev, sc->sc_addr, &temp, chan) < 0) return (-1); return (temp); } static int ad7417_sensor_read(struct ad7417_sensor *sens) { struct ad7417_softc *sc; int temp; sc = device_get_softc(sens->dev); /* Init the ADC if not already done.*/ if (!sc->init_done) if (ad7417_init_adc(sc->sc_dev, sc->sc_addr) < 0) return (-1); if (sens->type == ADC7417_TEMP_SENSOR) { if (ad7417_get_temp(sc->sc_dev, sc->sc_addr, &temp) < 0) return (-1); temp += ZERO_C_TO_K; } else { temp = ad7417_adc_read(sens); } return (temp); } static int ad7417_sensor_sysctl(SYSCTL_HANDLER_ARGS) { device_t dev; struct ad7417_softc *sc; struct ad7417_sensor *sens; int value = 0; int error; dev = arg1; sc = device_get_softc(dev); sens = &sc->sc_sensors[arg2]; value = sens->therm.read(&sens->therm); if (value < 0) return (ENXIO); error = sysctl_handle_int(oidp, &value, 0, req); return (error); } diff --git a/sys/dev/iicbus/ds1631.c b/sys/dev/iicbus/ds1631.c index bce4e36a240e..f552005a739b 100644 --- a/sys/dev/iicbus/ds1631.c +++ b/sys/dev/iicbus/ds1631.c @@ -1,415 +1,414 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2012 Andreas Tobler - * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* Sensor: Maxim DS1631 */ #define DS1631_STOP 0x22 #define DS1631_START 0x51 #define DS1631_RESET 0x54 #define DS1631_TEMP 0xAA #define DS1631_CONTROL 0xAC #define DS1631_CONTROL_1SHOT 0x01 #define DS1631_CONTROL_9BIT 0x00 #define DS1631_CONTROL_10BIT 0x04 #define DS1631_CONTROL_11BIT 0x08 #define DS1631_CONTROL_12BIT 0x0C /* Regular bus attachment functions */ static int ds1631_probe(device_t); static int ds1631_attach(device_t); struct ds1631_softc { struct pmac_therm sc_sensor; device_t sc_dev; struct intr_config_hook enum_hook; uint32_t sc_addr; uint32_t init_done; }; struct write_data { uint8_t reg; uint8_t val; }; struct read_data { uint8_t reg; uint16_t val; }; /* Utility functions */ static int ds1631_sensor_read(struct ds1631_softc *sc); static int ds1631_sensor_sysctl(SYSCTL_HANDLER_ARGS); static void ds1631_start(void *xdev); static int ds1631_read_1(device_t dev, uint32_t addr, uint8_t reg, uint8_t *data); static int ds1631_read_2(device_t dev, uint32_t addr, uint8_t reg, uint16_t *data); static int ds1631_write(device_t dev, uint32_t addr, uint8_t reg, uint8_t *buff, int len); static device_method_t ds1631_methods[] = { /* Device interface */ DEVMETHOD(device_probe, ds1631_probe), DEVMETHOD(device_attach, ds1631_attach), { 0, 0 }, }; static driver_t ds1631_driver = { "ds1631", ds1631_methods, sizeof(struct ds1631_softc) }; static devclass_t ds1631_devclass; DRIVER_MODULE(ds1631, iicbus, ds1631_driver, ds1631_devclass, 0, 0); static int ds1631_write(device_t dev, uint32_t addr, uint8_t reg, uint8_t *buff, int len) { uint8_t buf[4]; int try = 0; struct iic_msg msg[] = { { addr, IIC_M_WR, 0, buf } }; /* Prepare the write msg. */ msg[0].len = len + 1; buf[0] = reg; memcpy(buf + 1, buff, len); for (;;) { - if (iicbus_transfer(dev, msg, 1) == 0) + if (iicbus_transfer(dev, msg, nitems(msg)) == 0) return (0); if (++try > 5) { device_printf(dev, "iicbus write failed\n"); return (-1); } pause("ds1631_write", hz); } } static int ds1631_read_1(device_t dev, uint32_t addr, uint8_t reg, uint8_t *data) { uint8_t buf[4]; int err, try = 0; struct iic_msg msg[2] = { { addr, IIC_M_WR, 1, ® }, { addr, IIC_M_RD, 1, buf }, }; for (;;) { - err = iicbus_transfer(dev, msg, 2); + err = iicbus_transfer(dev, msg, nitems(msg)); if (err != 0) goto retry; *data = *((uint8_t*)buf); return (0); retry: if (++try > 5) { device_printf(dev, "iicbus read failed\n"); return (-1); } pause("ds1631_read_1", hz); } } static int ds1631_read_2(device_t dev, uint32_t addr, uint8_t reg, uint16_t *data) { uint8_t buf[4]; int err, try = 0; struct iic_msg msg[2] = { { addr, IIC_M_WR, 1, ® }, { addr, IIC_M_RD, 2, buf }, }; for (;;) { - err = iicbus_transfer(dev, msg, 2); + err = iicbus_transfer(dev, msg, nitems(msg)); if (err != 0) goto retry; *data = *((uint16_t*)buf); return (0); retry: if (++try > 5) { device_printf(dev, "iicbus read failed\n"); return (-1); } pause("ds1631_read_2", hz); } } static int ds1631_probe(device_t dev) { const char *name, *compatible; struct ds1631_softc *sc; name = ofw_bus_get_name(dev); compatible = ofw_bus_get_compat(dev); if (!name) return (ENXIO); if (strcmp(name, "temp-monitor") != 0 || strcmp(compatible, "ds1631") != 0 ) return (ENXIO); sc = device_get_softc(dev); sc->sc_dev = dev; sc->sc_addr = iicbus_get_addr(dev); device_set_desc(dev, "Temp-Monitor DS1631"); return (0); } static int ds1631_attach(device_t dev) { struct ds1631_softc *sc; sc = device_get_softc(dev); sc->enum_hook.ich_func = ds1631_start; sc->enum_hook.ich_arg = dev; /* * We have to wait until interrupts are enabled. I2C read and write * only works if the interrupts are available. * The unin/i2c is controlled by the htpic on unin. But this is not * the master. The openpic on mac-io is controlling the htpic. * This one gets attached after the mac-io probing and then the * interrupts will be available. */ if (config_intrhook_establish(&sc->enum_hook) != 0) return (ENOMEM); return (0); } static int ds1631_init(device_t dev, uint32_t addr) { uint8_t conf; int err; struct ds1631_softc *sc; sc = device_get_softc(dev); err = ds1631_read_1(dev, addr, DS1631_CONTROL, &conf); if (err < 0) { device_printf(dev, "ds1631 read config failed: %x\n", err); return (-1); } /* Stop the conversion if not in 1SHOT mode. */ if (conf & ~DS1631_CONTROL_1SHOT) err = ds1631_write(dev, addr, DS1631_STOP, &conf, 0); /* * Setup the resolution, 10-bit is enough. Each bit increase in * resolution doubles the conversion time. */ conf = DS1631_CONTROL_10BIT; - err = ds1631_write(dev, addr, DS1631_CONTROL, &conf, 1); + err = ds1631_write(dev, addr, DS1631_CONTROL, &conf, sizeof(conf)); if (err < 0) { device_printf(dev, "ds1631 write config failed: %x\n", err); return (-1); } /* And now start....*/ err = ds1631_write(dev, addr, DS1631_START, &conf, 0); if (err < 0) { device_printf(dev, "ds1631 write start failed: %x\n", err); return (-1); } sc->init_done = 1; return (0); } static void ds1631_start(void *xdev) { phandle_t child, node; struct ds1631_softc *sc; struct sysctl_oid *oid, *sensroot_oid; struct sysctl_ctx_list *ctx; ssize_t plen; int i; char sysctl_desc[40], sysctl_name[40]; device_t dev = (device_t)xdev; sc = device_get_softc(dev); child = ofw_bus_get_node(dev); ctx = device_get_sysctl_ctx(dev); sensroot_oid = SYSCTL_ADD_NODE(ctx, SYSCTL_CHILDREN(device_get_sysctl_tree(dev)), OID_AUTO, "sensor", CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "DS1631 Sensor Information"); if (OF_getprop(child, "hwsensor-zone", &sc->sc_sensor.zone, sizeof(int)) < 0) sc->sc_sensor.zone = 0; plen = OF_getprop(child, "hwsensor-location", sc->sc_sensor.name, sizeof(sc->sc_sensor.name)); if (plen == -1) { /* * Ok, no hwsensor-location property, so let's look for a * location property on a sub node. */ for (node = OF_child(child); node; node = OF_peer(node)) plen = OF_getprop(node, "location", sc->sc_sensor.name, sizeof(sc->sc_sensor.name)); } if (plen == -1) { strcpy(sysctl_name, "sensor"); } else { for (i = 0; i < strlen(sc->sc_sensor.name); i++) { sysctl_name[i] = tolower(sc->sc_sensor.name[i]); if (isspace(sysctl_name[i])) sysctl_name[i] = '_'; } sysctl_name[i] = 0; } /* Make up target temperatures. These are low, for the drive bay. */ if (sc->sc_sensor.zone == 0) { sc->sc_sensor.target_temp = 400 + ZERO_C_TO_K; sc->sc_sensor.max_temp = 500 + ZERO_C_TO_K; } else { sc->sc_sensor.target_temp = 300 + ZERO_C_TO_K; sc->sc_sensor.max_temp = 500 + ZERO_C_TO_K; } sc->sc_sensor.read = (int (*)(struct pmac_therm *sc))(ds1631_sensor_read); pmac_thermal_sensor_register(&sc->sc_sensor); sprintf(sysctl_desc,"%s %s", sc->sc_sensor.name, "(C)"); oid = SYSCTL_ADD_NODE(ctx, SYSCTL_CHILDREN(sensroot_oid), OID_AUTO, sysctl_name, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "Sensor Information"); SYSCTL_ADD_PROC(ctx, SYSCTL_CHILDREN(oid), OID_AUTO, "temp", CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, dev, 0, ds1631_sensor_sysctl, "IK", sysctl_desc); config_intrhook_disestablish(&sc->enum_hook); } static int ds1631_sensor_read(struct ds1631_softc *sc) { uint16_t buf[2]; uint16_t read; int err; if (!sc->init_done) ds1631_init(sc->sc_dev, sc->sc_addr); err = ds1631_read_2(sc->sc_dev, sc->sc_addr, DS1631_TEMP, buf); if (err < 0) { device_printf(sc->sc_dev, "ds1631 read TEMP failed: %x\n", err); return (-1); } read = *((int16_t *)buf); /* * The default mode of the ADC is 12-bit, the resolution is 0.0625 C * per bit. The temperature is in tenth kelvin. * We use 10-bit resolution which seems enough, resolution is 0.25 C. */ return (((int16_t)(read) >> 6) * 25 / 10 + ZERO_C_TO_K); } static int ds1631_sensor_sysctl(SYSCTL_HANDLER_ARGS) { device_t dev; struct ds1631_softc *sc; int error; int temp; dev = arg1; sc = device_get_softc(dev); temp = ds1631_sensor_read(sc); if (temp < 0) return (EIO); error = sysctl_handle_int(oidp, &temp, 0, req); return (error); } diff --git a/sys/dev/iicbus/ds1775.c b/sys/dev/iicbus/ds1775.c index 2df1b13abf2e..223ce4410b72 100644 --- a/sys/dev/iicbus/ds1775.c +++ b/sys/dev/iicbus/ds1775.c @@ -1,274 +1,273 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2010 Andreas Tobler - * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* Drivebay sensor: LM75/DS1775. */ #define DS1775_TEMP 0x0 /* Regular bus attachment functions */ static int ds1775_probe(device_t); static int ds1775_attach(device_t); struct ds1775_softc { struct pmac_therm sc_sensor; device_t sc_dev; struct intr_config_hook enum_hook; uint32_t sc_addr; }; /* Utility functions */ static int ds1775_sensor_read(struct ds1775_softc *sc); static int ds1775_sensor_sysctl(SYSCTL_HANDLER_ARGS); static void ds1775_start(void *xdev); static int ds1775_read_2(device_t dev, uint32_t addr, uint8_t reg, uint16_t *data); static device_method_t ds1775_methods[] = { /* Device interface */ DEVMETHOD(device_probe, ds1775_probe), DEVMETHOD(device_attach, ds1775_attach), { 0, 0 }, }; static driver_t ds1775_driver = { "ds1775", ds1775_methods, sizeof(struct ds1775_softc) }; static devclass_t ds1775_devclass; DRIVER_MODULE(ds1775, iicbus, ds1775_driver, ds1775_devclass, 0, 0); static int ds1775_read_2(device_t dev, uint32_t addr, uint8_t reg, uint16_t *data) { uint8_t buf[4]; int err, try = 0; struct iic_msg msg[2] = { { addr, IIC_M_WR | IIC_M_NOSTOP, 1, ® }, { addr, IIC_M_RD, 2, buf }, }; for (;;) { - err = iicbus_transfer(dev, msg, 2); + err = iicbus_transfer(dev, msg, nitems(msg)); if (err != 0) goto retry; *data = *((uint16_t*)buf); return (0); retry: if (++try > 5) { device_printf(dev, "iicbus read failed\n"); return (-1); } pause("ds1775_read_2", hz); } } static int ds1775_probe(device_t dev) { const char *name, *compatible; struct ds1775_softc *sc; name = ofw_bus_get_name(dev); compatible = ofw_bus_get_compat(dev); if (!name) return (ENXIO); if (strcmp(name, "temp-monitor") != 0 || (strcmp(compatible, "ds1775") != 0 && strcmp(compatible, "lm75") != 0)) return (ENXIO); sc = device_get_softc(dev); sc->sc_dev = dev; sc->sc_addr = iicbus_get_addr(dev); device_set_desc(dev, "Temp-Monitor DS1775"); return (0); } static int ds1775_attach(device_t dev) { struct ds1775_softc *sc; sc = device_get_softc(dev); sc->enum_hook.ich_func = ds1775_start; sc->enum_hook.ich_arg = dev; /* We have to wait until interrupts are enabled. I2C read and write * only works if the interrupts are available. * The unin/i2c is controlled by the htpic on unin. But this is not * the master. The openpic on mac-io is controlling the htpic. * This one gets attached after the mac-io probing and then the * interrupts will be available. */ if (config_intrhook_establish(&sc->enum_hook) != 0) return (ENOMEM); return (0); } static void ds1775_start(void *xdev) { phandle_t child; struct ds1775_softc *sc; struct sysctl_oid *oid, *sensroot_oid; struct sysctl_ctx_list *ctx; ssize_t plen; int i; char sysctl_name[40], sysctl_desc[40]; device_t dev = (device_t)xdev; sc = device_get_softc(dev); child = ofw_bus_get_node(dev); ctx = device_get_sysctl_ctx(dev); sensroot_oid = SYSCTL_ADD_NODE(ctx, SYSCTL_CHILDREN(device_get_sysctl_tree(dev)), OID_AUTO, "sensor", CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "DS1775 Sensor Information"); if (OF_getprop(child, "hwsensor-zone", &sc->sc_sensor.zone, sizeof(int)) < 0) sc->sc_sensor.zone = 0; plen = OF_getprop(child, "hwsensor-location", sc->sc_sensor.name, sizeof(sc->sc_sensor.name)); if (plen == -1) { strcpy(sysctl_name, "sensor"); } else { for (i = 0; i < strlen(sc->sc_sensor.name); i++) { sysctl_name[i] = tolower(sc->sc_sensor.name[i]); if (isspace(sysctl_name[i])) sysctl_name[i] = '_'; } sysctl_name[i] = 0; } /* Make up target temperatures. These are low, for the drive bay. */ if (sc->sc_sensor.zone == 0) { sc->sc_sensor.target_temp = 500 + ZERO_C_TO_K; sc->sc_sensor.max_temp = 600 + ZERO_C_TO_K; } else { sc->sc_sensor.target_temp = 300 + ZERO_C_TO_K; sc->sc_sensor.max_temp = 600 + ZERO_C_TO_K; } sc->sc_sensor.read = (int (*)(struct pmac_therm *sc))(ds1775_sensor_read); pmac_thermal_sensor_register(&sc->sc_sensor); sprintf(sysctl_desc,"%s %s", sc->sc_sensor.name, "(C)"); oid = SYSCTL_ADD_NODE(ctx, SYSCTL_CHILDREN(sensroot_oid), OID_AUTO, sysctl_name, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "Sensor Information"); SYSCTL_ADD_PROC(ctx, SYSCTL_CHILDREN(oid), OID_AUTO, "temp", CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, dev, 0, ds1775_sensor_sysctl, "IK", sysctl_desc); config_intrhook_disestablish(&sc->enum_hook); } static int ds1775_sensor_read(struct ds1775_softc *sc) { uint16_t buf[2]; uint16_t read; int err; err = ds1775_read_2(sc->sc_dev, sc->sc_addr, DS1775_TEMP, buf); if (err < 0) return (-1); read = *((int16_t *)buf); /* The default mode of the ADC is 9 bit, the resolution is 0.5 C per bit. The temperature is in tenth kelvin. */ return (((int16_t)(read) >> 7) * 5 + ZERO_C_TO_K); } static int ds1775_sensor_sysctl(SYSCTL_HANDLER_ARGS) { device_t dev; struct ds1775_softc *sc; int error; int temp; dev = arg1; sc = device_get_softc(dev); temp = ds1775_sensor_read(sc); if (temp < 0) return (EIO); error = sysctl_handle_int(oidp, &temp, 0, req); return (error); } diff --git a/sys/dev/iicbus/max6690.c b/sys/dev/iicbus/max6690.c index b0c1a9f8ddcb..7847c7e237e6 100644 --- a/sys/dev/iicbus/max6690.c +++ b/sys/dev/iicbus/max6690.c @@ -1,383 +1,383 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2010 Andreas Tobler - * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* Inlet, Backside, U3 Heatsink sensor: MAX6690. */ #define MAX6690_INT_TEMP 0x0 #define MAX6690_EXT_TEMP 0x1 #define MAX6690_RSL_STATUS 0x2 #define MAX6690_EEXT_TEMP 0x10 #define MAX6690_IEXT_TEMP 0x11 #define MAX6690_TEMP_MASK 0xe0 struct max6690_sensor { struct pmac_therm therm; device_t dev; int id; }; /* Regular bus attachment functions */ static int max6690_probe(device_t); static int max6690_attach(device_t); /* Utility functions */ static int max6690_sensor_read(struct max6690_sensor *sens); static int max6690_sensor_sysctl(SYSCTL_HANDLER_ARGS); static void max6690_start(void *xdev); static int max6690_read(device_t dev, uint32_t addr, uint8_t reg, uint8_t *data); struct max6690_softc { device_t sc_dev; struct intr_config_hook enum_hook; uint32_t sc_addr; struct max6690_sensor *sc_sensors; int sc_nsensors; }; static device_method_t max6690_methods[] = { /* Device interface */ DEVMETHOD(device_probe, max6690_probe), DEVMETHOD(device_attach, max6690_attach), { 0, 0 }, }; static driver_t max6690_driver = { "max6690", max6690_methods, sizeof(struct max6690_softc) }; static devclass_t max6690_devclass; DRIVER_MODULE(max6690, iicbus, max6690_driver, max6690_devclass, 0, 0); static MALLOC_DEFINE(M_MAX6690, "max6690", "Temp-Monitor MAX6690"); static int max6690_read(device_t dev, uint32_t addr, uint8_t reg, uint8_t *data) { uint8_t buf[4]; uint8_t busy[1], rsl; int err, try = 0; /* Busy register RSL. */ rsl = MAX6690_RSL_STATUS; /* first read the status register, 0x2. If busy, retry. */ struct iic_msg msg[4] = { { addr, IIC_M_WR | IIC_M_NOSTOP, 1, &rsl }, { addr, IIC_M_RD, 1, busy }, { addr, IIC_M_WR | IIC_M_NOSTOP, 1, ® }, { addr, IIC_M_RD, 1, buf }, }; for (;;) { - err = iicbus_transfer(dev, msg, 4); + err = iicbus_transfer(dev, msg, nitems(msg)); if (err != 0) goto retry; if (busy[0] & 0x80) goto retry; /* Check for invalid value and retry. */ if (buf[0] == 0xff) goto retry; *data = *((uint8_t*)buf); return (0); retry: if (++try > 5) { device_printf(dev, "iicbus read failed\n"); return (-1); } pause("max6690_read", hz); } } static int max6690_probe(device_t dev) { const char *name, *compatible; struct max6690_softc *sc; name = ofw_bus_get_name(dev); compatible = ofw_bus_get_compat(dev); if (!name) return (ENXIO); if (strcmp(name, "temp-monitor") != 0 || strcmp(compatible, "max6690") != 0) return (ENXIO); sc = device_get_softc(dev); sc->sc_dev = dev; sc->sc_addr = iicbus_get_addr(dev); device_set_desc(dev, "Temp-Monitor MAX6690"); return (0); } /* * This function returns the number of sensors. If we call it the second time * and we have allocated memory for sc->sc_sensors, we fill in the properties. */ static int max6690_fill_sensor_prop(device_t dev) { phandle_t child; struct max6690_softc *sc; u_int id[8]; char location[96]; int i = 0, j, len = 0, prop_len, prev_len = 0; sc = device_get_softc(dev); child = ofw_bus_get_node(dev); /* Fill the sensor location property. */ prop_len = OF_getprop(child, "hwsensor-location", location, sizeof(location)); while (len < prop_len) { if (sc->sc_sensors != NULL) strcpy(sc->sc_sensors[i].therm.name, location + len); prev_len = strlen(location + len) + 1; len += prev_len; i++; } if (sc->sc_sensors == NULL) return (i); /* Fill the sensor id property. */ prop_len = OF_getprop(child, "hwsensor-id", id, sizeof(id)); for (j = 0; j < i; j++) sc->sc_sensors[j].id = (id[j] & 0xf); /* Fill the sensor zone property. */ prop_len = OF_getprop(child, "hwsensor-zone", id, sizeof(id)); for (j = 0; j < i; j++) sc->sc_sensors[j].therm.zone = id[j]; /* Set up remaining sensor properties */ for (j = 0; j < i; j++) { sc->sc_sensors[j].dev = dev; sc->sc_sensors[j].therm.target_temp = 400 + ZERO_C_TO_K; sc->sc_sensors[j].therm.max_temp = 800 + ZERO_C_TO_K; sc->sc_sensors[j].therm.read = (int (*)(struct pmac_therm *))(max6690_sensor_read); } return (i); } static int max6690_attach(device_t dev) { struct max6690_softc *sc; sc = device_get_softc(dev); sc->enum_hook.ich_func = max6690_start; sc->enum_hook.ich_arg = dev; /* We have to wait until interrupts are enabled. I2C read and write * only works if the interrupts are available. * The unin/i2c is controlled by the htpic on unin. But this is not * the master. The openpic on mac-io is controlling the htpic. * This one gets attached after the mac-io probing and then the * interrupts will be available. */ if (config_intrhook_establish(&sc->enum_hook) != 0) return (ENOMEM); return (0); } static void max6690_start(void *xdev) { struct max6690_softc *sc; struct sysctl_oid *oid, *sensroot_oid; struct sysctl_ctx_list *ctx; char sysctl_desc[40], sysctl_name[32]; int i, j; device_t dev = (device_t)xdev; sc = device_get_softc(dev); sc->sc_nsensors = 0; /* Count the actual number of sensors. */ sc->sc_nsensors = max6690_fill_sensor_prop(dev); device_printf(dev, "%d sensors detected.\n", sc->sc_nsensors); if (sc->sc_nsensors == 0) device_printf(dev, "WARNING: No MAX6690 sensors detected!\n"); sc->sc_sensors = malloc (sc->sc_nsensors * sizeof(struct max6690_sensor), M_MAX6690, M_WAITOK | M_ZERO); ctx = device_get_sysctl_ctx(dev); sensroot_oid = SYSCTL_ADD_NODE(ctx, SYSCTL_CHILDREN(device_get_sysctl_tree(dev)), OID_AUTO, "sensor", CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "MAX6690 Sensor Information"); /* Now we can fill the properties into the allocated struct. */ sc->sc_nsensors = max6690_fill_sensor_prop(dev); /* Register with powermac_thermal */ for (i = 0; i < sc->sc_nsensors; i++) pmac_thermal_sensor_register(&sc->sc_sensors[i].therm); /* Add sysctls for the sensors. */ for (i = 0; i < sc->sc_nsensors; i++) { for (j = 0; j < strlen(sc->sc_sensors[i].therm.name); j++) { sysctl_name[j] = tolower(sc->sc_sensors[i].therm.name[j]); if (isspace(sysctl_name[j])) sysctl_name[j] = '_'; } sysctl_name[j] = 0; sprintf(sysctl_desc,"%s %s", sc->sc_sensors[i].therm.name, "(C)"); oid = SYSCTL_ADD_NODE(ctx, SYSCTL_CHILDREN(sensroot_oid), OID_AUTO, sysctl_name, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "Sensor Information"); /* I use i to pass the sensor id. */ SYSCTL_ADD_PROC(ctx, SYSCTL_CHILDREN(oid), OID_AUTO, "temp", - CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_NEEDGIANT, dev, i % 2, - max6690_sensor_sysctl, "IK", sysctl_desc); + CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE, + dev, i % 2, + max6690_sensor_sysctl, "IK", sysctl_desc); } /* Dump sensor location & ID. */ if (bootverbose) { device_printf(dev, "Sensors\n"); for (i = 0; i < sc->sc_nsensors; i++) { device_printf(dev, "Location : %s ID: %d\n", sc->sc_sensors[i].therm.name, sc->sc_sensors[i].id); } } config_intrhook_disestablish(&sc->enum_hook); } static int max6690_sensor_read(struct max6690_sensor *sens) { uint8_t reg_int = 0, reg_ext = 0; uint8_t integer = 0; uint8_t fraction = 0; int err, temp; struct max6690_softc *sc; sc = device_get_softc(sens->dev); /* The internal sensor id's are even, the external are odd. */ if ((sens->id % 2) == 0) { reg_int = MAX6690_INT_TEMP; reg_ext = MAX6690_IEXT_TEMP; } else { reg_int = MAX6690_EXT_TEMP; reg_ext = MAX6690_EEXT_TEMP; } err = max6690_read(sc->sc_dev, sc->sc_addr, reg_int, &integer); if (err < 0) return (-1); err = max6690_read(sc->sc_dev, sc->sc_addr, reg_ext, &fraction); if (err < 0) return (-1); fraction &= MAX6690_TEMP_MASK; /* The temperature is in tenth kelvin, the fractional part resolution is 0.125. */ temp = (integer * 10) + (fraction >> 5) * 10 / 8; return (temp + ZERO_C_TO_K); } static int max6690_sensor_sysctl(SYSCTL_HANDLER_ARGS) { device_t dev; struct max6690_softc *sc; struct max6690_sensor *sens; int error; int temp; dev = arg1; sc = device_get_softc(dev); sens = &sc->sc_sensors[arg2]; temp = max6690_sensor_read(sens); if (temp < 0) return (EIO); error = sysctl_handle_int(oidp, &temp, 0, req); return (error); } diff --git a/sys/fs/mntfs/mntfs_vnops.c b/sys/fs/mntfs/mntfs_vnops.c new file mode 100644 index 000000000000..2708bc5d319f --- /dev/null +++ b/sys/fs/mntfs/mntfs_vnops.c @@ -0,0 +1,95 @@ +/*- + * SPDX-License-Identifier: BSD-2-Clause-FreeBSD + * + * Copyright (c) 2020 Netflix, Inc. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY THE AUTHORS ``AS IS'' AND ANY EXPRESS OR + * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES + * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. + * IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY DIRECT, INDIRECT, + * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, + * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS + * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND + * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR + * TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE + * USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + * + * $FreeBSD$ + */ + +#include +#include +#include +#include + +/* + * The "mntfs" VCHR vnodes implemented here provide a safe way for file systems + * to access their disk devices. Using the normal devfs vnode has the problem + * that if the device disappears, the devfs vnode is vgone'd as part of + * removing it from the application-visible namespace, and some file systems + * (notably FFS with softdep) get very unhappy if their dirty buffers are + * invalidated out from under them. By using a separate, private vnode, + * file systems are able to clean up their buffer state in a controlled fashion + * when the underlying device disappears. + */ + +static int +mntfs_reclaim(struct vop_reclaim_args *ap) +{ + struct vnode *vp = ap->a_vp; + + dev_rel(vp->v_rdev); + return (0); +} + +struct vop_vector mntfs_vnodeops = { + .vop_default = &default_vnodeops, + + .vop_fsync = vop_stdfsync, + .vop_strategy = VOP_PANIC, + .vop_reclaim = mntfs_reclaim, +}; +VFS_VOP_VECTOR_REGISTER(mntfs_vnodeops); + +/* + * Allocate a private VCHR vnode for use by a mounted fs. + * The underlying device will be the same as for the given vnode. + * This mntfs vnode must be freed with mntfs_freevp() rather than just + * releasing the reference. + */ +struct vnode * +mntfs_allocvp(struct mount *mp, struct vnode *ovp) +{ + struct vnode *vp; + struct cdev *dev; + + ASSERT_VOP_ELOCKED(ovp, __func__); + + dev = ovp->v_rdev; + + getnewvnode("mntfs", mp, &mntfs_vnodeops, &vp); + vp->v_type = VCHR; + vp->v_data = NULL; + dev_ref(dev); + vp->v_rdev = dev; + + return (vp); +} + +void +mntfs_freevp(struct vnode *vp) +{ + + vgone(vp); + vrele(vp); +} diff --git a/sys/kern/subr_counter.c b/sys/kern/subr_counter.c index 8eb9f19e158c..fec51b6b635a 100644 --- a/sys/kern/subr_counter.c +++ b/sys/kern/subr_counter.c @@ -1,174 +1,192 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2012 Gleb Smirnoff * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #define IN_SUBR_COUNTER_C #include void counter_u64_zero(counter_u64_t c) { counter_u64_zero_inline(c); } uint64_t counter_u64_fetch(counter_u64_t c) { return (counter_u64_fetch_inline(c)); } counter_u64_t counter_u64_alloc(int flags) { return (uma_zalloc_pcpu(pcpu_zone_64, flags | M_ZERO)); } void counter_u64_free(counter_u64_t c) { uma_zfree_pcpu(pcpu_zone_64, c); } int sysctl_handle_counter_u64(SYSCTL_HANDLER_ARGS) { uint64_t out; int error; out = counter_u64_fetch(*(counter_u64_t *)arg1); error = SYSCTL_OUT(req, &out, sizeof(uint64_t)); if (error || !req->newptr) return (error); /* * Any write attempt to a counter zeroes it. */ counter_u64_zero(*(counter_u64_t *)arg1); return (0); } int sysctl_handle_counter_u64_array(SYSCTL_HANDLER_ARGS) { uint64_t *out; int error; out = malloc(arg2 * sizeof(uint64_t), M_TEMP, M_WAITOK); for (int i = 0; i < arg2; i++) out[i] = counter_u64_fetch(((counter_u64_t *)arg1)[i]); error = SYSCTL_OUT(req, out, arg2 * sizeof(uint64_t)); free(out, M_TEMP); if (error || !req->newptr) return (error); /* * Any write attempt to a counter zeroes it. */ for (int i = 0; i < arg2; i++) counter_u64_zero(((counter_u64_t *)arg1)[i]); return (0); } /* * MP-friendly version of ppsratecheck(). * * Returns non-negative if we are in the rate, negative otherwise. * 0 - rate limit not reached. * -1 - rate limit reached. * >0 - rate limit was reached before, and was just reset. The return value * is number of events since last reset. */ int64_t counter_ratecheck(struct counter_rate *cr, int64_t limit) { int64_t val; int now; val = cr->cr_over; now = ticks; if ((u_int)(now - cr->cr_ticks) >= hz) { /* * Time to clear the structure, we are in the next second. * First try unlocked read, and then proceed with atomic. */ if ((cr->cr_lock == 0) && atomic_cmpset_acq_int(&cr->cr_lock, 0, 1)) { /* * Check if other thread has just went through the * reset sequence before us. */ if ((u_int)(now - cr->cr_ticks) >= hz) { val = counter_u64_fetch(cr->cr_rate); counter_u64_zero(cr->cr_rate); cr->cr_over = 0; cr->cr_ticks = now; if (val <= limit) val = 0; } atomic_store_rel_int(&cr->cr_lock, 0); } else /* * We failed to lock, in this case other thread may * be running counter_u64_zero(), so it is not safe * to do an update, we skip it. */ return (val); } counter_u64_add(cr->cr_rate, 1); if (cr->cr_over != 0) return (-1); if (counter_u64_fetch(cr->cr_rate) > limit) val = cr->cr_over = -1; return (val); } + +void +counter_u64_sysinit(void *arg) +{ + counter_u64_t *cp; + + cp = arg; + *cp = counter_u64_alloc(M_WAITOK); +} + +void +counter_u64_sysuninit(void *arg) +{ + counter_u64_t *cp; + + cp = arg; + counter_u64_free(*cp); +} diff --git a/sys/kern/subr_pcpu.c b/sys/kern/subr_pcpu.c index 101440e3b1ec..f37b9324a919 100644 --- a/sys/kern/subr_pcpu.c +++ b/sys/kern/subr_pcpu.c @@ -1,419 +1,419 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 2001 Wind River Systems, Inc. * All rights reserved. * Written by: John Baldwin * * Copyright (c) 2009 Jeffrey Roberson * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the author nor the names of any co-contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ /* * This module provides MI support for per-cpu data. * * Each architecture determines the mapping of logical CPU IDs to physical * CPUs. The requirements of this mapping are as follows: * - Logical CPU IDs must reside in the range 0 ... MAXCPU - 1. * - The mapping is not required to be dense. That is, there may be * gaps in the mappings. * - The platform sets the value of MAXCPU in . * - It is suggested, but not required, that in the non-SMP case, the * platform define MAXCPU to be 1 and define the logical ID of the * sole CPU as 0. */ #include __FBSDID("$FreeBSD$"); #include "opt_ddb.h" #include #include #include #include #include #include #include #include #include #include #include static MALLOC_DEFINE(M_PCPU, "Per-cpu", "Per-cpu resource accouting."); struct dpcpu_free { uintptr_t df_start; int df_len; TAILQ_ENTRY(dpcpu_free) df_link; }; DPCPU_DEFINE_STATIC(char, modspace[DPCPU_MODMIN] __aligned(__alignof(void *))); static TAILQ_HEAD(, dpcpu_free) dpcpu_head = TAILQ_HEAD_INITIALIZER(dpcpu_head); static struct sx dpcpu_lock; uintptr_t dpcpu_off[MAXCPU]; struct pcpu *cpuid_to_pcpu[MAXCPU]; struct cpuhead cpuhead = STAILQ_HEAD_INITIALIZER(cpuhead); /* * Initialize the MI portions of a struct pcpu. */ void pcpu_init(struct pcpu *pcpu, int cpuid, size_t size) { bzero(pcpu, size); KASSERT(cpuid >= 0 && cpuid < MAXCPU, ("pcpu_init: invalid cpuid %d", cpuid)); pcpu->pc_cpuid = cpuid; cpuid_to_pcpu[cpuid] = pcpu; STAILQ_INSERT_TAIL(&cpuhead, pcpu, pc_allcpu); cpu_pcpu_init(pcpu, cpuid, size); pcpu->pc_rm_queue.rmq_next = &pcpu->pc_rm_queue; pcpu->pc_rm_queue.rmq_prev = &pcpu->pc_rm_queue; pcpu->pc_zpcpu_offset = zpcpu_offset_cpu(cpuid); } void dpcpu_init(void *dpcpu, int cpuid) { struct pcpu *pcpu; pcpu = pcpu_find(cpuid); pcpu->pc_dynamic = (uintptr_t)dpcpu - DPCPU_START; /* * Initialize defaults from our linker section. */ memcpy(dpcpu, (void *)DPCPU_START, DPCPU_BYTES); /* * Place it in the global pcpu offset array. */ dpcpu_off[cpuid] = pcpu->pc_dynamic; } static void dpcpu_startup(void *dummy __unused) { struct dpcpu_free *df; df = malloc(sizeof(*df), M_PCPU, M_WAITOK | M_ZERO); df->df_start = (uintptr_t)&DPCPU_NAME(modspace); df->df_len = DPCPU_MODMIN; TAILQ_INSERT_HEAD(&dpcpu_head, df, df_link); sx_init(&dpcpu_lock, "dpcpu alloc lock"); } SYSINIT(dpcpu, SI_SUB_KLD, SI_ORDER_FIRST, dpcpu_startup, NULL); /* * UMA_PCPU_ZONE zones, that are available for all kernel * consumers. Right now 64 bit zone is used for counter(9) * and int zone is used for mount point counters. */ uma_zone_t pcpu_zone_int; uma_zone_t pcpu_zone_64; static void pcpu_zones_startup(void) { pcpu_zone_int = uma_zcreate("int pcpu", sizeof(int), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_PCPU); pcpu_zone_64 = uma_zcreate("64 pcpu", sizeof(uint64_t), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_PCPU); } -SYSINIT(pcpu_zones, SI_SUB_VM, SI_ORDER_ANY, pcpu_zones_startup, NULL); +SYSINIT(pcpu_zones, SI_SUB_COUNTER, SI_ORDER_FIRST, pcpu_zones_startup, NULL); /* * First-fit extent based allocator for allocating space in the per-cpu * region reserved for modules. This is only intended for use by the * kernel linkers to place module linker sets. */ void * dpcpu_alloc(int size) { struct dpcpu_free *df; void *s; s = NULL; size = roundup2(size, sizeof(void *)); sx_xlock(&dpcpu_lock); TAILQ_FOREACH(df, &dpcpu_head, df_link) { if (df->df_len < size) continue; if (df->df_len == size) { s = (void *)df->df_start; TAILQ_REMOVE(&dpcpu_head, df, df_link); free(df, M_PCPU); break; } s = (void *)df->df_start; df->df_len -= size; df->df_start = df->df_start + size; break; } sx_xunlock(&dpcpu_lock); return (s); } /* * Free dynamic per-cpu space at module unload time. */ void dpcpu_free(void *s, int size) { struct dpcpu_free *df; struct dpcpu_free *dn; uintptr_t start; uintptr_t end; size = roundup2(size, sizeof(void *)); start = (uintptr_t)s; end = start + size; /* * Free a region of space and merge it with as many neighbors as * possible. Keeping the list sorted simplifies this operation. */ sx_xlock(&dpcpu_lock); TAILQ_FOREACH(df, &dpcpu_head, df_link) { if (df->df_start > end) break; /* * If we expand at the end of an entry we may have to * merge it with the one following it as well. */ if (df->df_start + df->df_len == start) { df->df_len += size; dn = TAILQ_NEXT(df, df_link); if (df->df_start + df->df_len == dn->df_start) { df->df_len += dn->df_len; TAILQ_REMOVE(&dpcpu_head, dn, df_link); free(dn, M_PCPU); } sx_xunlock(&dpcpu_lock); return; } if (df->df_start == end) { df->df_start = start; df->df_len += size; sx_xunlock(&dpcpu_lock); return; } } dn = malloc(sizeof(*df), M_PCPU, M_WAITOK | M_ZERO); dn->df_start = start; dn->df_len = size; if (df) TAILQ_INSERT_BEFORE(df, dn, df_link); else TAILQ_INSERT_TAIL(&dpcpu_head, dn, df_link); sx_xunlock(&dpcpu_lock); } /* * Initialize the per-cpu storage from an updated linker-set region. */ void dpcpu_copy(void *s, int size) { #ifdef SMP uintptr_t dpcpu; int i; CPU_FOREACH(i) { dpcpu = dpcpu_off[i]; if (dpcpu == 0) continue; memcpy((void *)(dpcpu + (uintptr_t)s), s, size); } #else memcpy((void *)(dpcpu_off[0] + (uintptr_t)s), s, size); #endif } /* * Destroy a struct pcpu. */ void pcpu_destroy(struct pcpu *pcpu) { STAILQ_REMOVE(&cpuhead, pcpu, pcpu, pc_allcpu); cpuid_to_pcpu[pcpu->pc_cpuid] = NULL; dpcpu_off[pcpu->pc_cpuid] = 0; } /* * Locate a struct pcpu by cpu id. */ struct pcpu * pcpu_find(u_int cpuid) { return (cpuid_to_pcpu[cpuid]); } int sysctl_dpcpu_quad(SYSCTL_HANDLER_ARGS) { uintptr_t dpcpu; int64_t count; int i; count = 0; CPU_FOREACH(i) { dpcpu = dpcpu_off[i]; if (dpcpu == 0) continue; count += *(int64_t *)(dpcpu + (uintptr_t)arg1); } return (SYSCTL_OUT(req, &count, sizeof(count))); } int sysctl_dpcpu_long(SYSCTL_HANDLER_ARGS) { uintptr_t dpcpu; long count; int i; count = 0; CPU_FOREACH(i) { dpcpu = dpcpu_off[i]; if (dpcpu == 0) continue; count += *(long *)(dpcpu + (uintptr_t)arg1); } return (SYSCTL_OUT(req, &count, sizeof(count))); } int sysctl_dpcpu_int(SYSCTL_HANDLER_ARGS) { uintptr_t dpcpu; int count; int i; count = 0; CPU_FOREACH(i) { dpcpu = dpcpu_off[i]; if (dpcpu == 0) continue; count += *(int *)(dpcpu + (uintptr_t)arg1); } return (SYSCTL_OUT(req, &count, sizeof(count))); } #ifdef DDB DB_SHOW_COMMAND(dpcpu_off, db_show_dpcpu_off) { int id; CPU_FOREACH(id) { db_printf("dpcpu_off[%2d] = 0x%jx (+ DPCPU_START = %p)\n", id, (uintmax_t)dpcpu_off[id], (void *)(uintptr_t)(dpcpu_off[id] + DPCPU_START)); } } static void show_pcpu(struct pcpu *pc) { struct thread *td; db_printf("cpuid = %d\n", pc->pc_cpuid); db_printf("dynamic pcpu = %p\n", (void *)pc->pc_dynamic); db_printf("curthread = "); td = pc->pc_curthread; if (td != NULL) db_printf("%p: pid %d tid %d critnest %d \"%s\"\n", td, td->td_proc->p_pid, td->td_tid, td->td_critnest, td->td_name); else db_printf("none\n"); db_printf("curpcb = %p\n", pc->pc_curpcb); db_printf("fpcurthread = "); td = pc->pc_fpcurthread; if (td != NULL) db_printf("%p: pid %d \"%s\"\n", td, td->td_proc->p_pid, td->td_name); else db_printf("none\n"); db_printf("idlethread = "); td = pc->pc_idlethread; if (td != NULL) db_printf("%p: tid %d \"%s\"\n", td, td->td_tid, td->td_name); else db_printf("none\n"); db_show_mdpcpu(pc); #ifdef VIMAGE db_printf("curvnet = %p\n", pc->pc_curthread->td_vnet); #endif #ifdef WITNESS db_printf("spin locks held:\n"); witness_list_locks(&pc->pc_spinlocks, db_printf); #endif } DB_SHOW_COMMAND(pcpu, db_show_pcpu) { struct pcpu *pc; int id; if (have_addr) id = ((addr >> 4) % 16) * 10 + (addr % 16); else id = PCPU_GET(cpuid); pc = pcpu_find(id); if (pc == NULL) { db_printf("CPU %d not found\n", id); return; } show_pcpu(pc); } DB_SHOW_ALL_COMMAND(pcpu, db_show_cpu_all) { struct pcpu *pc; int id; db_printf("Current CPU: %d\n\n", PCPU_GET(cpuid)); CPU_FOREACH(id) { pc = pcpu_find(id); if (pc != NULL) { show_pcpu(pc); db_printf("\n"); } } } DB_SHOW_ALIAS(allpcpu, db_show_cpu_all); #endif diff --git a/sys/kern/subr_smr.c b/sys/kern/subr_smr.c index f5f30ac9fedb..cbbf185fee79 100644 --- a/sys/kern/subr_smr.c +++ b/sys/kern/subr_smr.c @@ -1,645 +1,633 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2019,2020 Jeffrey Roberson * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice unmodified, this list of conditions, and the following * disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include /* * Global Unbounded Sequences (GUS) * * This is a novel safe memory reclamation technique inspired by * epoch based reclamation from Samy Al Bahra's concurrency kit which * in turn was based on work described in: * Fraser, K. 2004. Practical Lock-Freedom. PhD Thesis, University * of Cambridge Computing Laboratory. * And shares some similarities with: * Wang, Stamler, Parmer. 2016 Parallel Sections: Scaling System-Level * Data-Structures * * This is not an implementation of hazard pointers or related * techniques. The term safe memory reclamation is used as a * generic descriptor for algorithms that defer frees to avoid * use-after-free errors with lockless datastructures or as * a mechanism to detect quiescence for writer synchronization. * * The basic approach is to maintain a monotonic write sequence * number that is updated on some application defined granularity. * Readers record the most recent write sequence number they have * observed. A shared read sequence number records the lowest * sequence number observed by any reader as of the last poll. Any * write older than this value has been observed by all readers * and memory can be reclaimed. Like Epoch we also detect idle * readers by storing an invalid sequence number in the per-cpu * state when the read section exits. Like Parsec we establish * a global write clock that is used to mark memory on free. * * The write and read sequence numbers can be thought of as a two * handed clock with readers always advancing towards writers. GUS * maintains the invariant that all readers can safely access memory * that was visible at the time they loaded their copy of the sequence * number. Periodically the read sequence or hand is polled and * advanced as far towards the write sequence as active readers allow. * Memory which was freed between the old and new global read sequence * number can now be reclaimed. When the system is idle the two hands * meet and no deferred memory is outstanding. Readers never advance * any sequence number, they only observe them. The shared read * sequence number is consequently never higher than the write sequence. * A stored sequence number that falls outside of this range has expired * and needs no scan to reclaim. * * A notable distinction between GUS and Epoch, qsbr, rcu, etc. is * that advancing the sequence number is decoupled from detecting its * observation. That is to say, the delta between read and write * sequence numbers is not bound. This can be thought of as a more * generalized form of epoch which requires them at most one step * apart. This results in a more granular assignment of sequence * numbers even as read latencies prohibit all or some expiration. * It also allows writers to advance the sequence number and save the * poll for expiration until a later time when it is likely to * complete without waiting. The batch granularity and free-to-use * latency is dynamic and can be significantly smaller than in more * strict systems. * * This mechanism is primarily intended to be used in coordination with * UMA. By integrating with the allocator we avoid all of the callout * queue machinery and are provided with an efficient way to batch * sequence advancement and waiting. The allocator accumulates a full * per-cpu cache of memory before advancing the sequence. It then * delays waiting for this sequence to expire until the memory is * selected for reuse. In this way we only increment the sequence * value once for n=cache-size frees and the waits are done long * after the sequence has been expired so they need only be verified * to account for pathological conditions and to advance the read * sequence. Tying the sequence number to the bucket size has the * nice property that as the zone gets busier the buckets get larger * and the sequence writes become fewer. If the coherency of advancing * the write sequence number becomes too costly we can advance * it for every N buckets in exchange for higher free-to-use * latency and consequently higher memory consumption. * * If the read overhead of accessing the shared cacheline becomes * especially burdensome an invariant TSC could be used in place of the * sequence. The algorithm would then only need to maintain the minimum * observed tsc. This would trade potential cache synchronization * overhead for local serialization and cpu timestamp overhead. */ /* * A simplified diagram: * * 0 UINT_MAX * | -------------------- sequence number space -------------------- | * ^ rd seq ^ wr seq * | ----- valid sequence numbers ---- | * ^cpuA ^cpuC * | -- free -- | --------- deferred frees -------- | ---- free ---- | * * * In this example cpuA has the lowest sequence number and poll can * advance rd seq. cpuB is not running and is considered to observe * wr seq. * * Freed memory that is tagged with a sequence number between rd seq and * wr seq can not be safely reclaimed because cpuA may hold a reference to * it. Any other memory is guaranteed to be unreferenced. * * Any writer is free to advance wr seq at any time however it may busy * poll in pathological cases. */ static uma_zone_t smr_shared_zone; static uma_zone_t smr_zone; #ifndef INVARIANTS #define SMR_SEQ_INIT 1 /* All valid sequence numbers are odd. */ #define SMR_SEQ_INCR 2 /* * SMR_SEQ_MAX_DELTA is the maximum distance allowed between rd_seq and * wr_seq. For the modular arithmetic to work a value of UNIT_MAX / 2 * would be possible but it is checked after we increment the wr_seq so * a safety margin is left to prevent overflow. * * We will block until SMR_SEQ_MAX_ADVANCE sequence numbers have progressed * to prevent integer wrapping. See smr_advance() for more details. */ #define SMR_SEQ_MAX_DELTA (UINT_MAX / 4) #define SMR_SEQ_MAX_ADVANCE (SMR_SEQ_MAX_DELTA - 1024) #else /* We want to test the wrapping feature in invariants kernels. */ #define SMR_SEQ_INCR (UINT_MAX / 10000) #define SMR_SEQ_INIT (UINT_MAX - 100000) /* Force extra polls to test the integer overflow detection. */ #define SMR_SEQ_MAX_DELTA (SMR_SEQ_INCR * 32) #define SMR_SEQ_MAX_ADVANCE SMR_SEQ_MAX_DELTA / 2 #endif /* * The grace period for lazy (tick based) SMR. * * Hardclock is responsible for advancing ticks on a single CPU while every * CPU receives a regular clock interrupt. The clock interrupts are flushing * the store buffers and any speculative loads that may violate our invariants. * Because these interrupts are not synchronized we must wait one additional * tick in the future to be certain that all processors have had their state * synchronized by an interrupt. * * This assumes that the clock interrupt will only be delayed by other causes * that will flush the store buffer or prevent access to the section protected * data. For example, an idle processor, or an system management interrupt, * or a vm exit. */ #define SMR_LAZY_GRACE 2 #define SMR_LAZY_INCR (SMR_LAZY_GRACE * SMR_SEQ_INCR) /* * The maximum sequence number ahead of wr_seq that may still be valid. The * sequence may not be advanced on write for lazy or deferred SMRs. In this * case poll needs to attempt to forward the sequence number if the goal is * within wr_seq + SMR_SEQ_ADVANCE. */ #define SMR_SEQ_ADVANCE SMR_LAZY_INCR static SYSCTL_NODE(_debug, OID_AUTO, smr, CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, "SMR Stats"); -static counter_u64_t advance = EARLY_COUNTER; +static COUNTER_U64_DEFINE_EARLY(advance); SYSCTL_COUNTER_U64(_debug_smr, OID_AUTO, advance, CTLFLAG_RW, &advance, ""); -static counter_u64_t advance_wait = EARLY_COUNTER; +static COUNTER_U64_DEFINE_EARLY(advance_wait); SYSCTL_COUNTER_U64(_debug_smr, OID_AUTO, advance_wait, CTLFLAG_RW, &advance_wait, ""); -static counter_u64_t poll = EARLY_COUNTER; +static COUNTER_U64_DEFINE_EARLY(poll); SYSCTL_COUNTER_U64(_debug_smr, OID_AUTO, poll, CTLFLAG_RW, &poll, ""); -static counter_u64_t poll_scan = EARLY_COUNTER; +static COUNTER_U64_DEFINE_EARLY(poll_scan); SYSCTL_COUNTER_U64(_debug_smr, OID_AUTO, poll_scan, CTLFLAG_RW, &poll_scan, ""); -static counter_u64_t poll_fail = EARLY_COUNTER; +static COUNTER_U64_DEFINE_EARLY(poll_fail); SYSCTL_COUNTER_U64(_debug_smr, OID_AUTO, poll_fail, CTLFLAG_RW, &poll_fail, ""); /* * Advance a lazy write sequence number. These move forward at the rate of * ticks. Grace is SMR_LAZY_INCR (2 ticks) in the future. * * This returns the goal write sequence number. */ static smr_seq_t smr_lazy_advance(smr_t smr, smr_shared_t s) { union s_wr s_wr, old; int t, d; CRITICAL_ASSERT(curthread); /* * Load the stored ticks value before the current one. This way the * current value can only be the same or larger. */ old._pair = s_wr._pair = atomic_load_acq_64(&s->s_wr._pair); t = ticks; /* * The most probable condition that the update already took place. */ d = t - s_wr.ticks; if (__predict_true(d == 0)) goto out; /* Cap the rate of advancement and handle long idle periods. */ if (d > SMR_LAZY_GRACE || d < 0) d = SMR_LAZY_GRACE; s_wr.ticks = t; s_wr.seq += d * SMR_SEQ_INCR; /* * This can only fail if another thread races to call advance(). * Strong cmpset semantics mean we are guaranteed that the update * happened. */ atomic_cmpset_64(&s->s_wr._pair, old._pair, s_wr._pair); out: return (s_wr.seq + SMR_LAZY_INCR); } /* * Increment the shared write sequence by 2. Since it is initialized * to 1 this means the only valid values are odd and an observed value * of 0 in a particular CPU means it is not currently in a read section. */ static smr_seq_t smr_shared_advance(smr_shared_t s) { return (atomic_fetchadd_int(&s->s_wr.seq, SMR_SEQ_INCR) + SMR_SEQ_INCR); } /* * Advance the write sequence number for a normal smr section. If the * write sequence is too far behind the read sequence we have to poll * to advance rd_seq and prevent undetectable wraps. */ static smr_seq_t smr_default_advance(smr_t smr, smr_shared_t s) { smr_seq_t goal, s_rd_seq; CRITICAL_ASSERT(curthread); KASSERT((zpcpu_get(smr)->c_flags & SMR_LAZY) == 0, ("smr_default_advance: called with lazy smr.")); /* * Load the current read seq before incrementing the goal so * we are guaranteed it is always < goal. */ s_rd_seq = atomic_load_acq_int(&s->s_rd_seq); goal = smr_shared_advance(s); /* * Force a synchronization here if the goal is getting too * far ahead of the read sequence number. This keeps the * wrap detecting arithmetic working in pathological cases. */ if (SMR_SEQ_DELTA(goal, s_rd_seq) >= SMR_SEQ_MAX_DELTA) { counter_u64_add(advance_wait, 1); smr_wait(smr, goal - SMR_SEQ_MAX_ADVANCE); } counter_u64_add(advance, 1); return (goal); } /* * Deferred SMRs conditionally update s_wr_seq based on an * cpu local interval count. */ static smr_seq_t smr_deferred_advance(smr_t smr, smr_shared_t s, smr_t self) { if (++self->c_deferred < self->c_limit) return (smr_shared_current(s) + SMR_SEQ_INCR); self->c_deferred = 0; return (smr_default_advance(smr, s)); } /* * Advance the write sequence and return the value for use as the * wait goal. This guarantees that any changes made by the calling * thread prior to this call will be visible to all threads after * rd_seq meets or exceeds the return value. * * This function may busy loop if the readers are roughly 1 billion * sequence numbers behind the writers. * * Lazy SMRs will not busy loop and the wrap happens every 25 days * at 1khz and 60 hours at 10khz. Readers can block for no longer * than half of this for SMR_SEQ_ macros to continue working. */ smr_seq_t smr_advance(smr_t smr) { smr_t self; smr_shared_t s; smr_seq_t goal; int flags; /* * It is illegal to enter while in an smr section. */ SMR_ASSERT_NOT_ENTERED(smr); /* * Modifications not done in a smr section need to be visible * before advancing the seq. */ atomic_thread_fence_rel(); critical_enter(); /* Try to touch the line once. */ self = zpcpu_get(smr); s = self->c_shared; flags = self->c_flags; goal = SMR_SEQ_INVALID; if ((flags & (SMR_LAZY | SMR_DEFERRED)) == 0) goal = smr_default_advance(smr, s); else if ((flags & SMR_LAZY) != 0) goal = smr_lazy_advance(smr, s); else if ((flags & SMR_DEFERRED) != 0) goal = smr_deferred_advance(smr, s, self); critical_exit(); return (goal); } /* * Poll to determine the currently observed sequence number on a cpu * and spinwait if the 'wait' argument is true. */ static smr_seq_t smr_poll_cpu(smr_t c, smr_seq_t s_rd_seq, smr_seq_t goal, bool wait) { smr_seq_t c_seq; c_seq = SMR_SEQ_INVALID; for (;;) { c_seq = atomic_load_int(&c->c_seq); if (c_seq == SMR_SEQ_INVALID) break; /* * There is a race described in smr.h:smr_enter that * can lead to a stale seq value but not stale data * access. If we find a value out of range here we * pin it to the current min to prevent it from * advancing until that stale section has expired. * * The race is created when a cpu loads the s_wr_seq * value in a local register and then another thread * advances s_wr_seq and calls smr_poll() which will * oberve no value yet in c_seq and advance s_rd_seq * up to s_wr_seq which is beyond the register * cached value. This is only likely to happen on * hypervisor or with a system management interrupt. */ if (SMR_SEQ_LT(c_seq, s_rd_seq)) c_seq = s_rd_seq; /* * If the sequence number meets the goal we are done * with this cpu. */ if (SMR_SEQ_LEQ(goal, c_seq)) break; if (!wait) break; cpu_spinwait(); } return (c_seq); } /* * Loop until all cores have observed the goal sequence or have * gone inactive. Returns the oldest sequence currently active; * * This function assumes a snapshot of sequence values has * been obtained and validated by smr_poll(). */ static smr_seq_t smr_poll_scan(smr_t smr, smr_shared_t s, smr_seq_t s_rd_seq, smr_seq_t s_wr_seq, smr_seq_t goal, bool wait) { smr_seq_t rd_seq, c_seq; int i; CRITICAL_ASSERT(curthread); counter_u64_add_protected(poll_scan, 1); /* * The read sequence can be no larger than the write sequence at * the start of the poll. */ rd_seq = s_wr_seq; CPU_FOREACH(i) { /* * Query the active sequence on this cpu. If we're not * waiting and we don't meet the goal we will still scan * the rest of the cpus to update s_rd_seq before returning * failure. */ c_seq = smr_poll_cpu(zpcpu_get_cpu(smr, i), s_rd_seq, goal, wait); /* * Limit the minimum observed rd_seq whether we met the goal * or not. */ if (c_seq != SMR_SEQ_INVALID) rd_seq = SMR_SEQ_MIN(rd_seq, c_seq); } /* * Advance the rd_seq as long as we observed a more recent value. */ s_rd_seq = atomic_load_int(&s->s_rd_seq); if (SMR_SEQ_GT(rd_seq, s_rd_seq)) { atomic_cmpset_int(&s->s_rd_seq, s_rd_seq, rd_seq); s_rd_seq = rd_seq; } return (s_rd_seq); } /* * Poll to determine whether all readers have observed the 'goal' write * sequence number. * * If wait is true this will spin until the goal is met. * * This routine will updated the minimum observed read sequence number in * s_rd_seq if it does a scan. It may not do a scan if another call has * advanced s_rd_seq beyond the callers goal already. * * Returns true if the goal is met and false if not. */ bool smr_poll(smr_t smr, smr_seq_t goal, bool wait) { smr_shared_t s; smr_t self; smr_seq_t s_wr_seq, s_rd_seq; smr_delta_t delta; int flags; bool success; /* * It is illegal to enter while in an smr section. */ KASSERT(!wait || !SMR_ENTERED(smr), ("smr_poll: Blocking not allowed in a SMR section.")); KASSERT(!wait || (zpcpu_get(smr)->c_flags & SMR_LAZY) == 0, ("smr_poll: Blocking not allowed on lazy smrs.")); /* * Use a critical section so that we can avoid ABA races * caused by long preemption sleeps. */ success = true; critical_enter(); /* Attempt to load from self only once. */ self = zpcpu_get(smr); s = self->c_shared; flags = self->c_flags; counter_u64_add_protected(poll, 1); /* * Conditionally advance the lazy write clock on any writer * activity. */ if ((flags & SMR_LAZY) != 0) smr_lazy_advance(smr, s); /* * Acquire barrier loads s_wr_seq after s_rd_seq so that we can not * observe an updated read sequence that is larger than write. */ s_rd_seq = atomic_load_acq_int(&s->s_rd_seq); /* * If we have already observed the sequence number we can immediately * return success. Most polls should meet this criterion. */ if (SMR_SEQ_LEQ(goal, s_rd_seq)) goto out; /* * wr_seq must be loaded prior to any c_seq value so that a * stale c_seq can only reference time after this wr_seq. */ s_wr_seq = atomic_load_acq_int(&s->s_wr.seq); /* * This is the distance from s_wr_seq to goal. Positive values * are in the future. */ delta = SMR_SEQ_DELTA(goal, s_wr_seq); /* * Detect a stale wr_seq. * * This goal may have come from a deferred advance or a lazy * smr. If we are not blocking we can not succeed but the * sequence number is valid. */ if (delta > 0 && delta <= SMR_SEQ_ADVANCE && (flags & (SMR_LAZY | SMR_DEFERRED)) != 0) { if (!wait) { success = false; goto out; } /* LAZY is always !wait. */ s_wr_seq = smr_shared_advance(s); delta = 0; } /* * Detect an invalid goal. * * The goal must be in the range of s_wr_seq >= goal >= s_rd_seq for * it to be valid. If it is not then the caller held on to it and * the integer wrapped. If we wrapped back within range the caller * will harmlessly scan. */ if (delta > 0) goto out; /* Determine the lowest visible sequence number. */ s_rd_seq = smr_poll_scan(smr, s, s_rd_seq, s_wr_seq, goal, wait); success = SMR_SEQ_LEQ(goal, s_rd_seq); out: if (!success) counter_u64_add_protected(poll_fail, 1); critical_exit(); /* * Serialize with smr_advance()/smr_exit(). The caller is now free * to modify memory as expected. */ atomic_thread_fence_acq(); return (success); } smr_t smr_create(const char *name, int limit, int flags) { smr_t smr, c; smr_shared_t s; int i; s = uma_zalloc(smr_shared_zone, M_WAITOK); smr = uma_zalloc_pcpu(smr_zone, M_WAITOK); s->s_name = name; s->s_rd_seq = s->s_wr.seq = SMR_SEQ_INIT; s->s_wr.ticks = ticks; /* Initialize all CPUS, not just those running. */ for (i = 0; i <= mp_maxid; i++) { c = zpcpu_get_cpu(smr, i); c->c_seq = SMR_SEQ_INVALID; c->c_shared = s; c->c_deferred = 0; c->c_limit = limit; c->c_flags = flags; } atomic_thread_fence_seq_cst(); return (smr); } void smr_destroy(smr_t smr) { smr_synchronize(smr); uma_zfree(smr_shared_zone, smr->c_shared); uma_zfree_pcpu(smr_zone, smr); } /* * Initialize the UMA slab zone. */ void smr_init(void) { smr_shared_zone = uma_zcreate("SMR SHARED", sizeof(struct smr_shared), NULL, NULL, NULL, NULL, (CACHE_LINE_SIZE * 2) - 1, 0); smr_zone = uma_zcreate("SMR CPU", sizeof(struct smr), NULL, NULL, NULL, NULL, (CACHE_LINE_SIZE * 2) - 1, UMA_ZONE_PCPU); } - -static void -smr_init_counters(void *unused) -{ - - advance = counter_u64_alloc(M_WAITOK); - advance_wait = counter_u64_alloc(M_WAITOK); - poll = counter_u64_alloc(M_WAITOK); - poll_scan = counter_u64_alloc(M_WAITOK); - poll_fail = counter_u64_alloc(M_WAITOK); -} -SYSINIT(smr_counters, SI_SUB_CPU, SI_ORDER_ANY, smr_init_counters, NULL); diff --git a/sys/kern/vfs_cache.c b/sys/kern/vfs_cache.c index f581dedcde7a..da15237ab3c6 100644 --- a/sys/kern/vfs_cache.c +++ b/sys/kern/vfs_cache.c @@ -1,2742 +1,2723 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1989, 1993, 1995 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed to Berkeley by * Poul-Henning Kamp of the FreeBSD Project. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)vfs_cache.c 8.5 (Berkeley) 3/22/95 */ #include __FBSDID("$FreeBSD$"); #include "opt_ddb.h" #include "opt_ktrace.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef KTRACE #include #endif #ifdef DDB #include #endif #include SDT_PROVIDER_DECLARE(vfs); SDT_PROBE_DEFINE3(vfs, namecache, enter, done, "struct vnode *", "char *", "struct vnode *"); SDT_PROBE_DEFINE2(vfs, namecache, enter_negative, done, "struct vnode *", "char *"); SDT_PROBE_DEFINE1(vfs, namecache, fullpath, entry, "struct vnode *"); SDT_PROBE_DEFINE3(vfs, namecache, fullpath, hit, "struct vnode *", "char *", "struct vnode *"); SDT_PROBE_DEFINE1(vfs, namecache, fullpath, miss, "struct vnode *"); SDT_PROBE_DEFINE3(vfs, namecache, fullpath, return, "int", "struct vnode *", "char *"); SDT_PROBE_DEFINE3(vfs, namecache, lookup, hit, "struct vnode *", "char *", "struct vnode *"); SDT_PROBE_DEFINE2(vfs, namecache, lookup, hit__negative, "struct vnode *", "char *"); SDT_PROBE_DEFINE2(vfs, namecache, lookup, miss, "struct vnode *", "char *"); SDT_PROBE_DEFINE1(vfs, namecache, purge, done, "struct vnode *"); SDT_PROBE_DEFINE1(vfs, namecache, purge_negative, done, "struct vnode *"); SDT_PROBE_DEFINE1(vfs, namecache, purgevfs, done, "struct mount *"); SDT_PROBE_DEFINE3(vfs, namecache, zap, done, "struct vnode *", "char *", "struct vnode *"); SDT_PROBE_DEFINE2(vfs, namecache, zap_negative, done, "struct vnode *", "char *"); SDT_PROBE_DEFINE2(vfs, namecache, shrink_negative, done, "struct vnode *", "char *"); /* * This structure describes the elements in the cache of recent * names looked up by namei. */ struct namecache { LIST_ENTRY(namecache) nc_hash; /* hash chain */ LIST_ENTRY(namecache) nc_src; /* source vnode list */ TAILQ_ENTRY(namecache) nc_dst; /* destination vnode list */ struct vnode *nc_dvp; /* vnode of parent of name */ union { struct vnode *nu_vp; /* vnode the name refers to */ } n_un; u_char nc_flag; /* flag bits */ u_char nc_nlen; /* length of name */ char nc_name[0]; /* segment name + nul */ }; /* * struct namecache_ts repeats struct namecache layout up to the * nc_nlen member. * struct namecache_ts is used in place of struct namecache when time(s) need * to be stored. The nc_dotdottime field is used when a cache entry is mapping * both a non-dotdot directory name plus dotdot for the directory's * parent. */ struct namecache_ts { struct timespec nc_time; /* timespec provided by fs */ struct timespec nc_dotdottime; /* dotdot timespec provided by fs */ int nc_ticks; /* ticks value when entry was added */ struct namecache nc_nc; }; #define nc_vp n_un.nu_vp /* * Flags in namecache.nc_flag */ #define NCF_WHITE 0x01 #define NCF_ISDOTDOT 0x02 #define NCF_TS 0x04 #define NCF_DTS 0x08 #define NCF_DVDROP 0x10 #define NCF_NEGATIVE 0x20 #define NCF_HOTNEGATIVE 0x40 /* * Name caching works as follows: * * Names found by directory scans are retained in a cache * for future reference. It is managed LRU, so frequently * used names will hang around. Cache is indexed by hash value * obtained from (dvp, name) where dvp refers to the directory * containing name. * * If it is a "negative" entry, (i.e. for a name that is known NOT to * exist) the vnode pointer will be NULL. * * Upon reaching the last segment of a path, if the reference * is for DELETE, or NOCACHE is set (rewrite), and the * name is located in the cache, it will be dropped. * * These locks are used (in the order in which they can be taken): * NAME TYPE ROLE * vnodelock mtx vnode lists and v_cache_dd field protection * bucketlock rwlock for access to given set of hash buckets * neglist mtx negative entry LRU management * * Additionally, ncneg_shrink_lock mtx is used to have at most one thread * shrinking the LRU list. * * It is legal to take multiple vnodelock and bucketlock locks. The locking * order is lower address first. Both are recursive. * * "." lookups are lockless. * * ".." and vnode -> name lookups require vnodelock. * * name -> vnode lookup requires the relevant bucketlock to be held for reading. * * Insertions and removals of entries require involved vnodes and bucketlocks * to be write-locked to prevent other threads from seeing the entry. * * Some lookups result in removal of the found entry (e.g. getting rid of a * negative entry with the intent to create a positive one), which poses a * problem when multiple threads reach the state. Similarly, two different * threads can purge two different vnodes and try to remove the same name. * * If the already held vnode lock is lower than the second required lock, we * can just take the other lock. However, in the opposite case, this could * deadlock. As such, this is resolved by trylocking and if that fails unlocking * the first node, locking everything in order and revalidating the state. */ /* * Structures associated with name caching. */ #define NCHHASH(hash) \ (&nchashtbl[(hash) & nchash]) static __read_mostly LIST_HEAD(nchashhead, namecache) *nchashtbl;/* Hash Table */ static u_long __read_mostly nchash; /* size of hash table */ SYSCTL_ULONG(_debug, OID_AUTO, nchash, CTLFLAG_RD, &nchash, 0, "Size of namecache hash table"); static u_long __read_mostly ncnegfactor = 5; /* ratio of negative entries */ SYSCTL_ULONG(_vfs, OID_AUTO, ncnegfactor, CTLFLAG_RW, &ncnegfactor, 0, "Ratio of negative namecache entries"); static u_long __exclusive_cache_line numneg; /* number of negative entries allocated */ static u_long __exclusive_cache_line numcache;/* number of cache entries allocated */ u_int ncsizefactor = 2; SYSCTL_UINT(_vfs, OID_AUTO, ncsizefactor, CTLFLAG_RW, &ncsizefactor, 0, "Size factor for namecache"); static u_int __read_mostly ncpurgeminvnodes; SYSCTL_UINT(_vfs, OID_AUTO, ncpurgeminvnodes, CTLFLAG_RW, &ncpurgeminvnodes, 0, "Number of vnodes below which purgevfs ignores the request"); static u_int __read_mostly ncsize; /* the size as computed on creation or resizing */ struct nchstats nchstats; /* cache effectiveness statistics */ static struct mtx __exclusive_cache_line ncneg_shrink_lock; static int shrink_list_turn; struct neglist { struct mtx nl_lock; TAILQ_HEAD(, namecache) nl_list; } __aligned(CACHE_LINE_SIZE); static struct neglist __read_mostly *neglists; static struct neglist ncneg_hot; static u_long numhotneg; #define numneglists (ncneghash + 1) static u_int __read_mostly ncneghash; static inline struct neglist * NCP2NEGLIST(struct namecache *ncp) { return (&neglists[(((uintptr_t)(ncp) >> 8) & ncneghash)]); } #define numbucketlocks (ncbuckethash + 1) static u_int __read_mostly ncbuckethash; static struct rwlock_padalign __read_mostly *bucketlocks; #define HASH2BUCKETLOCK(hash) \ ((struct rwlock *)(&bucketlocks[((hash) & ncbuckethash)])) #define numvnodelocks (ncvnodehash + 1) static u_int __read_mostly ncvnodehash; static struct mtx __read_mostly *vnodelocks; static inline struct mtx * VP2VNODELOCK(struct vnode *vp) { return (&vnodelocks[(((uintptr_t)(vp) >> 8) & ncvnodehash)]); } /* * UMA zones for the VFS cache. * * The small cache is used for entries with short names, which are the * most common. The large cache is used for entries which are too big to * fit in the small cache. */ static uma_zone_t __read_mostly cache_zone_small; static uma_zone_t __read_mostly cache_zone_small_ts; static uma_zone_t __read_mostly cache_zone_large; static uma_zone_t __read_mostly cache_zone_large_ts; #define CACHE_PATH_CUTOFF 35 static struct namecache * cache_alloc(int len, int ts) { struct namecache_ts *ncp_ts; struct namecache *ncp; if (__predict_false(ts)) { if (len <= CACHE_PATH_CUTOFF) ncp_ts = uma_zalloc(cache_zone_small_ts, M_WAITOK); else ncp_ts = uma_zalloc(cache_zone_large_ts, M_WAITOK); ncp = &ncp_ts->nc_nc; } else { if (len <= CACHE_PATH_CUTOFF) ncp = uma_zalloc(cache_zone_small, M_WAITOK); else ncp = uma_zalloc(cache_zone_large, M_WAITOK); } return (ncp); } static void cache_free(struct namecache *ncp) { struct namecache_ts *ncp_ts; if (ncp == NULL) return; if ((ncp->nc_flag & NCF_DVDROP) != 0) vdrop(ncp->nc_dvp); if (__predict_false(ncp->nc_flag & NCF_TS)) { ncp_ts = __containerof(ncp, struct namecache_ts, nc_nc); if (ncp->nc_nlen <= CACHE_PATH_CUTOFF) uma_zfree(cache_zone_small_ts, ncp_ts); else uma_zfree(cache_zone_large_ts, ncp_ts); } else { if (ncp->nc_nlen <= CACHE_PATH_CUTOFF) uma_zfree(cache_zone_small, ncp); else uma_zfree(cache_zone_large, ncp); } } static void cache_out_ts(struct namecache *ncp, struct timespec *tsp, int *ticksp) { struct namecache_ts *ncp_ts; KASSERT((ncp->nc_flag & NCF_TS) != 0 || (tsp == NULL && ticksp == NULL), ("No NCF_TS")); if (tsp == NULL && ticksp == NULL) return; ncp_ts = __containerof(ncp, struct namecache_ts, nc_nc); if (tsp != NULL) *tsp = ncp_ts->nc_time; if (ticksp != NULL) *ticksp = ncp_ts->nc_ticks; } #ifdef DEBUG_CACHE static int __read_mostly doingcache = 1; /* 1 => enable the cache */ SYSCTL_INT(_debug, OID_AUTO, vfscache, CTLFLAG_RW, &doingcache, 0, "VFS namecache enabled"); #endif /* Export size information to userland */ SYSCTL_INT(_debug_sizeof, OID_AUTO, namecache, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, sizeof(struct namecache), "sizeof(struct namecache)"); /* * The new name cache statistics */ static SYSCTL_NODE(_vfs, OID_AUTO, cache, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "Name cache statistics"); -#define STATNODE_ULONG(name, descr) \ +#define STATNODE_ULONG(name, descr) \ SYSCTL_ULONG(_vfs_cache, OID_AUTO, name, CTLFLAG_RD, &name, 0, descr); -#define STATNODE_COUNTER(name, descr) \ - static counter_u64_t __read_mostly name; \ - SYSCTL_COUNTER_U64(_vfs_cache, OID_AUTO, name, CTLFLAG_RD, &name, descr); +#define STATNODE_COUNTER(name, descr) \ + static COUNTER_U64_DEFINE_EARLY(name); \ + SYSCTL_COUNTER_U64(_vfs_cache, OID_AUTO, name, CTLFLAG_RD, &name, \ + descr); STATNODE_ULONG(numneg, "Number of negative cache entries"); STATNODE_ULONG(numcache, "Number of cache entries"); STATNODE_COUNTER(numcachehv, "Number of namecache entries with vnodes held"); STATNODE_COUNTER(numcalls, "Number of cache lookups"); STATNODE_COUNTER(dothits, "Number of '.' hits"); STATNODE_COUNTER(dotdothits, "Number of '..' hits"); STATNODE_COUNTER(numchecks, "Number of checks in lookup"); STATNODE_COUNTER(nummiss, "Number of cache misses"); STATNODE_COUNTER(nummisszap, "Number of cache misses we do not want to cache"); STATNODE_COUNTER(numposzaps, "Number of cache hits (positive) we do not want to cache"); STATNODE_COUNTER(numposhits, "Number of cache hits (positive)"); STATNODE_COUNTER(numnegzaps, "Number of cache hits (negative) we do not want to cache"); STATNODE_COUNTER(numneghits, "Number of cache hits (negative)"); /* These count for vn_getcwd(), too. */ STATNODE_COUNTER(numfullpathcalls, "Number of fullpath search calls"); STATNODE_COUNTER(numfullpathfail1, "Number of fullpath search errors (ENOTDIR)"); STATNODE_COUNTER(numfullpathfail2, "Number of fullpath search errors (VOP_VPTOCNP failures)"); STATNODE_COUNTER(numfullpathfail4, "Number of fullpath search errors (ENOMEM)"); STATNODE_COUNTER(numfullpathfound, "Number of successful fullpath calls"); STATNODE_COUNTER(zap_and_exit_bucket_relock_success, "Number of successful removals after relocking"); static long zap_and_exit_bucket_fail; STATNODE_ULONG(zap_and_exit_bucket_fail, "Number of times zap_and_exit failed to lock"); static long zap_and_exit_bucket_fail2; STATNODE_ULONG(zap_and_exit_bucket_fail2, "Number of times zap_and_exit failed to lock"); static long cache_lock_vnodes_cel_3_failures; STATNODE_ULONG(cache_lock_vnodes_cel_3_failures, "Number of times 3-way vnode locking failed"); STATNODE_ULONG(numhotneg, "Number of hot negative entries"); STATNODE_COUNTER(numneg_evicted, "Number of negative entries evicted when adding a new entry"); STATNODE_COUNTER(shrinking_skipped, "Number of times shrinking was already in progress"); static void cache_zap_locked(struct namecache *ncp, bool neg_locked); static int vn_fullpath_hardlink(struct thread *td, struct nameidata *ndp, char **retbuf, char **freebuf, size_t *buflen); static int vn_fullpath_any(struct thread *td, struct vnode *vp, struct vnode *rdir, char *buf, char **retbuf, size_t *buflen); static int vn_fullpath_dir(struct thread *td, struct vnode *vp, struct vnode *rdir, char *buf, char **retbuf, size_t *len, bool slash_prefixed, size_t addend); static MALLOC_DEFINE(M_VFSCACHE, "vfscache", "VFS name cache entries"); static int cache_yield; SYSCTL_INT(_vfs_cache, OID_AUTO, yield, CTLFLAG_RD, &cache_yield, 0, "Number of times cache called yield"); static void __noinline cache_maybe_yield(void) { if (should_yield()) { cache_yield++; kern_yield(PRI_USER); } } static inline void cache_assert_vlp_locked(struct mtx *vlp) { if (vlp != NULL) mtx_assert(vlp, MA_OWNED); } static inline void cache_assert_vnode_locked(struct vnode *vp) { struct mtx *vlp; vlp = VP2VNODELOCK(vp); cache_assert_vlp_locked(vlp); } static uint32_t cache_get_hash(char *name, u_char len, struct vnode *dvp) { uint32_t hash; hash = fnv_32_buf(name, len, FNV1_32_INIT); hash = fnv_32_buf(&dvp, sizeof(dvp), hash); return (hash); } static inline struct rwlock * NCP2BUCKETLOCK(struct namecache *ncp) { uint32_t hash; hash = cache_get_hash(ncp->nc_name, ncp->nc_nlen, ncp->nc_dvp); return (HASH2BUCKETLOCK(hash)); } #ifdef INVARIANTS static void cache_assert_bucket_locked(struct namecache *ncp, int mode) { struct rwlock *blp; blp = NCP2BUCKETLOCK(ncp); rw_assert(blp, mode); } #else #define cache_assert_bucket_locked(x, y) do { } while (0) #endif #define cache_sort_vnodes(x, y) _cache_sort_vnodes((void **)(x), (void **)(y)) static void _cache_sort_vnodes(void **p1, void **p2) { void *tmp; MPASS(*p1 != NULL || *p2 != NULL); if (*p1 > *p2) { tmp = *p2; *p2 = *p1; *p1 = tmp; } } static void cache_lock_all_buckets(void) { u_int i; for (i = 0; i < numbucketlocks; i++) rw_wlock(&bucketlocks[i]); } static void cache_unlock_all_buckets(void) { u_int i; for (i = 0; i < numbucketlocks; i++) rw_wunlock(&bucketlocks[i]); } static void cache_lock_all_vnodes(void) { u_int i; for (i = 0; i < numvnodelocks; i++) mtx_lock(&vnodelocks[i]); } static void cache_unlock_all_vnodes(void) { u_int i; for (i = 0; i < numvnodelocks; i++) mtx_unlock(&vnodelocks[i]); } static int cache_trylock_vnodes(struct mtx *vlp1, struct mtx *vlp2) { cache_sort_vnodes(&vlp1, &vlp2); if (vlp1 != NULL) { if (!mtx_trylock(vlp1)) return (EAGAIN); } if (!mtx_trylock(vlp2)) { if (vlp1 != NULL) mtx_unlock(vlp1); return (EAGAIN); } return (0); } static void cache_lock_vnodes(struct mtx *vlp1, struct mtx *vlp2) { MPASS(vlp1 != NULL || vlp2 != NULL); MPASS(vlp1 <= vlp2); if (vlp1 != NULL) mtx_lock(vlp1); if (vlp2 != NULL) mtx_lock(vlp2); } static void cache_unlock_vnodes(struct mtx *vlp1, struct mtx *vlp2) { MPASS(vlp1 != NULL || vlp2 != NULL); if (vlp1 != NULL) mtx_unlock(vlp1); if (vlp2 != NULL) mtx_unlock(vlp2); } static int sysctl_nchstats(SYSCTL_HANDLER_ARGS) { struct nchstats snap; if (req->oldptr == NULL) return (SYSCTL_OUT(req, 0, sizeof(snap))); snap = nchstats; snap.ncs_goodhits = counter_u64_fetch(numposhits); snap.ncs_neghits = counter_u64_fetch(numneghits); snap.ncs_badhits = counter_u64_fetch(numposzaps) + counter_u64_fetch(numnegzaps); snap.ncs_miss = counter_u64_fetch(nummisszap) + counter_u64_fetch(nummiss); return (SYSCTL_OUT(req, &snap, sizeof(snap))); } SYSCTL_PROC(_vfs_cache, OID_AUTO, nchstats, CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, 0, 0, sysctl_nchstats, "LU", "VFS cache effectiveness statistics"); #ifdef DIAGNOSTIC /* * Grab an atomic snapshot of the name cache hash chain lengths */ static SYSCTL_NODE(_debug, OID_AUTO, hashstat, CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, "hash table stats"); static int sysctl_debug_hashstat_rawnchash(SYSCTL_HANDLER_ARGS) { struct nchashhead *ncpp; struct namecache *ncp; int i, error, n_nchash, *cntbuf; retry: n_nchash = nchash + 1; /* nchash is max index, not count */ if (req->oldptr == NULL) return SYSCTL_OUT(req, 0, n_nchash * sizeof(int)); cntbuf = malloc(n_nchash * sizeof(int), M_TEMP, M_ZERO | M_WAITOK); cache_lock_all_buckets(); if (n_nchash != nchash + 1) { cache_unlock_all_buckets(); free(cntbuf, M_TEMP); goto retry; } /* Scan hash tables counting entries */ for (ncpp = nchashtbl, i = 0; i < n_nchash; ncpp++, i++) LIST_FOREACH(ncp, ncpp, nc_hash) cntbuf[i]++; cache_unlock_all_buckets(); for (error = 0, i = 0; i < n_nchash; i++) if ((error = SYSCTL_OUT(req, &cntbuf[i], sizeof(int))) != 0) break; free(cntbuf, M_TEMP); return (error); } SYSCTL_PROC(_debug_hashstat, OID_AUTO, rawnchash, CTLTYPE_INT|CTLFLAG_RD| CTLFLAG_MPSAFE, 0, 0, sysctl_debug_hashstat_rawnchash, "S,int", "nchash chain lengths"); static int sysctl_debug_hashstat_nchash(SYSCTL_HANDLER_ARGS) { int error; struct nchashhead *ncpp; struct namecache *ncp; int n_nchash; int count, maxlength, used, pct; if (!req->oldptr) return SYSCTL_OUT(req, 0, 4 * sizeof(int)); cache_lock_all_buckets(); n_nchash = nchash + 1; /* nchash is max index, not count */ used = 0; maxlength = 0; /* Scan hash tables for applicable entries */ for (ncpp = nchashtbl; n_nchash > 0; n_nchash--, ncpp++) { count = 0; LIST_FOREACH(ncp, ncpp, nc_hash) { count++; } if (count) used++; if (maxlength < count) maxlength = count; } n_nchash = nchash + 1; cache_unlock_all_buckets(); pct = (used * 100) / (n_nchash / 100); error = SYSCTL_OUT(req, &n_nchash, sizeof(n_nchash)); if (error) return (error); error = SYSCTL_OUT(req, &used, sizeof(used)); if (error) return (error); error = SYSCTL_OUT(req, &maxlength, sizeof(maxlength)); if (error) return (error); error = SYSCTL_OUT(req, &pct, sizeof(pct)); if (error) return (error); return (0); } SYSCTL_PROC(_debug_hashstat, OID_AUTO, nchash, CTLTYPE_INT|CTLFLAG_RD| CTLFLAG_MPSAFE, 0, 0, sysctl_debug_hashstat_nchash, "I", "nchash statistics (number of total/used buckets, maximum chain length, usage percentage)"); #endif /* * Negative entries management * * A variation of LRU scheme is used. New entries are hashed into one of * numneglists cold lists. Entries get promoted to the hot list on first hit. * * The shrinker will demote hot list head and evict from the cold list in a * round-robin manner. */ static void cache_negative_hit(struct namecache *ncp) { struct neglist *neglist; MPASS(ncp->nc_flag & NCF_NEGATIVE); if (ncp->nc_flag & NCF_HOTNEGATIVE) return; neglist = NCP2NEGLIST(ncp); mtx_lock(&ncneg_hot.nl_lock); mtx_lock(&neglist->nl_lock); if (!(ncp->nc_flag & NCF_HOTNEGATIVE)) { numhotneg++; TAILQ_REMOVE(&neglist->nl_list, ncp, nc_dst); TAILQ_INSERT_TAIL(&ncneg_hot.nl_list, ncp, nc_dst); ncp->nc_flag |= NCF_HOTNEGATIVE; } mtx_unlock(&neglist->nl_lock); mtx_unlock(&ncneg_hot.nl_lock); } static void cache_negative_insert(struct namecache *ncp, bool neg_locked) { struct neglist *neglist; MPASS(ncp->nc_flag & NCF_NEGATIVE); cache_assert_bucket_locked(ncp, RA_WLOCKED); neglist = NCP2NEGLIST(ncp); if (!neg_locked) { mtx_lock(&neglist->nl_lock); } else { mtx_assert(&neglist->nl_lock, MA_OWNED); } TAILQ_INSERT_TAIL(&neglist->nl_list, ncp, nc_dst); if (!neg_locked) mtx_unlock(&neglist->nl_lock); atomic_add_rel_long(&numneg, 1); } static void cache_negative_remove(struct namecache *ncp, bool neg_locked) { struct neglist *neglist; bool hot_locked = false; bool list_locked = false; MPASS(ncp->nc_flag & NCF_NEGATIVE); cache_assert_bucket_locked(ncp, RA_WLOCKED); neglist = NCP2NEGLIST(ncp); if (!neg_locked) { if (ncp->nc_flag & NCF_HOTNEGATIVE) { hot_locked = true; mtx_lock(&ncneg_hot.nl_lock); if (!(ncp->nc_flag & NCF_HOTNEGATIVE)) { list_locked = true; mtx_lock(&neglist->nl_lock); } } else { list_locked = true; mtx_lock(&neglist->nl_lock); } } if (ncp->nc_flag & NCF_HOTNEGATIVE) { mtx_assert(&ncneg_hot.nl_lock, MA_OWNED); TAILQ_REMOVE(&ncneg_hot.nl_list, ncp, nc_dst); numhotneg--; } else { mtx_assert(&neglist->nl_lock, MA_OWNED); TAILQ_REMOVE(&neglist->nl_list, ncp, nc_dst); } if (list_locked) mtx_unlock(&neglist->nl_lock); if (hot_locked) mtx_unlock(&ncneg_hot.nl_lock); atomic_subtract_rel_long(&numneg, 1); } static void cache_negative_shrink_select(int start, struct namecache **ncpp, struct neglist **neglistpp) { struct neglist *neglist; struct namecache *ncp; int i; *ncpp = ncp = NULL; neglist = NULL; for (i = start; i < numneglists; i++) { neglist = &neglists[i]; if (TAILQ_FIRST(&neglist->nl_list) == NULL) continue; mtx_lock(&neglist->nl_lock); ncp = TAILQ_FIRST(&neglist->nl_list); if (ncp != NULL) break; mtx_unlock(&neglist->nl_lock); } *neglistpp = neglist; *ncpp = ncp; } static void cache_negative_zap_one(void) { struct namecache *ncp, *ncp2; struct neglist *neglist; struct mtx *dvlp; struct rwlock *blp; if (mtx_owner(&ncneg_shrink_lock) != NULL || !mtx_trylock(&ncneg_shrink_lock)) { counter_u64_add(shrinking_skipped, 1); return; } mtx_lock(&ncneg_hot.nl_lock); ncp = TAILQ_FIRST(&ncneg_hot.nl_list); if (ncp != NULL) { neglist = NCP2NEGLIST(ncp); mtx_lock(&neglist->nl_lock); TAILQ_REMOVE(&ncneg_hot.nl_list, ncp, nc_dst); TAILQ_INSERT_TAIL(&neglist->nl_list, ncp, nc_dst); ncp->nc_flag &= ~NCF_HOTNEGATIVE; numhotneg--; mtx_unlock(&neglist->nl_lock); } mtx_unlock(&ncneg_hot.nl_lock); cache_negative_shrink_select(shrink_list_turn, &ncp, &neglist); shrink_list_turn++; if (shrink_list_turn == numneglists) shrink_list_turn = 0; if (ncp == NULL && shrink_list_turn == 0) cache_negative_shrink_select(shrink_list_turn, &ncp, &neglist); mtx_unlock(&ncneg_shrink_lock); if (ncp == NULL) return; MPASS(ncp->nc_flag & NCF_NEGATIVE); dvlp = VP2VNODELOCK(ncp->nc_dvp); blp = NCP2BUCKETLOCK(ncp); mtx_unlock(&neglist->nl_lock); mtx_lock(dvlp); rw_wlock(blp); mtx_lock(&neglist->nl_lock); ncp2 = TAILQ_FIRST(&neglist->nl_list); if (ncp != ncp2 || dvlp != VP2VNODELOCK(ncp2->nc_dvp) || blp != NCP2BUCKETLOCK(ncp2) || !(ncp2->nc_flag & NCF_NEGATIVE)) { ncp = NULL; } else { SDT_PROBE2(vfs, namecache, shrink_negative, done, ncp->nc_dvp, ncp->nc_name); cache_zap_locked(ncp, true); counter_u64_add(numneg_evicted, 1); } mtx_unlock(&neglist->nl_lock); rw_wunlock(blp); mtx_unlock(dvlp); cache_free(ncp); } /* * cache_zap_locked(): * * Removes a namecache entry from cache, whether it contains an actual * pointer to a vnode or if it is just a negative cache entry. */ static void cache_zap_locked(struct namecache *ncp, bool neg_locked) { if (!(ncp->nc_flag & NCF_NEGATIVE)) cache_assert_vnode_locked(ncp->nc_vp); cache_assert_vnode_locked(ncp->nc_dvp); cache_assert_bucket_locked(ncp, RA_WLOCKED); CTR2(KTR_VFS, "cache_zap(%p) vp %p", ncp, (ncp->nc_flag & NCF_NEGATIVE) ? NULL : ncp->nc_vp); LIST_REMOVE(ncp, nc_hash); if (!(ncp->nc_flag & NCF_NEGATIVE)) { SDT_PROBE3(vfs, namecache, zap, done, ncp->nc_dvp, ncp->nc_name, ncp->nc_vp); TAILQ_REMOVE(&ncp->nc_vp->v_cache_dst, ncp, nc_dst); if (ncp == ncp->nc_vp->v_cache_dd) ncp->nc_vp->v_cache_dd = NULL; } else { SDT_PROBE2(vfs, namecache, zap_negative, done, ncp->nc_dvp, ncp->nc_name); cache_negative_remove(ncp, neg_locked); } if (ncp->nc_flag & NCF_ISDOTDOT) { if (ncp == ncp->nc_dvp->v_cache_dd) ncp->nc_dvp->v_cache_dd = NULL; } else { LIST_REMOVE(ncp, nc_src); if (LIST_EMPTY(&ncp->nc_dvp->v_cache_src)) { ncp->nc_flag |= NCF_DVDROP; counter_u64_add(numcachehv, -1); } } atomic_subtract_rel_long(&numcache, 1); } static void cache_zap_negative_locked_vnode_kl(struct namecache *ncp, struct vnode *vp) { struct rwlock *blp; MPASS(ncp->nc_dvp == vp); MPASS(ncp->nc_flag & NCF_NEGATIVE); cache_assert_vnode_locked(vp); blp = NCP2BUCKETLOCK(ncp); rw_wlock(blp); cache_zap_locked(ncp, false); rw_wunlock(blp); } static bool cache_zap_locked_vnode_kl2(struct namecache *ncp, struct vnode *vp, struct mtx **vlpp) { struct mtx *pvlp, *vlp1, *vlp2, *to_unlock; struct rwlock *blp; MPASS(vp == ncp->nc_dvp || vp == ncp->nc_vp); cache_assert_vnode_locked(vp); if (ncp->nc_flag & NCF_NEGATIVE) { if (*vlpp != NULL) { mtx_unlock(*vlpp); *vlpp = NULL; } cache_zap_negative_locked_vnode_kl(ncp, vp); return (true); } pvlp = VP2VNODELOCK(vp); blp = NCP2BUCKETLOCK(ncp); vlp1 = VP2VNODELOCK(ncp->nc_dvp); vlp2 = VP2VNODELOCK(ncp->nc_vp); if (*vlpp == vlp1 || *vlpp == vlp2) { to_unlock = *vlpp; *vlpp = NULL; } else { if (*vlpp != NULL) { mtx_unlock(*vlpp); *vlpp = NULL; } cache_sort_vnodes(&vlp1, &vlp2); if (vlp1 == pvlp) { mtx_lock(vlp2); to_unlock = vlp2; } else { if (!mtx_trylock(vlp1)) goto out_relock; to_unlock = vlp1; } } rw_wlock(blp); cache_zap_locked(ncp, false); rw_wunlock(blp); if (to_unlock != NULL) mtx_unlock(to_unlock); return (true); out_relock: mtx_unlock(vlp2); mtx_lock(vlp1); mtx_lock(vlp2); MPASS(*vlpp == NULL); *vlpp = vlp1; return (false); } static int __noinline cache_zap_locked_vnode(struct namecache *ncp, struct vnode *vp) { struct mtx *pvlp, *vlp1, *vlp2, *to_unlock; struct rwlock *blp; int error = 0; MPASS(vp == ncp->nc_dvp || vp == ncp->nc_vp); cache_assert_vnode_locked(vp); pvlp = VP2VNODELOCK(vp); if (ncp->nc_flag & NCF_NEGATIVE) { cache_zap_negative_locked_vnode_kl(ncp, vp); goto out; } blp = NCP2BUCKETLOCK(ncp); vlp1 = VP2VNODELOCK(ncp->nc_dvp); vlp2 = VP2VNODELOCK(ncp->nc_vp); cache_sort_vnodes(&vlp1, &vlp2); if (vlp1 == pvlp) { mtx_lock(vlp2); to_unlock = vlp2; } else { if (!mtx_trylock(vlp1)) { error = EAGAIN; goto out; } to_unlock = vlp1; } rw_wlock(blp); cache_zap_locked(ncp, false); rw_wunlock(blp); mtx_unlock(to_unlock); out: mtx_unlock(pvlp); return (error); } /* * If trylocking failed we can get here. We know enough to take all needed locks * in the right order and re-lookup the entry. */ static int cache_zap_unlocked_bucket(struct namecache *ncp, struct componentname *cnp, struct vnode *dvp, struct mtx *dvlp, struct mtx *vlp, uint32_t hash, struct rwlock *blp) { struct namecache *rncp; cache_assert_bucket_locked(ncp, RA_UNLOCKED); cache_sort_vnodes(&dvlp, &vlp); cache_lock_vnodes(dvlp, vlp); rw_wlock(blp); LIST_FOREACH(rncp, (NCHHASH(hash)), nc_hash) { if (rncp == ncp && rncp->nc_dvp == dvp && rncp->nc_nlen == cnp->cn_namelen && !bcmp(rncp->nc_name, cnp->cn_nameptr, rncp->nc_nlen)) break; } if (rncp != NULL) { cache_zap_locked(rncp, false); rw_wunlock(blp); cache_unlock_vnodes(dvlp, vlp); counter_u64_add(zap_and_exit_bucket_relock_success, 1); return (0); } rw_wunlock(blp); cache_unlock_vnodes(dvlp, vlp); return (EAGAIN); } static int __noinline cache_zap_wlocked_bucket(struct namecache *ncp, struct componentname *cnp, uint32_t hash, struct rwlock *blp) { struct mtx *dvlp, *vlp; struct vnode *dvp; cache_assert_bucket_locked(ncp, RA_WLOCKED); dvlp = VP2VNODELOCK(ncp->nc_dvp); vlp = NULL; if (!(ncp->nc_flag & NCF_NEGATIVE)) vlp = VP2VNODELOCK(ncp->nc_vp); if (cache_trylock_vnodes(dvlp, vlp) == 0) { cache_zap_locked(ncp, false); rw_wunlock(blp); cache_unlock_vnodes(dvlp, vlp); return (0); } dvp = ncp->nc_dvp; rw_wunlock(blp); return (cache_zap_unlocked_bucket(ncp, cnp, dvp, dvlp, vlp, hash, blp)); } static int __noinline cache_zap_rlocked_bucket(struct namecache *ncp, struct componentname *cnp, uint32_t hash, struct rwlock *blp) { struct mtx *dvlp, *vlp; struct vnode *dvp; cache_assert_bucket_locked(ncp, RA_RLOCKED); dvlp = VP2VNODELOCK(ncp->nc_dvp); vlp = NULL; if (!(ncp->nc_flag & NCF_NEGATIVE)) vlp = VP2VNODELOCK(ncp->nc_vp); if (cache_trylock_vnodes(dvlp, vlp) == 0) { rw_runlock(blp); rw_wlock(blp); cache_zap_locked(ncp, false); rw_wunlock(blp); cache_unlock_vnodes(dvlp, vlp); return (0); } dvp = ncp->nc_dvp; rw_runlock(blp); return (cache_zap_unlocked_bucket(ncp, cnp, dvp, dvlp, vlp, hash, blp)); } static int cache_zap_wlocked_bucket_kl(struct namecache *ncp, struct rwlock *blp, struct mtx **vlpp1, struct mtx **vlpp2) { struct mtx *dvlp, *vlp; cache_assert_bucket_locked(ncp, RA_WLOCKED); dvlp = VP2VNODELOCK(ncp->nc_dvp); vlp = NULL; if (!(ncp->nc_flag & NCF_NEGATIVE)) vlp = VP2VNODELOCK(ncp->nc_vp); cache_sort_vnodes(&dvlp, &vlp); if (*vlpp1 == dvlp && *vlpp2 == vlp) { cache_zap_locked(ncp, false); cache_unlock_vnodes(dvlp, vlp); *vlpp1 = NULL; *vlpp2 = NULL; return (0); } if (*vlpp1 != NULL) mtx_unlock(*vlpp1); if (*vlpp2 != NULL) mtx_unlock(*vlpp2); *vlpp1 = NULL; *vlpp2 = NULL; if (cache_trylock_vnodes(dvlp, vlp) == 0) { cache_zap_locked(ncp, false); cache_unlock_vnodes(dvlp, vlp); return (0); } rw_wunlock(blp); *vlpp1 = dvlp; *vlpp2 = vlp; if (*vlpp1 != NULL) mtx_lock(*vlpp1); mtx_lock(*vlpp2); rw_wlock(blp); return (EAGAIN); } static void cache_lookup_unlock(struct rwlock *blp, struct mtx *vlp) { if (blp != NULL) { rw_runlock(blp); } else { mtx_unlock(vlp); } } static int __noinline cache_lookup_dot(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, struct timespec *tsp, int *ticksp) { int ltype; *vpp = dvp; CTR2(KTR_VFS, "cache_lookup(%p, %s) found via .", dvp, cnp->cn_nameptr); counter_u64_add(dothits, 1); SDT_PROBE3(vfs, namecache, lookup, hit, dvp, ".", *vpp); if (tsp != NULL) timespecclear(tsp); if (ticksp != NULL) *ticksp = ticks; vrefact(*vpp); /* * When we lookup "." we still can be asked to lock it * differently... */ ltype = cnp->cn_lkflags & LK_TYPE_MASK; if (ltype != VOP_ISLOCKED(*vpp)) { if (ltype == LK_EXCLUSIVE) { vn_lock(*vpp, LK_UPGRADE | LK_RETRY); if (VN_IS_DOOMED((*vpp))) { /* forced unmount */ vrele(*vpp); *vpp = NULL; return (ENOENT); } } else vn_lock(*vpp, LK_DOWNGRADE | LK_RETRY); } return (-1); } static __noinline int cache_lookup_nomakeentry(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, struct timespec *tsp, int *ticksp) { struct namecache *ncp; struct rwlock *blp; struct mtx *dvlp, *dvlp2; uint32_t hash; int error; if (cnp->cn_namelen == 2 && cnp->cn_nameptr[0] == '.' && cnp->cn_nameptr[1] == '.') { counter_u64_add(dotdothits, 1); dvlp = VP2VNODELOCK(dvp); dvlp2 = NULL; mtx_lock(dvlp); retry_dotdot: ncp = dvp->v_cache_dd; if (ncp == NULL) { SDT_PROBE3(vfs, namecache, lookup, miss, dvp, "..", NULL); mtx_unlock(dvlp); if (dvlp2 != NULL) mtx_unlock(dvlp2); return (0); } if ((ncp->nc_flag & NCF_ISDOTDOT) != 0) { if (ncp->nc_dvp != dvp) panic("dvp %p v_cache_dd %p\n", dvp, ncp); if (!cache_zap_locked_vnode_kl2(ncp, dvp, &dvlp2)) goto retry_dotdot; MPASS(dvp->v_cache_dd == NULL); mtx_unlock(dvlp); if (dvlp2 != NULL) mtx_unlock(dvlp2); cache_free(ncp); } else { dvp->v_cache_dd = NULL; mtx_unlock(dvlp); if (dvlp2 != NULL) mtx_unlock(dvlp2); } return (0); } hash = cache_get_hash(cnp->cn_nameptr, cnp->cn_namelen, dvp); blp = HASH2BUCKETLOCK(hash); retry: if (LIST_EMPTY(NCHHASH(hash))) goto out_no_entry; rw_wlock(blp); LIST_FOREACH(ncp, (NCHHASH(hash)), nc_hash) { counter_u64_add(numchecks, 1); if (ncp->nc_dvp == dvp && ncp->nc_nlen == cnp->cn_namelen && !bcmp(ncp->nc_name, cnp->cn_nameptr, ncp->nc_nlen)) break; } /* We failed to find an entry */ if (ncp == NULL) { rw_wunlock(blp); goto out_no_entry; } error = cache_zap_wlocked_bucket(ncp, cnp, hash, blp); if (__predict_false(error != 0)) { zap_and_exit_bucket_fail++; cache_maybe_yield(); goto retry; } counter_u64_add(numposzaps, 1); cache_free(ncp); return (0); out_no_entry: SDT_PROBE3(vfs, namecache, lookup, miss, dvp, cnp->cn_nameptr, NULL); counter_u64_add(nummisszap, 1); return (0); } /** * Lookup a name in the name cache * * # Arguments * * - dvp: Parent directory in which to search. * - vpp: Return argument. Will contain desired vnode on cache hit. * - cnp: Parameters of the name search. The most interesting bits of * the cn_flags field have the following meanings: * - MAKEENTRY: If clear, free an entry from the cache rather than look * it up. * - ISDOTDOT: Must be set if and only if cn_nameptr == ".." * - tsp: Return storage for cache timestamp. On a successful (positive * or negative) lookup, tsp will be filled with any timespec that * was stored when this cache entry was created. However, it will * be clear for "." entries. * - ticks: Return storage for alternate cache timestamp. On a successful * (positive or negative) lookup, it will contain the ticks value * that was current when the cache entry was created, unless cnp * was ".". * * # Returns * * - -1: A positive cache hit. vpp will contain the desired vnode. * - ENOENT: A negative cache hit, or dvp was recycled out from under us due * to a forced unmount. vpp will not be modified. If the entry * is a whiteout, then the ISWHITEOUT flag will be set in * cnp->cn_flags. * - 0: A cache miss. vpp will not be modified. * * # Locking * * On a cache hit, vpp will be returned locked and ref'd. If we're looking up * .., dvp is unlocked. If we're looking up . an extra ref is taken, but the * lock is not recursively acquired. */ int cache_lookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, struct timespec *tsp, int *ticksp) { struct namecache_ts *ncp_ts; struct namecache *ncp; struct rwlock *blp; struct mtx *dvlp; uint32_t hash; enum vgetstate vs; int error, ltype; #ifdef DEBUG_CACHE if (__predict_false(!doingcache)) { cnp->cn_flags &= ~MAKEENTRY; return (0); } #endif counter_u64_add(numcalls, 1); if (__predict_false(cnp->cn_namelen == 1 && cnp->cn_nameptr[0] == '.')) return (cache_lookup_dot(dvp, vpp, cnp, tsp, ticksp)); if ((cnp->cn_flags & MAKEENTRY) == 0) return (cache_lookup_nomakeentry(dvp, vpp, cnp, tsp, ticksp)); retry: blp = NULL; dvlp = NULL; error = 0; if (cnp->cn_namelen == 2 && cnp->cn_nameptr[0] == '.' && cnp->cn_nameptr[1] == '.') { counter_u64_add(dotdothits, 1); dvlp = VP2VNODELOCK(dvp); mtx_lock(dvlp); ncp = dvp->v_cache_dd; if (ncp == NULL) { SDT_PROBE3(vfs, namecache, lookup, miss, dvp, "..", NULL); mtx_unlock(dvlp); return (0); } if ((ncp->nc_flag & NCF_ISDOTDOT) != 0) { if (ncp->nc_flag & NCF_NEGATIVE) *vpp = NULL; else *vpp = ncp->nc_vp; } else *vpp = ncp->nc_dvp; /* Return failure if negative entry was found. */ if (*vpp == NULL) goto negative_success; CTR3(KTR_VFS, "cache_lookup(%p, %s) found %p via ..", dvp, cnp->cn_nameptr, *vpp); SDT_PROBE3(vfs, namecache, lookup, hit, dvp, "..", *vpp); cache_out_ts(ncp, tsp, ticksp); if ((ncp->nc_flag & (NCF_ISDOTDOT | NCF_DTS)) == NCF_DTS && tsp != NULL) { ncp_ts = __containerof(ncp, struct namecache_ts, nc_nc); *tsp = ncp_ts->nc_dotdottime; } goto success; } hash = cache_get_hash(cnp->cn_nameptr, cnp->cn_namelen, dvp); blp = HASH2BUCKETLOCK(hash); rw_rlock(blp); LIST_FOREACH(ncp, (NCHHASH(hash)), nc_hash) { counter_u64_add(numchecks, 1); if (ncp->nc_dvp == dvp && ncp->nc_nlen == cnp->cn_namelen && !bcmp(ncp->nc_name, cnp->cn_nameptr, ncp->nc_nlen)) break; } /* We failed to find an entry */ if (__predict_false(ncp == NULL)) { rw_runlock(blp); SDT_PROBE3(vfs, namecache, lookup, miss, dvp, cnp->cn_nameptr, NULL); counter_u64_add(nummiss, 1); return (0); } if (ncp->nc_flag & NCF_NEGATIVE) goto negative_success; /* We found a "positive" match, return the vnode */ counter_u64_add(numposhits, 1); *vpp = ncp->nc_vp; CTR4(KTR_VFS, "cache_lookup(%p, %s) found %p via ncp %p", dvp, cnp->cn_nameptr, *vpp, ncp); SDT_PROBE3(vfs, namecache, lookup, hit, dvp, ncp->nc_name, *vpp); cache_out_ts(ncp, tsp, ticksp); success: /* * On success we return a locked and ref'd vnode as per the lookup * protocol. */ MPASS(dvp != *vpp); ltype = 0; /* silence gcc warning */ if (cnp->cn_flags & ISDOTDOT) { ltype = VOP_ISLOCKED(dvp); VOP_UNLOCK(dvp); } vs = vget_prep(*vpp); cache_lookup_unlock(blp, dvlp); error = vget_finish(*vpp, cnp->cn_lkflags, vs); if (cnp->cn_flags & ISDOTDOT) { vn_lock(dvp, ltype | LK_RETRY); if (VN_IS_DOOMED(dvp)) { if (error == 0) vput(*vpp); *vpp = NULL; return (ENOENT); } } if (error) { *vpp = NULL; goto retry; } if ((cnp->cn_flags & ISLASTCN) && (cnp->cn_lkflags & LK_TYPE_MASK) == LK_EXCLUSIVE) { ASSERT_VOP_ELOCKED(*vpp, "cache_lookup"); } return (-1); negative_success: /* We found a negative match, and want to create it, so purge */ if (cnp->cn_nameiop == CREATE) { counter_u64_add(numnegzaps, 1); goto zap_and_exit; } counter_u64_add(numneghits, 1); cache_negative_hit(ncp); if (ncp->nc_flag & NCF_WHITE) cnp->cn_flags |= ISWHITEOUT; SDT_PROBE2(vfs, namecache, lookup, hit__negative, dvp, ncp->nc_name); cache_out_ts(ncp, tsp, ticksp); cache_lookup_unlock(blp, dvlp); return (ENOENT); zap_and_exit: if (blp != NULL) error = cache_zap_rlocked_bucket(ncp, cnp, hash, blp); else error = cache_zap_locked_vnode(ncp, dvp); if (__predict_false(error != 0)) { zap_and_exit_bucket_fail2++; cache_maybe_yield(); goto retry; } cache_free(ncp); return (0); } struct celockstate { struct mtx *vlp[3]; struct rwlock *blp[2]; }; CTASSERT((nitems(((struct celockstate *)0)->vlp) == 3)); CTASSERT((nitems(((struct celockstate *)0)->blp) == 2)); static inline void cache_celockstate_init(struct celockstate *cel) { bzero(cel, sizeof(*cel)); } static void cache_lock_vnodes_cel(struct celockstate *cel, struct vnode *vp, struct vnode *dvp) { struct mtx *vlp1, *vlp2; MPASS(cel->vlp[0] == NULL); MPASS(cel->vlp[1] == NULL); MPASS(cel->vlp[2] == NULL); MPASS(vp != NULL || dvp != NULL); vlp1 = VP2VNODELOCK(vp); vlp2 = VP2VNODELOCK(dvp); cache_sort_vnodes(&vlp1, &vlp2); if (vlp1 != NULL) { mtx_lock(vlp1); cel->vlp[0] = vlp1; } mtx_lock(vlp2); cel->vlp[1] = vlp2; } static void cache_unlock_vnodes_cel(struct celockstate *cel) { MPASS(cel->vlp[0] != NULL || cel->vlp[1] != NULL); if (cel->vlp[0] != NULL) mtx_unlock(cel->vlp[0]); if (cel->vlp[1] != NULL) mtx_unlock(cel->vlp[1]); if (cel->vlp[2] != NULL) mtx_unlock(cel->vlp[2]); } static bool cache_lock_vnodes_cel_3(struct celockstate *cel, struct vnode *vp) { struct mtx *vlp; bool ret; cache_assert_vlp_locked(cel->vlp[0]); cache_assert_vlp_locked(cel->vlp[1]); MPASS(cel->vlp[2] == NULL); MPASS(vp != NULL); vlp = VP2VNODELOCK(vp); ret = true; if (vlp >= cel->vlp[1]) { mtx_lock(vlp); } else { if (mtx_trylock(vlp)) goto out; cache_lock_vnodes_cel_3_failures++; cache_unlock_vnodes_cel(cel); if (vlp < cel->vlp[0]) { mtx_lock(vlp); mtx_lock(cel->vlp[0]); mtx_lock(cel->vlp[1]); } else { if (cel->vlp[0] != NULL) mtx_lock(cel->vlp[0]); mtx_lock(vlp); mtx_lock(cel->vlp[1]); } ret = false; } out: cel->vlp[2] = vlp; return (ret); } static void cache_lock_buckets_cel(struct celockstate *cel, struct rwlock *blp1, struct rwlock *blp2) { MPASS(cel->blp[0] == NULL); MPASS(cel->blp[1] == NULL); cache_sort_vnodes(&blp1, &blp2); if (blp1 != NULL) { rw_wlock(blp1); cel->blp[0] = blp1; } rw_wlock(blp2); cel->blp[1] = blp2; } static void cache_unlock_buckets_cel(struct celockstate *cel) { if (cel->blp[0] != NULL) rw_wunlock(cel->blp[0]); rw_wunlock(cel->blp[1]); } /* * Lock part of the cache affected by the insertion. * * This means vnodelocks for dvp, vp and the relevant bucketlock. * However, insertion can result in removal of an old entry. In this * case we have an additional vnode and bucketlock pair to lock. If the * entry is negative, ncelock is locked instead of the vnode. * * That is, in the worst case we have to lock 3 vnodes and 2 bucketlocks, while * preserving the locking order (smaller address first). */ static void cache_enter_lock(struct celockstate *cel, struct vnode *dvp, struct vnode *vp, uint32_t hash) { struct namecache *ncp; struct rwlock *blps[2]; blps[0] = HASH2BUCKETLOCK(hash); for (;;) { blps[1] = NULL; cache_lock_vnodes_cel(cel, dvp, vp); if (vp == NULL || vp->v_type != VDIR) break; ncp = vp->v_cache_dd; if (ncp == NULL) break; if ((ncp->nc_flag & NCF_ISDOTDOT) == 0) break; MPASS(ncp->nc_dvp == vp); blps[1] = NCP2BUCKETLOCK(ncp); if (ncp->nc_flag & NCF_NEGATIVE) break; if (cache_lock_vnodes_cel_3(cel, ncp->nc_vp)) break; /* * All vnodes got re-locked. Re-validate the state and if * nothing changed we are done. Otherwise restart. */ if (ncp == vp->v_cache_dd && (ncp->nc_flag & NCF_ISDOTDOT) != 0 && blps[1] == NCP2BUCKETLOCK(ncp) && VP2VNODELOCK(ncp->nc_vp) == cel->vlp[2]) break; cache_unlock_vnodes_cel(cel); cel->vlp[0] = NULL; cel->vlp[1] = NULL; cel->vlp[2] = NULL; } cache_lock_buckets_cel(cel, blps[0], blps[1]); } static void cache_enter_lock_dd(struct celockstate *cel, struct vnode *dvp, struct vnode *vp, uint32_t hash) { struct namecache *ncp; struct rwlock *blps[2]; blps[0] = HASH2BUCKETLOCK(hash); for (;;) { blps[1] = NULL; cache_lock_vnodes_cel(cel, dvp, vp); ncp = dvp->v_cache_dd; if (ncp == NULL) break; if ((ncp->nc_flag & NCF_ISDOTDOT) == 0) break; MPASS(ncp->nc_dvp == dvp); blps[1] = NCP2BUCKETLOCK(ncp); if (ncp->nc_flag & NCF_NEGATIVE) break; if (cache_lock_vnodes_cel_3(cel, ncp->nc_vp)) break; if (ncp == dvp->v_cache_dd && (ncp->nc_flag & NCF_ISDOTDOT) != 0 && blps[1] == NCP2BUCKETLOCK(ncp) && VP2VNODELOCK(ncp->nc_vp) == cel->vlp[2]) break; cache_unlock_vnodes_cel(cel); cel->vlp[0] = NULL; cel->vlp[1] = NULL; cel->vlp[2] = NULL; } cache_lock_buckets_cel(cel, blps[0], blps[1]); } static void cache_enter_unlock(struct celockstate *cel) { cache_unlock_buckets_cel(cel); cache_unlock_vnodes_cel(cel); } static void __noinline cache_enter_dotdot_prep(struct vnode *dvp, struct vnode *vp, struct componentname *cnp) { struct celockstate cel; struct namecache *ncp; uint32_t hash; int len; if (dvp->v_cache_dd == NULL) return; len = cnp->cn_namelen; cache_celockstate_init(&cel); hash = cache_get_hash(cnp->cn_nameptr, len, dvp); cache_enter_lock_dd(&cel, dvp, vp, hash); ncp = dvp->v_cache_dd; if (ncp != NULL && (ncp->nc_flag & NCF_ISDOTDOT)) { KASSERT(ncp->nc_dvp == dvp, ("wrong isdotdot parent")); cache_zap_locked(ncp, false); } else { ncp = NULL; } dvp->v_cache_dd = NULL; cache_enter_unlock(&cel); cache_free(ncp); } /* * Add an entry to the cache. */ void cache_enter_time(struct vnode *dvp, struct vnode *vp, struct componentname *cnp, struct timespec *tsp, struct timespec *dtsp) { struct celockstate cel; struct namecache *ncp, *n2, *ndd; struct namecache_ts *ncp_ts, *n2_ts; struct nchashhead *ncpp; uint32_t hash; int flag; int len; u_long lnumcache; CTR3(KTR_VFS, "cache_enter(%p, %p, %s)", dvp, vp, cnp->cn_nameptr); VNASSERT(vp == NULL || !VN_IS_DOOMED(vp), vp, ("cache_enter: Adding a doomed vnode")); VNASSERT(dvp == NULL || !VN_IS_DOOMED(dvp), dvp, ("cache_enter: Doomed vnode used as src")); #ifdef DEBUG_CACHE if (__predict_false(!doingcache)) return; #endif flag = 0; if (__predict_false(cnp->cn_nameptr[0] == '.')) { if (cnp->cn_namelen == 1) return; if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') { cache_enter_dotdot_prep(dvp, vp, cnp); flag = NCF_ISDOTDOT; } } /* * Avoid blowout in namecache entries. */ lnumcache = atomic_fetchadd_long(&numcache, 1) + 1; if (__predict_false(lnumcache >= ncsize)) { atomic_add_long(&numcache, -1); return; } cache_celockstate_init(&cel); ndd = NULL; ncp_ts = NULL; /* * Calculate the hash key and setup as much of the new * namecache entry as possible before acquiring the lock. */ ncp = cache_alloc(cnp->cn_namelen, tsp != NULL); ncp->nc_flag = flag; ncp->nc_vp = vp; if (vp == NULL) ncp->nc_flag |= NCF_NEGATIVE; ncp->nc_dvp = dvp; if (tsp != NULL) { ncp_ts = __containerof(ncp, struct namecache_ts, nc_nc); ncp_ts->nc_time = *tsp; ncp_ts->nc_ticks = ticks; ncp_ts->nc_nc.nc_flag |= NCF_TS; if (dtsp != NULL) { ncp_ts->nc_dotdottime = *dtsp; ncp_ts->nc_nc.nc_flag |= NCF_DTS; } } len = ncp->nc_nlen = cnp->cn_namelen; hash = cache_get_hash(cnp->cn_nameptr, len, dvp); strlcpy(ncp->nc_name, cnp->cn_nameptr, len + 1); cache_enter_lock(&cel, dvp, vp, hash); /* * See if this vnode or negative entry is already in the cache * with this name. This can happen with concurrent lookups of * the same path name. */ ncpp = NCHHASH(hash); LIST_FOREACH(n2, ncpp, nc_hash) { if (n2->nc_dvp == dvp && n2->nc_nlen == cnp->cn_namelen && !bcmp(n2->nc_name, cnp->cn_nameptr, n2->nc_nlen)) { if (tsp != NULL) { KASSERT((n2->nc_flag & NCF_TS) != 0, ("no NCF_TS")); n2_ts = __containerof(n2, struct namecache_ts, nc_nc); n2_ts->nc_time = ncp_ts->nc_time; n2_ts->nc_ticks = ncp_ts->nc_ticks; if (dtsp != NULL) { n2_ts->nc_dotdottime = ncp_ts->nc_dotdottime; if (ncp->nc_flag & NCF_NEGATIVE) mtx_lock(&ncneg_hot.nl_lock); n2_ts->nc_nc.nc_flag |= NCF_DTS; if (ncp->nc_flag & NCF_NEGATIVE) mtx_unlock(&ncneg_hot.nl_lock); } } goto out_unlock_free; } } if (flag == NCF_ISDOTDOT) { /* * See if we are trying to add .. entry, but some other lookup * has populated v_cache_dd pointer already. */ if (dvp->v_cache_dd != NULL) goto out_unlock_free; KASSERT(vp == NULL || vp->v_type == VDIR, ("wrong vnode type %p", vp)); dvp->v_cache_dd = ncp; } if (vp != NULL) { if (vp->v_type == VDIR) { if (flag != NCF_ISDOTDOT) { /* * For this case, the cache entry maps both the * directory name in it and the name ".." for the * directory's parent. */ if ((ndd = vp->v_cache_dd) != NULL) { if ((ndd->nc_flag & NCF_ISDOTDOT) != 0) cache_zap_locked(ndd, false); else ndd = NULL; } vp->v_cache_dd = ncp; } } else { vp->v_cache_dd = NULL; } } if (flag != NCF_ISDOTDOT) { if (LIST_EMPTY(&dvp->v_cache_src)) { vhold(dvp); counter_u64_add(numcachehv, 1); } LIST_INSERT_HEAD(&dvp->v_cache_src, ncp, nc_src); } /* * Insert the new namecache entry into the appropriate chain * within the cache entries table. */ LIST_INSERT_HEAD(ncpp, ncp, nc_hash); /* * If the entry is "negative", we place it into the * "negative" cache queue, otherwise, we place it into the * destination vnode's cache entries queue. */ if (vp != NULL) { TAILQ_INSERT_HEAD(&vp->v_cache_dst, ncp, nc_dst); SDT_PROBE3(vfs, namecache, enter, done, dvp, ncp->nc_name, vp); } else { if (cnp->cn_flags & ISWHITEOUT) ncp->nc_flag |= NCF_WHITE; cache_negative_insert(ncp, false); SDT_PROBE2(vfs, namecache, enter_negative, done, dvp, ncp->nc_name); } cache_enter_unlock(&cel); if (numneg * ncnegfactor > lnumcache) cache_negative_zap_one(); cache_free(ndd); return; out_unlock_free: cache_enter_unlock(&cel); cache_free(ncp); return; } static u_int cache_roundup_2(u_int val) { u_int res; for (res = 1; res <= val; res <<= 1) continue; return (res); } /* * Name cache initialization, from vfs_init() when we are booting */ static void nchinit(void *dummy __unused) { u_int i; cache_zone_small = uma_zcreate("S VFS Cache", sizeof(struct namecache) + CACHE_PATH_CUTOFF + 1, NULL, NULL, NULL, NULL, UMA_ALIGNOF(struct namecache), UMA_ZONE_ZINIT); cache_zone_small_ts = uma_zcreate("STS VFS Cache", sizeof(struct namecache_ts) + CACHE_PATH_CUTOFF + 1, NULL, NULL, NULL, NULL, UMA_ALIGNOF(struct namecache_ts), UMA_ZONE_ZINIT); cache_zone_large = uma_zcreate("L VFS Cache", sizeof(struct namecache) + NAME_MAX + 1, NULL, NULL, NULL, NULL, UMA_ALIGNOF(struct namecache), UMA_ZONE_ZINIT); cache_zone_large_ts = uma_zcreate("LTS VFS Cache", sizeof(struct namecache_ts) + NAME_MAX + 1, NULL, NULL, NULL, NULL, UMA_ALIGNOF(struct namecache_ts), UMA_ZONE_ZINIT); ncsize = desiredvnodes * ncsizefactor; nchashtbl = hashinit(desiredvnodes * 2, M_VFSCACHE, &nchash); ncbuckethash = cache_roundup_2(mp_ncpus * mp_ncpus) - 1; if (ncbuckethash < 7) /* arbitrarily chosen to avoid having one lock */ ncbuckethash = 7; if (ncbuckethash > nchash) ncbuckethash = nchash; bucketlocks = malloc(sizeof(*bucketlocks) * numbucketlocks, M_VFSCACHE, M_WAITOK | M_ZERO); for (i = 0; i < numbucketlocks; i++) rw_init_flags(&bucketlocks[i], "ncbuc", RW_DUPOK | RW_RECURSE); ncvnodehash = ncbuckethash; vnodelocks = malloc(sizeof(*vnodelocks) * numvnodelocks, M_VFSCACHE, M_WAITOK | M_ZERO); for (i = 0; i < numvnodelocks; i++) mtx_init(&vnodelocks[i], "ncvn", NULL, MTX_DUPOK | MTX_RECURSE); ncpurgeminvnodes = numbucketlocks * 2; ncneghash = 3; neglists = malloc(sizeof(*neglists) * numneglists, M_VFSCACHE, M_WAITOK | M_ZERO); for (i = 0; i < numneglists; i++) { mtx_init(&neglists[i].nl_lock, "ncnegl", NULL, MTX_DEF); TAILQ_INIT(&neglists[i].nl_list); } mtx_init(&ncneg_hot.nl_lock, "ncneglh", NULL, MTX_DEF); TAILQ_INIT(&ncneg_hot.nl_list); mtx_init(&ncneg_shrink_lock, "ncnegs", NULL, MTX_DEF); - - numcachehv = counter_u64_alloc(M_WAITOK); - numcalls = counter_u64_alloc(M_WAITOK); - dothits = counter_u64_alloc(M_WAITOK); - dotdothits = counter_u64_alloc(M_WAITOK); - numchecks = counter_u64_alloc(M_WAITOK); - nummiss = counter_u64_alloc(M_WAITOK); - nummisszap = counter_u64_alloc(M_WAITOK); - numposzaps = counter_u64_alloc(M_WAITOK); - numposhits = counter_u64_alloc(M_WAITOK); - numnegzaps = counter_u64_alloc(M_WAITOK); - numneghits = counter_u64_alloc(M_WAITOK); - numfullpathcalls = counter_u64_alloc(M_WAITOK); - numfullpathfail1 = counter_u64_alloc(M_WAITOK); - numfullpathfail2 = counter_u64_alloc(M_WAITOK); - numfullpathfail4 = counter_u64_alloc(M_WAITOK); - numfullpathfound = counter_u64_alloc(M_WAITOK); - zap_and_exit_bucket_relock_success = counter_u64_alloc(M_WAITOK); - numneg_evicted = counter_u64_alloc(M_WAITOK); - shrinking_skipped = counter_u64_alloc(M_WAITOK); } SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nchinit, NULL); void cache_changesize(u_long newmaxvnodes) { struct nchashhead *new_nchashtbl, *old_nchashtbl; u_long new_nchash, old_nchash; struct namecache *ncp; uint32_t hash; u_long newncsize; int i; newncsize = newmaxvnodes * ncsizefactor; newmaxvnodes = cache_roundup_2(newmaxvnodes * 2); if (newmaxvnodes < numbucketlocks) newmaxvnodes = numbucketlocks; new_nchashtbl = hashinit(newmaxvnodes, M_VFSCACHE, &new_nchash); /* If same hash table size, nothing to do */ if (nchash == new_nchash) { free(new_nchashtbl, M_VFSCACHE); return; } /* * Move everything from the old hash table to the new table. * None of the namecache entries in the table can be removed * because to do so, they have to be removed from the hash table. */ cache_lock_all_vnodes(); cache_lock_all_buckets(); old_nchashtbl = nchashtbl; old_nchash = nchash; nchashtbl = new_nchashtbl; nchash = new_nchash; for (i = 0; i <= old_nchash; i++) { while ((ncp = LIST_FIRST(&old_nchashtbl[i])) != NULL) { hash = cache_get_hash(ncp->nc_name, ncp->nc_nlen, ncp->nc_dvp); LIST_REMOVE(ncp, nc_hash); LIST_INSERT_HEAD(NCHHASH(hash), ncp, nc_hash); } } ncsize = newncsize; cache_unlock_all_buckets(); cache_unlock_all_vnodes(); free(old_nchashtbl, M_VFSCACHE); } /* * Invalidate all entries from and to a particular vnode. */ void cache_purge(struct vnode *vp) { TAILQ_HEAD(, namecache) ncps; struct namecache *ncp, *nnp; struct mtx *vlp, *vlp2; CTR1(KTR_VFS, "cache_purge(%p)", vp); SDT_PROBE1(vfs, namecache, purge, done, vp); if (LIST_EMPTY(&vp->v_cache_src) && TAILQ_EMPTY(&vp->v_cache_dst) && vp->v_cache_dd == NULL) return; TAILQ_INIT(&ncps); vlp = VP2VNODELOCK(vp); vlp2 = NULL; mtx_lock(vlp); retry: while (!LIST_EMPTY(&vp->v_cache_src)) { ncp = LIST_FIRST(&vp->v_cache_src); if (!cache_zap_locked_vnode_kl2(ncp, vp, &vlp2)) goto retry; TAILQ_INSERT_TAIL(&ncps, ncp, nc_dst); } while (!TAILQ_EMPTY(&vp->v_cache_dst)) { ncp = TAILQ_FIRST(&vp->v_cache_dst); if (!cache_zap_locked_vnode_kl2(ncp, vp, &vlp2)) goto retry; TAILQ_INSERT_TAIL(&ncps, ncp, nc_dst); } ncp = vp->v_cache_dd; if (ncp != NULL) { KASSERT(ncp->nc_flag & NCF_ISDOTDOT, ("lost dotdot link")); if (!cache_zap_locked_vnode_kl2(ncp, vp, &vlp2)) goto retry; TAILQ_INSERT_TAIL(&ncps, ncp, nc_dst); } KASSERT(vp->v_cache_dd == NULL, ("incomplete purge")); mtx_unlock(vlp); if (vlp2 != NULL) mtx_unlock(vlp2); TAILQ_FOREACH_SAFE(ncp, &ncps, nc_dst, nnp) { cache_free(ncp); } } /* * Invalidate all negative entries for a particular directory vnode. */ void cache_purge_negative(struct vnode *vp) { TAILQ_HEAD(, namecache) ncps; struct namecache *ncp, *nnp; struct mtx *vlp; CTR1(KTR_VFS, "cache_purge_negative(%p)", vp); SDT_PROBE1(vfs, namecache, purge_negative, done, vp); if (LIST_EMPTY(&vp->v_cache_src)) return; TAILQ_INIT(&ncps); vlp = VP2VNODELOCK(vp); mtx_lock(vlp); LIST_FOREACH_SAFE(ncp, &vp->v_cache_src, nc_src, nnp) { if (!(ncp->nc_flag & NCF_NEGATIVE)) continue; cache_zap_negative_locked_vnode_kl(ncp, vp); TAILQ_INSERT_TAIL(&ncps, ncp, nc_dst); } mtx_unlock(vlp); TAILQ_FOREACH_SAFE(ncp, &ncps, nc_dst, nnp) { cache_free(ncp); } } /* * Flush all entries referencing a particular filesystem. */ void cache_purgevfs(struct mount *mp, bool force) { TAILQ_HEAD(, namecache) ncps; struct mtx *vlp1, *vlp2; struct rwlock *blp; struct nchashhead *bucket; struct namecache *ncp, *nnp; u_long i, j, n_nchash; int error; /* Scan hash tables for applicable entries */ SDT_PROBE1(vfs, namecache, purgevfs, done, mp); if (!force && mp->mnt_nvnodelistsize <= ncpurgeminvnodes) return; TAILQ_INIT(&ncps); n_nchash = nchash + 1; vlp1 = vlp2 = NULL; for (i = 0; i < numbucketlocks; i++) { blp = (struct rwlock *)&bucketlocks[i]; rw_wlock(blp); for (j = i; j < n_nchash; j += numbucketlocks) { retry: bucket = &nchashtbl[j]; LIST_FOREACH_SAFE(ncp, bucket, nc_hash, nnp) { cache_assert_bucket_locked(ncp, RA_WLOCKED); if (ncp->nc_dvp->v_mount != mp) continue; error = cache_zap_wlocked_bucket_kl(ncp, blp, &vlp1, &vlp2); if (error != 0) goto retry; TAILQ_INSERT_HEAD(&ncps, ncp, nc_dst); } } rw_wunlock(blp); if (vlp1 == NULL && vlp2 == NULL) cache_maybe_yield(); } if (vlp1 != NULL) mtx_unlock(vlp1); if (vlp2 != NULL) mtx_unlock(vlp2); TAILQ_FOREACH_SAFE(ncp, &ncps, nc_dst, nnp) { cache_free(ncp); } } /* * Perform canonical checks and cache lookup and pass on to filesystem * through the vop_cachedlookup only if needed. */ int vfs_cache_lookup(struct vop_lookup_args *ap) { struct vnode *dvp; int error; struct vnode **vpp = ap->a_vpp; struct componentname *cnp = ap->a_cnp; int flags = cnp->cn_flags; *vpp = NULL; dvp = ap->a_dvp; if (dvp->v_type != VDIR) return (ENOTDIR); if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) && (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME)) return (EROFS); error = vn_dir_check_exec(dvp, cnp); if (error != 0) return (error); error = cache_lookup(dvp, vpp, cnp, NULL, NULL); if (error == 0) return (VOP_CACHEDLOOKUP(dvp, vpp, cnp)); if (error == -1) return (0); return (error); } /* Implementation of the getcwd syscall. */ int sys___getcwd(struct thread *td, struct __getcwd_args *uap) { char *buf, *retbuf; size_t buflen; int error; buflen = uap->buflen; if (__predict_false(buflen < 2)) return (EINVAL); if (buflen > MAXPATHLEN) buflen = MAXPATHLEN; buf = malloc(buflen, M_TEMP, M_WAITOK); error = vn_getcwd(td, buf, &retbuf, &buflen); if (error == 0) error = copyout(retbuf, uap->buf, buflen); free(buf, M_TEMP); return (error); } int vn_getcwd(struct thread *td, char *buf, char **retbuf, size_t *buflen) { struct pwd *pwd; int error; pwd = pwd_hold(td); error = vn_fullpath_any(td, pwd->pwd_cdir, pwd->pwd_rdir, buf, retbuf, buflen); pwd_drop(pwd); #ifdef KTRACE if (KTRPOINT(curthread, KTR_NAMEI) && error == 0) ktrnamei(*retbuf); #endif return (error); } static int kern___realpathat(struct thread *td, int fd, const char *path, char *buf, size_t size, int flags, enum uio_seg pathseg) { struct nameidata nd; char *retbuf, *freebuf; int error; if (flags != 0) return (EINVAL); NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | SAVENAME | WANTPARENT | AUDITVNODE1, pathseg, path, fd, &cap_fstat_rights, td); if ((error = namei(&nd)) != 0) return (error); error = vn_fullpath_hardlink(td, &nd, &retbuf, &freebuf, &size); if (error == 0) { error = copyout(retbuf, buf, size); free(freebuf, M_TEMP); } NDFREE(&nd, 0); return (error); } int sys___realpathat(struct thread *td, struct __realpathat_args *uap) { return (kern___realpathat(td, uap->fd, uap->path, uap->buf, uap->size, uap->flags, UIO_USERSPACE)); } /* * Retrieve the full filesystem path that correspond to a vnode from the name * cache (if available) */ int vn_fullpath(struct thread *td, struct vnode *vn, char **retbuf, char **freebuf) { struct pwd *pwd; char *buf; size_t buflen; int error; if (__predict_false(vn == NULL)) return (EINVAL); buflen = MAXPATHLEN; buf = malloc(buflen, M_TEMP, M_WAITOK); pwd = pwd_hold(td); error = vn_fullpath_any(td, vn, pwd->pwd_rdir, buf, retbuf, &buflen); pwd_drop(pwd); if (!error) *freebuf = buf; else free(buf, M_TEMP); return (error); } /* * This function is similar to vn_fullpath, but it attempts to lookup the * pathname relative to the global root mount point. This is required for the * auditing sub-system, as audited pathnames must be absolute, relative to the * global root mount point. */ int vn_fullpath_global(struct thread *td, struct vnode *vn, char **retbuf, char **freebuf) { char *buf; size_t buflen; int error; if (__predict_false(vn == NULL)) return (EINVAL); buflen = MAXPATHLEN; buf = malloc(buflen, M_TEMP, M_WAITOK); error = vn_fullpath_any(td, vn, rootvnode, buf, retbuf, &buflen); if (!error) *freebuf = buf; else free(buf, M_TEMP); return (error); } int vn_vptocnp(struct vnode **vp, struct ucred *cred, char *buf, size_t *buflen) { struct vnode *dvp; struct namecache *ncp; struct mtx *vlp; int error; vlp = VP2VNODELOCK(*vp); mtx_lock(vlp); TAILQ_FOREACH(ncp, &((*vp)->v_cache_dst), nc_dst) { if ((ncp->nc_flag & NCF_ISDOTDOT) == 0) break; } if (ncp != NULL) { if (*buflen < ncp->nc_nlen) { mtx_unlock(vlp); vrele(*vp); counter_u64_add(numfullpathfail4, 1); error = ENOMEM; SDT_PROBE3(vfs, namecache, fullpath, return, error, vp, NULL); return (error); } *buflen -= ncp->nc_nlen; memcpy(buf + *buflen, ncp->nc_name, ncp->nc_nlen); SDT_PROBE3(vfs, namecache, fullpath, hit, ncp->nc_dvp, ncp->nc_name, vp); dvp = *vp; *vp = ncp->nc_dvp; vref(*vp); mtx_unlock(vlp); vrele(dvp); return (0); } SDT_PROBE1(vfs, namecache, fullpath, miss, vp); mtx_unlock(vlp); vn_lock(*vp, LK_SHARED | LK_RETRY); error = VOP_VPTOCNP(*vp, &dvp, cred, buf, buflen); vput(*vp); if (error) { counter_u64_add(numfullpathfail2, 1); SDT_PROBE3(vfs, namecache, fullpath, return, error, vp, NULL); return (error); } *vp = dvp; if (VN_IS_DOOMED(dvp)) { /* forced unmount */ vrele(dvp); error = ENOENT; SDT_PROBE3(vfs, namecache, fullpath, return, error, vp, NULL); return (error); } /* * *vp has its use count incremented still. */ return (0); } /* * Resolve a directory to a pathname. * * The name of the directory can always be found in the namecache or fetched * from the filesystem. There is also guaranteed to be only one parent, meaning * we can just follow vnodes up until we find the root. * * The vnode must be referenced. */ static int vn_fullpath_dir(struct thread *td, struct vnode *vp, struct vnode *rdir, char *buf, char **retbuf, size_t *len, bool slash_prefixed, size_t addend) { #ifdef KDTRACE_HOOKS struct vnode *startvp = vp; #endif struct vnode *vp1; size_t buflen; int error; VNPASS(vp->v_type == VDIR || VN_IS_DOOMED(vp), vp); VNPASS(vp->v_usecount > 0, vp); buflen = *len; if (!slash_prefixed) { MPASS(*len >= 2); buflen--; buf[buflen] = '\0'; } error = 0; SDT_PROBE1(vfs, namecache, fullpath, entry, vp); counter_u64_add(numfullpathcalls, 1); while (vp != rdir && vp != rootvnode) { /* * The vp vnode must be already fully constructed, * since it is either found in namecache or obtained * from VOP_VPTOCNP(). We may test for VV_ROOT safely * without obtaining the vnode lock. */ if ((vp->v_vflag & VV_ROOT) != 0) { vn_lock(vp, LK_RETRY | LK_SHARED); /* * With the vnode locked, check for races with * unmount, forced or not. Note that we * already verified that vp is not equal to * the root vnode, which means that * mnt_vnodecovered can be NULL only for the * case of unmount. */ if (VN_IS_DOOMED(vp) || (vp1 = vp->v_mount->mnt_vnodecovered) == NULL || vp1->v_mountedhere != vp->v_mount) { vput(vp); error = ENOENT; SDT_PROBE3(vfs, namecache, fullpath, return, error, vp, NULL); break; } vref(vp1); vput(vp); vp = vp1; continue; } if (vp->v_type != VDIR) { vrele(vp); counter_u64_add(numfullpathfail1, 1); error = ENOTDIR; SDT_PROBE3(vfs, namecache, fullpath, return, error, vp, NULL); break; } error = vn_vptocnp(&vp, td->td_ucred, buf, &buflen); if (error) break; if (buflen == 0) { vrele(vp); error = ENOMEM; SDT_PROBE3(vfs, namecache, fullpath, return, error, startvp, NULL); break; } buf[--buflen] = '/'; slash_prefixed = true; } if (error) return (error); if (!slash_prefixed) { if (buflen == 0) { vrele(vp); counter_u64_add(numfullpathfail4, 1); SDT_PROBE3(vfs, namecache, fullpath, return, ENOMEM, startvp, NULL); return (ENOMEM); } buf[--buflen] = '/'; } counter_u64_add(numfullpathfound, 1); vrele(vp); *retbuf = buf + buflen; SDT_PROBE3(vfs, namecache, fullpath, return, 0, startvp, *retbuf); *len -= buflen; *len += addend; return (0); } /* * Resolve an arbitrary vnode to a pathname. * * Note 2 caveats: * - hardlinks are not tracked, thus if the vnode is not a directory this can * resolve to a different path than the one used to find it * - namecache is not mandatory, meaning names are not guaranteed to be added * (in which case resolving fails) */ static int vn_fullpath_any(struct thread *td, struct vnode *vp, struct vnode *rdir, char *buf, char **retbuf, size_t *buflen) { size_t orig_buflen; bool slash_prefixed; int error; if (*buflen < 2) return (EINVAL); orig_buflen = *buflen; vref(vp); slash_prefixed = false; if (vp->v_type != VDIR) { *buflen -= 1; buf[*buflen] = '\0'; error = vn_vptocnp(&vp, td->td_ucred, buf, buflen); if (error) return (error); if (*buflen == 0) { vrele(vp); return (ENOMEM); } *buflen -= 1; buf[*buflen] = '/'; slash_prefixed = true; } return (vn_fullpath_dir(td, vp, rdir, buf, retbuf, buflen, slash_prefixed, orig_buflen - *buflen)); } /* * Resolve an arbitrary vnode to a pathname (taking care of hardlinks). * * Since the namecache does not track handlings, the caller is expected to first * look up the target vnode with SAVENAME | WANTPARENT flags passed to namei. * * Then we have 2 cases: * - if the found vnode is a directory, the path can be constructed just by * fullowing names up the chain * - otherwise we populate the buffer with the saved name and start resolving * from the parent */ static int vn_fullpath_hardlink(struct thread *td, struct nameidata *ndp, char **retbuf, char **freebuf, size_t *buflen) { char *buf, *tmpbuf; struct pwd *pwd; struct componentname *cnp; struct vnode *vp; size_t addend; int error; bool slash_prefixed; if (*buflen < 2) return (EINVAL); if (*buflen > MAXPATHLEN) *buflen = MAXPATHLEN; slash_prefixed = false; buf = malloc(*buflen, M_TEMP, M_WAITOK); pwd = pwd_hold(td); addend = 0; vp = ndp->ni_vp; if (vp->v_type != VDIR) { cnp = &ndp->ni_cnd; addend = cnp->cn_namelen + 2; if (*buflen < addend) { error = ENOMEM; goto out_bad; } *buflen -= addend; tmpbuf = buf + *buflen; tmpbuf[0] = '/'; memcpy(&tmpbuf[1], cnp->cn_nameptr, cnp->cn_namelen); tmpbuf[addend - 1] = '\0'; slash_prefixed = true; vp = ndp->ni_dvp; } vref(vp); error = vn_fullpath_dir(td, vp, pwd->pwd_rdir, buf, retbuf, buflen, slash_prefixed, addend); if (error != 0) goto out_bad; pwd_drop(pwd); *freebuf = buf; return (0); out_bad: pwd_drop(pwd); free(buf, M_TEMP); return (error); } struct vnode * vn_dir_dd_ino(struct vnode *vp) { struct namecache *ncp; struct vnode *ddvp; struct mtx *vlp; enum vgetstate vs; ASSERT_VOP_LOCKED(vp, "vn_dir_dd_ino"); vlp = VP2VNODELOCK(vp); mtx_lock(vlp); TAILQ_FOREACH(ncp, &(vp->v_cache_dst), nc_dst) { if ((ncp->nc_flag & NCF_ISDOTDOT) != 0) continue; ddvp = ncp->nc_dvp; vs = vget_prep(ddvp); mtx_unlock(vlp); if (vget_finish(ddvp, LK_SHARED | LK_NOWAIT, vs)) return (NULL); return (ddvp); } mtx_unlock(vlp); return (NULL); } int vn_commname(struct vnode *vp, char *buf, u_int buflen) { struct namecache *ncp; struct mtx *vlp; int l; vlp = VP2VNODELOCK(vp); mtx_lock(vlp); TAILQ_FOREACH(ncp, &vp->v_cache_dst, nc_dst) if ((ncp->nc_flag & NCF_ISDOTDOT) == 0) break; if (ncp == NULL) { mtx_unlock(vlp); return (ENOENT); } l = min(ncp->nc_nlen, buflen - 1); memcpy(buf, ncp->nc_name, l); mtx_unlock(vlp); buf[l] = '\0'; return (0); } /* * This function updates path string to vnode's full global path * and checks the size of the new path string against the pathlen argument. * * Requires a locked, referenced vnode. * Vnode is re-locked on success or ENODEV, otherwise unlocked. * * If vp is a directory, the call to vn_fullpath_global() always succeeds * because it falls back to the ".." lookup if the namecache lookup fails. */ int vn_path_to_global_path(struct thread *td, struct vnode *vp, char *path, u_int pathlen) { struct nameidata nd; struct vnode *vp1; char *rpath, *fbuf; int error; ASSERT_VOP_ELOCKED(vp, __func__); /* Construct global filesystem path from vp. */ VOP_UNLOCK(vp); error = vn_fullpath_global(td, vp, &rpath, &fbuf); if (error != 0) { vrele(vp); return (error); } if (strlen(rpath) >= pathlen) { vrele(vp); error = ENAMETOOLONG; goto out; } /* * Re-lookup the vnode by path to detect a possible rename. * As a side effect, the vnode is relocked. * If vnode was renamed, return ENOENT. */ NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1, UIO_SYSSPACE, path, td); error = namei(&nd); if (error != 0) { vrele(vp); goto out; } NDFREE(&nd, NDF_ONLY_PNBUF); vp1 = nd.ni_vp; vrele(vp); if (vp1 == vp) strcpy(path, rpath); else { vput(vp1); error = ENOENT; } out: free(fbuf, M_TEMP); return (error); } #ifdef DDB static void db_print_vpath(struct vnode *vp) { while (vp != NULL) { db_printf("%p: ", vp); if (vp == rootvnode) { db_printf("/"); vp = NULL; } else { if (vp->v_vflag & VV_ROOT) { db_printf(""); vp = vp->v_mount->mnt_vnodecovered; } else { struct namecache *ncp; char *ncn; int i; ncp = TAILQ_FIRST(&vp->v_cache_dst); if (ncp != NULL) { ncn = ncp->nc_name; for (i = 0; i < ncp->nc_nlen; i++) db_printf("%c", *ncn++); vp = ncp->nc_dvp; } else { vp = NULL; } } } db_printf("\n"); } return; } DB_SHOW_COMMAND(vpath, db_show_vpath) { struct vnode *vp; if (!have_addr) { db_printf("usage: show vpath \n"); return; } vp = (struct vnode *)addr; db_print_vpath(vp); } #endif diff --git a/sys/kern/vfs_subr.c b/sys/kern/vfs_subr.c index 258ccb2d1ed2..bc38e7a83188 100644 --- a/sys/kern/vfs_subr.c +++ b/sys/kern/vfs_subr.c @@ -1,6469 +1,6471 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1989, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)vfs_subr.c 8.31 (Berkeley) 5/26/95 */ /* * External virtual filesystem routines */ #include __FBSDID("$FreeBSD$"); #include "opt_ddb.h" #include "opt_watchdog.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef DDB #include #endif static void delmntque(struct vnode *vp); static int flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo, int slpflag, int slptimeo); static void syncer_shutdown(void *arg, int howto); static int vtryrecycle(struct vnode *vp); static void v_init_counters(struct vnode *); static void v_incr_devcount(struct vnode *); static void v_decr_devcount(struct vnode *); static void vgonel(struct vnode *); static void vfs_knllock(void *arg); static void vfs_knlunlock(void *arg); static void vfs_knl_assert_locked(void *arg); static void vfs_knl_assert_unlocked(void *arg); static void destroy_vpollinfo(struct vpollinfo *vi); static int v_inval_buf_range_locked(struct vnode *vp, struct bufobj *bo, daddr_t startlbn, daddr_t endlbn); static void vnlru_recalc(void); /* * These fences are intended for cases where some synchronization is * needed between access of v_iflags and lockless vnode refcount (v_holdcnt * and v_usecount) updates. Access to v_iflags is generally synchronized * by the interlock, but we have some internal assertions that check vnode * flags without acquiring the lock. Thus, these fences are INVARIANTS-only * for now. */ #ifdef INVARIANTS #define VNODE_REFCOUNT_FENCE_ACQ() atomic_thread_fence_acq() #define VNODE_REFCOUNT_FENCE_REL() atomic_thread_fence_rel() #else #define VNODE_REFCOUNT_FENCE_ACQ() #define VNODE_REFCOUNT_FENCE_REL() #endif /* * Number of vnodes in existence. Increased whenever getnewvnode() * allocates a new vnode, decreased in vdropl() for VIRF_DOOMED vnode. */ static u_long __exclusive_cache_line numvnodes; SYSCTL_ULONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "Number of vnodes in existence"); static counter_u64_t vnodes_created; SYSCTL_COUNTER_U64(_vfs, OID_AUTO, vnodes_created, CTLFLAG_RD, &vnodes_created, "Number of vnodes created by getnewvnode"); /* * Conversion tables for conversion from vnode types to inode formats * and back. */ enum vtype iftovt_tab[16] = { VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON, VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VNON }; int vttoif_tab[10] = { 0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK, S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT }; /* * List of allocates vnodes in the system. */ static TAILQ_HEAD(freelst, vnode) vnode_list; static struct vnode *vnode_list_free_marker; static struct vnode *vnode_list_reclaim_marker; /* * "Free" vnode target. Free vnodes are rarely completely free, but are * just ones that are cheap to recycle. Usually they are for files which * have been stat'd but not read; these usually have inode and namecache * data attached to them. This target is the preferred minimum size of a * sub-cache consisting mostly of such files. The system balances the size * of this sub-cache with its complement to try to prevent either from * thrashing while the other is relatively inactive. The targets express * a preference for the best balance. * * "Above" this target there are 2 further targets (watermarks) related * to recyling of free vnodes. In the best-operating case, the cache is * exactly full, the free list has size between vlowat and vhiwat above the * free target, and recycling from it and normal use maintains this state. * Sometimes the free list is below vlowat or even empty, but this state * is even better for immediate use provided the cache is not full. * Otherwise, vnlru_proc() runs to reclaim enough vnodes (usually non-free * ones) to reach one of these states. The watermarks are currently hard- * coded as 4% and 9% of the available space higher. These and the default * of 25% for wantfreevnodes are too large if the memory size is large. * E.g., 9% of 75% of MAXVNODES is more than 566000 vnodes to reclaim * whenever vnlru_proc() becomes active. */ static long wantfreevnodes; static long __exclusive_cache_line freevnodes; SYSCTL_ULONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, "Number of \"free\" vnodes"); static long freevnodes_old; static counter_u64_t recycles_count; SYSCTL_COUNTER_U64(_vfs, OID_AUTO, recycles, CTLFLAG_RD, &recycles_count, "Number of vnodes recycled to meet vnode cache targets"); static counter_u64_t recycles_free_count; SYSCTL_COUNTER_U64(_vfs, OID_AUTO, recycles_free, CTLFLAG_RD, &recycles_free_count, "Number of free vnodes recycled to meet vnode cache targets"); /* * Various variables used for debugging the new implementation of * reassignbuf(). * XXX these are probably of (very) limited utility now. */ static int reassignbufcalls; SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW | CTLFLAG_STATS, &reassignbufcalls, 0, "Number of calls to reassignbuf"); static counter_u64_t deferred_inact; SYSCTL_COUNTER_U64(_vfs, OID_AUTO, deferred_inact, CTLFLAG_RD, &deferred_inact, "Number of times inactive processing was deferred"); /* To keep more than one thread at a time from running vfs_getnewfsid */ static struct mtx mntid_mtx; /* * Lock for any access to the following: * vnode_list * numvnodes * freevnodes */ static struct mtx __exclusive_cache_line vnode_list_mtx; /* Publicly exported FS */ struct nfs_public nfs_pub; static uma_zone_t buf_trie_zone; /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */ static uma_zone_t vnode_zone; static uma_zone_t vnodepoll_zone; /* * The workitem queue. * * It is useful to delay writes of file data and filesystem metadata * for tens of seconds so that quickly created and deleted files need * not waste disk bandwidth being created and removed. To realize this, * we append vnodes to a "workitem" queue. When running with a soft * updates implementation, most pending metadata dependencies should * not wait for more than a few seconds. Thus, mounted on block devices * are delayed only about a half the time that file data is delayed. * Similarly, directory updates are more critical, so are only delayed * about a third the time that file data is delayed. Thus, there are * SYNCER_MAXDELAY queues that are processed round-robin at a rate of * one each second (driven off the filesystem syncer process). The * syncer_delayno variable indicates the next queue that is to be processed. * Items that need to be processed soon are placed in this queue: * * syncer_workitem_pending[syncer_delayno] * * A delay of fifteen seconds is done by placing the request fifteen * entries later in the queue: * * syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask] * */ static int syncer_delayno; static long syncer_mask; LIST_HEAD(synclist, bufobj); static struct synclist *syncer_workitem_pending; /* * The sync_mtx protects: * bo->bo_synclist * sync_vnode_count * syncer_delayno * syncer_state * syncer_workitem_pending * syncer_worklist_len * rushjob */ static struct mtx sync_mtx; static struct cv sync_wakeup; #define SYNCER_MAXDELAY 32 static int syncer_maxdelay = SYNCER_MAXDELAY; /* maximum delay time */ static int syncdelay = 30; /* max time to delay syncing data */ static int filedelay = 30; /* time to delay syncing files */ SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, "Time to delay syncing files (in seconds)"); static int dirdelay = 29; /* time to delay syncing directories */ SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, "Time to delay syncing directories (in seconds)"); static int metadelay = 28; /* time to delay syncing metadata */ SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, "Time to delay syncing metadata (in seconds)"); static int rushjob; /* number of slots to run ASAP */ static int stat_rush_requests; /* number of times I/O speeded up */ SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, "Number of times I/O speeded up (rush requests)"); #define VDBATCH_SIZE 8 struct vdbatch { u_int index; long freevnodes; struct mtx lock; struct vnode *tab[VDBATCH_SIZE]; }; DPCPU_DEFINE_STATIC(struct vdbatch, vd); static void vdbatch_dequeue(struct vnode *vp); /* * When shutting down the syncer, run it at four times normal speed. */ #define SYNCER_SHUTDOWN_SPEEDUP 4 static int sync_vnode_count; static int syncer_worklist_len; static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY } syncer_state; /* Target for maximum number of vnodes. */ u_long desiredvnodes; static u_long gapvnodes; /* gap between wanted and desired */ static u_long vhiwat; /* enough extras after expansion */ static u_long vlowat; /* minimal extras before expansion */ static u_long vstir; /* nonzero to stir non-free vnodes */ static volatile int vsmalltrigger = 8; /* pref to keep if > this many pages */ static u_long vnlru_read_freevnodes(void); /* * Note that no attempt is made to sanitize these parameters. */ static int sysctl_maxvnodes(SYSCTL_HANDLER_ARGS) { u_long val; int error; val = desiredvnodes; error = sysctl_handle_long(oidp, &val, 0, req); if (error != 0 || req->newptr == NULL) return (error); if (val == desiredvnodes) return (0); mtx_lock(&vnode_list_mtx); desiredvnodes = val; wantfreevnodes = desiredvnodes / 4; vnlru_recalc(); mtx_unlock(&vnode_list_mtx); /* * XXX There is no protection against multiple threads changing * desiredvnodes at the same time. Locking above only helps vnlru and * getnewvnode. */ vfs_hash_changesize(desiredvnodes); cache_changesize(desiredvnodes); return (0); } SYSCTL_PROC(_kern, KERN_MAXVNODES, maxvnodes, CTLTYPE_ULONG | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0, sysctl_maxvnodes, "LU", "Target for maximum number of vnodes"); static int sysctl_wantfreevnodes(SYSCTL_HANDLER_ARGS) { u_long val; int error; val = wantfreevnodes; error = sysctl_handle_long(oidp, &val, 0, req); if (error != 0 || req->newptr == NULL) return (error); if (val == wantfreevnodes) return (0); mtx_lock(&vnode_list_mtx); wantfreevnodes = val; vnlru_recalc(); mtx_unlock(&vnode_list_mtx); return (0); } SYSCTL_PROC(_vfs, OID_AUTO, wantfreevnodes, CTLTYPE_ULONG | CTLFLAG_MPSAFE | CTLFLAG_RW, NULL, 0, sysctl_wantfreevnodes, "LU", "Target for minimum number of \"free\" vnodes"); SYSCTL_ULONG(_kern, OID_AUTO, minvnodes, CTLFLAG_RW, &wantfreevnodes, 0, "Old name for vfs.wantfreevnodes (legacy)"); static int vnlru_nowhere; SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW, &vnlru_nowhere, 0, "Number of times the vnlru process ran without success"); static int sysctl_try_reclaim_vnode(SYSCTL_HANDLER_ARGS) { struct vnode *vp; struct nameidata nd; char *buf; unsigned long ndflags; int error; if (req->newptr == NULL) return (EINVAL); if (req->newlen >= PATH_MAX) return (E2BIG); buf = malloc(PATH_MAX, M_TEMP, M_WAITOK); error = SYSCTL_IN(req, buf, req->newlen); if (error != 0) goto out; buf[req->newlen] = '\0'; ndflags = LOCKLEAF | NOFOLLOW | AUDITVNODE1 | NOCACHE | SAVENAME; NDINIT(&nd, LOOKUP, ndflags, UIO_SYSSPACE, buf, curthread); if ((error = namei(&nd)) != 0) goto out; vp = nd.ni_vp; if (VN_IS_DOOMED(vp)) { /* * This vnode is being recycled. Return != 0 to let the caller * know that the sysctl had no effect. Return EAGAIN because a * subsequent call will likely succeed (since namei will create * a new vnode if necessary) */ error = EAGAIN; goto putvnode; } counter_u64_add(recycles_count, 1); vgone(vp); putvnode: NDFREE(&nd, 0); out: free(buf, M_TEMP); return (error); } static int sysctl_ftry_reclaim_vnode(SYSCTL_HANDLER_ARGS) { struct thread *td = curthread; struct vnode *vp; struct file *fp; int error; int fd; if (req->newptr == NULL) return (EBADF); error = sysctl_handle_int(oidp, &fd, 0, req); if (error != 0) return (error); error = getvnode(curthread, fd, &cap_fcntl_rights, &fp); if (error != 0) return (error); vp = fp->f_vnode; error = vn_lock(vp, LK_EXCLUSIVE); if (error != 0) goto drop; counter_u64_add(recycles_count, 1); vgone(vp); VOP_UNLOCK(vp); drop: fdrop(fp, td); return (error); } SYSCTL_PROC(_debug, OID_AUTO, try_reclaim_vnode, CTLTYPE_STRING | CTLFLAG_MPSAFE | CTLFLAG_WR, NULL, 0, sysctl_try_reclaim_vnode, "A", "Try to reclaim a vnode by its pathname"); SYSCTL_PROC(_debug, OID_AUTO, ftry_reclaim_vnode, CTLTYPE_INT | CTLFLAG_MPSAFE | CTLFLAG_WR, NULL, 0, sysctl_ftry_reclaim_vnode, "I", "Try to reclaim a vnode by its file descriptor"); /* Shift count for (uintptr_t)vp to initialize vp->v_hash. */ static int vnsz2log; /* * Support for the bufobj clean & dirty pctrie. */ static void * buf_trie_alloc(struct pctrie *ptree) { return uma_zalloc(buf_trie_zone, M_NOWAIT); } static void buf_trie_free(struct pctrie *ptree, void *node) { uma_zfree(buf_trie_zone, node); } PCTRIE_DEFINE(BUF, buf, b_lblkno, buf_trie_alloc, buf_trie_free); /* * Initialize the vnode management data structures. * * Reevaluate the following cap on the number of vnodes after the physical * memory size exceeds 512GB. In the limit, as the physical memory size * grows, the ratio of the memory size in KB to vnodes approaches 64:1. */ #ifndef MAXVNODES_MAX #define MAXVNODES_MAX (512UL * 1024 * 1024 / 64) /* 8M */ #endif static MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker"); static struct vnode * vn_alloc_marker(struct mount *mp) { struct vnode *vp; vp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO); vp->v_type = VMARKER; vp->v_mount = mp; return (vp); } static void vn_free_marker(struct vnode *vp) { MPASS(vp->v_type == VMARKER); free(vp, M_VNODE_MARKER); } /* * Initialize a vnode as it first enters the zone. */ static int vnode_init(void *mem, int size, int flags) { struct vnode *vp; vp = mem; bzero(vp, size); /* * Setup locks. */ vp->v_vnlock = &vp->v_lock; mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF); /* * By default, don't allow shared locks unless filesystems opt-in. */ lockinit(vp->v_vnlock, PVFS, "vnode", VLKTIMEOUT, LK_NOSHARE | LK_IS_VNODE); /* * Initialize bufobj. */ bufobj_init(&vp->v_bufobj, vp); /* * Initialize namecache. */ LIST_INIT(&vp->v_cache_src); TAILQ_INIT(&vp->v_cache_dst); /* * Initialize rangelocks. */ rangelock_init(&vp->v_rl); vp->v_dbatchcpu = NOCPU; mtx_lock(&vnode_list_mtx); TAILQ_INSERT_BEFORE(vnode_list_free_marker, vp, v_vnodelist); mtx_unlock(&vnode_list_mtx); return (0); } /* * Free a vnode when it is cleared from the zone. */ static void vnode_fini(void *mem, int size) { struct vnode *vp; struct bufobj *bo; vp = mem; vdbatch_dequeue(vp); mtx_lock(&vnode_list_mtx); TAILQ_REMOVE(&vnode_list, vp, v_vnodelist); mtx_unlock(&vnode_list_mtx); rangelock_destroy(&vp->v_rl); lockdestroy(vp->v_vnlock); mtx_destroy(&vp->v_interlock); bo = &vp->v_bufobj; rw_destroy(BO_LOCKPTR(bo)); } /* * Provide the size of NFS nclnode and NFS fh for calculation of the * vnode memory consumption. The size is specified directly to * eliminate dependency on NFS-private header. * * Other filesystems may use bigger or smaller (like UFS and ZFS) * private inode data, but the NFS-based estimation is ample enough. * Still, we care about differences in the size between 64- and 32-bit * platforms. * * Namecache structure size is heuristically * sizeof(struct namecache_ts) + CACHE_PATH_CUTOFF + 1. */ #ifdef _LP64 #define NFS_NCLNODE_SZ (528 + 64) #define NC_SZ 148 #else #define NFS_NCLNODE_SZ (360 + 32) #define NC_SZ 92 #endif static void vntblinit(void *dummy __unused) { struct vdbatch *vd; int cpu, physvnodes, virtvnodes; u_int i; /* * Desiredvnodes is a function of the physical memory size and the * kernel's heap size. Generally speaking, it scales with the * physical memory size. The ratio of desiredvnodes to the physical * memory size is 1:16 until desiredvnodes exceeds 98,304. * Thereafter, the * marginal ratio of desiredvnodes to the physical memory size is * 1:64. However, desiredvnodes is limited by the kernel's heap * size. The memory required by desiredvnodes vnodes and vm objects * must not exceed 1/10th of the kernel's heap size. */ physvnodes = maxproc + pgtok(vm_cnt.v_page_count) / 64 + 3 * min(98304 * 16, pgtok(vm_cnt.v_page_count)) / 64; virtvnodes = vm_kmem_size / (10 * (sizeof(struct vm_object) + sizeof(struct vnode) + NC_SZ * ncsizefactor + NFS_NCLNODE_SZ)); desiredvnodes = min(physvnodes, virtvnodes); if (desiredvnodes > MAXVNODES_MAX) { if (bootverbose) printf("Reducing kern.maxvnodes %lu -> %lu\n", desiredvnodes, MAXVNODES_MAX); desiredvnodes = MAXVNODES_MAX; } wantfreevnodes = desiredvnodes / 4; mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF); TAILQ_INIT(&vnode_list); mtx_init(&vnode_list_mtx, "vnode_list", NULL, MTX_DEF); /* * The lock is taken to appease WITNESS. */ mtx_lock(&vnode_list_mtx); vnlru_recalc(); mtx_unlock(&vnode_list_mtx); vnode_list_free_marker = vn_alloc_marker(NULL); TAILQ_INSERT_HEAD(&vnode_list, vnode_list_free_marker, v_vnodelist); vnode_list_reclaim_marker = vn_alloc_marker(NULL); TAILQ_INSERT_HEAD(&vnode_list, vnode_list_reclaim_marker, v_vnodelist); vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL, vnode_init, vnode_fini, UMA_ALIGN_PTR, 0); vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); /* * Preallocate enough nodes to support one-per buf so that * we can not fail an insert. reassignbuf() callers can not * tolerate the insertion failure. */ buf_trie_zone = uma_zcreate("BUF TRIE", pctrie_node_size(), NULL, NULL, pctrie_zone_init, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); uma_prealloc(buf_trie_zone, nbuf); vnodes_created = counter_u64_alloc(M_WAITOK); recycles_count = counter_u64_alloc(M_WAITOK); recycles_free_count = counter_u64_alloc(M_WAITOK); deferred_inact = counter_u64_alloc(M_WAITOK); /* * Initialize the filesystem syncer. */ syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE, &syncer_mask); syncer_maxdelay = syncer_mask + 1; mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF); cv_init(&sync_wakeup, "syncer"); for (i = 1; i <= sizeof(struct vnode); i <<= 1) vnsz2log++; vnsz2log--; CPU_FOREACH(cpu) { vd = DPCPU_ID_PTR((cpu), vd); bzero(vd, sizeof(*vd)); mtx_init(&vd->lock, "vdbatch", NULL, MTX_DEF); } } SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL); /* * Mark a mount point as busy. Used to synchronize access and to delay * unmounting. Eventually, mountlist_mtx is not released on failure. * * vfs_busy() is a custom lock, it can block the caller. * vfs_busy() only sleeps if the unmount is active on the mount point. * For a mountpoint mp, vfs_busy-enforced lock is before lock of any * vnode belonging to mp. * * Lookup uses vfs_busy() to traverse mount points. * root fs var fs * / vnode lock A / vnode lock (/var) D * /var vnode lock B /log vnode lock(/var/log) E * vfs_busy lock C vfs_busy lock F * * Within each file system, the lock order is C->A->B and F->D->E. * * When traversing across mounts, the system follows that lock order: * * C->A->B * | * +->F->D->E * * The lookup() process for namei("/var") illustrates the process: * VOP_LOOKUP() obtains B while A is held * vfs_busy() obtains a shared lock on F while A and B are held * vput() releases lock on B * vput() releases lock on A * VFS_ROOT() obtains lock on D while shared lock on F is held * vfs_unbusy() releases shared lock on F * vn_lock() obtains lock on deadfs vnode vp_crossmp instead of A. * Attempt to lock A (instead of vp_crossmp) while D is held would * violate the global order, causing deadlocks. * * dounmount() locks B while F is drained. */ int vfs_busy(struct mount *mp, int flags) { MPASS((flags & ~MBF_MASK) == 0); CTR3(KTR_VFS, "%s: mp %p with flags %d", __func__, mp, flags); if (vfs_op_thread_enter(mp)) { MPASS((mp->mnt_kern_flag & MNTK_DRAINING) == 0); MPASS((mp->mnt_kern_flag & MNTK_UNMOUNT) == 0); MPASS((mp->mnt_kern_flag & MNTK_REFEXPIRE) == 0); vfs_mp_count_add_pcpu(mp, ref, 1); vfs_mp_count_add_pcpu(mp, lockref, 1); vfs_op_thread_exit(mp); if (flags & MBF_MNTLSTLOCK) mtx_unlock(&mountlist_mtx); return (0); } MNT_ILOCK(mp); vfs_assert_mount_counters(mp); MNT_REF(mp); /* * If mount point is currently being unmounted, sleep until the * mount point fate is decided. If thread doing the unmounting fails, * it will clear MNTK_UNMOUNT flag before waking us up, indicating * that this mount point has survived the unmount attempt and vfs_busy * should retry. Otherwise the unmounter thread will set MNTK_REFEXPIRE * flag in addition to MNTK_UNMOUNT, indicating that mount point is * about to be really destroyed. vfs_busy needs to release its * reference on the mount point in this case and return with ENOENT, * telling the caller that mount mount it tried to busy is no longer * valid. */ while (mp->mnt_kern_flag & MNTK_UNMOUNT) { if (flags & MBF_NOWAIT || mp->mnt_kern_flag & MNTK_REFEXPIRE) { MNT_REL(mp); MNT_IUNLOCK(mp); CTR1(KTR_VFS, "%s: failed busying before sleeping", __func__); return (ENOENT); } if (flags & MBF_MNTLSTLOCK) mtx_unlock(&mountlist_mtx); mp->mnt_kern_flag |= MNTK_MWAIT; msleep(mp, MNT_MTX(mp), PVFS | PDROP, "vfs_busy", 0); if (flags & MBF_MNTLSTLOCK) mtx_lock(&mountlist_mtx); MNT_ILOCK(mp); } if (flags & MBF_MNTLSTLOCK) mtx_unlock(&mountlist_mtx); mp->mnt_lockref++; MNT_IUNLOCK(mp); return (0); } /* * Free a busy filesystem. */ void vfs_unbusy(struct mount *mp) { int c; CTR2(KTR_VFS, "%s: mp %p", __func__, mp); if (vfs_op_thread_enter(mp)) { MPASS((mp->mnt_kern_flag & MNTK_DRAINING) == 0); vfs_mp_count_sub_pcpu(mp, lockref, 1); vfs_mp_count_sub_pcpu(mp, ref, 1); vfs_op_thread_exit(mp); return; } MNT_ILOCK(mp); vfs_assert_mount_counters(mp); MNT_REL(mp); c = --mp->mnt_lockref; if (mp->mnt_vfs_ops == 0) { MPASS((mp->mnt_kern_flag & MNTK_DRAINING) == 0); MNT_IUNLOCK(mp); return; } if (c < 0) vfs_dump_mount_counters(mp); if (c == 0 && (mp->mnt_kern_flag & MNTK_DRAINING) != 0) { MPASS(mp->mnt_kern_flag & MNTK_UNMOUNT); CTR1(KTR_VFS, "%s: waking up waiters", __func__); mp->mnt_kern_flag &= ~MNTK_DRAINING; wakeup(&mp->mnt_lockref); } MNT_IUNLOCK(mp); } /* * Lookup a mount point by filesystem identifier. */ struct mount * vfs_getvfs(fsid_t *fsid) { struct mount *mp; CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid); mtx_lock(&mountlist_mtx); TAILQ_FOREACH(mp, &mountlist, mnt_list) { if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] && mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) { vfs_ref(mp); mtx_unlock(&mountlist_mtx); return (mp); } } mtx_unlock(&mountlist_mtx); CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid); return ((struct mount *) 0); } /* * Lookup a mount point by filesystem identifier, busying it before * returning. * * To avoid congestion on mountlist_mtx, implement simple direct-mapped * cache for popular filesystem identifiers. The cache is lockess, using * the fact that struct mount's are never freed. In worst case we may * get pointer to unmounted or even different filesystem, so we have to * check what we got, and go slow way if so. */ struct mount * vfs_busyfs(fsid_t *fsid) { #define FSID_CACHE_SIZE 256 typedef struct mount * volatile vmp_t; static vmp_t cache[FSID_CACHE_SIZE]; struct mount *mp; int error; uint32_t hash; CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid); hash = fsid->val[0] ^ fsid->val[1]; hash = (hash >> 16 ^ hash) & (FSID_CACHE_SIZE - 1); mp = cache[hash]; if (mp == NULL || mp->mnt_stat.f_fsid.val[0] != fsid->val[0] || mp->mnt_stat.f_fsid.val[1] != fsid->val[1]) goto slow; if (vfs_busy(mp, 0) != 0) { cache[hash] = NULL; goto slow; } if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] && mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) return (mp); else vfs_unbusy(mp); slow: mtx_lock(&mountlist_mtx); TAILQ_FOREACH(mp, &mountlist, mnt_list) { if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] && mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) { error = vfs_busy(mp, MBF_MNTLSTLOCK); if (error) { cache[hash] = NULL; mtx_unlock(&mountlist_mtx); return (NULL); } cache[hash] = mp; return (mp); } } CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid); mtx_unlock(&mountlist_mtx); return ((struct mount *) 0); } /* * Check if a user can access privileged mount options. */ int vfs_suser(struct mount *mp, struct thread *td) { int error; if (jailed(td->td_ucred)) { /* * If the jail of the calling thread lacks permission for * this type of file system, deny immediately. */ if (!prison_allow(td->td_ucred, mp->mnt_vfc->vfc_prison_flag)) return (EPERM); /* * If the file system was mounted outside the jail of the * calling thread, deny immediately. */ if (prison_check(td->td_ucred, mp->mnt_cred) != 0) return (EPERM); } /* * If file system supports delegated administration, we don't check * for the PRIV_VFS_MOUNT_OWNER privilege - it will be better verified * by the file system itself. * If this is not the user that did original mount, we check for * the PRIV_VFS_MOUNT_OWNER privilege. */ if (!(mp->mnt_vfc->vfc_flags & VFCF_DELEGADMIN) && mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) { if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0) return (error); } return (0); } /* * Get a new unique fsid. Try to make its val[0] unique, since this value * will be used to create fake device numbers for stat(). Also try (but * not so hard) make its val[0] unique mod 2^16, since some emulators only * support 16-bit device numbers. We end up with unique val[0]'s for the * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls. * * Keep in mind that several mounts may be running in parallel. Starting * the search one past where the previous search terminated is both a * micro-optimization and a defense against returning the same fsid to * different mounts. */ void vfs_getnewfsid(struct mount *mp) { static uint16_t mntid_base; struct mount *nmp; fsid_t tfsid; int mtype; CTR2(KTR_VFS, "%s: mp %p", __func__, mp); mtx_lock(&mntid_mtx); mtype = mp->mnt_vfc->vfc_typenum; tfsid.val[1] = mtype; mtype = (mtype & 0xFF) << 24; for (;;) { tfsid.val[0] = makedev(255, mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF)); mntid_base++; if ((nmp = vfs_getvfs(&tfsid)) == NULL) break; vfs_rel(nmp); } mp->mnt_stat.f_fsid.val[0] = tfsid.val[0]; mp->mnt_stat.f_fsid.val[1] = tfsid.val[1]; mtx_unlock(&mntid_mtx); } /* * Knob to control the precision of file timestamps: * * 0 = seconds only; nanoseconds zeroed. * 1 = seconds and nanoseconds, accurate within 1/HZ. * 2 = seconds and nanoseconds, truncated to microseconds. * >=3 = seconds and nanoseconds, maximum precision. */ enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC }; static int timestamp_precision = TSP_USEC; SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW, ×tamp_precision, 0, "File timestamp precision (0: seconds, " "1: sec + ns accurate to 1/HZ, 2: sec + ns truncated to us, " "3+: sec + ns (max. precision))"); /* * Get a current timestamp. */ void vfs_timestamp(struct timespec *tsp) { struct timeval tv; switch (timestamp_precision) { case TSP_SEC: tsp->tv_sec = time_second; tsp->tv_nsec = 0; break; case TSP_HZ: getnanotime(tsp); break; case TSP_USEC: microtime(&tv); TIMEVAL_TO_TIMESPEC(&tv, tsp); break; case TSP_NSEC: default: nanotime(tsp); break; } } /* * Set vnode attributes to VNOVAL */ void vattr_null(struct vattr *vap) { vap->va_type = VNON; vap->va_size = VNOVAL; vap->va_bytes = VNOVAL; vap->va_mode = VNOVAL; vap->va_nlink = VNOVAL; vap->va_uid = VNOVAL; vap->va_gid = VNOVAL; vap->va_fsid = VNOVAL; vap->va_fileid = VNOVAL; vap->va_blocksize = VNOVAL; vap->va_rdev = VNOVAL; vap->va_atime.tv_sec = VNOVAL; vap->va_atime.tv_nsec = VNOVAL; vap->va_mtime.tv_sec = VNOVAL; vap->va_mtime.tv_nsec = VNOVAL; vap->va_ctime.tv_sec = VNOVAL; vap->va_ctime.tv_nsec = VNOVAL; vap->va_birthtime.tv_sec = VNOVAL; vap->va_birthtime.tv_nsec = VNOVAL; vap->va_flags = VNOVAL; vap->va_gen = VNOVAL; vap->va_vaflags = 0; } /* * Try to reduce the total number of vnodes. * * This routine (and its user) are buggy in at least the following ways: * - all parameters were picked years ago when RAM sizes were significantly * smaller * - it can pick vnodes based on pages used by the vm object, but filesystems * like ZFS don't use it making the pick broken * - since ZFS has its own aging policy it gets partially combated by this one * - a dedicated method should be provided for filesystems to let them decide * whether the vnode should be recycled * * This routine is called when we have too many vnodes. It attempts * to free vnodes and will potentially free vnodes that still * have VM backing store (VM backing store is typically the cause * of a vnode blowout so we want to do this). Therefore, this operation * is not considered cheap. * * A number of conditions may prevent a vnode from being reclaimed. * the buffer cache may have references on the vnode, a directory * vnode may still have references due to the namei cache representing * underlying files, or the vnode may be in active use. It is not * desirable to reuse such vnodes. These conditions may cause the * number of vnodes to reach some minimum value regardless of what * you set kern.maxvnodes to. Do not set kern.maxvnodes too low. * * @param reclaim_nc_src Only reclaim directories with outgoing namecache * entries if this argument is strue * @param trigger Only reclaim vnodes with fewer than this many resident * pages. * @param target How many vnodes to reclaim. * @return The number of vnodes that were reclaimed. */ static int vlrureclaim(bool reclaim_nc_src, int trigger, u_long target) { struct vnode *vp, *mvp; struct mount *mp; struct vm_object *object; u_long done; bool retried; mtx_assert(&vnode_list_mtx, MA_OWNED); retried = false; done = 0; mvp = vnode_list_reclaim_marker; restart: vp = mvp; while (done < target) { vp = TAILQ_NEXT(vp, v_vnodelist); if (__predict_false(vp == NULL)) break; if (__predict_false(vp->v_type == VMARKER)) continue; /* * If it's been deconstructed already, it's still * referenced, or it exceeds the trigger, skip it. * Also skip free vnodes. We are trying to make space * to expand the free list, not reduce it. */ if (vp->v_usecount > 0 || vp->v_holdcnt == 0 || (!reclaim_nc_src && !LIST_EMPTY(&vp->v_cache_src))) goto next_iter; if (vp->v_type == VBAD || vp->v_type == VNON) goto next_iter; if (!VI_TRYLOCK(vp)) goto next_iter; if (vp->v_usecount > 0 || vp->v_holdcnt == 0 || (!reclaim_nc_src && !LIST_EMPTY(&vp->v_cache_src)) || VN_IS_DOOMED(vp) || vp->v_type == VNON) { VI_UNLOCK(vp); goto next_iter; } object = atomic_load_ptr(&vp->v_object); if (object == NULL || object->resident_page_count > trigger) { VI_UNLOCK(vp); goto next_iter; } vholdl(vp); VI_UNLOCK(vp); TAILQ_REMOVE(&vnode_list, mvp, v_vnodelist); TAILQ_INSERT_AFTER(&vnode_list, vp, mvp, v_vnodelist); mtx_unlock(&vnode_list_mtx); if (vn_start_write(vp, &mp, V_NOWAIT) != 0) { vdrop(vp); goto next_iter_unlocked; } if (VOP_LOCK(vp, LK_EXCLUSIVE|LK_NOWAIT) != 0) { vdrop(vp); vn_finished_write(mp); goto next_iter_unlocked; } VI_LOCK(vp); if (vp->v_usecount > 0 || (!reclaim_nc_src && !LIST_EMPTY(&vp->v_cache_src)) || (vp->v_object != NULL && vp->v_object->resident_page_count > trigger)) { VOP_UNLOCK(vp); vdropl(vp); vn_finished_write(mp); goto next_iter_unlocked; } counter_u64_add(recycles_count, 1); vgonel(vp); VOP_UNLOCK(vp); vdropl(vp); vn_finished_write(mp); done++; next_iter_unlocked: if (should_yield()) kern_yield(PRI_USER); mtx_lock(&vnode_list_mtx); goto restart; next_iter: MPASS(vp->v_type != VMARKER); if (!should_yield()) continue; TAILQ_REMOVE(&vnode_list, mvp, v_vnodelist); TAILQ_INSERT_AFTER(&vnode_list, vp, mvp, v_vnodelist); mtx_unlock(&vnode_list_mtx); kern_yield(PRI_USER); mtx_lock(&vnode_list_mtx); goto restart; } if (done == 0 && !retried) { TAILQ_REMOVE(&vnode_list, mvp, v_vnodelist); TAILQ_INSERT_HEAD(&vnode_list, mvp, v_vnodelist); retried = true; goto restart; } return (done); } static int max_vnlru_free = 10000; /* limit on vnode free requests per call */ SYSCTL_INT(_debug, OID_AUTO, max_vnlru_free, CTLFLAG_RW, &max_vnlru_free, 0, "limit on vnode free requests per call to the vnlru_free routine"); /* * Attempt to reduce the free list by the requested amount. */ static int vnlru_free_locked(int count, struct vfsops *mnt_op) { struct vnode *vp, *mvp; struct mount *mp; int ocount; mtx_assert(&vnode_list_mtx, MA_OWNED); if (count > max_vnlru_free) count = max_vnlru_free; ocount = count; mvp = vnode_list_free_marker; restart: vp = mvp; while (count > 0) { vp = TAILQ_NEXT(vp, v_vnodelist); if (__predict_false(vp == NULL)) { TAILQ_REMOVE(&vnode_list, mvp, v_vnodelist); TAILQ_INSERT_TAIL(&vnode_list, mvp, v_vnodelist); break; } if (__predict_false(vp->v_type == VMARKER)) continue; /* * Don't recycle if our vnode is from different type * of mount point. Note that mp is type-safe, the * check does not reach unmapped address even if * vnode is reclaimed. * Don't recycle if we can't get the interlock without * blocking. */ if (vp->v_holdcnt > 0 || (mnt_op != NULL && (mp = vp->v_mount) != NULL && mp->mnt_op != mnt_op) || !VI_TRYLOCK(vp)) { continue; } TAILQ_REMOVE(&vnode_list, mvp, v_vnodelist); TAILQ_INSERT_AFTER(&vnode_list, vp, mvp, v_vnodelist); if (__predict_false(vp->v_type == VBAD || vp->v_type == VNON)) { VI_UNLOCK(vp); continue; } vholdl(vp); count--; mtx_unlock(&vnode_list_mtx); VI_UNLOCK(vp); vtryrecycle(vp); vdrop(vp); mtx_lock(&vnode_list_mtx); goto restart; } return (ocount - count); } void vnlru_free(int count, struct vfsops *mnt_op) { mtx_lock(&vnode_list_mtx); vnlru_free_locked(count, mnt_op); mtx_unlock(&vnode_list_mtx); } static void vnlru_recalc(void) { mtx_assert(&vnode_list_mtx, MA_OWNED); gapvnodes = imax(desiredvnodes - wantfreevnodes, 100); vhiwat = gapvnodes / 11; /* 9% -- just under the 10% in vlrureclaim() */ vlowat = vhiwat / 2; } /* * Attempt to recycle vnodes in a context that is always safe to block. * Calling vlrurecycle() from the bowels of filesystem code has some * interesting deadlock problems. */ static struct proc *vnlruproc; static int vnlruproc_sig; /* * The main freevnodes counter is only updated when threads requeue their vnode * batches. CPUs are conditionally walked to compute a more accurate total. * * Limit how much of a slop are we willing to tolerate. Note: the actual value * at any given moment can still exceed slop, but it should not be by significant * margin in practice. */ #define VNLRU_FREEVNODES_SLOP 128 static u_long vnlru_read_freevnodes(void) { struct vdbatch *vd; long slop; int cpu; mtx_assert(&vnode_list_mtx, MA_OWNED); if (freevnodes > freevnodes_old) slop = freevnodes - freevnodes_old; else slop = freevnodes_old - freevnodes; if (slop < VNLRU_FREEVNODES_SLOP) return (freevnodes >= 0 ? freevnodes : 0); freevnodes_old = freevnodes; CPU_FOREACH(cpu) { vd = DPCPU_ID_PTR((cpu), vd); freevnodes_old += vd->freevnodes; } return (freevnodes_old >= 0 ? freevnodes_old : 0); } static bool vnlru_under(u_long rnumvnodes, u_long limit) { u_long rfreevnodes, space; if (__predict_false(rnumvnodes > desiredvnodes)) return (true); space = desiredvnodes - rnumvnodes; if (space < limit) { rfreevnodes = vnlru_read_freevnodes(); if (rfreevnodes > wantfreevnodes) space += rfreevnodes - wantfreevnodes; } return (space < limit); } static bool vnlru_under_unlocked(u_long rnumvnodes, u_long limit) { long rfreevnodes, space; if (__predict_false(rnumvnodes > desiredvnodes)) return (true); space = desiredvnodes - rnumvnodes; if (space < limit) { rfreevnodes = atomic_load_long(&freevnodes); if (rfreevnodes > wantfreevnodes) space += rfreevnodes - wantfreevnodes; } return (space < limit); } static void vnlru_kick(void) { mtx_assert(&vnode_list_mtx, MA_OWNED); if (vnlruproc_sig == 0) { vnlruproc_sig = 1; wakeup(vnlruproc); } } static void vnlru_proc(void) { u_long rnumvnodes, rfreevnodes, target; unsigned long onumvnodes; int done, force, trigger, usevnodes; bool reclaim_nc_src, want_reread; EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, vnlruproc, SHUTDOWN_PRI_FIRST); force = 0; want_reread = false; for (;;) { kproc_suspend_check(vnlruproc); mtx_lock(&vnode_list_mtx); rnumvnodes = atomic_load_long(&numvnodes); if (want_reread) { force = vnlru_under(numvnodes, vhiwat) ? 1 : 0; want_reread = false; } /* * If numvnodes is too large (due to desiredvnodes being * adjusted using its sysctl, or emergency growth), first * try to reduce it by discarding from the free list. */ if (rnumvnodes > desiredvnodes) { vnlru_free_locked(rnumvnodes - desiredvnodes, NULL); rnumvnodes = atomic_load_long(&numvnodes); } /* * Sleep if the vnode cache is in a good state. This is * when it is not over-full and has space for about a 4% * or 9% expansion (by growing its size or inexcessively * reducing its free list). Otherwise, try to reclaim * space for a 10% expansion. */ if (vstir && force == 0) { force = 1; vstir = 0; } if (force == 0 && !vnlru_under(rnumvnodes, vlowat)) { vnlruproc_sig = 0; wakeup(&vnlruproc_sig); msleep(vnlruproc, &vnode_list_mtx, PVFS|PDROP, "vlruwt", hz); continue; } rfreevnodes = vnlru_read_freevnodes(); onumvnodes = rnumvnodes; /* * Calculate parameters for recycling. These are the same * throughout the loop to give some semblance of fairness. * The trigger point is to avoid recycling vnodes with lots * of resident pages. We aren't trying to free memory; we * are trying to recycle or at least free vnodes. */ if (rnumvnodes <= desiredvnodes) usevnodes = rnumvnodes - rfreevnodes; else usevnodes = rnumvnodes; if (usevnodes <= 0) usevnodes = 1; /* * The trigger value is is chosen to give a conservatively * large value to ensure that it alone doesn't prevent * making progress. The value can easily be so large that * it is effectively infinite in some congested and * misconfigured cases, and this is necessary. Normally * it is about 8 to 100 (pages), which is quite large. */ trigger = vm_cnt.v_page_count * 2 / usevnodes; if (force < 2) trigger = vsmalltrigger; reclaim_nc_src = force >= 3; target = rnumvnodes * (int64_t)gapvnodes / imax(desiredvnodes, 1); target = target / 10 + 1; done = vlrureclaim(reclaim_nc_src, trigger, target); mtx_unlock(&vnode_list_mtx); if (onumvnodes > desiredvnodes && numvnodes <= desiredvnodes) uma_reclaim(UMA_RECLAIM_DRAIN); if (done == 0) { if (force == 0 || force == 1) { force = 2; continue; } if (force == 2) { force = 3; continue; } want_reread = true; force = 0; vnlru_nowhere++; tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3); } else { want_reread = true; kern_yield(PRI_USER); } } } static struct kproc_desc vnlru_kp = { "vnlru", vnlru_proc, &vnlruproc }; SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp); /* * Routines having to do with the management of the vnode table. */ /* * Try to recycle a freed vnode. We abort if anyone picks up a reference * before we actually vgone(). This function must be called with the vnode * held to prevent the vnode from being returned to the free list midway * through vgone(). */ static int vtryrecycle(struct vnode *vp) { struct mount *vnmp; CTR2(KTR_VFS, "%s: vp %p", __func__, vp); VNASSERT(vp->v_holdcnt, vp, ("vtryrecycle: Recycling vp %p without a reference.", vp)); /* * This vnode may found and locked via some other list, if so we * can't recycle it yet. */ if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0) { CTR2(KTR_VFS, "%s: impossible to recycle, vp %p lock is already held", __func__, vp); return (EWOULDBLOCK); } /* * Don't recycle if its filesystem is being suspended. */ if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) { VOP_UNLOCK(vp); CTR2(KTR_VFS, "%s: impossible to recycle, cannot start the write for %p", __func__, vp); return (EBUSY); } /* * If we got this far, we need to acquire the interlock and see if * anyone picked up this vnode from another list. If not, we will * mark it with DOOMED via vgonel() so that anyone who does find it * will skip over it. */ VI_LOCK(vp); if (vp->v_usecount) { VOP_UNLOCK(vp); VI_UNLOCK(vp); vn_finished_write(vnmp); CTR2(KTR_VFS, "%s: impossible to recycle, %p is already referenced", __func__, vp); return (EBUSY); } if (!VN_IS_DOOMED(vp)) { counter_u64_add(recycles_free_count, 1); vgonel(vp); } VOP_UNLOCK(vp); VI_UNLOCK(vp); vn_finished_write(vnmp); return (0); } /* * Allocate a new vnode. * * The operation never returns an error. Returning an error was disabled * in r145385 (dated 2005) with the following comment: * * XXX Not all VFS_VGET/ffs_vget callers check returns. * * Given the age of this commit (almost 15 years at the time of writing this * comment) restoring the ability to fail requires a significant audit of * all codepaths. * * The routine can try to free a vnode or stall for up to 1 second waiting for * vnlru to clear things up, but ultimately always performs a M_WAITOK allocation. */ static u_long vn_alloc_cyclecount; static struct vnode * __noinline vn_alloc_hard(struct mount *mp) { u_long rnumvnodes, rfreevnodes; mtx_lock(&vnode_list_mtx); rnumvnodes = atomic_load_long(&numvnodes); if (rnumvnodes + 1 < desiredvnodes) { vn_alloc_cyclecount = 0; goto alloc; } rfreevnodes = vnlru_read_freevnodes(); if (vn_alloc_cyclecount++ >= rfreevnodes) { vn_alloc_cyclecount = 0; vstir = 1; } /* * Grow the vnode cache if it will not be above its target max * after growing. Otherwise, if the free list is nonempty, try * to reclaim 1 item from it before growing the cache (possibly * above its target max if the reclamation failed or is delayed). * Otherwise, wait for some space. In all cases, schedule * vnlru_proc() if we are getting short of space. The watermarks * should be chosen so that we never wait or even reclaim from * the free list to below its target minimum. */ if (vnlru_free_locked(1, NULL) > 0) goto alloc; if (mp == NULL || (mp->mnt_kern_flag & MNTK_SUSPEND) == 0) { /* * Wait for space for a new vnode. */ vnlru_kick(); msleep(&vnlruproc_sig, &vnode_list_mtx, PVFS, "vlruwk", hz); if (atomic_load_long(&numvnodes) + 1 > desiredvnodes && vnlru_read_freevnodes() > 1) vnlru_free_locked(1, NULL); } alloc: rnumvnodes = atomic_fetchadd_long(&numvnodes, 1) + 1; if (vnlru_under(rnumvnodes, vlowat)) vnlru_kick(); mtx_unlock(&vnode_list_mtx); return (uma_zalloc(vnode_zone, M_WAITOK)); } static struct vnode * vn_alloc(struct mount *mp) { u_long rnumvnodes; if (__predict_false(vn_alloc_cyclecount != 0)) return (vn_alloc_hard(mp)); rnumvnodes = atomic_fetchadd_long(&numvnodes, 1) + 1; if (__predict_false(vnlru_under_unlocked(rnumvnodes, vlowat))) { atomic_subtract_long(&numvnodes, 1); return (vn_alloc_hard(mp)); } return (uma_zalloc(vnode_zone, M_WAITOK)); } static void vn_free(struct vnode *vp) { atomic_subtract_long(&numvnodes, 1); uma_zfree(vnode_zone, vp); } /* * Return the next vnode from the free list. */ int getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops, struct vnode **vpp) { struct vnode *vp; struct thread *td; struct lock_object *lo; CTR3(KTR_VFS, "%s: mp %p with tag %s", __func__, mp, tag); KASSERT(vops->registered, ("%s: not registered vector op %p\n", __func__, vops)); td = curthread; if (td->td_vp_reserved != NULL) { vp = td->td_vp_reserved; td->td_vp_reserved = NULL; } else { vp = vn_alloc(mp); } counter_u64_add(vnodes_created, 1); /* * Locks are given the generic name "vnode" when created. * Follow the historic practice of using the filesystem * name when they allocated, e.g., "zfs", "ufs", "nfs, etc. * * Locks live in a witness group keyed on their name. Thus, * when a lock is renamed, it must also move from the witness * group of its old name to the witness group of its new name. * * The change only needs to be made when the vnode moves * from one filesystem type to another. We ensure that each * filesystem use a single static name pointer for its tag so * that we can compare pointers rather than doing a strcmp(). */ lo = &vp->v_vnlock->lock_object; #ifdef WITNESS if (lo->lo_name != tag) { #endif lo->lo_name = tag; #ifdef WITNESS WITNESS_DESTROY(lo); WITNESS_INIT(lo, tag); } #endif /* * By default, don't allow shared locks unless filesystems opt-in. */ vp->v_vnlock->lock_object.lo_flags |= LK_NOSHARE; /* * Finalize various vnode identity bits. */ KASSERT(vp->v_object == NULL, ("stale v_object %p", vp)); KASSERT(vp->v_lockf == NULL, ("stale v_lockf %p", vp)); KASSERT(vp->v_pollinfo == NULL, ("stale v_pollinfo %p", vp)); vp->v_type = VNON; vp->v_op = vops; v_init_counters(vp); vp->v_bufobj.bo_ops = &buf_ops_bio; #ifdef DIAGNOSTIC if (mp == NULL && vops != &dead_vnodeops) printf("NULL mp in getnewvnode(9), tag %s\n", tag); #endif #ifdef MAC mac_vnode_init(vp); if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0) mac_vnode_associate_singlelabel(mp, vp); #endif if (mp != NULL) { vp->v_bufobj.bo_bsize = mp->mnt_stat.f_iosize; if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0) vp->v_vflag |= VV_NOKNOTE; } /* * For the filesystems which do not use vfs_hash_insert(), * still initialize v_hash to have vfs_hash_index() useful. * E.g., nullfs uses vfs_hash_index() on the lower vnode for * its own hashing. */ vp->v_hash = (uintptr_t)vp >> vnsz2log; *vpp = vp; return (0); } void getnewvnode_reserve(void) { struct thread *td; td = curthread; MPASS(td->td_vp_reserved == NULL); td->td_vp_reserved = vn_alloc(NULL); } void getnewvnode_drop_reserve(void) { struct thread *td; td = curthread; if (td->td_vp_reserved != NULL) { vn_free(td->td_vp_reserved); td->td_vp_reserved = NULL; } } static void freevnode(struct vnode *vp) { struct bufobj *bo; /* * The vnode has been marked for destruction, so free it. * * The vnode will be returned to the zone where it will * normally remain until it is needed for another vnode. We * need to cleanup (or verify that the cleanup has already * been done) any residual data left from its current use * so as not to contaminate the freshly allocated vnode. */ CTR2(KTR_VFS, "%s: destroying the vnode %p", __func__, vp); bo = &vp->v_bufobj; VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't")); VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count")); VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count")); VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count")); VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's")); VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0")); VNASSERT(pctrie_is_empty(&bo->bo_clean.bv_root), vp, ("clean blk trie not empty")); VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0")); VNASSERT(pctrie_is_empty(&bo->bo_dirty.bv_root), vp, ("dirty blk trie not empty")); VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst")); VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src")); VNASSERT(vp->v_cache_dd == NULL, vp, ("vp has namecache for ..")); VNASSERT(TAILQ_EMPTY(&vp->v_rl.rl_waiters), vp, ("Dangling rangelock waiters")); VI_UNLOCK(vp); #ifdef MAC mac_vnode_destroy(vp); #endif if (vp->v_pollinfo != NULL) { destroy_vpollinfo(vp->v_pollinfo); vp->v_pollinfo = NULL; } #ifdef INVARIANTS /* XXX Elsewhere we detect an already freed vnode via NULL v_op. */ vp->v_op = NULL; #endif vp->v_mountedhere = NULL; vp->v_unpcb = NULL; vp->v_rdev = NULL; vp->v_fifoinfo = NULL; vp->v_lasta = vp->v_clen = vp->v_cstart = vp->v_lastw = 0; vp->v_irflag = 0; vp->v_iflag = 0; vp->v_vflag = 0; bo->bo_flag = 0; vn_free(vp); } /* * Delete from old mount point vnode list, if on one. */ static void delmntque(struct vnode *vp) { struct mount *mp; VNPASS((vp->v_mflag & VMP_LAZYLIST) == 0, vp); mp = vp->v_mount; if (mp == NULL) return; MNT_ILOCK(mp); VI_LOCK(vp); vp->v_mount = NULL; VI_UNLOCK(vp); VNASSERT(mp->mnt_nvnodelistsize > 0, vp, ("bad mount point vnode list size")); TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes); mp->mnt_nvnodelistsize--; MNT_REL(mp); MNT_IUNLOCK(mp); } static void insmntque_stddtr(struct vnode *vp, void *dtr_arg) { vp->v_data = NULL; vp->v_op = &dead_vnodeops; vgone(vp); vput(vp); } /* * Insert into list of vnodes for the new mount point, if available. */ int insmntque1(struct vnode *vp, struct mount *mp, void (*dtr)(struct vnode *, void *), void *dtr_arg) { KASSERT(vp->v_mount == NULL, ("insmntque: vnode already on per mount vnode list")); VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)")); ASSERT_VOP_ELOCKED(vp, "insmntque: non-locked vp"); /* * We acquire the vnode interlock early to ensure that the * vnode cannot be recycled by another process releasing a * holdcnt on it before we get it on both the vnode list * and the active vnode list. The mount mutex protects only * manipulation of the vnode list and the vnode freelist * mutex protects only manipulation of the active vnode list. * Hence the need to hold the vnode interlock throughout. */ MNT_ILOCK(mp); VI_LOCK(vp); if (((mp->mnt_kern_flag & MNTK_UNMOUNT) != 0 && ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 || mp->mnt_nvnodelistsize == 0)) && (vp->v_vflag & VV_FORCEINSMQ) == 0) { VI_UNLOCK(vp); MNT_IUNLOCK(mp); if (dtr != NULL) dtr(vp, dtr_arg); return (EBUSY); } vp->v_mount = mp; MNT_REF(mp); TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes); VNASSERT(mp->mnt_nvnodelistsize >= 0, vp, ("neg mount point vnode list size")); mp->mnt_nvnodelistsize++; VI_UNLOCK(vp); MNT_IUNLOCK(mp); return (0); } int insmntque(struct vnode *vp, struct mount *mp) { return (insmntque1(vp, mp, insmntque_stddtr, NULL)); } /* * Flush out and invalidate all buffers associated with a bufobj * Called with the underlying object locked. */ int bufobj_invalbuf(struct bufobj *bo, int flags, int slpflag, int slptimeo) { int error; BO_LOCK(bo); if (flags & V_SAVE) { error = bufobj_wwait(bo, slpflag, slptimeo); if (error) { BO_UNLOCK(bo); return (error); } if (bo->bo_dirty.bv_cnt > 0) { BO_UNLOCK(bo); if ((error = BO_SYNC(bo, MNT_WAIT)) != 0) return (error); /* * XXX We could save a lock/unlock if this was only * enabled under INVARIANTS */ BO_LOCK(bo); if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0) panic("vinvalbuf: dirty bufs"); } } /* * If you alter this loop please notice that interlock is dropped and * reacquired in flushbuflist. Special care is needed to ensure that * no race conditions occur from this. */ do { error = flushbuflist(&bo->bo_clean, flags, bo, slpflag, slptimeo); if (error == 0 && !(flags & V_CLEANONLY)) error = flushbuflist(&bo->bo_dirty, flags, bo, slpflag, slptimeo); if (error != 0 && error != EAGAIN) { BO_UNLOCK(bo); return (error); } } while (error != 0); /* * Wait for I/O to complete. XXX needs cleaning up. The vnode can * have write I/O in-progress but if there is a VM object then the * VM object can also have read-I/O in-progress. */ do { bufobj_wwait(bo, 0, 0); if ((flags & V_VMIO) == 0 && bo->bo_object != NULL) { BO_UNLOCK(bo); vm_object_pip_wait_unlocked(bo->bo_object, "bovlbx"); BO_LOCK(bo); } } while (bo->bo_numoutput > 0); BO_UNLOCK(bo); /* * Destroy the copy in the VM cache, too. */ if (bo->bo_object != NULL && (flags & (V_ALT | V_NORMAL | V_CLEANONLY | V_VMIO)) == 0) { VM_OBJECT_WLOCK(bo->bo_object); vm_object_page_remove(bo->bo_object, 0, 0, (flags & V_SAVE) ? OBJPR_CLEANONLY : 0); VM_OBJECT_WUNLOCK(bo->bo_object); } #ifdef INVARIANTS BO_LOCK(bo); if ((flags & (V_ALT | V_NORMAL | V_CLEANONLY | V_VMIO | V_ALLOWCLEAN)) == 0 && (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0)) panic("vinvalbuf: flush failed"); if ((flags & (V_ALT | V_NORMAL | V_CLEANONLY | V_VMIO)) == 0 && bo->bo_dirty.bv_cnt > 0) panic("vinvalbuf: flush dirty failed"); BO_UNLOCK(bo); #endif return (0); } /* * Flush out and invalidate all buffers associated with a vnode. * Called with the underlying object locked. */ int vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo) { CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags); ASSERT_VOP_LOCKED(vp, "vinvalbuf"); if (vp->v_object != NULL && vp->v_object->handle != vp) return (0); return (bufobj_invalbuf(&vp->v_bufobj, flags, slpflag, slptimeo)); } /* * Flush out buffers on the specified list. * */ static int flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo, int slpflag, int slptimeo) { struct buf *bp, *nbp; int retval, error; daddr_t lblkno; b_xflags_t xflags; ASSERT_BO_WLOCKED(bo); retval = 0; TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) { /* * If we are flushing both V_NORMAL and V_ALT buffers then * do not skip any buffers. If we are flushing only V_NORMAL * buffers then skip buffers marked as BX_ALTDATA. If we are * flushing only V_ALT buffers then skip buffers not marked * as BX_ALTDATA. */ if (((flags & (V_NORMAL | V_ALT)) != (V_NORMAL | V_ALT)) && (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA) != 0) || ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0))) { continue; } if (nbp != NULL) { lblkno = nbp->b_lblkno; xflags = nbp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN); } retval = EAGAIN; error = BUF_TIMELOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo), "flushbuf", slpflag, slptimeo); if (error) { BO_LOCK(bo); return (error != ENOLCK ? error : EAGAIN); } KASSERT(bp->b_bufobj == bo, ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); /* * XXX Since there are no node locks for NFS, I * believe there is a slight chance that a delayed * write will occur while sleeping just above, so * check for it. */ if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) && (flags & V_SAVE)) { bremfree(bp); bp->b_flags |= B_ASYNC; bwrite(bp); BO_LOCK(bo); return (EAGAIN); /* XXX: why not loop ? */ } bremfree(bp); bp->b_flags |= (B_INVAL | B_RELBUF); bp->b_flags &= ~B_ASYNC; brelse(bp); BO_LOCK(bo); if (nbp == NULL) break; nbp = gbincore(bo, lblkno); if (nbp == NULL || (nbp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) != xflags) break; /* nbp invalid */ } return (retval); } int bnoreuselist(struct bufv *bufv, struct bufobj *bo, daddr_t startn, daddr_t endn) { struct buf *bp; int error; daddr_t lblkno; ASSERT_BO_LOCKED(bo); for (lblkno = startn;;) { again: bp = BUF_PCTRIE_LOOKUP_GE(&bufv->bv_root, lblkno); if (bp == NULL || bp->b_lblkno >= endn || bp->b_lblkno < startn) break; error = BUF_TIMELOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo), "brlsfl", 0, 0); if (error != 0) { BO_RLOCK(bo); if (error == ENOLCK) goto again; return (error); } KASSERT(bp->b_bufobj == bo, ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); lblkno = bp->b_lblkno + 1; if ((bp->b_flags & B_MANAGED) == 0) bremfree(bp); bp->b_flags |= B_RELBUF; /* * In the VMIO case, use the B_NOREUSE flag to hint that the * pages backing each buffer in the range are unlikely to be * reused. Dirty buffers will have the hint applied once * they've been written. */ if ((bp->b_flags & B_VMIO) != 0) bp->b_flags |= B_NOREUSE; brelse(bp); BO_RLOCK(bo); } return (0); } /* * Truncate a file's buffer and pages to a specified length. This * is in lieu of the old vinvalbuf mechanism, which performed unneeded * sync activity. */ int vtruncbuf(struct vnode *vp, off_t length, int blksize) { struct buf *bp, *nbp; struct bufobj *bo; daddr_t startlbn; CTR4(KTR_VFS, "%s: vp %p with block %d:%ju", __func__, vp, blksize, (uintmax_t)length); /* * Round up to the *next* lbn. */ startlbn = howmany(length, blksize); ASSERT_VOP_LOCKED(vp, "vtruncbuf"); bo = &vp->v_bufobj; restart_unlocked: BO_LOCK(bo); while (v_inval_buf_range_locked(vp, bo, startlbn, INT64_MAX) == EAGAIN) ; if (length > 0) { restartsync: TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { if (bp->b_lblkno > 0) continue; /* * Since we hold the vnode lock this should only * fail if we're racing with the buf daemon. */ if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) goto restart_unlocked; VNASSERT((bp->b_flags & B_DELWRI), vp, ("buf(%p) on dirty queue without DELWRI", bp)); bremfree(bp); bawrite(bp); BO_LOCK(bo); goto restartsync; } } bufobj_wwait(bo, 0, 0); BO_UNLOCK(bo); vnode_pager_setsize(vp, length); return (0); } /* * Invalidate the cached pages of a file's buffer within the range of block * numbers [startlbn, endlbn). */ void v_inval_buf_range(struct vnode *vp, daddr_t startlbn, daddr_t endlbn, int blksize) { struct bufobj *bo; off_t start, end; ASSERT_VOP_LOCKED(vp, "v_inval_buf_range"); start = blksize * startlbn; end = blksize * endlbn; bo = &vp->v_bufobj; BO_LOCK(bo); MPASS(blksize == bo->bo_bsize); while (v_inval_buf_range_locked(vp, bo, startlbn, endlbn) == EAGAIN) ; BO_UNLOCK(bo); vn_pages_remove(vp, OFF_TO_IDX(start), OFF_TO_IDX(end + PAGE_SIZE - 1)); } static int v_inval_buf_range_locked(struct vnode *vp, struct bufobj *bo, daddr_t startlbn, daddr_t endlbn) { struct buf *bp, *nbp; bool anyfreed; ASSERT_VOP_LOCKED(vp, "v_inval_buf_range_locked"); ASSERT_BO_LOCKED(bo); do { anyfreed = false; TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) { if (bp->b_lblkno < startlbn || bp->b_lblkno >= endlbn) continue; if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) { BO_LOCK(bo); return (EAGAIN); } bremfree(bp); bp->b_flags |= B_INVAL | B_RELBUF; bp->b_flags &= ~B_ASYNC; brelse(bp); anyfreed = true; BO_LOCK(bo); if (nbp != NULL && (((nbp->b_xflags & BX_VNCLEAN) == 0) || nbp->b_vp != vp || (nbp->b_flags & B_DELWRI) != 0)) return (EAGAIN); } TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) { if (bp->b_lblkno < startlbn || bp->b_lblkno >= endlbn) continue; if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo)) == ENOLCK) { BO_LOCK(bo); return (EAGAIN); } bremfree(bp); bp->b_flags |= B_INVAL | B_RELBUF; bp->b_flags &= ~B_ASYNC; brelse(bp); anyfreed = true; BO_LOCK(bo); if (nbp != NULL && (((nbp->b_xflags & BX_VNDIRTY) == 0) || (nbp->b_vp != vp) || (nbp->b_flags & B_DELWRI) == 0)) return (EAGAIN); } } while (anyfreed); return (0); } static void buf_vlist_remove(struct buf *bp) { struct bufv *bv; KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp)); ASSERT_BO_WLOCKED(bp->b_bufobj); KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) != (BX_VNDIRTY|BX_VNCLEAN), ("buf_vlist_remove: Buf %p is on two lists", bp)); if (bp->b_xflags & BX_VNDIRTY) bv = &bp->b_bufobj->bo_dirty; else bv = &bp->b_bufobj->bo_clean; BUF_PCTRIE_REMOVE(&bv->bv_root, bp->b_lblkno); TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs); bv->bv_cnt--; bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN); } /* * Add the buffer to the sorted clean or dirty block list. * * NOTE: xflags is passed as a constant, optimizing this inline function! */ static void buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags) { struct bufv *bv; struct buf *n; int error; ASSERT_BO_WLOCKED(bo); + KASSERT((bo->bo_flag & BO_NOBUFS) == 0, + ("buf_vlist_add: bo %p does not allow bufs", bo)); KASSERT((xflags & BX_VNDIRTY) == 0 || (bo->bo_flag & BO_DEAD) == 0, ("dead bo %p", bo)); KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags)); bp->b_xflags |= xflags; if (xflags & BX_VNDIRTY) bv = &bo->bo_dirty; else bv = &bo->bo_clean; /* * Keep the list ordered. Optimize empty list insertion. Assume * we tend to grow at the tail so lookup_le should usually be cheaper * than _ge. */ if (bv->bv_cnt == 0 || bp->b_lblkno > TAILQ_LAST(&bv->bv_hd, buflists)->b_lblkno) TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs); else if ((n = BUF_PCTRIE_LOOKUP_LE(&bv->bv_root, bp->b_lblkno)) == NULL) TAILQ_INSERT_HEAD(&bv->bv_hd, bp, b_bobufs); else TAILQ_INSERT_AFTER(&bv->bv_hd, n, bp, b_bobufs); error = BUF_PCTRIE_INSERT(&bv->bv_root, bp); if (error) panic("buf_vlist_add: Preallocated nodes insufficient."); bv->bv_cnt++; } /* * Look up a buffer using the buffer tries. */ struct buf * gbincore(struct bufobj *bo, daddr_t lblkno) { struct buf *bp; ASSERT_BO_LOCKED(bo); bp = BUF_PCTRIE_LOOKUP(&bo->bo_clean.bv_root, lblkno); if (bp != NULL) return (bp); return BUF_PCTRIE_LOOKUP(&bo->bo_dirty.bv_root, lblkno); } /* * Associate a buffer with a vnode. */ void bgetvp(struct vnode *vp, struct buf *bp) { struct bufobj *bo; bo = &vp->v_bufobj; ASSERT_BO_WLOCKED(bo); VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free")); CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags); VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp, ("bgetvp: bp already attached! %p", bp)); vhold(vp); bp->b_vp = vp; bp->b_bufobj = bo; /* * Insert onto list for new vnode. */ buf_vlist_add(bp, bo, BX_VNCLEAN); } /* * Disassociate a buffer from a vnode. */ void brelvp(struct buf *bp) { struct bufobj *bo; struct vnode *vp; CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); KASSERT(bp->b_vp != NULL, ("brelvp: NULL")); /* * Delete from old vnode list, if on one. */ vp = bp->b_vp; /* XXX */ bo = bp->b_bufobj; BO_LOCK(bo); if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) buf_vlist_remove(bp); else panic("brelvp: Buffer %p not on queue.", bp); if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) { bo->bo_flag &= ~BO_ONWORKLST; mtx_lock(&sync_mtx); LIST_REMOVE(bo, bo_synclist); syncer_worklist_len--; mtx_unlock(&sync_mtx); } bp->b_vp = NULL; bp->b_bufobj = NULL; BO_UNLOCK(bo); vdrop(vp); } /* * Add an item to the syncer work queue. */ static void vn_syncer_add_to_worklist(struct bufobj *bo, int delay) { int slot; ASSERT_BO_WLOCKED(bo); mtx_lock(&sync_mtx); if (bo->bo_flag & BO_ONWORKLST) LIST_REMOVE(bo, bo_synclist); else { bo->bo_flag |= BO_ONWORKLST; syncer_worklist_len++; } if (delay > syncer_maxdelay - 2) delay = syncer_maxdelay - 2; slot = (syncer_delayno + delay) & syncer_mask; LIST_INSERT_HEAD(&syncer_workitem_pending[slot], bo, bo_synclist); mtx_unlock(&sync_mtx); } static int sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS) { int error, len; mtx_lock(&sync_mtx); len = syncer_worklist_len - sync_vnode_count; mtx_unlock(&sync_mtx); error = SYSCTL_OUT(req, &len, sizeof(len)); return (error); } SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_MPSAFE| CTLFLAG_RD, NULL, 0, sysctl_vfs_worklist_len, "I", "Syncer thread worklist length"); static struct proc *updateproc; static void sched_sync(void); static struct kproc_desc up_kp = { "syncer", sched_sync, &updateproc }; SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp); static int sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td) { struct vnode *vp; struct mount *mp; *bo = LIST_FIRST(slp); if (*bo == NULL) return (0); vp = bo2vnode(*bo); if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0) return (1); /* * We use vhold in case the vnode does not * successfully sync. vhold prevents the vnode from * going away when we unlock the sync_mtx so that * we can acquire the vnode interlock. */ vholdl(vp); mtx_unlock(&sync_mtx); VI_UNLOCK(vp); if (vn_start_write(vp, &mp, V_NOWAIT) != 0) { vdrop(vp); mtx_lock(&sync_mtx); return (*bo == LIST_FIRST(slp)); } vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); (void) VOP_FSYNC(vp, MNT_LAZY, td); VOP_UNLOCK(vp); vn_finished_write(mp); BO_LOCK(*bo); if (((*bo)->bo_flag & BO_ONWORKLST) != 0) { /* * Put us back on the worklist. The worklist * routine will remove us from our current * position and then add us back in at a later * position. */ vn_syncer_add_to_worklist(*bo, syncdelay); } BO_UNLOCK(*bo); vdrop(vp); mtx_lock(&sync_mtx); return (0); } static int first_printf = 1; /* * System filesystem synchronizer daemon. */ static void sched_sync(void) { struct synclist *next, *slp; struct bufobj *bo; long starttime; struct thread *td = curthread; int last_work_seen; int net_worklist_len; int syncer_final_iter; int error; last_work_seen = 0; syncer_final_iter = 0; syncer_state = SYNCER_RUNNING; starttime = time_uptime; td->td_pflags |= TDP_NORUNNINGBUF; EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc, SHUTDOWN_PRI_LAST); mtx_lock(&sync_mtx); for (;;) { if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter == 0) { mtx_unlock(&sync_mtx); kproc_suspend_check(td->td_proc); mtx_lock(&sync_mtx); } net_worklist_len = syncer_worklist_len - sync_vnode_count; if (syncer_state != SYNCER_RUNNING && starttime != time_uptime) { if (first_printf) { printf("\nSyncing disks, vnodes remaining... "); first_printf = 0; } printf("%d ", net_worklist_len); } starttime = time_uptime; /* * Push files whose dirty time has expired. Be careful * of interrupt race on slp queue. * * Skip over empty worklist slots when shutting down. */ do { slp = &syncer_workitem_pending[syncer_delayno]; syncer_delayno += 1; if (syncer_delayno == syncer_maxdelay) syncer_delayno = 0; next = &syncer_workitem_pending[syncer_delayno]; /* * If the worklist has wrapped since the * it was emptied of all but syncer vnodes, * switch to the FINAL_DELAY state and run * for one more second. */ if (syncer_state == SYNCER_SHUTTING_DOWN && net_worklist_len == 0 && last_work_seen == syncer_delayno) { syncer_state = SYNCER_FINAL_DELAY; syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP; } } while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) && syncer_worklist_len > 0); /* * Keep track of the last time there was anything * on the worklist other than syncer vnodes. * Return to the SHUTTING_DOWN state if any * new work appears. */ if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING) last_work_seen = syncer_delayno; if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY) syncer_state = SYNCER_SHUTTING_DOWN; while (!LIST_EMPTY(slp)) { error = sync_vnode(slp, &bo, td); if (error == 1) { LIST_REMOVE(bo, bo_synclist); LIST_INSERT_HEAD(next, bo, bo_synclist); continue; } if (first_printf == 0) { /* * Drop the sync mutex, because some watchdog * drivers need to sleep while patting */ mtx_unlock(&sync_mtx); wdog_kern_pat(WD_LASTVAL); mtx_lock(&sync_mtx); } } if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0) syncer_final_iter--; /* * The variable rushjob allows the kernel to speed up the * processing of the filesystem syncer process. A rushjob * value of N tells the filesystem syncer to process the next * N seconds worth of work on its queue ASAP. Currently rushjob * is used by the soft update code to speed up the filesystem * syncer process when the incore state is getting so far * ahead of the disk that the kernel memory pool is being * threatened with exhaustion. */ if (rushjob > 0) { rushjob -= 1; continue; } /* * Just sleep for a short period of time between * iterations when shutting down to allow some I/O * to happen. * * If it has taken us less than a second to process the * current work, then wait. Otherwise start right over * again. We can still lose time if any single round * takes more than two seconds, but it does not really * matter as we are just trying to generally pace the * filesystem activity. */ if (syncer_state != SYNCER_RUNNING || time_uptime == starttime) { thread_lock(td); sched_prio(td, PPAUSE); thread_unlock(td); } if (syncer_state != SYNCER_RUNNING) cv_timedwait(&sync_wakeup, &sync_mtx, hz / SYNCER_SHUTDOWN_SPEEDUP); else if (time_uptime == starttime) cv_timedwait(&sync_wakeup, &sync_mtx, hz); } } /* * Request the syncer daemon to speed up its work. * We never push it to speed up more than half of its * normal turn time, otherwise it could take over the cpu. */ int speedup_syncer(void) { int ret = 0; mtx_lock(&sync_mtx); if (rushjob < syncdelay / 2) { rushjob += 1; stat_rush_requests += 1; ret = 1; } mtx_unlock(&sync_mtx); cv_broadcast(&sync_wakeup); return (ret); } /* * Tell the syncer to speed up its work and run though its work * list several times, then tell it to shut down. */ static void syncer_shutdown(void *arg, int howto) { if (howto & RB_NOSYNC) return; mtx_lock(&sync_mtx); syncer_state = SYNCER_SHUTTING_DOWN; rushjob = 0; mtx_unlock(&sync_mtx); cv_broadcast(&sync_wakeup); kproc_shutdown(arg, howto); } void syncer_suspend(void) { syncer_shutdown(updateproc, 0); } void syncer_resume(void) { mtx_lock(&sync_mtx); first_printf = 1; syncer_state = SYNCER_RUNNING; mtx_unlock(&sync_mtx); cv_broadcast(&sync_wakeup); kproc_resume(updateproc); } /* * Reassign a buffer from one vnode to another. * Used to assign file specific control information * (indirect blocks) to the vnode to which they belong. */ void reassignbuf(struct buf *bp) { struct vnode *vp; struct bufobj *bo; int delay; #ifdef INVARIANTS struct bufv *bv; #endif vp = bp->b_vp; bo = bp->b_bufobj; ++reassignbufcalls; CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); /* * B_PAGING flagged buffers cannot be reassigned because their vp * is not fully linked in. */ if (bp->b_flags & B_PAGING) panic("cannot reassign paging buffer"); /* * Delete from old vnode list, if on one. */ BO_LOCK(bo); if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) buf_vlist_remove(bp); else panic("reassignbuf: Buffer %p not on queue.", bp); /* * If dirty, put on list of dirty buffers; otherwise insert onto list * of clean buffers. */ if (bp->b_flags & B_DELWRI) { if ((bo->bo_flag & BO_ONWORKLST) == 0) { switch (vp->v_type) { case VDIR: delay = dirdelay; break; case VCHR: delay = metadelay; break; default: delay = filedelay; } vn_syncer_add_to_worklist(bo, delay); } buf_vlist_add(bp, bo, BX_VNDIRTY); } else { buf_vlist_add(bp, bo, BX_VNCLEAN); if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) { mtx_lock(&sync_mtx); LIST_REMOVE(bo, bo_synclist); syncer_worklist_len--; mtx_unlock(&sync_mtx); bo->bo_flag &= ~BO_ONWORKLST; } } #ifdef INVARIANTS bv = &bo->bo_clean; bp = TAILQ_FIRST(&bv->bv_hd); KASSERT(bp == NULL || bp->b_bufobj == bo, ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); bp = TAILQ_LAST(&bv->bv_hd, buflists); KASSERT(bp == NULL || bp->b_bufobj == bo, ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); bv = &bo->bo_dirty; bp = TAILQ_FIRST(&bv->bv_hd); KASSERT(bp == NULL || bp->b_bufobj == bo, ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); bp = TAILQ_LAST(&bv->bv_hd, buflists); KASSERT(bp == NULL || bp->b_bufobj == bo, ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo)); #endif BO_UNLOCK(bo); } static void v_init_counters(struct vnode *vp) { VNASSERT(vp->v_type == VNON && vp->v_data == NULL && vp->v_iflag == 0, vp, ("%s called for an initialized vnode", __FUNCTION__)); ASSERT_VI_UNLOCKED(vp, __FUNCTION__); refcount_init(&vp->v_holdcnt, 1); refcount_init(&vp->v_usecount, 1); } /* * Increment si_usecount of the associated device, if any. */ static void v_incr_devcount(struct vnode *vp) { ASSERT_VI_LOCKED(vp, __FUNCTION__); if (vp->v_type == VCHR && vp->v_rdev != NULL) { dev_lock(); vp->v_rdev->si_usecount++; dev_unlock(); } } /* * Decrement si_usecount of the associated device, if any. * * The caller is required to hold the interlock when transitioning a VCHR use * count to zero. This prevents a race with devfs_reclaim_vchr() that would * leak a si_usecount reference. The vnode lock will also prevent this race * if it is held while dropping the last ref. * * The race is: * * CPU1 CPU2 * devfs_reclaim_vchr * make v_usecount == 0 * VI_LOCK * sees v_usecount == 0, no updates * vp->v_rdev = NULL; * ... * VI_UNLOCK * VI_LOCK * v_decr_devcount * sees v_rdev == NULL, no updates * * In this scenario si_devcount decrement is not performed. */ static void v_decr_devcount(struct vnode *vp) { ASSERT_VOP_LOCKED(vp, __func__); ASSERT_VI_LOCKED(vp, __FUNCTION__); if (vp->v_type == VCHR && vp->v_rdev != NULL) { dev_lock(); VNPASS(vp->v_rdev->si_usecount > 0, vp); vp->v_rdev->si_usecount--; dev_unlock(); } } /* * Grab a particular vnode from the free list, increment its * reference count and lock it. VIRF_DOOMED is set if the vnode * is being destroyed. Only callers who specify LK_RETRY will * see doomed vnodes. If inactive processing was delayed in * vput try to do it here. * * usecount is manipulated using atomics without holding any locks. * * holdcnt can be manipulated using atomics without holding any locks, * except when transitioning 1<->0, in which case the interlock is held. */ enum vgetstate vget_prep(struct vnode *vp) { enum vgetstate vs; if (refcount_acquire_if_not_zero(&vp->v_usecount)) { vs = VGET_USECOUNT; } else { vhold(vp); vs = VGET_HOLDCNT; } return (vs); } int vget(struct vnode *vp, int flags, struct thread *td) { enum vgetstate vs; MPASS(td == curthread); vs = vget_prep(vp); return (vget_finish(vp, flags, vs)); } static int __noinline vget_finish_vchr(struct vnode *vp) { VNASSERT(vp->v_type == VCHR, vp, ("type != VCHR)")); /* * See the comment in vget_finish before usecount bump. */ if (refcount_acquire_if_not_zero(&vp->v_usecount)) { #ifdef INVARIANTS int old = atomic_fetchadd_int(&vp->v_holdcnt, -1); VNASSERT(old > 0, vp, ("%s: wrong hold count %d", __func__, old)); #else refcount_release(&vp->v_holdcnt); #endif return (0); } VI_LOCK(vp); if (refcount_acquire_if_not_zero(&vp->v_usecount)) { #ifdef INVARIANTS int old = atomic_fetchadd_int(&vp->v_holdcnt, -1); VNASSERT(old > 1, vp, ("%s: wrong hold count %d", __func__, old)); #else refcount_release(&vp->v_holdcnt); #endif VI_UNLOCK(vp); return (0); } v_incr_devcount(vp); refcount_acquire(&vp->v_usecount); VI_UNLOCK(vp); return (0); } int vget_finish(struct vnode *vp, int flags, enum vgetstate vs) { int error, old; if ((flags & LK_INTERLOCK) != 0) ASSERT_VI_LOCKED(vp, __func__); else ASSERT_VI_UNLOCKED(vp, __func__); VNPASS(vp->v_holdcnt > 0, vp); VNPASS(vs == VGET_HOLDCNT || vp->v_usecount > 0, vp); error = vn_lock(vp, flags); if (__predict_false(error != 0)) { if (vs == VGET_USECOUNT) vrele(vp); else vdrop(vp); CTR2(KTR_VFS, "%s: impossible to lock vnode %p", __func__, vp); return (error); } if (vs == VGET_USECOUNT) return (0); if (__predict_false(vp->v_type == VCHR)) return (vget_finish_vchr(vp)); /* * We hold the vnode. If the usecount is 0 it will be utilized to keep * the vnode around. Otherwise someone else lended their hold count and * we have to drop ours. */ old = atomic_fetchadd_int(&vp->v_usecount, 1); VNASSERT(old >= 0, vp, ("%s: wrong use count %d", __func__, old)); if (old != 0) { #ifdef INVARIANTS old = atomic_fetchadd_int(&vp->v_holdcnt, -1); VNASSERT(old > 1, vp, ("%s: wrong hold count %d", __func__, old)); #else refcount_release(&vp->v_holdcnt); #endif } return (0); } /* * Increase the reference (use) and hold count of a vnode. * This will also remove the vnode from the free list if it is presently free. */ static void __noinline vref_vchr(struct vnode *vp, bool interlock) { /* * See the comment in vget_finish before usecount bump. */ if (!interlock) { if (refcount_acquire_if_not_zero(&vp->v_usecount)) { VNODE_REFCOUNT_FENCE_ACQ(); VNASSERT(vp->v_holdcnt > 0, vp, ("%s: active vnode not held", __func__)); return; } VI_LOCK(vp); /* * By the time we get here the vnode might have been doomed, at * which point the 0->1 use count transition is no longer * protected by the interlock. Since it can't bounce back to * VCHR and requires vref semantics, punt it back */ if (__predict_false(vp->v_type == VBAD)) { VI_UNLOCK(vp); vref(vp); return; } } VNASSERT(vp->v_type == VCHR, vp, ("type != VCHR)")); if (refcount_acquire_if_not_zero(&vp->v_usecount)) { VNODE_REFCOUNT_FENCE_ACQ(); VNASSERT(vp->v_holdcnt > 0, vp, ("%s: active vnode not held", __func__)); if (!interlock) VI_UNLOCK(vp); return; } vhold(vp); v_incr_devcount(vp); refcount_acquire(&vp->v_usecount); if (!interlock) VI_UNLOCK(vp); return; } void vref(struct vnode *vp) { int old; CTR2(KTR_VFS, "%s: vp %p", __func__, vp); if (__predict_false(vp->v_type == VCHR)) { vref_vchr(vp, false); return; } if (refcount_acquire_if_not_zero(&vp->v_usecount)) { VNODE_REFCOUNT_FENCE_ACQ(); VNASSERT(vp->v_holdcnt > 0, vp, ("%s: active vnode not held", __func__)); return; } vhold(vp); /* * See the comment in vget_finish. */ old = atomic_fetchadd_int(&vp->v_usecount, 1); VNASSERT(old >= 0, vp, ("%s: wrong use count %d", __func__, old)); if (old != 0) { #ifdef INVARIANTS old = atomic_fetchadd_int(&vp->v_holdcnt, -1); VNASSERT(old > 1, vp, ("%s: wrong hold count %d", __func__, old)); #else refcount_release(&vp->v_holdcnt); #endif } } void vrefl(struct vnode *vp) { ASSERT_VI_LOCKED(vp, __func__); CTR2(KTR_VFS, "%s: vp %p", __func__, vp); if (__predict_false(vp->v_type == VCHR)) { vref_vchr(vp, true); return; } vref(vp); } void vrefact(struct vnode *vp) { CTR2(KTR_VFS, "%s: vp %p", __func__, vp); #ifdef INVARIANTS int old = atomic_fetchadd_int(&vp->v_usecount, 1); VNASSERT(old > 0, vp, ("%s: wrong use count %d", __func__, old)); #else refcount_acquire(&vp->v_usecount); #endif } void vrefactn(struct vnode *vp, u_int n) { CTR2(KTR_VFS, "%s: vp %p", __func__, vp); #ifdef INVARIANTS int old = atomic_fetchadd_int(&vp->v_usecount, n); VNASSERT(old > 0, vp, ("%s: wrong use count %d", __func__, old)); #else atomic_add_int(&vp->v_usecount, n); #endif } /* * Return reference count of a vnode. * * The results of this call are only guaranteed when some mechanism is used to * stop other processes from gaining references to the vnode. This may be the * case if the caller holds the only reference. This is also useful when stale * data is acceptable as race conditions may be accounted for by some other * means. */ int vrefcnt(struct vnode *vp) { return (vp->v_usecount); } void vlazy(struct vnode *vp) { struct mount *mp; VNASSERT(vp->v_holdcnt > 0, vp, ("%s: vnode not held", __func__)); if ((vp->v_mflag & VMP_LAZYLIST) != 0) return; /* * We may get here for inactive routines after the vnode got doomed. */ if (VN_IS_DOOMED(vp)) return; mp = vp->v_mount; mtx_lock(&mp->mnt_listmtx); if ((vp->v_mflag & VMP_LAZYLIST) == 0) { vp->v_mflag |= VMP_LAZYLIST; TAILQ_INSERT_TAIL(&mp->mnt_lazyvnodelist, vp, v_lazylist); mp->mnt_lazyvnodelistsize++; } mtx_unlock(&mp->mnt_listmtx); } /* * This routine is only meant to be called from vgonel prior to dooming * the vnode. */ static void vunlazy_gone(struct vnode *vp) { struct mount *mp; ASSERT_VOP_ELOCKED(vp, __func__); ASSERT_VI_LOCKED(vp, __func__); VNPASS(!VN_IS_DOOMED(vp), vp); if (vp->v_mflag & VMP_LAZYLIST) { mp = vp->v_mount; mtx_lock(&mp->mnt_listmtx); VNPASS(vp->v_mflag & VMP_LAZYLIST, vp); vp->v_mflag &= ~VMP_LAZYLIST; TAILQ_REMOVE(&mp->mnt_lazyvnodelist, vp, v_lazylist); mp->mnt_lazyvnodelistsize--; mtx_unlock(&mp->mnt_listmtx); } } static void vdefer_inactive(struct vnode *vp) { ASSERT_VI_LOCKED(vp, __func__); VNASSERT(vp->v_holdcnt > 0, vp, ("%s: vnode without hold count", __func__)); if (VN_IS_DOOMED(vp)) { vdropl(vp); return; } if (vp->v_iflag & VI_DEFINACT) { VNASSERT(vp->v_holdcnt > 1, vp, ("lost hold count")); vdropl(vp); return; } if (vp->v_usecount > 0) { vp->v_iflag &= ~VI_OWEINACT; vdropl(vp); return; } vlazy(vp); vp->v_iflag |= VI_DEFINACT; VI_UNLOCK(vp); counter_u64_add(deferred_inact, 1); } static void vdefer_inactive_unlocked(struct vnode *vp) { VI_LOCK(vp); if ((vp->v_iflag & VI_OWEINACT) == 0) { vdropl(vp); return; } vdefer_inactive(vp); } enum vput_op { VRELE, VPUT, VUNREF }; /* * Handle ->v_usecount transitioning to 0. * * By releasing the last usecount we take ownership of the hold count which * provides liveness of the vnode, meaning we have to vdrop. * * If the vnode is of type VCHR we may need to decrement si_usecount, see * v_decr_devcount for details. * * For all vnodes we may need to perform inactive processing. It requires an * exclusive lock on the vnode, while it is legal to call here with only a * shared lock (or no locks). If locking the vnode in an expected manner fails, * inactive processing gets deferred to the syncer. * * XXX Some filesystems pass in an exclusively locked vnode and strongly depend * on the lock being held all the way until VOP_INACTIVE. This in particular * happens with UFS which adds half-constructed vnodes to the hash, where they * can be found by other code. */ static void vput_final(struct vnode *vp, enum vput_op func) { int error; bool want_unlock; CTR2(KTR_VFS, "%s: vp %p", __func__, vp); VNPASS(vp->v_holdcnt > 0, vp); VI_LOCK(vp); if (__predict_false(vp->v_type == VCHR && func != VRELE)) v_decr_devcount(vp); /* * By the time we got here someone else might have transitioned * the count back to > 0. */ if (vp->v_usecount > 0) goto out; /* * If the vnode is doomed vgone already performed inactive processing * (if needed). */ if (VN_IS_DOOMED(vp)) goto out; if (__predict_true(VOP_NEED_INACTIVE(vp) == 0)) goto out; if (vp->v_iflag & VI_DOINGINACT) goto out; /* * Locking operations here will drop the interlock and possibly the * vnode lock, opening a window where the vnode can get doomed all the * while ->v_usecount is 0. Set VI_OWEINACT to let vgone know to * perform inactive. */ vp->v_iflag |= VI_OWEINACT; want_unlock = false; error = 0; switch (func) { case VRELE: switch (VOP_ISLOCKED(vp)) { case LK_EXCLUSIVE: break; case LK_EXCLOTHER: case 0: want_unlock = true; error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK); VI_LOCK(vp); break; default: /* * The lock has at least one sharer, but we have no way * to conclude whether this is us. Play it safe and * defer processing. */ error = EAGAIN; break; } break; case VPUT: want_unlock = true; if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { error = VOP_LOCK(vp, LK_UPGRADE | LK_INTERLOCK | LK_NOWAIT); VI_LOCK(vp); } break; case VUNREF: if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) { error = VOP_LOCK(vp, LK_TRYUPGRADE | LK_INTERLOCK); VI_LOCK(vp); } break; } if (error == 0) { vinactive(vp); if (want_unlock) VOP_UNLOCK(vp); vdropl(vp); } else { vdefer_inactive(vp); } return; out: if (func == VPUT) VOP_UNLOCK(vp); vdropl(vp); } /* * Decrement ->v_usecount for a vnode. * * Releasing the last use count requires additional processing, see vput_final * above for details. * * Note that releasing use count without the vnode lock requires special casing * for VCHR, see v_decr_devcount for details. * * Comment above each variant denotes lock state on entry and exit. */ static void __noinline vrele_vchr(struct vnode *vp) { if (refcount_release_if_not_last(&vp->v_usecount)) return; VI_LOCK(vp); if (!refcount_release(&vp->v_usecount)) { VI_UNLOCK(vp); return; } v_decr_devcount(vp); VI_UNLOCK(vp); vput_final(vp, VRELE); } /* * in: any * out: same as passed in */ void vrele(struct vnode *vp) { ASSERT_VI_UNLOCKED(vp, __func__); if (__predict_false(vp->v_type == VCHR)) { vrele_vchr(vp); return; } if (!refcount_release(&vp->v_usecount)) return; vput_final(vp, VRELE); } /* * in: locked * out: unlocked */ void vput(struct vnode *vp) { ASSERT_VOP_LOCKED(vp, __func__); ASSERT_VI_UNLOCKED(vp, __func__); if (!refcount_release(&vp->v_usecount)) { VOP_UNLOCK(vp); return; } vput_final(vp, VPUT); } /* * in: locked * out: locked */ void vunref(struct vnode *vp) { ASSERT_VOP_LOCKED(vp, __func__); ASSERT_VI_UNLOCKED(vp, __func__); if (!refcount_release(&vp->v_usecount)) return; vput_final(vp, VUNREF); } void vhold(struct vnode *vp) { struct vdbatch *vd; int old; CTR2(KTR_VFS, "%s: vp %p", __func__, vp); old = atomic_fetchadd_int(&vp->v_holdcnt, 1); VNASSERT(old >= 0, vp, ("%s: wrong hold count %d", __func__, old)); if (old != 0) return; critical_enter(); vd = DPCPU_PTR(vd); vd->freevnodes--; critical_exit(); } void vholdl(struct vnode *vp) { ASSERT_VI_LOCKED(vp, __func__); CTR2(KTR_VFS, "%s: vp %p", __func__, vp); vhold(vp); } void vholdnz(struct vnode *vp) { CTR2(KTR_VFS, "%s: vp %p", __func__, vp); #ifdef INVARIANTS int old = atomic_fetchadd_int(&vp->v_holdcnt, 1); VNASSERT(old > 0, vp, ("%s: wrong hold count %d", __func__, old)); #else atomic_add_int(&vp->v_holdcnt, 1); #endif } static void __noinline vdbatch_process(struct vdbatch *vd) { struct vnode *vp; int i; mtx_assert(&vd->lock, MA_OWNED); MPASS(curthread->td_pinned > 0); MPASS(vd->index == VDBATCH_SIZE); mtx_lock(&vnode_list_mtx); critical_enter(); freevnodes += vd->freevnodes; for (i = 0; i < VDBATCH_SIZE; i++) { vp = vd->tab[i]; TAILQ_REMOVE(&vnode_list, vp, v_vnodelist); TAILQ_INSERT_TAIL(&vnode_list, vp, v_vnodelist); MPASS(vp->v_dbatchcpu != NOCPU); vp->v_dbatchcpu = NOCPU; } mtx_unlock(&vnode_list_mtx); vd->freevnodes = 0; bzero(vd->tab, sizeof(vd->tab)); vd->index = 0; critical_exit(); } static void vdbatch_enqueue(struct vnode *vp) { struct vdbatch *vd; ASSERT_VI_LOCKED(vp, __func__); VNASSERT(!VN_IS_DOOMED(vp), vp, ("%s: deferring requeue of a doomed vnode", __func__)); critical_enter(); vd = DPCPU_PTR(vd); vd->freevnodes++; if (vp->v_dbatchcpu != NOCPU) { VI_UNLOCK(vp); critical_exit(); return; } sched_pin(); critical_exit(); mtx_lock(&vd->lock); MPASS(vd->index < VDBATCH_SIZE); MPASS(vd->tab[vd->index] == NULL); /* * A hack: we depend on being pinned so that we know what to put in * ->v_dbatchcpu. */ vp->v_dbatchcpu = curcpu; vd->tab[vd->index] = vp; vd->index++; VI_UNLOCK(vp); if (vd->index == VDBATCH_SIZE) vdbatch_process(vd); mtx_unlock(&vd->lock); sched_unpin(); } /* * This routine must only be called for vnodes which are about to be * deallocated. Supporting dequeue for arbitrary vndoes would require * validating that the locked batch matches. */ static void vdbatch_dequeue(struct vnode *vp) { struct vdbatch *vd; int i; short cpu; VNASSERT(vp->v_type == VBAD || vp->v_type == VNON, vp, ("%s: called for a used vnode\n", __func__)); cpu = vp->v_dbatchcpu; if (cpu == NOCPU) return; vd = DPCPU_ID_PTR(cpu, vd); mtx_lock(&vd->lock); for (i = 0; i < vd->index; i++) { if (vd->tab[i] != vp) continue; vp->v_dbatchcpu = NOCPU; vd->index--; vd->tab[i] = vd->tab[vd->index]; vd->tab[vd->index] = NULL; break; } mtx_unlock(&vd->lock); /* * Either we dequeued the vnode above or the target CPU beat us to it. */ MPASS(vp->v_dbatchcpu == NOCPU); } /* * Drop the hold count of the vnode. If this is the last reference to * the vnode we place it on the free list unless it has been vgone'd * (marked VIRF_DOOMED) in which case we will free it. * * Because the vnode vm object keeps a hold reference on the vnode if * there is at least one resident non-cached page, the vnode cannot * leave the active list without the page cleanup done. */ static void vdrop_deactivate(struct vnode *vp) { struct mount *mp; ASSERT_VI_LOCKED(vp, __func__); /* * Mark a vnode as free: remove it from its active list * and put it up for recycling on the freelist. */ VNASSERT(!VN_IS_DOOMED(vp), vp, ("vdrop: returning doomed vnode")); VNASSERT(vp->v_op != NULL, vp, ("vdrop: vnode already reclaimed.")); VNASSERT((vp->v_iflag & VI_OWEINACT) == 0, vp, ("vnode with VI_OWEINACT set")); VNASSERT((vp->v_iflag & VI_DEFINACT) == 0, vp, ("vnode with VI_DEFINACT set")); if (vp->v_mflag & VMP_LAZYLIST) { mp = vp->v_mount; mtx_lock(&mp->mnt_listmtx); VNASSERT(vp->v_mflag & VMP_LAZYLIST, vp, ("lost VMP_LAZYLIST")); /* * Don't remove the vnode from the lazy list if another thread * has increased the hold count. It may have re-enqueued the * vnode to the lazy list and is now responsible for its * removal. */ if (vp->v_holdcnt == 0) { vp->v_mflag &= ~VMP_LAZYLIST; TAILQ_REMOVE(&mp->mnt_lazyvnodelist, vp, v_lazylist); mp->mnt_lazyvnodelistsize--; } mtx_unlock(&mp->mnt_listmtx); } vdbatch_enqueue(vp); } void vdrop(struct vnode *vp) { ASSERT_VI_UNLOCKED(vp, __func__); CTR2(KTR_VFS, "%s: vp %p", __func__, vp); if (refcount_release_if_not_last(&vp->v_holdcnt)) return; VI_LOCK(vp); vdropl(vp); } void vdropl(struct vnode *vp) { ASSERT_VI_LOCKED(vp, __func__); CTR2(KTR_VFS, "%s: vp %p", __func__, vp); if (!refcount_release(&vp->v_holdcnt)) { VI_UNLOCK(vp); return; } if (VN_IS_DOOMED(vp)) { freevnode(vp); return; } vdrop_deactivate(vp); } /* * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT * flags. DOINGINACT prevents us from recursing in calls to vinactive. */ static void vinactivef(struct vnode *vp) { struct vm_object *obj; ASSERT_VOP_ELOCKED(vp, "vinactive"); ASSERT_VI_LOCKED(vp, "vinactive"); VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp, ("vinactive: recursed on VI_DOINGINACT")); CTR2(KTR_VFS, "%s: vp %p", __func__, vp); vp->v_iflag |= VI_DOINGINACT; vp->v_iflag &= ~VI_OWEINACT; VI_UNLOCK(vp); /* * Before moving off the active list, we must be sure that any * modified pages are converted into the vnode's dirty * buffers, since these will no longer be checked once the * vnode is on the inactive list. * * The write-out of the dirty pages is asynchronous. At the * point that VOP_INACTIVE() is called, there could still be * pending I/O and dirty pages in the object. */ if ((obj = vp->v_object) != NULL && (vp->v_vflag & VV_NOSYNC) == 0 && vm_object_mightbedirty(obj)) { VM_OBJECT_WLOCK(obj); vm_object_page_clean(obj, 0, 0, 0); VM_OBJECT_WUNLOCK(obj); } VOP_INACTIVE(vp, curthread); VI_LOCK(vp); VNASSERT(vp->v_iflag & VI_DOINGINACT, vp, ("vinactive: lost VI_DOINGINACT")); vp->v_iflag &= ~VI_DOINGINACT; } void vinactive(struct vnode *vp) { ASSERT_VOP_ELOCKED(vp, "vinactive"); ASSERT_VI_LOCKED(vp, "vinactive"); CTR2(KTR_VFS, "%s: vp %p", __func__, vp); if ((vp->v_iflag & VI_OWEINACT) == 0) return; if (vp->v_iflag & VI_DOINGINACT) return; if (vp->v_usecount > 0) { vp->v_iflag &= ~VI_OWEINACT; return; } vinactivef(vp); } /* * Remove any vnodes in the vnode table belonging to mount point mp. * * If FORCECLOSE is not specified, there should not be any active ones, * return error if any are found (nb: this is a user error, not a * system error). If FORCECLOSE is specified, detach any active vnodes * that are found. * * If WRITECLOSE is set, only flush out regular file vnodes open for * writing. * * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped. * * `rootrefs' specifies the base reference count for the root vnode * of this filesystem. The root vnode is considered busy if its * v_usecount exceeds this value. On a successful return, vflush(, td) * will call vrele() on the root vnode exactly rootrefs times. * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must * be zero. */ #ifdef DIAGNOSTIC static int busyprt = 0; /* print out busy vnodes */ SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "Print out busy vnodes"); #endif int vflush(struct mount *mp, int rootrefs, int flags, struct thread *td) { struct vnode *vp, *mvp, *rootvp = NULL; struct vattr vattr; int busy = 0, error; CTR4(KTR_VFS, "%s: mp %p with rootrefs %d and flags %d", __func__, mp, rootrefs, flags); if (rootrefs > 0) { KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0, ("vflush: bad args")); /* * Get the filesystem root vnode. We can vput() it * immediately, since with rootrefs > 0, it won't go away. */ if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp)) != 0) { CTR2(KTR_VFS, "%s: vfs_root lookup failed with %d", __func__, error); return (error); } vput(rootvp); } loop: MNT_VNODE_FOREACH_ALL(vp, mp, mvp) { vholdl(vp); error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE); if (error) { vdrop(vp); MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); goto loop; } /* * Skip over a vnodes marked VV_SYSTEM. */ if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) { VOP_UNLOCK(vp); vdrop(vp); continue; } /* * If WRITECLOSE is set, flush out unlinked but still open * files (even if open only for reading) and regular file * vnodes open for writing. */ if (flags & WRITECLOSE) { if (vp->v_object != NULL) { VM_OBJECT_WLOCK(vp->v_object); vm_object_page_clean(vp->v_object, 0, 0, 0); VM_OBJECT_WUNLOCK(vp->v_object); } error = VOP_FSYNC(vp, MNT_WAIT, td); if (error != 0) { VOP_UNLOCK(vp); vdrop(vp); MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); return (error); } error = VOP_GETATTR(vp, &vattr, td->td_ucred); VI_LOCK(vp); if ((vp->v_type == VNON || (error == 0 && vattr.va_nlink > 0)) && (vp->v_writecount <= 0 || vp->v_type != VREG)) { VOP_UNLOCK(vp); vdropl(vp); continue; } } else VI_LOCK(vp); /* * With v_usecount == 0, all we need to do is clear out the * vnode data structures and we are done. * * If FORCECLOSE is set, forcibly close the vnode. */ if (vp->v_usecount == 0 || (flags & FORCECLOSE)) { vgonel(vp); } else { busy++; #ifdef DIAGNOSTIC if (busyprt) vn_printf(vp, "vflush: busy vnode "); #endif } VOP_UNLOCK(vp); vdropl(vp); } if (rootrefs > 0 && (flags & FORCECLOSE) == 0) { /* * If just the root vnode is busy, and if its refcount * is equal to `rootrefs', then go ahead and kill it. */ VI_LOCK(rootvp); KASSERT(busy > 0, ("vflush: not busy")); VNASSERT(rootvp->v_usecount >= rootrefs, rootvp, ("vflush: usecount %d < rootrefs %d", rootvp->v_usecount, rootrefs)); if (busy == 1 && rootvp->v_usecount == rootrefs) { VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK); vgone(rootvp); VOP_UNLOCK(rootvp); busy = 0; } else VI_UNLOCK(rootvp); } if (busy) { CTR2(KTR_VFS, "%s: failing as %d vnodes are busy", __func__, busy); return (EBUSY); } for (; rootrefs > 0; rootrefs--) vrele(rootvp); return (0); } /* * Recycle an unused vnode to the front of the free list. */ int vrecycle(struct vnode *vp) { int recycled; VI_LOCK(vp); recycled = vrecyclel(vp); VI_UNLOCK(vp); return (recycled); } /* * vrecycle, with the vp interlock held. */ int vrecyclel(struct vnode *vp) { int recycled; ASSERT_VOP_ELOCKED(vp, __func__); ASSERT_VI_LOCKED(vp, __func__); CTR2(KTR_VFS, "%s: vp %p", __func__, vp); recycled = 0; if (vp->v_usecount == 0) { recycled = 1; vgonel(vp); } return (recycled); } /* * Eliminate all activity associated with a vnode * in preparation for reuse. */ void vgone(struct vnode *vp) { VI_LOCK(vp); vgonel(vp); VI_UNLOCK(vp); } static void notify_lowervp_vfs_dummy(struct mount *mp __unused, struct vnode *lowervp __unused) { } /* * Notify upper mounts about reclaimed or unlinked vnode. */ void vfs_notify_upper(struct vnode *vp, int event) { static struct vfsops vgonel_vfsops = { .vfs_reclaim_lowervp = notify_lowervp_vfs_dummy, .vfs_unlink_lowervp = notify_lowervp_vfs_dummy, }; struct mount *mp, *ump, *mmp; mp = vp->v_mount; if (mp == NULL) return; if (TAILQ_EMPTY(&mp->mnt_uppers)) return; mmp = malloc(sizeof(struct mount), M_TEMP, M_WAITOK | M_ZERO); mmp->mnt_op = &vgonel_vfsops; mmp->mnt_kern_flag |= MNTK_MARKER; MNT_ILOCK(mp); mp->mnt_kern_flag |= MNTK_VGONE_UPPER; for (ump = TAILQ_FIRST(&mp->mnt_uppers); ump != NULL;) { if ((ump->mnt_kern_flag & MNTK_MARKER) != 0) { ump = TAILQ_NEXT(ump, mnt_upper_link); continue; } TAILQ_INSERT_AFTER(&mp->mnt_uppers, ump, mmp, mnt_upper_link); MNT_IUNLOCK(mp); switch (event) { case VFS_NOTIFY_UPPER_RECLAIM: VFS_RECLAIM_LOWERVP(ump, vp); break; case VFS_NOTIFY_UPPER_UNLINK: VFS_UNLINK_LOWERVP(ump, vp); break; default: KASSERT(0, ("invalid event %d", event)); break; } MNT_ILOCK(mp); ump = TAILQ_NEXT(mmp, mnt_upper_link); TAILQ_REMOVE(&mp->mnt_uppers, mmp, mnt_upper_link); } free(mmp, M_TEMP); mp->mnt_kern_flag &= ~MNTK_VGONE_UPPER; if ((mp->mnt_kern_flag & MNTK_VGONE_WAITER) != 0) { mp->mnt_kern_flag &= ~MNTK_VGONE_WAITER; wakeup(&mp->mnt_uppers); } MNT_IUNLOCK(mp); } /* * vgone, with the vp interlock held. */ static void vgonel(struct vnode *vp) { struct thread *td; struct mount *mp; vm_object_t object; bool active, oweinact; ASSERT_VOP_ELOCKED(vp, "vgonel"); ASSERT_VI_LOCKED(vp, "vgonel"); VNASSERT(vp->v_holdcnt, vp, ("vgonel: vp %p has no reference.", vp)); CTR2(KTR_VFS, "%s: vp %p", __func__, vp); td = curthread; /* * Don't vgonel if we're already doomed. */ if (vp->v_irflag & VIRF_DOOMED) return; vunlazy_gone(vp); vp->v_irflag |= VIRF_DOOMED; /* * Check to see if the vnode is in use. If so, we have to call * VOP_CLOSE() and VOP_INACTIVE(). */ active = vp->v_usecount > 0; oweinact = (vp->v_iflag & VI_OWEINACT) != 0; /* * If we need to do inactive VI_OWEINACT will be set. */ if (vp->v_iflag & VI_DEFINACT) { VNASSERT(vp->v_holdcnt > 1, vp, ("lost hold count")); vp->v_iflag &= ~VI_DEFINACT; vdropl(vp); } else { VNASSERT(vp->v_holdcnt > 0, vp, ("vnode without hold count")); VI_UNLOCK(vp); } vfs_notify_upper(vp, VFS_NOTIFY_UPPER_RECLAIM); /* * If purging an active vnode, it must be closed and * deactivated before being reclaimed. */ if (active) VOP_CLOSE(vp, FNONBLOCK, NOCRED, td); if (oweinact || active) { VI_LOCK(vp); vinactivef(vp); VI_UNLOCK(vp); } if (vp->v_type == VSOCK) vfs_unp_reclaim(vp); /* * Clean out any buffers associated with the vnode. * If the flush fails, just toss the buffers. */ mp = NULL; if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd)) (void) vn_start_secondary_write(vp, &mp, V_WAIT); if (vinvalbuf(vp, V_SAVE, 0, 0) != 0) { while (vinvalbuf(vp, 0, 0, 0) != 0) ; } BO_LOCK(&vp->v_bufobj); KASSERT(TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd) && vp->v_bufobj.bo_dirty.bv_cnt == 0 && TAILQ_EMPTY(&vp->v_bufobj.bo_clean.bv_hd) && vp->v_bufobj.bo_clean.bv_cnt == 0, ("vp %p bufobj not invalidated", vp)); /* * For VMIO bufobj, BO_DEAD is set later, or in * vm_object_terminate() after the object's page queue is * flushed. */ object = vp->v_bufobj.bo_object; if (object == NULL) vp->v_bufobj.bo_flag |= BO_DEAD; BO_UNLOCK(&vp->v_bufobj); /* * Handle the VM part. Tmpfs handles v_object on its own (the * OBJT_VNODE check). Nullfs or other bypassing filesystems * should not touch the object borrowed from the lower vnode * (the handle check). */ if (object != NULL && object->type == OBJT_VNODE && object->handle == vp) vnode_destroy_vobject(vp); /* * Reclaim the vnode. */ if (VOP_RECLAIM(vp, td)) panic("vgone: cannot reclaim"); if (mp != NULL) vn_finished_secondary_write(mp); VNASSERT(vp->v_object == NULL, vp, ("vop_reclaim left v_object vp=%p", vp)); /* * Clear the advisory locks and wake up waiting threads. */ (void)VOP_ADVLOCKPURGE(vp); vp->v_lockf = NULL; /* * Delete from old mount point vnode list. */ delmntque(vp); cache_purge(vp); /* * Done with purge, reset to the standard lock and invalidate * the vnode. */ VI_LOCK(vp); vp->v_vnlock = &vp->v_lock; vp->v_op = &dead_vnodeops; vp->v_type = VBAD; } /* * Calculate the total number of references to a special device. */ int vcount(struct vnode *vp) { int count; dev_lock(); count = vp->v_rdev->si_usecount; dev_unlock(); return (count); } /* * Print out a description of a vnode. */ static const char * const typename[] = {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD", "VMARKER"}; void vn_printf(struct vnode *vp, const char *fmt, ...) { va_list ap; char buf[256], buf2[16]; u_long flags; va_start(ap, fmt); vprintf(fmt, ap); va_end(ap); printf("%p: ", (void *)vp); printf("type %s\n", typename[vp->v_type]); printf(" usecount %d, writecount %d, refcount %d", vp->v_usecount, vp->v_writecount, vp->v_holdcnt); switch (vp->v_type) { case VDIR: printf(" mountedhere %p\n", vp->v_mountedhere); break; case VCHR: printf(" rdev %p\n", vp->v_rdev); break; case VSOCK: printf(" socket %p\n", vp->v_unpcb); break; case VFIFO: printf(" fifoinfo %p\n", vp->v_fifoinfo); break; default: printf("\n"); break; } buf[0] = '\0'; buf[1] = '\0'; if (vp->v_irflag & VIRF_DOOMED) strlcat(buf, "|VIRF_DOOMED", sizeof(buf)); flags = vp->v_irflag & ~(VIRF_DOOMED); if (flags != 0) { snprintf(buf2, sizeof(buf2), "|VIRF(0x%lx)", flags); strlcat(buf, buf2, sizeof(buf)); } if (vp->v_vflag & VV_ROOT) strlcat(buf, "|VV_ROOT", sizeof(buf)); if (vp->v_vflag & VV_ISTTY) strlcat(buf, "|VV_ISTTY", sizeof(buf)); if (vp->v_vflag & VV_NOSYNC) strlcat(buf, "|VV_NOSYNC", sizeof(buf)); if (vp->v_vflag & VV_ETERNALDEV) strlcat(buf, "|VV_ETERNALDEV", sizeof(buf)); if (vp->v_vflag & VV_CACHEDLABEL) strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf)); if (vp->v_vflag & VV_VMSIZEVNLOCK) strlcat(buf, "|VV_VMSIZEVNLOCK", sizeof(buf)); if (vp->v_vflag & VV_COPYONWRITE) strlcat(buf, "|VV_COPYONWRITE", sizeof(buf)); if (vp->v_vflag & VV_SYSTEM) strlcat(buf, "|VV_SYSTEM", sizeof(buf)); if (vp->v_vflag & VV_PROCDEP) strlcat(buf, "|VV_PROCDEP", sizeof(buf)); if (vp->v_vflag & VV_NOKNOTE) strlcat(buf, "|VV_NOKNOTE", sizeof(buf)); if (vp->v_vflag & VV_DELETED) strlcat(buf, "|VV_DELETED", sizeof(buf)); if (vp->v_vflag & VV_MD) strlcat(buf, "|VV_MD", sizeof(buf)); if (vp->v_vflag & VV_FORCEINSMQ) strlcat(buf, "|VV_FORCEINSMQ", sizeof(buf)); if (vp->v_vflag & VV_READLINK) strlcat(buf, "|VV_READLINK", sizeof(buf)); flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC | VV_ETERNALDEV | VV_CACHEDLABEL | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP | VV_NOKNOTE | VV_DELETED | VV_MD | VV_FORCEINSMQ); if (flags != 0) { snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags); strlcat(buf, buf2, sizeof(buf)); } if (vp->v_iflag & VI_TEXT_REF) strlcat(buf, "|VI_TEXT_REF", sizeof(buf)); if (vp->v_iflag & VI_MOUNT) strlcat(buf, "|VI_MOUNT", sizeof(buf)); if (vp->v_iflag & VI_DOINGINACT) strlcat(buf, "|VI_DOINGINACT", sizeof(buf)); if (vp->v_iflag & VI_OWEINACT) strlcat(buf, "|VI_OWEINACT", sizeof(buf)); if (vp->v_iflag & VI_DEFINACT) strlcat(buf, "|VI_DEFINACT", sizeof(buf)); flags = vp->v_iflag & ~(VI_TEXT_REF | VI_MOUNT | VI_DOINGINACT | VI_OWEINACT | VI_DEFINACT); if (flags != 0) { snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags); strlcat(buf, buf2, sizeof(buf)); } if (vp->v_mflag & VMP_LAZYLIST) strlcat(buf, "|VMP_LAZYLIST", sizeof(buf)); flags = vp->v_mflag & ~(VMP_LAZYLIST); if (flags != 0) { snprintf(buf2, sizeof(buf2), "|VMP(0x%lx)", flags); strlcat(buf, buf2, sizeof(buf)); } printf(" flags (%s)\n", buf + 1); if (mtx_owned(VI_MTX(vp))) printf(" VI_LOCKed"); if (vp->v_object != NULL) printf(" v_object %p ref %d pages %d " "cleanbuf %d dirtybuf %d\n", vp->v_object, vp->v_object->ref_count, vp->v_object->resident_page_count, vp->v_bufobj.bo_clean.bv_cnt, vp->v_bufobj.bo_dirty.bv_cnt); printf(" "); lockmgr_printinfo(vp->v_vnlock); if (vp->v_data != NULL) VOP_PRINT(vp); } #ifdef DDB /* * List all of the locked vnodes in the system. * Called when debugging the kernel. */ DB_SHOW_COMMAND(lockedvnods, lockedvnodes) { struct mount *mp; struct vnode *vp; /* * Note: because this is DDB, we can't obey the locking semantics * for these structures, which means we could catch an inconsistent * state and dereference a nasty pointer. Not much to be done * about that. */ db_printf("Locked vnodes\n"); TAILQ_FOREACH(mp, &mountlist, mnt_list) { TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { if (vp->v_type != VMARKER && VOP_ISLOCKED(vp)) vn_printf(vp, "vnode "); } } } /* * Show details about the given vnode. */ DB_SHOW_COMMAND(vnode, db_show_vnode) { struct vnode *vp; if (!have_addr) return; vp = (struct vnode *)addr; vn_printf(vp, "vnode "); } /* * Show details about the given mount point. */ DB_SHOW_COMMAND(mount, db_show_mount) { struct mount *mp; struct vfsopt *opt; struct statfs *sp; struct vnode *vp; char buf[512]; uint64_t mflags; u_int flags; if (!have_addr) { /* No address given, print short info about all mount points. */ TAILQ_FOREACH(mp, &mountlist, mnt_list) { db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname, mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename); if (db_pager_quit) break; } db_printf("\nMore info: show mount \n"); return; } mp = (struct mount *)addr; db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname, mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename); buf[0] = '\0'; mflags = mp->mnt_flag; #define MNT_FLAG(flag) do { \ if (mflags & (flag)) { \ if (buf[0] != '\0') \ strlcat(buf, ", ", sizeof(buf)); \ strlcat(buf, (#flag) + 4, sizeof(buf)); \ mflags &= ~(flag); \ } \ } while (0) MNT_FLAG(MNT_RDONLY); MNT_FLAG(MNT_SYNCHRONOUS); MNT_FLAG(MNT_NOEXEC); MNT_FLAG(MNT_NOSUID); MNT_FLAG(MNT_NFS4ACLS); MNT_FLAG(MNT_UNION); MNT_FLAG(MNT_ASYNC); MNT_FLAG(MNT_SUIDDIR); MNT_FLAG(MNT_SOFTDEP); MNT_FLAG(MNT_NOSYMFOLLOW); MNT_FLAG(MNT_GJOURNAL); MNT_FLAG(MNT_MULTILABEL); MNT_FLAG(MNT_ACLS); MNT_FLAG(MNT_NOATIME); MNT_FLAG(MNT_NOCLUSTERR); MNT_FLAG(MNT_NOCLUSTERW); MNT_FLAG(MNT_SUJ); MNT_FLAG(MNT_EXRDONLY); MNT_FLAG(MNT_EXPORTED); MNT_FLAG(MNT_DEFEXPORTED); MNT_FLAG(MNT_EXPORTANON); MNT_FLAG(MNT_EXKERB); MNT_FLAG(MNT_EXPUBLIC); MNT_FLAG(MNT_LOCAL); MNT_FLAG(MNT_QUOTA); MNT_FLAG(MNT_ROOTFS); MNT_FLAG(MNT_USER); MNT_FLAG(MNT_IGNORE); MNT_FLAG(MNT_UPDATE); MNT_FLAG(MNT_DELEXPORT); MNT_FLAG(MNT_RELOAD); MNT_FLAG(MNT_FORCE); MNT_FLAG(MNT_SNAPSHOT); MNT_FLAG(MNT_BYFSID); #undef MNT_FLAG if (mflags != 0) { if (buf[0] != '\0') strlcat(buf, ", ", sizeof(buf)); snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf), "0x%016jx", mflags); } db_printf(" mnt_flag = %s\n", buf); buf[0] = '\0'; flags = mp->mnt_kern_flag; #define MNT_KERN_FLAG(flag) do { \ if (flags & (flag)) { \ if (buf[0] != '\0') \ strlcat(buf, ", ", sizeof(buf)); \ strlcat(buf, (#flag) + 5, sizeof(buf)); \ flags &= ~(flag); \ } \ } while (0) MNT_KERN_FLAG(MNTK_UNMOUNTF); MNT_KERN_FLAG(MNTK_ASYNC); MNT_KERN_FLAG(MNTK_SOFTDEP); MNT_KERN_FLAG(MNTK_DRAINING); MNT_KERN_FLAG(MNTK_REFEXPIRE); MNT_KERN_FLAG(MNTK_EXTENDED_SHARED); MNT_KERN_FLAG(MNTK_SHARED_WRITES); MNT_KERN_FLAG(MNTK_NO_IOPF); MNT_KERN_FLAG(MNTK_VGONE_UPPER); MNT_KERN_FLAG(MNTK_VGONE_WAITER); MNT_KERN_FLAG(MNTK_LOOKUP_EXCL_DOTDOT); MNT_KERN_FLAG(MNTK_MARKER); MNT_KERN_FLAG(MNTK_USES_BCACHE); MNT_KERN_FLAG(MNTK_NOASYNC); MNT_KERN_FLAG(MNTK_UNMOUNT); MNT_KERN_FLAG(MNTK_MWAIT); MNT_KERN_FLAG(MNTK_SUSPEND); MNT_KERN_FLAG(MNTK_SUSPEND2); MNT_KERN_FLAG(MNTK_SUSPENDED); MNT_KERN_FLAG(MNTK_LOOKUP_SHARED); MNT_KERN_FLAG(MNTK_NOKNOTE); #undef MNT_KERN_FLAG if (flags != 0) { if (buf[0] != '\0') strlcat(buf, ", ", sizeof(buf)); snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf), "0x%08x", flags); } db_printf(" mnt_kern_flag = %s\n", buf); db_printf(" mnt_opt = "); opt = TAILQ_FIRST(mp->mnt_opt); if (opt != NULL) { db_printf("%s", opt->name); opt = TAILQ_NEXT(opt, link); while (opt != NULL) { db_printf(", %s", opt->name); opt = TAILQ_NEXT(opt, link); } } db_printf("\n"); sp = &mp->mnt_stat; db_printf(" mnt_stat = { version=%u type=%u flags=0x%016jx " "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju " "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju " "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n", (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags, (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize, (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree, (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files, (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites, (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads, (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax, (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]); db_printf(" mnt_cred = { uid=%u ruid=%u", (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid); if (jailed(mp->mnt_cred)) db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id); db_printf(" }\n"); db_printf(" mnt_ref = %d (with %d in the struct)\n", vfs_mount_fetch_counter(mp, MNT_COUNT_REF), mp->mnt_ref); db_printf(" mnt_gen = %d\n", mp->mnt_gen); db_printf(" mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize); db_printf(" mnt_lazyvnodelistsize = %d\n", mp->mnt_lazyvnodelistsize); db_printf(" mnt_writeopcount = %d (with %d in the struct)\n", vfs_mount_fetch_counter(mp, MNT_COUNT_WRITEOPCOUNT), mp->mnt_writeopcount); db_printf(" mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen); db_printf(" mnt_iosize_max = %d\n", mp->mnt_iosize_max); db_printf(" mnt_hashseed = %u\n", mp->mnt_hashseed); db_printf(" mnt_lockref = %d (with %d in the struct)\n", vfs_mount_fetch_counter(mp, MNT_COUNT_LOCKREF), mp->mnt_lockref); db_printf(" mnt_secondary_writes = %d\n", mp->mnt_secondary_writes); db_printf(" mnt_secondary_accwrites = %d\n", mp->mnt_secondary_accwrites); db_printf(" mnt_gjprovider = %s\n", mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL"); db_printf(" mnt_vfs_ops = %d\n", mp->mnt_vfs_ops); db_printf("\n\nList of active vnodes\n"); TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { if (vp->v_type != VMARKER && vp->v_holdcnt > 0) { vn_printf(vp, "vnode "); if (db_pager_quit) break; } } db_printf("\n\nList of inactive vnodes\n"); TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { if (vp->v_type != VMARKER && vp->v_holdcnt == 0) { vn_printf(vp, "vnode "); if (db_pager_quit) break; } } } #endif /* DDB */ /* * Fill in a struct xvfsconf based on a struct vfsconf. */ static int vfsconf2x(struct sysctl_req *req, struct vfsconf *vfsp) { struct xvfsconf xvfsp; bzero(&xvfsp, sizeof(xvfsp)); strcpy(xvfsp.vfc_name, vfsp->vfc_name); xvfsp.vfc_typenum = vfsp->vfc_typenum; xvfsp.vfc_refcount = vfsp->vfc_refcount; xvfsp.vfc_flags = vfsp->vfc_flags; /* * These are unused in userland, we keep them * to not break binary compatibility. */ xvfsp.vfc_vfsops = NULL; xvfsp.vfc_next = NULL; return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp))); } #ifdef COMPAT_FREEBSD32 struct xvfsconf32 { uint32_t vfc_vfsops; char vfc_name[MFSNAMELEN]; int32_t vfc_typenum; int32_t vfc_refcount; int32_t vfc_flags; uint32_t vfc_next; }; static int vfsconf2x32(struct sysctl_req *req, struct vfsconf *vfsp) { struct xvfsconf32 xvfsp; bzero(&xvfsp, sizeof(xvfsp)); strcpy(xvfsp.vfc_name, vfsp->vfc_name); xvfsp.vfc_typenum = vfsp->vfc_typenum; xvfsp.vfc_refcount = vfsp->vfc_refcount; xvfsp.vfc_flags = vfsp->vfc_flags; return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp))); } #endif /* * Top level filesystem related information gathering. */ static int sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS) { struct vfsconf *vfsp; int error; error = 0; vfsconf_slock(); TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) { #ifdef COMPAT_FREEBSD32 if (req->flags & SCTL_MASK32) error = vfsconf2x32(req, vfsp); else #endif error = vfsconf2x(req, vfsp); if (error) break; } vfsconf_sunlock(); return (error); } SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0, sysctl_vfs_conflist, "S,xvfsconf", "List of all configured filesystems"); #ifndef BURN_BRIDGES static int sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS); static int vfs_sysctl(SYSCTL_HANDLER_ARGS) { int *name = (int *)arg1 - 1; /* XXX */ u_int namelen = arg2 + 1; /* XXX */ struct vfsconf *vfsp; log(LOG_WARNING, "userland calling deprecated sysctl, " "please rebuild world\n"); #if 1 || defined(COMPAT_PRELITE2) /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */ if (namelen == 1) return (sysctl_ovfs_conf(oidp, arg1, arg2, req)); #endif switch (name[1]) { case VFS_MAXTYPENUM: if (namelen != 2) return (ENOTDIR); return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int))); case VFS_CONF: if (namelen != 3) return (ENOTDIR); /* overloaded */ vfsconf_slock(); TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) { if (vfsp->vfc_typenum == name[2]) break; } vfsconf_sunlock(); if (vfsp == NULL) return (EOPNOTSUPP); #ifdef COMPAT_FREEBSD32 if (req->flags & SCTL_MASK32) return (vfsconf2x32(req, vfsp)); else #endif return (vfsconf2x(req, vfsp)); } return (EOPNOTSUPP); } static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP | CTLFLAG_MPSAFE, vfs_sysctl, "Generic filesystem"); #if 1 || defined(COMPAT_PRELITE2) static int sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS) { int error; struct vfsconf *vfsp; struct ovfsconf ovfs; vfsconf_slock(); TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) { bzero(&ovfs, sizeof(ovfs)); ovfs.vfc_vfsops = vfsp->vfc_vfsops; /* XXX used as flag */ strcpy(ovfs.vfc_name, vfsp->vfc_name); ovfs.vfc_index = vfsp->vfc_typenum; ovfs.vfc_refcount = vfsp->vfc_refcount; ovfs.vfc_flags = vfsp->vfc_flags; error = SYSCTL_OUT(req, &ovfs, sizeof ovfs); if (error != 0) { vfsconf_sunlock(); return (error); } } vfsconf_sunlock(); return (0); } #endif /* 1 || COMPAT_PRELITE2 */ #endif /* !BURN_BRIDGES */ #define KINFO_VNODESLOP 10 #ifdef notyet /* * Dump vnode list (via sysctl). */ /* ARGSUSED */ static int sysctl_vnode(SYSCTL_HANDLER_ARGS) { struct xvnode *xvn; struct mount *mp; struct vnode *vp; int error, len, n; /* * Stale numvnodes access is not fatal here. */ req->lock = 0; len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn; if (!req->oldptr) /* Make an estimate */ return (SYSCTL_OUT(req, 0, len)); error = sysctl_wire_old_buffer(req, 0); if (error != 0) return (error); xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK); n = 0; mtx_lock(&mountlist_mtx); TAILQ_FOREACH(mp, &mountlist, mnt_list) { if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) continue; MNT_ILOCK(mp); TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { if (n == len) break; vref(vp); xvn[n].xv_size = sizeof *xvn; xvn[n].xv_vnode = vp; xvn[n].xv_id = 0; /* XXX compat */ #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field XV_COPY(usecount); XV_COPY(writecount); XV_COPY(holdcnt); XV_COPY(mount); XV_COPY(numoutput); XV_COPY(type); #undef XV_COPY xvn[n].xv_flag = vp->v_vflag; switch (vp->v_type) { case VREG: case VDIR: case VLNK: break; case VBLK: case VCHR: if (vp->v_rdev == NULL) { vrele(vp); continue; } xvn[n].xv_dev = dev2udev(vp->v_rdev); break; case VSOCK: xvn[n].xv_socket = vp->v_socket; break; case VFIFO: xvn[n].xv_fifo = vp->v_fifoinfo; break; case VNON: case VBAD: default: /* shouldn't happen? */ vrele(vp); continue; } vrele(vp); ++n; } MNT_IUNLOCK(mp); mtx_lock(&mountlist_mtx); vfs_unbusy(mp); if (n == len) break; } mtx_unlock(&mountlist_mtx); error = SYSCTL_OUT(req, xvn, n * sizeof *xvn); free(xvn, M_TEMP); return (error); } SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE, 0, 0, sysctl_vnode, "S,xvnode", ""); #endif static void unmount_or_warn(struct mount *mp) { int error; error = dounmount(mp, MNT_FORCE, curthread); if (error != 0) { printf("unmount of %s failed (", mp->mnt_stat.f_mntonname); if (error == EBUSY) printf("BUSY)\n"); else printf("%d)\n", error); } } /* * Unmount all filesystems. The list is traversed in reverse order * of mounting to avoid dependencies. */ void vfs_unmountall(void) { struct mount *mp, *tmp; CTR1(KTR_VFS, "%s: unmounting all filesystems", __func__); /* * Since this only runs when rebooting, it is not interlocked. */ TAILQ_FOREACH_REVERSE_SAFE(mp, &mountlist, mntlist, mnt_list, tmp) { vfs_ref(mp); /* * Forcibly unmounting "/dev" before "/" would prevent clean * unmount of the latter. */ if (mp == rootdevmp) continue; unmount_or_warn(mp); } if (rootdevmp != NULL) unmount_or_warn(rootdevmp); } static void vfs_deferred_inactive(struct vnode *vp, int lkflags) { ASSERT_VI_LOCKED(vp, __func__); VNASSERT((vp->v_iflag & VI_DEFINACT) == 0, vp, ("VI_DEFINACT still set")); if ((vp->v_iflag & VI_OWEINACT) == 0) { vdropl(vp); return; } if (vn_lock(vp, lkflags) == 0) { VI_LOCK(vp); vinactive(vp); VOP_UNLOCK(vp); vdropl(vp); return; } vdefer_inactive_unlocked(vp); } static int vfs_periodic_inactive_filter(struct vnode *vp, void *arg) { return (vp->v_iflag & VI_DEFINACT); } static void __noinline vfs_periodic_inactive(struct mount *mp, int flags) { struct vnode *vp, *mvp; int lkflags; lkflags = LK_EXCLUSIVE | LK_INTERLOCK; if (flags != MNT_WAIT) lkflags |= LK_NOWAIT; MNT_VNODE_FOREACH_LAZY(vp, mp, mvp, vfs_periodic_inactive_filter, NULL) { if ((vp->v_iflag & VI_DEFINACT) == 0) { VI_UNLOCK(vp); continue; } vp->v_iflag &= ~VI_DEFINACT; vfs_deferred_inactive(vp, lkflags); } } static inline bool vfs_want_msync(struct vnode *vp) { struct vm_object *obj; /* * This test may be performed without any locks held. * We rely on vm_object's type stability. */ if (vp->v_vflag & VV_NOSYNC) return (false); obj = vp->v_object; return (obj != NULL && vm_object_mightbedirty(obj)); } static int vfs_periodic_msync_inactive_filter(struct vnode *vp, void *arg __unused) { if (vp->v_vflag & VV_NOSYNC) return (false); if (vp->v_iflag & VI_DEFINACT) return (true); return (vfs_want_msync(vp)); } static void __noinline vfs_periodic_msync_inactive(struct mount *mp, int flags) { struct vnode *vp, *mvp; struct vm_object *obj; struct thread *td; int lkflags, objflags; bool seen_defer; td = curthread; lkflags = LK_EXCLUSIVE | LK_INTERLOCK; if (flags != MNT_WAIT) { lkflags |= LK_NOWAIT; objflags = OBJPC_NOSYNC; } else { objflags = OBJPC_SYNC; } MNT_VNODE_FOREACH_LAZY(vp, mp, mvp, vfs_periodic_msync_inactive_filter, NULL) { seen_defer = false; if (vp->v_iflag & VI_DEFINACT) { vp->v_iflag &= ~VI_DEFINACT; seen_defer = true; } if (!vfs_want_msync(vp)) { if (seen_defer) vfs_deferred_inactive(vp, lkflags); else VI_UNLOCK(vp); continue; } if (vget(vp, lkflags, td) == 0) { obj = vp->v_object; if (obj != NULL && (vp->v_vflag & VV_NOSYNC) == 0) { VM_OBJECT_WLOCK(obj); vm_object_page_clean(obj, 0, 0, objflags); VM_OBJECT_WUNLOCK(obj); } vput(vp); if (seen_defer) vdrop(vp); } else { if (seen_defer) vdefer_inactive_unlocked(vp); } } } void vfs_periodic(struct mount *mp, int flags) { CTR2(KTR_VFS, "%s: mp %p", __func__, mp); if ((mp->mnt_kern_flag & MNTK_NOMSYNC) != 0) vfs_periodic_inactive(mp, flags); else vfs_periodic_msync_inactive(mp, flags); } static void destroy_vpollinfo_free(struct vpollinfo *vi) { knlist_destroy(&vi->vpi_selinfo.si_note); mtx_destroy(&vi->vpi_lock); uma_zfree(vnodepoll_zone, vi); } static void destroy_vpollinfo(struct vpollinfo *vi) { knlist_clear(&vi->vpi_selinfo.si_note, 1); seldrain(&vi->vpi_selinfo); destroy_vpollinfo_free(vi); } /* * Initialize per-vnode helper structure to hold poll-related state. */ void v_addpollinfo(struct vnode *vp) { struct vpollinfo *vi; if (vp->v_pollinfo != NULL) return; vi = uma_zalloc(vnodepoll_zone, M_WAITOK | M_ZERO); mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF); knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock, vfs_knlunlock, vfs_knl_assert_locked, vfs_knl_assert_unlocked); VI_LOCK(vp); if (vp->v_pollinfo != NULL) { VI_UNLOCK(vp); destroy_vpollinfo_free(vi); return; } vp->v_pollinfo = vi; VI_UNLOCK(vp); } /* * Record a process's interest in events which might happen to * a vnode. Because poll uses the historic select-style interface * internally, this routine serves as both the ``check for any * pending events'' and the ``record my interest in future events'' * functions. (These are done together, while the lock is held, * to avoid race conditions.) */ int vn_pollrecord(struct vnode *vp, struct thread *td, int events) { v_addpollinfo(vp); mtx_lock(&vp->v_pollinfo->vpi_lock); if (vp->v_pollinfo->vpi_revents & events) { /* * This leaves events we are not interested * in available for the other process which * which presumably had requested them * (otherwise they would never have been * recorded). */ events &= vp->v_pollinfo->vpi_revents; vp->v_pollinfo->vpi_revents &= ~events; mtx_unlock(&vp->v_pollinfo->vpi_lock); return (events); } vp->v_pollinfo->vpi_events |= events; selrecord(td, &vp->v_pollinfo->vpi_selinfo); mtx_unlock(&vp->v_pollinfo->vpi_lock); return (0); } /* * Routine to create and manage a filesystem syncer vnode. */ #define sync_close ((int (*)(struct vop_close_args *))nullop) static int sync_fsync(struct vop_fsync_args *); static int sync_inactive(struct vop_inactive_args *); static int sync_reclaim(struct vop_reclaim_args *); static struct vop_vector sync_vnodeops = { .vop_bypass = VOP_EOPNOTSUPP, .vop_close = sync_close, /* close */ .vop_fsync = sync_fsync, /* fsync */ .vop_inactive = sync_inactive, /* inactive */ .vop_need_inactive = vop_stdneed_inactive, /* need_inactive */ .vop_reclaim = sync_reclaim, /* reclaim */ .vop_lock1 = vop_stdlock, /* lock */ .vop_unlock = vop_stdunlock, /* unlock */ .vop_islocked = vop_stdislocked, /* islocked */ }; VFS_VOP_VECTOR_REGISTER(sync_vnodeops); /* * Create a new filesystem syncer vnode for the specified mount point. */ void vfs_allocate_syncvnode(struct mount *mp) { struct vnode *vp; struct bufobj *bo; static long start, incr, next; int error; /* Allocate a new vnode */ error = getnewvnode("syncer", mp, &sync_vnodeops, &vp); if (error != 0) panic("vfs_allocate_syncvnode: getnewvnode() failed"); vp->v_type = VNON; vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); vp->v_vflag |= VV_FORCEINSMQ; error = insmntque(vp, mp); if (error != 0) panic("vfs_allocate_syncvnode: insmntque() failed"); vp->v_vflag &= ~VV_FORCEINSMQ; VOP_UNLOCK(vp); /* * Place the vnode onto the syncer worklist. We attempt to * scatter them about on the list so that they will go off * at evenly distributed times even if all the filesystems * are mounted at once. */ next += incr; if (next == 0 || next > syncer_maxdelay) { start /= 2; incr /= 2; if (start == 0) { start = syncer_maxdelay / 2; incr = syncer_maxdelay; } next = start; } bo = &vp->v_bufobj; BO_LOCK(bo); vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0); /* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */ mtx_lock(&sync_mtx); sync_vnode_count++; if (mp->mnt_syncer == NULL) { mp->mnt_syncer = vp; vp = NULL; } mtx_unlock(&sync_mtx); BO_UNLOCK(bo); if (vp != NULL) { vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); vgone(vp); vput(vp); } } void vfs_deallocate_syncvnode(struct mount *mp) { struct vnode *vp; mtx_lock(&sync_mtx); vp = mp->mnt_syncer; if (vp != NULL) mp->mnt_syncer = NULL; mtx_unlock(&sync_mtx); if (vp != NULL) vrele(vp); } /* * Do a lazy sync of the filesystem. */ static int sync_fsync(struct vop_fsync_args *ap) { struct vnode *syncvp = ap->a_vp; struct mount *mp = syncvp->v_mount; int error, save; struct bufobj *bo; /* * We only need to do something if this is a lazy evaluation. */ if (ap->a_waitfor != MNT_LAZY) return (0); /* * Move ourselves to the back of the sync list. */ bo = &syncvp->v_bufobj; BO_LOCK(bo); vn_syncer_add_to_worklist(bo, syncdelay); BO_UNLOCK(bo); /* * Walk the list of vnodes pushing all that are dirty and * not already on the sync list. */ if (vfs_busy(mp, MBF_NOWAIT) != 0) return (0); if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) { vfs_unbusy(mp); return (0); } save = curthread_pflags_set(TDP_SYNCIO); /* * The filesystem at hand may be idle with free vnodes stored in the * batch. Return them instead of letting them stay there indefinitely. */ vfs_periodic(mp, MNT_NOWAIT); error = VFS_SYNC(mp, MNT_LAZY); curthread_pflags_restore(save); vn_finished_write(mp); vfs_unbusy(mp); return (error); } /* * The syncer vnode is no referenced. */ static int sync_inactive(struct vop_inactive_args *ap) { vgone(ap->a_vp); return (0); } /* * The syncer vnode is no longer needed and is being decommissioned. * * Modifications to the worklist must be protected by sync_mtx. */ static int sync_reclaim(struct vop_reclaim_args *ap) { struct vnode *vp = ap->a_vp; struct bufobj *bo; bo = &vp->v_bufobj; BO_LOCK(bo); mtx_lock(&sync_mtx); if (vp->v_mount->mnt_syncer == vp) vp->v_mount->mnt_syncer = NULL; if (bo->bo_flag & BO_ONWORKLST) { LIST_REMOVE(bo, bo_synclist); syncer_worklist_len--; sync_vnode_count--; bo->bo_flag &= ~BO_ONWORKLST; } mtx_unlock(&sync_mtx); BO_UNLOCK(bo); return (0); } int vn_need_pageq_flush(struct vnode *vp) { struct vm_object *obj; int need; MPASS(mtx_owned(VI_MTX(vp))); need = 0; if ((obj = vp->v_object) != NULL && (vp->v_vflag & VV_NOSYNC) == 0 && vm_object_mightbedirty(obj)) need = 1; return (need); } /* * Check if vnode represents a disk device */ int vn_isdisk(struct vnode *vp, int *errp) { int error; if (vp->v_type != VCHR) { error = ENOTBLK; goto out; } error = 0; dev_lock(); if (vp->v_rdev == NULL) error = ENXIO; else if (vp->v_rdev->si_devsw == NULL) error = ENXIO; else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK)) error = ENOTBLK; dev_unlock(); out: if (errp != NULL) *errp = error; return (error == 0); } /* * Common filesystem object access control check routine. Accepts a * vnode's type, "mode", uid and gid, requested access mode, credentials, * and optional call-by-reference privused argument allowing vaccess() * to indicate to the caller whether privilege was used to satisfy the * request (obsoleted). Returns 0 on success, or an errno on failure. */ int vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid, accmode_t accmode, struct ucred *cred, int *privused) { accmode_t dac_granted; accmode_t priv_granted; KASSERT((accmode & ~(VEXEC | VWRITE | VREAD | VADMIN | VAPPEND)) == 0, ("invalid bit in accmode")); KASSERT((accmode & VAPPEND) == 0 || (accmode & VWRITE), ("VAPPEND without VWRITE")); /* * Look for a normal, non-privileged way to access the file/directory * as requested. If it exists, go with that. */ if (privused != NULL) *privused = 0; dac_granted = 0; /* Check the owner. */ if (cred->cr_uid == file_uid) { dac_granted |= VADMIN; if (file_mode & S_IXUSR) dac_granted |= VEXEC; if (file_mode & S_IRUSR) dac_granted |= VREAD; if (file_mode & S_IWUSR) dac_granted |= (VWRITE | VAPPEND); if ((accmode & dac_granted) == accmode) return (0); goto privcheck; } /* Otherwise, check the groups (first match) */ if (groupmember(file_gid, cred)) { if (file_mode & S_IXGRP) dac_granted |= VEXEC; if (file_mode & S_IRGRP) dac_granted |= VREAD; if (file_mode & S_IWGRP) dac_granted |= (VWRITE | VAPPEND); if ((accmode & dac_granted) == accmode) return (0); goto privcheck; } /* Otherwise, check everyone else. */ if (file_mode & S_IXOTH) dac_granted |= VEXEC; if (file_mode & S_IROTH) dac_granted |= VREAD; if (file_mode & S_IWOTH) dac_granted |= (VWRITE | VAPPEND); if ((accmode & dac_granted) == accmode) return (0); privcheck: /* * Build a privilege mask to determine if the set of privileges * satisfies the requirements when combined with the granted mask * from above. For each privilege, if the privilege is required, * bitwise or the request type onto the priv_granted mask. */ priv_granted = 0; if (type == VDIR) { /* * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC * requests, instead of PRIV_VFS_EXEC. */ if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) && !priv_check_cred(cred, PRIV_VFS_LOOKUP)) priv_granted |= VEXEC; } else { /* * Ensure that at least one execute bit is on. Otherwise, * a privileged user will always succeed, and we don't want * this to happen unless the file really is executable. */ if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) && (file_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) != 0 && !priv_check_cred(cred, PRIV_VFS_EXEC)) priv_granted |= VEXEC; } if ((accmode & VREAD) && ((dac_granted & VREAD) == 0) && !priv_check_cred(cred, PRIV_VFS_READ)) priv_granted |= VREAD; if ((accmode & VWRITE) && ((dac_granted & VWRITE) == 0) && !priv_check_cred(cred, PRIV_VFS_WRITE)) priv_granted |= (VWRITE | VAPPEND); if ((accmode & VADMIN) && ((dac_granted & VADMIN) == 0) && !priv_check_cred(cred, PRIV_VFS_ADMIN)) priv_granted |= VADMIN; if ((accmode & (priv_granted | dac_granted)) == accmode) { /* XXX audit: privilege used */ if (privused != NULL) *privused = 1; return (0); } return ((accmode & VADMIN) ? EPERM : EACCES); } /* * Credential check based on process requesting service, and per-attribute * permissions. */ int extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred, struct thread *td, accmode_t accmode) { /* * Kernel-invoked always succeeds. */ if (cred == NOCRED) return (0); /* * Do not allow privileged processes in jail to directly manipulate * system attributes. */ switch (attrnamespace) { case EXTATTR_NAMESPACE_SYSTEM: /* Potentially should be: return (EPERM); */ return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM)); case EXTATTR_NAMESPACE_USER: return (VOP_ACCESS(vp, accmode, cred, td)); default: return (EPERM); } } #ifdef DEBUG_VFS_LOCKS /* * This only exists to suppress warnings from unlocked specfs accesses. It is * no longer ok to have an unlocked VFS. */ #define IGNORE_LOCK(vp) (KERNEL_PANICKED() || (vp) == NULL || \ (vp)->v_type == VCHR || (vp)->v_type == VBAD) int vfs_badlock_ddb = 1; /* Drop into debugger on violation. */ SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0, "Drop into debugger on lock violation"); int vfs_badlock_mutex = 1; /* Check for interlock across VOPs. */ SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex, 0, "Check for interlock across VOPs"); int vfs_badlock_print = 1; /* Print lock violations. */ SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print, 0, "Print lock violations"); int vfs_badlock_vnode = 1; /* Print vnode details on lock violations. */ SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_vnode, CTLFLAG_RW, &vfs_badlock_vnode, 0, "Print vnode details on lock violations"); #ifdef KDB int vfs_badlock_backtrace = 1; /* Print backtrace at lock violations. */ SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW, &vfs_badlock_backtrace, 0, "Print backtrace at lock violations"); #endif static void vfs_badlock(const char *msg, const char *str, struct vnode *vp) { #ifdef KDB if (vfs_badlock_backtrace) kdb_backtrace(); #endif if (vfs_badlock_vnode) vn_printf(vp, "vnode "); if (vfs_badlock_print) printf("%s: %p %s\n", str, (void *)vp, msg); if (vfs_badlock_ddb) kdb_enter(KDB_WHY_VFSLOCK, "lock violation"); } void assert_vi_locked(struct vnode *vp, const char *str) { if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp))) vfs_badlock("interlock is not locked but should be", str, vp); } void assert_vi_unlocked(struct vnode *vp, const char *str) { if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp))) vfs_badlock("interlock is locked but should not be", str, vp); } void assert_vop_locked(struct vnode *vp, const char *str) { int locked; if (!IGNORE_LOCK(vp)) { locked = VOP_ISLOCKED(vp); if (locked == 0 || locked == LK_EXCLOTHER) vfs_badlock("is not locked but should be", str, vp); } } void assert_vop_unlocked(struct vnode *vp, const char *str) { if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == LK_EXCLUSIVE) vfs_badlock("is locked but should not be", str, vp); } void assert_vop_elocked(struct vnode *vp, const char *str) { if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE) vfs_badlock("is not exclusive locked but should be", str, vp); } #endif /* DEBUG_VFS_LOCKS */ void vop_rename_fail(struct vop_rename_args *ap) { if (ap->a_tvp != NULL) vput(ap->a_tvp); if (ap->a_tdvp == ap->a_tvp) vrele(ap->a_tdvp); else vput(ap->a_tdvp); vrele(ap->a_fdvp); vrele(ap->a_fvp); } void vop_rename_pre(void *ap) { struct vop_rename_args *a = ap; #ifdef DEBUG_VFS_LOCKS if (a->a_tvp) ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME"); ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME"); ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME"); ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME"); /* Check the source (from). */ if (a->a_tdvp->v_vnlock != a->a_fdvp->v_vnlock && (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fdvp->v_vnlock)) ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked"); if (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fvp->v_vnlock) ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked"); /* Check the target. */ if (a->a_tvp) ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked"); ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked"); #endif if (a->a_tdvp != a->a_fdvp) vhold(a->a_fdvp); if (a->a_tvp != a->a_fvp) vhold(a->a_fvp); vhold(a->a_tdvp); if (a->a_tvp) vhold(a->a_tvp); } #ifdef DEBUG_VFS_LOCKS void vop_strategy_pre(void *ap) { struct vop_strategy_args *a; struct buf *bp; a = ap; bp = a->a_bp; /* * Cluster ops lock their component buffers but not the IO container. */ if ((bp->b_flags & B_CLUSTER) != 0) return; if (!KERNEL_PANICKED() && !BUF_ISLOCKED(bp)) { if (vfs_badlock_print) printf( "VOP_STRATEGY: bp is not locked but should be\n"); if (vfs_badlock_ddb) kdb_enter(KDB_WHY_VFSLOCK, "lock violation"); } } void vop_lock_pre(void *ap) { struct vop_lock1_args *a = ap; if ((a->a_flags & LK_INTERLOCK) == 0) ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK"); else ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK"); } void vop_lock_post(void *ap, int rc) { struct vop_lock1_args *a = ap; ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK"); if (rc == 0 && (a->a_flags & LK_EXCLOTHER) == 0) ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK"); } void vop_unlock_pre(void *ap) { struct vop_unlock_args *a = ap; ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK"); } void vop_need_inactive_pre(void *ap) { struct vop_need_inactive_args *a = ap; ASSERT_VI_LOCKED(a->a_vp, "VOP_NEED_INACTIVE"); } void vop_need_inactive_post(void *ap, int rc) { struct vop_need_inactive_args *a = ap; ASSERT_VI_LOCKED(a->a_vp, "VOP_NEED_INACTIVE"); } #endif void vop_create_post(void *ap, int rc) { struct vop_create_args *a = ap; if (!rc) VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE); } void vop_deleteextattr_post(void *ap, int rc) { struct vop_deleteextattr_args *a = ap; if (!rc) VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB); } void vop_link_post(void *ap, int rc) { struct vop_link_args *a = ap; if (!rc) { VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK); VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE); } } void vop_mkdir_post(void *ap, int rc) { struct vop_mkdir_args *a = ap; if (!rc) VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK); } void vop_mknod_post(void *ap, int rc) { struct vop_mknod_args *a = ap; if (!rc) VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE); } void vop_reclaim_post(void *ap, int rc) { struct vop_reclaim_args *a = ap; if (!rc) VFS_KNOTE_LOCKED(a->a_vp, NOTE_REVOKE); } void vop_remove_post(void *ap, int rc) { struct vop_remove_args *a = ap; if (!rc) { VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE); VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE); } } void vop_rename_post(void *ap, int rc) { struct vop_rename_args *a = ap; long hint; if (!rc) { hint = NOTE_WRITE; if (a->a_fdvp == a->a_tdvp) { if (a->a_tvp != NULL && a->a_tvp->v_type == VDIR) hint |= NOTE_LINK; VFS_KNOTE_UNLOCKED(a->a_fdvp, hint); VFS_KNOTE_UNLOCKED(a->a_tdvp, hint); } else { hint |= NOTE_EXTEND; if (a->a_fvp->v_type == VDIR) hint |= NOTE_LINK; VFS_KNOTE_UNLOCKED(a->a_fdvp, hint); if (a->a_fvp->v_type == VDIR && a->a_tvp != NULL && a->a_tvp->v_type == VDIR) hint &= ~NOTE_LINK; VFS_KNOTE_UNLOCKED(a->a_tdvp, hint); } VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME); if (a->a_tvp) VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE); } if (a->a_tdvp != a->a_fdvp) vdrop(a->a_fdvp); if (a->a_tvp != a->a_fvp) vdrop(a->a_fvp); vdrop(a->a_tdvp); if (a->a_tvp) vdrop(a->a_tvp); } void vop_rmdir_post(void *ap, int rc) { struct vop_rmdir_args *a = ap; if (!rc) { VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK); VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE); } } void vop_setattr_post(void *ap, int rc) { struct vop_setattr_args *a = ap; if (!rc) VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB); } void vop_setextattr_post(void *ap, int rc) { struct vop_setextattr_args *a = ap; if (!rc) VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB); } void vop_symlink_post(void *ap, int rc) { struct vop_symlink_args *a = ap; if (!rc) VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE); } void vop_open_post(void *ap, int rc) { struct vop_open_args *a = ap; if (!rc) VFS_KNOTE_LOCKED(a->a_vp, NOTE_OPEN); } void vop_close_post(void *ap, int rc) { struct vop_close_args *a = ap; if (!rc && (a->a_cred != NOCRED || /* filter out revokes */ !VN_IS_DOOMED(a->a_vp))) { VFS_KNOTE_LOCKED(a->a_vp, (a->a_fflag & FWRITE) != 0 ? NOTE_CLOSE_WRITE : NOTE_CLOSE); } } void vop_read_post(void *ap, int rc) { struct vop_read_args *a = ap; if (!rc) VFS_KNOTE_LOCKED(a->a_vp, NOTE_READ); } void vop_readdir_post(void *ap, int rc) { struct vop_readdir_args *a = ap; if (!rc) VFS_KNOTE_LOCKED(a->a_vp, NOTE_READ); } static struct knlist fs_knlist; static void vfs_event_init(void *arg) { knlist_init_mtx(&fs_knlist, NULL); } /* XXX - correct order? */ SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL); void vfs_event_signal(fsid_t *fsid, uint32_t event, intptr_t data __unused) { KNOTE_UNLOCKED(&fs_knlist, event); } static int filt_fsattach(struct knote *kn); static void filt_fsdetach(struct knote *kn); static int filt_fsevent(struct knote *kn, long hint); struct filterops fs_filtops = { .f_isfd = 0, .f_attach = filt_fsattach, .f_detach = filt_fsdetach, .f_event = filt_fsevent }; static int filt_fsattach(struct knote *kn) { kn->kn_flags |= EV_CLEAR; knlist_add(&fs_knlist, kn, 0); return (0); } static void filt_fsdetach(struct knote *kn) { knlist_remove(&fs_knlist, kn, 0); } static int filt_fsevent(struct knote *kn, long hint) { kn->kn_fflags |= hint; return (kn->kn_fflags != 0); } static int sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS) { struct vfsidctl vc; int error; struct mount *mp; error = SYSCTL_IN(req, &vc, sizeof(vc)); if (error) return (error); if (vc.vc_vers != VFS_CTL_VERS1) return (EINVAL); mp = vfs_getvfs(&vc.vc_fsid); if (mp == NULL) return (ENOENT); /* ensure that a specific sysctl goes to the right filesystem. */ if (strcmp(vc.vc_fstypename, "*") != 0 && strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) { vfs_rel(mp); return (EINVAL); } VCTLTOREQ(&vc, req); error = VFS_SYSCTL(mp, vc.vc_op, req); vfs_rel(mp); return (error); } SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLTYPE_OPAQUE | CTLFLAG_MPSAFE | CTLFLAG_WR, NULL, 0, sysctl_vfs_ctl, "", "Sysctl by fsid"); /* * Function to initialize a va_filerev field sensibly. * XXX: Wouldn't a random number make a lot more sense ?? */ u_quad_t init_va_filerev(void) { struct bintime bt; getbinuptime(&bt); return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL)); } static int filt_vfsread(struct knote *kn, long hint); static int filt_vfswrite(struct knote *kn, long hint); static int filt_vfsvnode(struct knote *kn, long hint); static void filt_vfsdetach(struct knote *kn); static struct filterops vfsread_filtops = { .f_isfd = 1, .f_detach = filt_vfsdetach, .f_event = filt_vfsread }; static struct filterops vfswrite_filtops = { .f_isfd = 1, .f_detach = filt_vfsdetach, .f_event = filt_vfswrite }; static struct filterops vfsvnode_filtops = { .f_isfd = 1, .f_detach = filt_vfsdetach, .f_event = filt_vfsvnode }; static void vfs_knllock(void *arg) { struct vnode *vp = arg; vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); } static void vfs_knlunlock(void *arg) { struct vnode *vp = arg; VOP_UNLOCK(vp); } static void vfs_knl_assert_locked(void *arg) { #ifdef DEBUG_VFS_LOCKS struct vnode *vp = arg; ASSERT_VOP_LOCKED(vp, "vfs_knl_assert_locked"); #endif } static void vfs_knl_assert_unlocked(void *arg) { #ifdef DEBUG_VFS_LOCKS struct vnode *vp = arg; ASSERT_VOP_UNLOCKED(vp, "vfs_knl_assert_unlocked"); #endif } int vfs_kqfilter(struct vop_kqfilter_args *ap) { struct vnode *vp = ap->a_vp; struct knote *kn = ap->a_kn; struct knlist *knl; switch (kn->kn_filter) { case EVFILT_READ: kn->kn_fop = &vfsread_filtops; break; case EVFILT_WRITE: kn->kn_fop = &vfswrite_filtops; break; case EVFILT_VNODE: kn->kn_fop = &vfsvnode_filtops; break; default: return (EINVAL); } kn->kn_hook = (caddr_t)vp; v_addpollinfo(vp); if (vp->v_pollinfo == NULL) return (ENOMEM); knl = &vp->v_pollinfo->vpi_selinfo.si_note; vhold(vp); knlist_add(knl, kn, 0); return (0); } /* * Detach knote from vnode */ static void filt_vfsdetach(struct knote *kn) { struct vnode *vp = (struct vnode *)kn->kn_hook; KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo")); knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0); vdrop(vp); } /*ARGSUSED*/ static int filt_vfsread(struct knote *kn, long hint) { struct vnode *vp = (struct vnode *)kn->kn_hook; struct vattr va; int res; /* * filesystem is gone, so set the EOF flag and schedule * the knote for deletion. */ if (hint == NOTE_REVOKE || (hint == 0 && vp->v_type == VBAD)) { VI_LOCK(vp); kn->kn_flags |= (EV_EOF | EV_ONESHOT); VI_UNLOCK(vp); return (1); } if (VOP_GETATTR(vp, &va, curthread->td_ucred)) return (0); VI_LOCK(vp); kn->kn_data = va.va_size - kn->kn_fp->f_offset; res = (kn->kn_sfflags & NOTE_FILE_POLL) != 0 || kn->kn_data != 0; VI_UNLOCK(vp); return (res); } /*ARGSUSED*/ static int filt_vfswrite(struct knote *kn, long hint) { struct vnode *vp = (struct vnode *)kn->kn_hook; VI_LOCK(vp); /* * filesystem is gone, so set the EOF flag and schedule * the knote for deletion. */ if (hint == NOTE_REVOKE || (hint == 0 && vp->v_type == VBAD)) kn->kn_flags |= (EV_EOF | EV_ONESHOT); kn->kn_data = 0; VI_UNLOCK(vp); return (1); } static int filt_vfsvnode(struct knote *kn, long hint) { struct vnode *vp = (struct vnode *)kn->kn_hook; int res; VI_LOCK(vp); if (kn->kn_sfflags & hint) kn->kn_fflags |= hint; if (hint == NOTE_REVOKE || (hint == 0 && vp->v_type == VBAD)) { kn->kn_flags |= EV_EOF; VI_UNLOCK(vp); return (1); } res = (kn->kn_fflags != 0); VI_UNLOCK(vp); return (res); } /* * Returns whether the directory is empty or not. * If it is empty, the return value is 0; otherwise * the return value is an error value (which may * be ENOTEMPTY). */ int vfs_emptydir(struct vnode *vp) { struct uio uio; struct iovec iov; struct dirent *dirent, *dp, *endp; int error, eof; error = 0; eof = 0; ASSERT_VOP_LOCKED(vp, "vfs_emptydir"); dirent = malloc(sizeof(struct dirent), M_TEMP, M_WAITOK); iov.iov_base = dirent; iov.iov_len = sizeof(struct dirent); uio.uio_iov = &iov; uio.uio_iovcnt = 1; uio.uio_offset = 0; uio.uio_resid = sizeof(struct dirent); uio.uio_segflg = UIO_SYSSPACE; uio.uio_rw = UIO_READ; uio.uio_td = curthread; while (eof == 0 && error == 0) { error = VOP_READDIR(vp, &uio, curthread->td_ucred, &eof, NULL, NULL); if (error != 0) break; endp = (void *)((uint8_t *)dirent + sizeof(struct dirent) - uio.uio_resid); for (dp = dirent; dp < endp; dp = (void *)((uint8_t *)dp + GENERIC_DIRSIZ(dp))) { if (dp->d_type == DT_WHT) continue; if (dp->d_namlen == 0) continue; if (dp->d_type != DT_DIR && dp->d_type != DT_UNKNOWN) { error = ENOTEMPTY; break; } if (dp->d_namlen > 2) { error = ENOTEMPTY; break; } if (dp->d_namlen == 1 && dp->d_name[0] != '.') { error = ENOTEMPTY; break; } if (dp->d_namlen == 2 && dp->d_name[1] != '.') { error = ENOTEMPTY; break; } uio.uio_resid = sizeof(struct dirent); } } free(dirent, M_TEMP); return (error); } int vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off) { int error; if (dp->d_reclen > ap->a_uio->uio_resid) return (ENAMETOOLONG); error = uiomove(dp, dp->d_reclen, ap->a_uio); if (error) { if (ap->a_ncookies != NULL) { if (ap->a_cookies != NULL) free(ap->a_cookies, M_TEMP); ap->a_cookies = NULL; *ap->a_ncookies = 0; } return (error); } if (ap->a_ncookies == NULL) return (0); KASSERT(ap->a_cookies, ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!")); *ap->a_cookies = realloc(*ap->a_cookies, (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO); (*ap->a_cookies)[*ap->a_ncookies] = off; *ap->a_ncookies += 1; return (0); } /* * The purpose of this routine is to remove granularity from accmode_t, * reducing it into standard unix access bits - VEXEC, VREAD, VWRITE, * VADMIN and VAPPEND. * * If it returns 0, the caller is supposed to continue with the usual * access checks using 'accmode' as modified by this routine. If it * returns nonzero value, the caller is supposed to return that value * as errno. * * Note that after this routine runs, accmode may be zero. */ int vfs_unixify_accmode(accmode_t *accmode) { /* * There is no way to specify explicit "deny" rule using * file mode or POSIX.1e ACLs. */ if (*accmode & VEXPLICIT_DENY) { *accmode = 0; return (0); } /* * None of these can be translated into usual access bits. * Also, the common case for NFSv4 ACLs is to not contain * either of these bits. Caller should check for VWRITE * on the containing directory instead. */ if (*accmode & (VDELETE_CHILD | VDELETE)) return (EPERM); if (*accmode & VADMIN_PERMS) { *accmode &= ~VADMIN_PERMS; *accmode |= VADMIN; } /* * There is no way to deny VREAD_ATTRIBUTES, VREAD_ACL * or VSYNCHRONIZE using file mode or POSIX.1e ACL. */ *accmode &= ~(VSTAT_PERMS | VSYNCHRONIZE); return (0); } /* * Clear out a doomed vnode (if any) and replace it with a new one as long * as the fs is not being unmounted. Return the root vnode to the caller. */ static int __noinline vfs_cache_root_fallback(struct mount *mp, int flags, struct vnode **vpp) { struct vnode *vp; int error; restart: if (mp->mnt_rootvnode != NULL) { MNT_ILOCK(mp); vp = mp->mnt_rootvnode; if (vp != NULL) { if (!VN_IS_DOOMED(vp)) { vrefact(vp); MNT_IUNLOCK(mp); error = vn_lock(vp, flags); if (error == 0) { *vpp = vp; return (0); } vrele(vp); goto restart; } /* * Clear the old one. */ mp->mnt_rootvnode = NULL; } MNT_IUNLOCK(mp); if (vp != NULL) { vfs_op_barrier_wait(mp); vrele(vp); } } error = VFS_CACHEDROOT(mp, flags, vpp); if (error != 0) return (error); if (mp->mnt_vfs_ops == 0) { MNT_ILOCK(mp); if (mp->mnt_vfs_ops != 0) { MNT_IUNLOCK(mp); return (0); } if (mp->mnt_rootvnode == NULL) { vrefact(*vpp); mp->mnt_rootvnode = *vpp; } else { if (mp->mnt_rootvnode != *vpp) { if (!VN_IS_DOOMED(mp->mnt_rootvnode)) { panic("%s: mismatch between vnode returned " " by VFS_CACHEDROOT and the one cached " " (%p != %p)", __func__, *vpp, mp->mnt_rootvnode); } } } MNT_IUNLOCK(mp); } return (0); } int vfs_cache_root(struct mount *mp, int flags, struct vnode **vpp) { struct vnode *vp; int error; if (!vfs_op_thread_enter(mp)) return (vfs_cache_root_fallback(mp, flags, vpp)); vp = atomic_load_ptr(&mp->mnt_rootvnode); if (vp == NULL || VN_IS_DOOMED(vp)) { vfs_op_thread_exit(mp); return (vfs_cache_root_fallback(mp, flags, vpp)); } vrefact(vp); vfs_op_thread_exit(mp); error = vn_lock(vp, flags); if (error != 0) { vrele(vp); return (vfs_cache_root_fallback(mp, flags, vpp)); } *vpp = vp; return (0); } struct vnode * vfs_cache_root_clear(struct mount *mp) { struct vnode *vp; /* * ops > 0 guarantees there is nobody who can see this vnode */ MPASS(mp->mnt_vfs_ops > 0); vp = mp->mnt_rootvnode; mp->mnt_rootvnode = NULL; return (vp); } void vfs_cache_root_set(struct mount *mp, struct vnode *vp) { MPASS(mp->mnt_vfs_ops > 0); vrefact(vp); mp->mnt_rootvnode = vp; } /* * These are helper functions for filesystems to traverse all * their vnodes. See MNT_VNODE_FOREACH_ALL() in sys/mount.h. * * This interface replaces MNT_VNODE_FOREACH. */ struct vnode * __mnt_vnode_next_all(struct vnode **mvp, struct mount *mp) { struct vnode *vp; if (should_yield()) kern_yield(PRI_USER); MNT_ILOCK(mp); KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); for (vp = TAILQ_NEXT(*mvp, v_nmntvnodes); vp != NULL; vp = TAILQ_NEXT(vp, v_nmntvnodes)) { /* Allow a racy peek at VIRF_DOOMED to save a lock acquisition. */ if (vp->v_type == VMARKER || VN_IS_DOOMED(vp)) continue; VI_LOCK(vp); if (VN_IS_DOOMED(vp)) { VI_UNLOCK(vp); continue; } break; } if (vp == NULL) { __mnt_vnode_markerfree_all(mvp, mp); /* MNT_IUNLOCK(mp); -- done in above function */ mtx_assert(MNT_MTX(mp), MA_NOTOWNED); return (NULL); } TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes); TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes); MNT_IUNLOCK(mp); return (vp); } struct vnode * __mnt_vnode_first_all(struct vnode **mvp, struct mount *mp) { struct vnode *vp; *mvp = vn_alloc_marker(mp); MNT_ILOCK(mp); MNT_REF(mp); TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) { /* Allow a racy peek at VIRF_DOOMED to save a lock acquisition. */ if (vp->v_type == VMARKER || VN_IS_DOOMED(vp)) continue; VI_LOCK(vp); if (VN_IS_DOOMED(vp)) { VI_UNLOCK(vp); continue; } break; } if (vp == NULL) { MNT_REL(mp); MNT_IUNLOCK(mp); vn_free_marker(*mvp); *mvp = NULL; return (NULL); } TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes); MNT_IUNLOCK(mp); return (vp); } void __mnt_vnode_markerfree_all(struct vnode **mvp, struct mount *mp) { if (*mvp == NULL) { MNT_IUNLOCK(mp); return; } mtx_assert(MNT_MTX(mp), MA_OWNED); KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes); MNT_REL(mp); MNT_IUNLOCK(mp); vn_free_marker(*mvp); *mvp = NULL; } /* * These are helper functions for filesystems to traverse their * lazy vnodes. See MNT_VNODE_FOREACH_LAZY() in sys/mount.h */ static void mnt_vnode_markerfree_lazy(struct vnode **mvp, struct mount *mp) { KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); MNT_ILOCK(mp); MNT_REL(mp); MNT_IUNLOCK(mp); vn_free_marker(*mvp); *mvp = NULL; } /* * Relock the mp mount vnode list lock with the vp vnode interlock in the * conventional lock order during mnt_vnode_next_lazy iteration. * * On entry, the mount vnode list lock is held and the vnode interlock is not. * The list lock is dropped and reacquired. On success, both locks are held. * On failure, the mount vnode list lock is held but the vnode interlock is * not, and the procedure may have yielded. */ static bool mnt_vnode_next_lazy_relock(struct vnode *mvp, struct mount *mp, struct vnode *vp) { VNASSERT(mvp->v_mount == mp && mvp->v_type == VMARKER && TAILQ_NEXT(mvp, v_lazylist) != NULL, mvp, ("%s: bad marker", __func__)); VNASSERT(vp->v_mount == mp && vp->v_type != VMARKER, vp, ("%s: inappropriate vnode", __func__)); ASSERT_VI_UNLOCKED(vp, __func__); mtx_assert(&mp->mnt_listmtx, MA_OWNED); TAILQ_REMOVE(&mp->mnt_lazyvnodelist, mvp, v_lazylist); TAILQ_INSERT_BEFORE(vp, mvp, v_lazylist); /* * Note we may be racing against vdrop which transitioned the hold * count to 0 and now waits for the ->mnt_listmtx lock. This is fine, * if we are the only user after we get the interlock we will just * vdrop. */ vhold(vp); mtx_unlock(&mp->mnt_listmtx); VI_LOCK(vp); if (VN_IS_DOOMED(vp)) { VNPASS((vp->v_mflag & VMP_LAZYLIST) == 0, vp); goto out_lost; } VNPASS(vp->v_mflag & VMP_LAZYLIST, vp); /* * There is nothing to do if we are the last user. */ if (!refcount_release_if_not_last(&vp->v_holdcnt)) goto out_lost; mtx_lock(&mp->mnt_listmtx); return (true); out_lost: vdropl(vp); maybe_yield(); mtx_lock(&mp->mnt_listmtx); return (false); } static struct vnode * mnt_vnode_next_lazy(struct vnode **mvp, struct mount *mp, mnt_lazy_cb_t *cb, void *cbarg) { struct vnode *vp; mtx_assert(&mp->mnt_listmtx, MA_OWNED); KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch")); restart: vp = TAILQ_NEXT(*mvp, v_lazylist); while (vp != NULL) { if (vp->v_type == VMARKER) { vp = TAILQ_NEXT(vp, v_lazylist); continue; } /* * See if we want to process the vnode. Note we may encounter a * long string of vnodes we don't care about and hog the list * as a result. Check for it and requeue the marker. */ VNPASS(!VN_IS_DOOMED(vp), vp); if (!cb(vp, cbarg)) { if (!should_yield()) { vp = TAILQ_NEXT(vp, v_lazylist); continue; } TAILQ_REMOVE(&mp->mnt_lazyvnodelist, *mvp, v_lazylist); TAILQ_INSERT_AFTER(&mp->mnt_lazyvnodelist, vp, *mvp, v_lazylist); mtx_unlock(&mp->mnt_listmtx); kern_yield(PRI_USER); mtx_lock(&mp->mnt_listmtx); goto restart; } /* * Try-lock because this is the wrong lock order. */ if (!VI_TRYLOCK(vp) && !mnt_vnode_next_lazy_relock(*mvp, mp, vp)) goto restart; KASSERT(vp->v_type != VMARKER, ("locked marker %p", vp)); KASSERT(vp->v_mount == mp || vp->v_mount == NULL, ("alien vnode on the lazy list %p %p", vp, mp)); VNPASS(vp->v_mount == mp, vp); VNPASS(!VN_IS_DOOMED(vp), vp); break; } TAILQ_REMOVE(&mp->mnt_lazyvnodelist, *mvp, v_lazylist); /* Check if we are done */ if (vp == NULL) { mtx_unlock(&mp->mnt_listmtx); mnt_vnode_markerfree_lazy(mvp, mp); return (NULL); } TAILQ_INSERT_AFTER(&mp->mnt_lazyvnodelist, vp, *mvp, v_lazylist); mtx_unlock(&mp->mnt_listmtx); ASSERT_VI_LOCKED(vp, "lazy iter"); return (vp); } struct vnode * __mnt_vnode_next_lazy(struct vnode **mvp, struct mount *mp, mnt_lazy_cb_t *cb, void *cbarg) { if (should_yield()) kern_yield(PRI_USER); mtx_lock(&mp->mnt_listmtx); return (mnt_vnode_next_lazy(mvp, mp, cb, cbarg)); } struct vnode * __mnt_vnode_first_lazy(struct vnode **mvp, struct mount *mp, mnt_lazy_cb_t *cb, void *cbarg) { struct vnode *vp; if (TAILQ_EMPTY(&mp->mnt_lazyvnodelist)) return (NULL); *mvp = vn_alloc_marker(mp); MNT_ILOCK(mp); MNT_REF(mp); MNT_IUNLOCK(mp); mtx_lock(&mp->mnt_listmtx); vp = TAILQ_FIRST(&mp->mnt_lazyvnodelist); if (vp == NULL) { mtx_unlock(&mp->mnt_listmtx); mnt_vnode_markerfree_lazy(mvp, mp); return (NULL); } TAILQ_INSERT_BEFORE(vp, *mvp, v_lazylist); return (mnt_vnode_next_lazy(mvp, mp, cb, cbarg)); } void __mnt_vnode_markerfree_lazy(struct vnode **mvp, struct mount *mp) { if (*mvp == NULL) return; mtx_lock(&mp->mnt_listmtx); TAILQ_REMOVE(&mp->mnt_lazyvnodelist, *mvp, v_lazylist); mtx_unlock(&mp->mnt_listmtx); mnt_vnode_markerfree_lazy(mvp, mp); } int vn_dir_check_exec(struct vnode *vp, struct componentname *cnp) { if ((cnp->cn_flags & NOEXECCHECK) != 0) { cnp->cn_flags &= ~NOEXECCHECK; return (0); } return (VOP_ACCESS(vp, VEXEC, cnp->cn_cred, cnp->cn_thread)); } diff --git a/sys/netinet/igmp.c b/sys/netinet/igmp.c index 2a08e73317a2..d619cfc5160e 100644 --- a/sys/netinet/igmp.c +++ b/sys/netinet/igmp.c @@ -1,3651 +1,3706 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 2007-2009 Bruce Simpson. * Copyright (c) 1988 Stephen Deering. * Copyright (c) 1992, 1993 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed to Berkeley by * Stephen Deering of Stanford University. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)igmp.c 8.1 (Berkeley) 7/19/93 */ /* * Internet Group Management Protocol (IGMP) routines. * [RFC1112, RFC2236, RFC3376] * * Written by Steve Deering, Stanford, May 1988. * Modified by Rosen Sharma, Stanford, Aug 1994. * Modified by Bill Fenner, Xerox PARC, Feb 1995. * Modified to fully comply to IGMPv2 by Bill Fenner, Oct 1995. * Significantly rewritten for IGMPv3, VIMAGE, and SMP by Bruce Simpson. * * MULTICAST Revision: 3.5.1.4 */ #include __FBSDID("$FreeBSD$"); #include "opt_ddb.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef DDB #include #endif #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifndef KTR_IGMPV3 #define KTR_IGMPV3 KTR_INET #endif static struct igmp_ifsoftc * igi_alloc_locked(struct ifnet *); static void igi_delete_locked(const struct ifnet *); static void igmp_dispatch_queue(struct mbufq *, int, const int); static void igmp_fasttimo_vnet(void); static void igmp_final_leave(struct in_multi *, struct igmp_ifsoftc *); static int igmp_handle_state_change(struct in_multi *, struct igmp_ifsoftc *); static int igmp_initial_join(struct in_multi *, struct igmp_ifsoftc *); static int igmp_input_v1_query(struct ifnet *, const struct ip *, const struct igmp *); static int igmp_input_v2_query(struct ifnet *, const struct ip *, const struct igmp *); static int igmp_input_v3_query(struct ifnet *, const struct ip *, /*const*/ struct igmpv3 *); static int igmp_input_v3_group_query(struct in_multi *, struct igmp_ifsoftc *, int, /*const*/ struct igmpv3 *); static int igmp_input_v1_report(struct ifnet *, /*const*/ struct ip *, /*const*/ struct igmp *); static int igmp_input_v2_report(struct ifnet *, /*const*/ struct ip *, /*const*/ struct igmp *); static void igmp_intr(struct mbuf *); static int igmp_isgroupreported(const struct in_addr); static struct mbuf * igmp_ra_alloc(void); #ifdef KTR static char * igmp_rec_type_to_str(const int); #endif static void igmp_set_version(struct igmp_ifsoftc *, const int); static void igmp_slowtimo_vnet(void); static int igmp_v1v2_queue_report(struct in_multi *, const int); static void igmp_v1v2_process_group_timer(struct in_multi *, const int); static void igmp_v1v2_process_querier_timers(struct igmp_ifsoftc *); static void igmp_v2_update_group(struct in_multi *, const int); static void igmp_v3_cancel_link_timers(struct igmp_ifsoftc *); static void igmp_v3_dispatch_general_query(struct igmp_ifsoftc *); static struct mbuf * igmp_v3_encap_report(struct ifnet *, struct mbuf *); static int igmp_v3_enqueue_group_record(struct mbufq *, struct in_multi *, const int, const int, const int); static int igmp_v3_enqueue_filter_change(struct mbufq *, struct in_multi *); static void igmp_v3_process_group_timers(struct in_multi_head *, struct mbufq *, struct mbufq *, struct in_multi *, const int); static int igmp_v3_merge_state_changes(struct in_multi *, struct mbufq *); static void igmp_v3_suppress_group_record(struct in_multi *); static int sysctl_igmp_default_version(SYSCTL_HANDLER_ARGS); static int sysctl_igmp_gsr(SYSCTL_HANDLER_ARGS); static int sysctl_igmp_ifinfo(SYSCTL_HANDLER_ARGS); +static int sysctl_igmp_stat(SYSCTL_HANDLER_ARGS); static const struct netisr_handler igmp_nh = { .nh_name = "igmp", .nh_handler = igmp_intr, .nh_proto = NETISR_IGMP, .nh_policy = NETISR_POLICY_SOURCE, }; /* * System-wide globals. * * Unlocked access to these is OK, except for the global IGMP output * queue. The IGMP subsystem lock ends up being system-wide for the moment, * because all VIMAGEs have to share a global output queue, as netisrs * themselves are not virtualized. * * Locking: * * The permitted lock order is: IN_MULTI_LIST_LOCK, IGMP_LOCK, IF_ADDR_LOCK. * Any may be taken independently; if any are held at the same * time, the above lock order must be followed. * * All output is delegated to the netisr. * Now that Giant has been eliminated, the netisr may be inlined. * * IN_MULTI_LIST_LOCK covers in_multi. * * IGMP_LOCK covers igmp_ifsoftc and any global variables in this file, * including the output queue. * * IF_ADDR_LOCK covers if_multiaddrs, which is used for a variety of * per-link state iterators. * * igmp_ifsoftc is valid as long as PF_INET is attached to the interface, * therefore it is not refcounted. * We allow unlocked reads of igmp_ifsoftc when accessed via in_multi. * * Reference counting * * IGMP acquires its own reference every time an in_multi is passed to * it and the group is being joined for the first time. * * IGMP releases its reference(s) on in_multi in a deferred way, * because the operations which process the release run as part of * a loop whose control variables are directly affected by the release * (that, and not recursing on the IF_ADDR_LOCK). * * VIMAGE: Each in_multi corresponds to an ifp, and each ifp corresponds * to a vnet in ifp->if_vnet. * * SMPng: XXX We may potentially race operations on ifma_protospec. * The problem is that we currently lack a clean way of taking the * IF_ADDR_LOCK() between the ifnet and in layers w/o recursing, * as anything which modifies ifma needs to be covered by that lock. * So check for ifma_protospec being NULL before proceeding. */ struct mtx igmp_mtx; struct mbuf *m_raopt; /* Router Alert option */ static MALLOC_DEFINE(M_IGMP, "igmp", "igmp state"); /* * VIMAGE-wide globals. * * The IGMPv3 timers themselves need to run per-image, however, * protosw timers run globally (see tcp). * An ifnet can only be in one vimage at a time, and the loopback * ifnet, loif, is itself virtualized. * It would otherwise be possible to seriously hose IGMP state, * and create inconsistencies in upstream multicast routing, if you have * multiple VIMAGEs running on the same link joining different multicast * groups, UNLESS the "primary IP address" is different. This is because * IGMP for IPv4 does not force link-local addresses to be used for each * node, unlike MLD for IPv6. * Obviously the IGMPv3 per-interface state has per-vimage granularity * also as a result. * * FUTURE: Stop using IFP_TO_IA/INADDR_ANY, and use source address selection * policy to control the address used by IGMP on the link. */ VNET_DEFINE_STATIC(int, interface_timers_running); /* IGMPv3 general * query response */ VNET_DEFINE_STATIC(int, state_change_timers_running); /* IGMPv3 state-change * retransmit */ VNET_DEFINE_STATIC(int, current_state_timers_running); /* IGMPv1/v2 host * report; IGMPv3 g/sg * query response */ #define V_interface_timers_running VNET(interface_timers_running) #define V_state_change_timers_running VNET(state_change_timers_running) #define V_current_state_timers_running VNET(current_state_timers_running) VNET_DEFINE_STATIC(LIST_HEAD(, igmp_ifsoftc), igi_head) = LIST_HEAD_INITIALIZER(igi_head); VNET_DEFINE_STATIC(struct igmpstat, igmpstat) = { .igps_version = IGPS_VERSION_3, .igps_len = sizeof(struct igmpstat), }; VNET_DEFINE_STATIC(struct timeval, igmp_gsrdelay) = {10, 0}; #define V_igi_head VNET(igi_head) #define V_igmpstat VNET(igmpstat) #define V_igmp_gsrdelay VNET(igmp_gsrdelay) VNET_DEFINE_STATIC(int, igmp_recvifkludge) = 1; VNET_DEFINE_STATIC(int, igmp_sendra) = 1; VNET_DEFINE_STATIC(int, igmp_sendlocal) = 1; VNET_DEFINE_STATIC(int, igmp_v1enable) = 1; VNET_DEFINE_STATIC(int, igmp_v2enable) = 1; VNET_DEFINE_STATIC(int, igmp_legacysupp); VNET_DEFINE_STATIC(int, igmp_default_version) = IGMP_VERSION_3; #define V_igmp_recvifkludge VNET(igmp_recvifkludge) #define V_igmp_sendra VNET(igmp_sendra) #define V_igmp_sendlocal VNET(igmp_sendlocal) #define V_igmp_v1enable VNET(igmp_v1enable) #define V_igmp_v2enable VNET(igmp_v2enable) #define V_igmp_legacysupp VNET(igmp_legacysupp) #define V_igmp_default_version VNET(igmp_default_version) /* * Virtualized sysctls. */ -SYSCTL_STRUCT(_net_inet_igmp, IGMPCTL_STATS, stats, CTLFLAG_VNET | CTLFLAG_RW, - &VNET_NAME(igmpstat), igmpstat, ""); +SYSCTL_PROC(_net_inet_igmp, IGMPCTL_STATS, stats, + CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_RW | CTLFLAG_MPSAFE, + &VNET_NAME(igmpstat), 0, sysctl_igmp_stat, "S,igmpstat", ""); SYSCTL_INT(_net_inet_igmp, OID_AUTO, recvifkludge, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(igmp_recvifkludge), 0, "Rewrite IGMPv1/v2 reports from 0.0.0.0 to contain subnet address"); SYSCTL_INT(_net_inet_igmp, OID_AUTO, sendra, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(igmp_sendra), 0, "Send IP Router Alert option in IGMPv2/v3 messages"); SYSCTL_INT(_net_inet_igmp, OID_AUTO, sendlocal, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(igmp_sendlocal), 0, "Send IGMP membership reports for 224.0.0.0/24 groups"); SYSCTL_INT(_net_inet_igmp, OID_AUTO, v1enable, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(igmp_v1enable), 0, "Enable backwards compatibility with IGMPv1"); SYSCTL_INT(_net_inet_igmp, OID_AUTO, v2enable, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(igmp_v2enable), 0, "Enable backwards compatibility with IGMPv2"); SYSCTL_INT(_net_inet_igmp, OID_AUTO, legacysupp, CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(igmp_legacysupp), 0, "Allow v1/v2 reports to suppress v3 group responses"); SYSCTL_PROC(_net_inet_igmp, OID_AUTO, default_version, CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, &VNET_NAME(igmp_default_version), 0, sysctl_igmp_default_version, "I", "Default version of IGMP to run on each interface"); SYSCTL_PROC(_net_inet_igmp, OID_AUTO, gsrdelay, CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, &VNET_NAME(igmp_gsrdelay.tv_sec), 0, sysctl_igmp_gsr, "I", "Rate limit for IGMPv3 Group-and-Source queries in seconds"); /* * Non-virtualized sysctls. */ static SYSCTL_NODE(_net_inet_igmp, OID_AUTO, ifinfo, CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_igmp_ifinfo, "Per-interface IGMPv3 state"); static __inline void igmp_save_context(struct mbuf *m, struct ifnet *ifp) { #ifdef VIMAGE m->m_pkthdr.PH_loc.ptr = ifp->if_vnet; #endif /* VIMAGE */ m->m_pkthdr.rcvif = ifp; m->m_pkthdr.flowid = ifp->if_index; } static __inline void igmp_scrub_context(struct mbuf *m) { m->m_pkthdr.PH_loc.ptr = NULL; m->m_pkthdr.flowid = 0; } /* * Restore context from a queued IGMP output chain. * Return saved ifindex. * * VIMAGE: The assertion is there to make sure that we * actually called CURVNET_SET() with what's in the mbuf chain. */ static __inline uint32_t igmp_restore_context(struct mbuf *m) { #ifdef notyet #if defined(VIMAGE) && defined(INVARIANTS) KASSERT(curvnet == (m->m_pkthdr.PH_loc.ptr), ("%s: called when curvnet was not restored", __func__)); #endif #endif return (m->m_pkthdr.flowid); } +/* + * IGMP statistics. + */ +static int +sysctl_igmp_stat(SYSCTL_HANDLER_ARGS) +{ + struct igmpstat igps0; + int error; + char *p; + + error = sysctl_wire_old_buffer(req, sizeof(V_igmpstat)); + if (error) + return (error); + + if (req->oldptr != NULL) { + if (req->oldlen < sizeof(V_igmpstat)) + error = ENOMEM; + else + error = SYSCTL_OUT(req, &V_igmpstat, + sizeof(V_igmpstat)); + } else + req->validlen = sizeof(V_igmpstat); + if (error) + goto out; + if (req->newptr != NULL) { + if (req->newlen < sizeof(V_igmpstat)) + error = ENOMEM; + else + error = SYSCTL_IN(req, &igps0, + sizeof(igps0)); + if (error) + goto out; + /* + * igps0 must be "all zero". + */ + p = (char *)&igps0; + while (*p == '\0' && p < (char *)&igps0 + sizeof(igps0)) + p++; + if (p != (char *)&igps0 + sizeof(igps0)) { + error = EINVAL; + goto out; + } + /* + * Avoid overwrite of the version and length field. + */ + igps0.igps_version = V_igmpstat.igps_version; + igps0.igps_len = V_igmpstat.igps_len; + bcopy(&igps0, &V_igmpstat, sizeof(V_igmpstat)); + } +out: + return (error); +} + /* * Retrieve or set default IGMP version. * * VIMAGE: Assume curvnet set by caller. * SMPng: NOTE: Serialized by IGMP lock. */ static int sysctl_igmp_default_version(SYSCTL_HANDLER_ARGS) { int error; int new; error = sysctl_wire_old_buffer(req, sizeof(int)); if (error) return (error); IGMP_LOCK(); new = V_igmp_default_version; error = sysctl_handle_int(oidp, &new, 0, req); if (error || !req->newptr) goto out_locked; if (new < IGMP_VERSION_1 || new > IGMP_VERSION_3) { error = EINVAL; goto out_locked; } CTR2(KTR_IGMPV3, "change igmp_default_version from %d to %d", V_igmp_default_version, new); V_igmp_default_version = new; out_locked: IGMP_UNLOCK(); return (error); } /* * Retrieve or set threshold between group-source queries in seconds. * * VIMAGE: Assume curvnet set by caller. * SMPng: NOTE: Serialized by IGMP lock. */ static int sysctl_igmp_gsr(SYSCTL_HANDLER_ARGS) { int error; int i; error = sysctl_wire_old_buffer(req, sizeof(int)); if (error) return (error); IGMP_LOCK(); i = V_igmp_gsrdelay.tv_sec; error = sysctl_handle_int(oidp, &i, 0, req); if (error || !req->newptr) goto out_locked; if (i < -1 || i >= 60) { error = EINVAL; goto out_locked; } CTR2(KTR_IGMPV3, "change igmp_gsrdelay from %d to %d", V_igmp_gsrdelay.tv_sec, i); V_igmp_gsrdelay.tv_sec = i; out_locked: IGMP_UNLOCK(); return (error); } /* * Expose struct igmp_ifsoftc to userland, keyed by ifindex. * For use by ifmcstat(8). * * SMPng: NOTE: Does an unlocked ifindex space read. * VIMAGE: Assume curvnet set by caller. The node handler itself * is not directly virtualized. */ static int sysctl_igmp_ifinfo(SYSCTL_HANDLER_ARGS) { int *name; int error; u_int namelen; struct ifnet *ifp; struct igmp_ifsoftc *igi; name = (int *)arg1; namelen = arg2; if (req->newptr != NULL) return (EPERM); if (namelen != 1) return (EINVAL); error = sysctl_wire_old_buffer(req, sizeof(struct igmp_ifinfo)); if (error) return (error); IN_MULTI_LIST_LOCK(); IGMP_LOCK(); if (name[0] <= 0 || name[0] > V_if_index) { error = ENOENT; goto out_locked; } error = ENOENT; ifp = ifnet_byindex(name[0]); if (ifp == NULL) goto out_locked; LIST_FOREACH(igi, &V_igi_head, igi_link) { if (ifp == igi->igi_ifp) { struct igmp_ifinfo info; info.igi_version = igi->igi_version; info.igi_v1_timer = igi->igi_v1_timer; info.igi_v2_timer = igi->igi_v2_timer; info.igi_v3_timer = igi->igi_v3_timer; info.igi_flags = igi->igi_flags; info.igi_rv = igi->igi_rv; info.igi_qi = igi->igi_qi; info.igi_qri = igi->igi_qri; info.igi_uri = igi->igi_uri; error = SYSCTL_OUT(req, &info, sizeof(info)); break; } } out_locked: IGMP_UNLOCK(); IN_MULTI_LIST_UNLOCK(); return (error); } /* * Dispatch an entire queue of pending packet chains * using the netisr. * VIMAGE: Assumes the vnet pointer has been set. */ static void igmp_dispatch_queue(struct mbufq *mq, int limit, const int loop) { struct epoch_tracker et; struct mbuf *m; NET_EPOCH_ENTER(et); while ((m = mbufq_dequeue(mq)) != NULL) { CTR3(KTR_IGMPV3, "%s: dispatch %p from %p", __func__, mq, m); if (loop) m->m_flags |= M_IGMP_LOOP; netisr_dispatch(NETISR_IGMP, m); if (--limit == 0) break; } NET_EPOCH_EXIT(et); } /* * Filter outgoing IGMP report state by group. * * Reports are ALWAYS suppressed for ALL-HOSTS (224.0.0.1). * If the net.inet.igmp.sendlocal sysctl is 0, then IGMP reports are * disabled for all groups in the 224.0.0.0/24 link-local scope. However, * this may break certain IGMP snooping switches which rely on the old * report behaviour. * * Return zero if the given group is one for which IGMP reports * should be suppressed, or non-zero if reports should be issued. */ static __inline int igmp_isgroupreported(const struct in_addr addr) { if (in_allhosts(addr) || ((!V_igmp_sendlocal && IN_LOCAL_GROUP(ntohl(addr.s_addr))))) return (0); return (1); } /* * Construct a Router Alert option to use in outgoing packets. */ static struct mbuf * igmp_ra_alloc(void) { struct mbuf *m; struct ipoption *p; m = m_get(M_WAITOK, MT_DATA); p = mtod(m, struct ipoption *); p->ipopt_dst.s_addr = INADDR_ANY; p->ipopt_list[0] = (char)IPOPT_RA; /* Router Alert Option */ p->ipopt_list[1] = 0x04; /* 4 bytes long */ p->ipopt_list[2] = IPOPT_EOL; /* End of IP option list */ p->ipopt_list[3] = 0x00; /* pad byte */ m->m_len = sizeof(p->ipopt_dst) + p->ipopt_list[1]; return (m); } /* * Attach IGMP when PF_INET is attached to an interface. */ struct igmp_ifsoftc * igmp_domifattach(struct ifnet *ifp) { struct igmp_ifsoftc *igi; CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)", __func__, ifp, ifp->if_xname); IGMP_LOCK(); igi = igi_alloc_locked(ifp); if (!(ifp->if_flags & IFF_MULTICAST)) igi->igi_flags |= IGIF_SILENT; IGMP_UNLOCK(); return (igi); } /* * VIMAGE: assume curvnet set by caller. */ static struct igmp_ifsoftc * igi_alloc_locked(/*const*/ struct ifnet *ifp) { struct igmp_ifsoftc *igi; IGMP_LOCK_ASSERT(); igi = malloc(sizeof(struct igmp_ifsoftc), M_IGMP, M_NOWAIT|M_ZERO); if (igi == NULL) goto out; igi->igi_ifp = ifp; igi->igi_version = V_igmp_default_version; igi->igi_flags = 0; igi->igi_rv = IGMP_RV_INIT; igi->igi_qi = IGMP_QI_INIT; igi->igi_qri = IGMP_QRI_INIT; igi->igi_uri = IGMP_URI_INIT; mbufq_init(&igi->igi_gq, IGMP_MAX_RESPONSE_PACKETS); LIST_INSERT_HEAD(&V_igi_head, igi, igi_link); CTR2(KTR_IGMPV3, "allocate igmp_ifsoftc for ifp %p(%s)", ifp, ifp->if_xname); out: return (igi); } /* * Hook for ifdetach. * * NOTE: Some finalization tasks need to run before the protocol domain * is detached, but also before the link layer does its cleanup. * * SMPNG: igmp_ifdetach() needs to take IF_ADDR_LOCK(). * XXX This is also bitten by unlocked ifma_protospec access. */ void igmp_ifdetach(struct ifnet *ifp) { struct igmp_ifsoftc *igi; struct ifmultiaddr *ifma, *next; struct in_multi *inm; struct in_multi_head inm_free_tmp; CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)", __func__, ifp, ifp->if_xname); SLIST_INIT(&inm_free_tmp); IGMP_LOCK(); igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp; if (igi->igi_version == IGMP_VERSION_3) { IF_ADDR_WLOCK(ifp); restart: CK_STAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next) { if (ifma->ifma_addr->sa_family != AF_INET || ifma->ifma_protospec == NULL) continue; inm = (struct in_multi *)ifma->ifma_protospec; if (inm->inm_state == IGMP_LEAVING_MEMBER) inm_rele_locked(&inm_free_tmp, inm); inm_clear_recorded(inm); if (__predict_false(ifma_restart)) { ifma_restart = false; goto restart; } } IF_ADDR_WUNLOCK(ifp); inm_release_list_deferred(&inm_free_tmp); } IGMP_UNLOCK(); } /* * Hook for domifdetach. */ void igmp_domifdetach(struct ifnet *ifp) { CTR3(KTR_IGMPV3, "%s: called for ifp %p(%s)", __func__, ifp, ifp->if_xname); IGMP_LOCK(); igi_delete_locked(ifp); IGMP_UNLOCK(); } static void igi_delete_locked(const struct ifnet *ifp) { struct igmp_ifsoftc *igi, *tigi; CTR3(KTR_IGMPV3, "%s: freeing igmp_ifsoftc for ifp %p(%s)", __func__, ifp, ifp->if_xname); IGMP_LOCK_ASSERT(); LIST_FOREACH_SAFE(igi, &V_igi_head, igi_link, tigi) { if (igi->igi_ifp == ifp) { /* * Free deferred General Query responses. */ mbufq_drain(&igi->igi_gq); LIST_REMOVE(igi, igi_link); free(igi, M_IGMP); return; } } } /* * Process a received IGMPv1 query. * Return non-zero if the message should be dropped. * * VIMAGE: The curvnet pointer is derived from the input ifp. */ static int igmp_input_v1_query(struct ifnet *ifp, const struct ip *ip, const struct igmp *igmp) { struct ifmultiaddr *ifma; struct igmp_ifsoftc *igi; struct in_multi *inm; NET_EPOCH_ASSERT(); /* * IGMPv1 Host Mmembership Queries SHOULD always be addressed to * 224.0.0.1. They are always treated as General Queries. * igmp_group is always ignored. Do not drop it as a userland * daemon may wish to see it. * XXX SMPng: unlocked increments in igmpstat assumed atomic. */ if (!in_allhosts(ip->ip_dst) || !in_nullhost(igmp->igmp_group)) { IGMPSTAT_INC(igps_rcv_badqueries); return (0); } IGMPSTAT_INC(igps_rcv_gen_queries); IN_MULTI_LIST_LOCK(); IGMP_LOCK(); igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp; KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp)); if (igi->igi_flags & IGIF_LOOPBACK) { CTR2(KTR_IGMPV3, "ignore v1 query on IGIF_LOOPBACK ifp %p(%s)", ifp, ifp->if_xname); goto out_locked; } /* * Switch to IGMPv1 host compatibility mode. */ igmp_set_version(igi, IGMP_VERSION_1); CTR2(KTR_IGMPV3, "process v1 query on ifp %p(%s)", ifp, ifp->if_xname); /* * Start the timers in all of our group records * for the interface on which the query arrived, * except those which are already running. */ CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { if (ifma->ifma_addr->sa_family != AF_INET || ifma->ifma_protospec == NULL) continue; inm = (struct in_multi *)ifma->ifma_protospec; if (inm->inm_timer != 0) continue; switch (inm->inm_state) { case IGMP_NOT_MEMBER: case IGMP_SILENT_MEMBER: break; case IGMP_G_QUERY_PENDING_MEMBER: case IGMP_SG_QUERY_PENDING_MEMBER: case IGMP_REPORTING_MEMBER: case IGMP_IDLE_MEMBER: case IGMP_LAZY_MEMBER: case IGMP_SLEEPING_MEMBER: case IGMP_AWAKENING_MEMBER: inm->inm_state = IGMP_REPORTING_MEMBER; inm->inm_timer = IGMP_RANDOM_DELAY( IGMP_V1V2_MAX_RI * PR_FASTHZ); V_current_state_timers_running = 1; break; case IGMP_LEAVING_MEMBER: break; } } out_locked: IGMP_UNLOCK(); IN_MULTI_LIST_UNLOCK(); return (0); } /* * Process a received IGMPv2 general or group-specific query. */ static int igmp_input_v2_query(struct ifnet *ifp, const struct ip *ip, const struct igmp *igmp) { struct ifmultiaddr *ifma; struct igmp_ifsoftc *igi; struct in_multi *inm; int is_general_query; uint16_t timer; NET_EPOCH_ASSERT(); is_general_query = 0; /* * Validate address fields upfront. * XXX SMPng: unlocked increments in igmpstat assumed atomic. */ if (in_nullhost(igmp->igmp_group)) { /* * IGMPv2 General Query. * If this was not sent to the all-hosts group, ignore it. */ if (!in_allhosts(ip->ip_dst)) return (0); IGMPSTAT_INC(igps_rcv_gen_queries); is_general_query = 1; } else { /* IGMPv2 Group-Specific Query. */ IGMPSTAT_INC(igps_rcv_group_queries); } IN_MULTI_LIST_LOCK(); IGMP_LOCK(); igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp; KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp)); if (igi->igi_flags & IGIF_LOOPBACK) { CTR2(KTR_IGMPV3, "ignore v2 query on IGIF_LOOPBACK ifp %p(%s)", ifp, ifp->if_xname); goto out_locked; } /* * Ignore v2 query if in v1 Compatibility Mode. */ if (igi->igi_version == IGMP_VERSION_1) goto out_locked; igmp_set_version(igi, IGMP_VERSION_2); timer = igmp->igmp_code * PR_FASTHZ / IGMP_TIMER_SCALE; if (timer == 0) timer = 1; if (is_general_query) { /* * For each reporting group joined on this * interface, kick the report timer. */ CTR2(KTR_IGMPV3, "process v2 general query on ifp %p(%s)", ifp, ifp->if_xname); CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { if (ifma->ifma_addr->sa_family != AF_INET || ifma->ifma_protospec == NULL) continue; inm = (struct in_multi *)ifma->ifma_protospec; igmp_v2_update_group(inm, timer); } } else { /* * Group-specific IGMPv2 query, we need only * look up the single group to process it. */ inm = inm_lookup(ifp, igmp->igmp_group); if (inm != NULL) { CTR3(KTR_IGMPV3, "process v2 query 0x%08x on ifp %p(%s)", ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname); igmp_v2_update_group(inm, timer); } } out_locked: IGMP_UNLOCK(); IN_MULTI_LIST_UNLOCK(); return (0); } /* * Update the report timer on a group in response to an IGMPv2 query. * * If we are becoming the reporting member for this group, start the timer. * If we already are the reporting member for this group, and timer is * below the threshold, reset it. * * We may be updating the group for the first time since we switched * to IGMPv3. If we are, then we must clear any recorded source lists, * and transition to REPORTING state; the group timer is overloaded * for group and group-source query responses. * * Unlike IGMPv3, the delay per group should be jittered * to avoid bursts of IGMPv2 reports. */ static void igmp_v2_update_group(struct in_multi *inm, const int timer) { CTR4(KTR_IGMPV3, "0x%08x: %s/%s timer=%d", __func__, ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname, timer); IN_MULTI_LIST_LOCK_ASSERT(); switch (inm->inm_state) { case IGMP_NOT_MEMBER: case IGMP_SILENT_MEMBER: break; case IGMP_REPORTING_MEMBER: if (inm->inm_timer != 0 && inm->inm_timer <= timer) { CTR1(KTR_IGMPV3, "%s: REPORTING and timer running, " "skipping.", __func__); break; } /* FALLTHROUGH */ case IGMP_SG_QUERY_PENDING_MEMBER: case IGMP_G_QUERY_PENDING_MEMBER: case IGMP_IDLE_MEMBER: case IGMP_LAZY_MEMBER: case IGMP_AWAKENING_MEMBER: CTR1(KTR_IGMPV3, "%s: ->REPORTING", __func__); inm->inm_state = IGMP_REPORTING_MEMBER; inm->inm_timer = IGMP_RANDOM_DELAY(timer); V_current_state_timers_running = 1; break; case IGMP_SLEEPING_MEMBER: CTR1(KTR_IGMPV3, "%s: ->AWAKENING", __func__); inm->inm_state = IGMP_AWAKENING_MEMBER; break; case IGMP_LEAVING_MEMBER: break; } } /* * Process a received IGMPv3 general, group-specific or * group-and-source-specific query. * Assumes m has already been pulled up to the full IGMP message length. * Return 0 if successful, otherwise an appropriate error code is returned. */ static int igmp_input_v3_query(struct ifnet *ifp, const struct ip *ip, /*const*/ struct igmpv3 *igmpv3) { struct igmp_ifsoftc *igi; struct in_multi *inm; int is_general_query; uint32_t maxresp, nsrc, qqi; uint16_t timer; uint8_t qrv; is_general_query = 0; CTR2(KTR_IGMPV3, "process v3 query on ifp %p(%s)", ifp, ifp->if_xname); maxresp = igmpv3->igmp_code; /* in 1/10ths of a second */ if (maxresp >= 128) { maxresp = IGMP_MANT(igmpv3->igmp_code) << (IGMP_EXP(igmpv3->igmp_code) + 3); } /* * Robustness must never be less than 2 for on-wire IGMPv3. * FUTURE: Check if ifp has IGIF_LOOPBACK set, as we will make * an exception for interfaces whose IGMPv3 state changes * are redirected to loopback (e.g. MANET). */ qrv = IGMP_QRV(igmpv3->igmp_misc); if (qrv < 2) { CTR3(KTR_IGMPV3, "%s: clamping qrv %d to %d", __func__, qrv, IGMP_RV_INIT); qrv = IGMP_RV_INIT; } qqi = igmpv3->igmp_qqi; if (qqi >= 128) { qqi = IGMP_MANT(igmpv3->igmp_qqi) << (IGMP_EXP(igmpv3->igmp_qqi) + 3); } timer = maxresp * PR_FASTHZ / IGMP_TIMER_SCALE; if (timer == 0) timer = 1; nsrc = ntohs(igmpv3->igmp_numsrc); /* * Validate address fields and versions upfront before * accepting v3 query. * XXX SMPng: Unlocked access to igmpstat counters here. */ if (in_nullhost(igmpv3->igmp_group)) { /* * IGMPv3 General Query. * * General Queries SHOULD be directed to 224.0.0.1. * A general query with a source list has undefined * behaviour; discard it. */ IGMPSTAT_INC(igps_rcv_gen_queries); if (!in_allhosts(ip->ip_dst) || nsrc > 0) { IGMPSTAT_INC(igps_rcv_badqueries); return (0); } is_general_query = 1; } else { /* Group or group-source specific query. */ if (nsrc == 0) IGMPSTAT_INC(igps_rcv_group_queries); else IGMPSTAT_INC(igps_rcv_gsr_queries); } IN_MULTI_LIST_LOCK(); IGMP_LOCK(); igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp; KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp)); if (igi->igi_flags & IGIF_LOOPBACK) { CTR2(KTR_IGMPV3, "ignore v3 query on IGIF_LOOPBACK ifp %p(%s)", ifp, ifp->if_xname); goto out_locked; } /* * Discard the v3 query if we're in Compatibility Mode. * The RFC is not obviously worded that hosts need to stay in * compatibility mode until the Old Version Querier Present * timer expires. */ if (igi->igi_version != IGMP_VERSION_3) { CTR3(KTR_IGMPV3, "ignore v3 query in v%d mode on ifp %p(%s)", igi->igi_version, ifp, ifp->if_xname); goto out_locked; } igmp_set_version(igi, IGMP_VERSION_3); igi->igi_rv = qrv; igi->igi_qi = qqi; igi->igi_qri = maxresp; CTR4(KTR_IGMPV3, "%s: qrv %d qi %d qri %d", __func__, qrv, qqi, maxresp); if (is_general_query) { /* * Schedule a current-state report on this ifp for * all groups, possibly containing source lists. * If there is a pending General Query response * scheduled earlier than the selected delay, do * not schedule any other reports. * Otherwise, reset the interface timer. */ CTR2(KTR_IGMPV3, "process v3 general query on ifp %p(%s)", ifp, ifp->if_xname); if (igi->igi_v3_timer == 0 || igi->igi_v3_timer >= timer) { igi->igi_v3_timer = IGMP_RANDOM_DELAY(timer); V_interface_timers_running = 1; } } else { /* * Group-source-specific queries are throttled on * a per-group basis to defeat denial-of-service attempts. * Queries for groups we are not a member of on this * link are simply ignored. */ inm = inm_lookup(ifp, igmpv3->igmp_group); if (inm == NULL) goto out_locked; if (nsrc > 0) { if (!ratecheck(&inm->inm_lastgsrtv, &V_igmp_gsrdelay)) { CTR1(KTR_IGMPV3, "%s: GS query throttled.", __func__); IGMPSTAT_INC(igps_drop_gsr_queries); goto out_locked; } } CTR3(KTR_IGMPV3, "process v3 0x%08x query on ifp %p(%s)", ntohl(igmpv3->igmp_group.s_addr), ifp, ifp->if_xname); /* * If there is a pending General Query response * scheduled sooner than the selected delay, no * further report need be scheduled. * Otherwise, prepare to respond to the * group-specific or group-and-source query. */ if (igi->igi_v3_timer == 0 || igi->igi_v3_timer >= timer) igmp_input_v3_group_query(inm, igi, timer, igmpv3); } out_locked: IGMP_UNLOCK(); IN_MULTI_LIST_UNLOCK(); return (0); } /* * Process a received IGMPv3 group-specific or group-and-source-specific * query. * Return <0 if any error occurred. Currently this is ignored. */ static int igmp_input_v3_group_query(struct in_multi *inm, struct igmp_ifsoftc *igi, int timer, /*const*/ struct igmpv3 *igmpv3) { int retval; uint16_t nsrc; IN_MULTI_LIST_LOCK_ASSERT(); IGMP_LOCK_ASSERT(); retval = 0; switch (inm->inm_state) { case IGMP_NOT_MEMBER: case IGMP_SILENT_MEMBER: case IGMP_SLEEPING_MEMBER: case IGMP_LAZY_MEMBER: case IGMP_AWAKENING_MEMBER: case IGMP_IDLE_MEMBER: case IGMP_LEAVING_MEMBER: return (retval); break; case IGMP_REPORTING_MEMBER: case IGMP_G_QUERY_PENDING_MEMBER: case IGMP_SG_QUERY_PENDING_MEMBER: break; } nsrc = ntohs(igmpv3->igmp_numsrc); /* * Deal with group-specific queries upfront. * If any group query is already pending, purge any recorded * source-list state if it exists, and schedule a query response * for this group-specific query. */ if (nsrc == 0) { if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER || inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER) { inm_clear_recorded(inm); timer = min(inm->inm_timer, timer); } inm->inm_state = IGMP_G_QUERY_PENDING_MEMBER; inm->inm_timer = IGMP_RANDOM_DELAY(timer); V_current_state_timers_running = 1; return (retval); } /* * Deal with the case where a group-and-source-specific query has * been received but a group-specific query is already pending. */ if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER) { timer = min(inm->inm_timer, timer); inm->inm_timer = IGMP_RANDOM_DELAY(timer); V_current_state_timers_running = 1; return (retval); } /* * Finally, deal with the case where a group-and-source-specific * query has been received, where a response to a previous g-s-r * query exists, or none exists. * In this case, we need to parse the source-list which the Querier * has provided us with and check if we have any source list filter * entries at T1 for these sources. If we do not, there is no need * schedule a report and the query may be dropped. * If we do, we must record them and schedule a current-state * report for those sources. * FIXME: Handling source lists larger than 1 mbuf requires that * we pass the mbuf chain pointer down to this function, and use * m_getptr() to walk the chain. */ if (inm->inm_nsrc > 0) { const struct in_addr *ap; int i, nrecorded; ap = (const struct in_addr *)(igmpv3 + 1); nrecorded = 0; for (i = 0; i < nsrc; i++, ap++) { retval = inm_record_source(inm, ap->s_addr); if (retval < 0) break; nrecorded += retval; } if (nrecorded > 0) { CTR1(KTR_IGMPV3, "%s: schedule response to SG query", __func__); inm->inm_state = IGMP_SG_QUERY_PENDING_MEMBER; inm->inm_timer = IGMP_RANDOM_DELAY(timer); V_current_state_timers_running = 1; } } return (retval); } /* * Process a received IGMPv1 host membership report. * * NOTE: 0.0.0.0 workaround breaks const correctness. */ static int igmp_input_v1_report(struct ifnet *ifp, /*const*/ struct ip *ip, /*const*/ struct igmp *igmp) { struct rm_priotracker in_ifa_tracker; struct in_ifaddr *ia; struct in_multi *inm; IGMPSTAT_INC(igps_rcv_reports); if (ifp->if_flags & IFF_LOOPBACK) return (0); if (!IN_MULTICAST(ntohl(igmp->igmp_group.s_addr)) || !in_hosteq(igmp->igmp_group, ip->ip_dst)) { IGMPSTAT_INC(igps_rcv_badreports); return (EINVAL); } /* * RFC 3376, Section 4.2.13, 9.2, 9.3: * Booting clients may use the source address 0.0.0.0. Some * IGMP daemons may not know how to use IP_RECVIF to determine * the interface upon which this message was received. * Replace 0.0.0.0 with the subnet address if told to do so. */ if (V_igmp_recvifkludge && in_nullhost(ip->ip_src)) { IFP_TO_IA(ifp, ia, &in_ifa_tracker); if (ia != NULL) ip->ip_src.s_addr = htonl(ia->ia_subnet); } CTR3(KTR_IGMPV3, "process v1 report 0x%08x on ifp %p(%s)", ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname); /* * IGMPv1 report suppression. * If we are a member of this group, and our membership should be * reported, stop our group timer and transition to the 'lazy' state. */ IN_MULTI_LIST_LOCK(); inm = inm_lookup(ifp, igmp->igmp_group); if (inm != NULL) { struct igmp_ifsoftc *igi; igi = inm->inm_igi; if (igi == NULL) { KASSERT(igi != NULL, ("%s: no igi for ifp %p", __func__, ifp)); goto out_locked; } IGMPSTAT_INC(igps_rcv_ourreports); /* * If we are in IGMPv3 host mode, do not allow the * other host's IGMPv1 report to suppress our reports * unless explicitly configured to do so. */ if (igi->igi_version == IGMP_VERSION_3) { if (V_igmp_legacysupp) igmp_v3_suppress_group_record(inm); goto out_locked; } inm->inm_timer = 0; switch (inm->inm_state) { case IGMP_NOT_MEMBER: case IGMP_SILENT_MEMBER: break; case IGMP_IDLE_MEMBER: case IGMP_LAZY_MEMBER: case IGMP_AWAKENING_MEMBER: CTR3(KTR_IGMPV3, "report suppressed for 0x%08x on ifp %p(%s)", ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname); case IGMP_SLEEPING_MEMBER: inm->inm_state = IGMP_SLEEPING_MEMBER; break; case IGMP_REPORTING_MEMBER: CTR3(KTR_IGMPV3, "report suppressed for 0x%08x on ifp %p(%s)", ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname); if (igi->igi_version == IGMP_VERSION_1) inm->inm_state = IGMP_LAZY_MEMBER; else if (igi->igi_version == IGMP_VERSION_2) inm->inm_state = IGMP_SLEEPING_MEMBER; break; case IGMP_G_QUERY_PENDING_MEMBER: case IGMP_SG_QUERY_PENDING_MEMBER: case IGMP_LEAVING_MEMBER: break; } } out_locked: IN_MULTI_LIST_UNLOCK(); return (0); } /* * Process a received IGMPv2 host membership report. * * NOTE: 0.0.0.0 workaround breaks const correctness. */ static int igmp_input_v2_report(struct ifnet *ifp, /*const*/ struct ip *ip, /*const*/ struct igmp *igmp) { struct rm_priotracker in_ifa_tracker; struct in_ifaddr *ia; struct in_multi *inm; /* * Make sure we don't hear our own membership report. Fast * leave requires knowing that we are the only member of a * group. */ IFP_TO_IA(ifp, ia, &in_ifa_tracker); if (ia != NULL && in_hosteq(ip->ip_src, IA_SIN(ia)->sin_addr)) { return (0); } IGMPSTAT_INC(igps_rcv_reports); if (ifp->if_flags & IFF_LOOPBACK) { return (0); } if (!IN_MULTICAST(ntohl(igmp->igmp_group.s_addr)) || !in_hosteq(igmp->igmp_group, ip->ip_dst)) { IGMPSTAT_INC(igps_rcv_badreports); return (EINVAL); } /* * RFC 3376, Section 4.2.13, 9.2, 9.3: * Booting clients may use the source address 0.0.0.0. Some * IGMP daemons may not know how to use IP_RECVIF to determine * the interface upon which this message was received. * Replace 0.0.0.0 with the subnet address if told to do so. */ if (V_igmp_recvifkludge && in_nullhost(ip->ip_src)) { if (ia != NULL) ip->ip_src.s_addr = htonl(ia->ia_subnet); } CTR3(KTR_IGMPV3, "process v2 report 0x%08x on ifp %p(%s)", ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname); /* * IGMPv2 report suppression. * If we are a member of this group, and our membership should be * reported, and our group timer is pending or about to be reset, * stop our group timer by transitioning to the 'lazy' state. */ IN_MULTI_LIST_LOCK(); inm = inm_lookup(ifp, igmp->igmp_group); if (inm != NULL) { struct igmp_ifsoftc *igi; igi = inm->inm_igi; KASSERT(igi != NULL, ("%s: no igi for ifp %p", __func__, ifp)); IGMPSTAT_INC(igps_rcv_ourreports); /* * If we are in IGMPv3 host mode, do not allow the * other host's IGMPv1 report to suppress our reports * unless explicitly configured to do so. */ if (igi->igi_version == IGMP_VERSION_3) { if (V_igmp_legacysupp) igmp_v3_suppress_group_record(inm); goto out_locked; } inm->inm_timer = 0; switch (inm->inm_state) { case IGMP_NOT_MEMBER: case IGMP_SILENT_MEMBER: case IGMP_SLEEPING_MEMBER: break; case IGMP_REPORTING_MEMBER: case IGMP_IDLE_MEMBER: case IGMP_AWAKENING_MEMBER: CTR3(KTR_IGMPV3, "report suppressed for 0x%08x on ifp %p(%s)", ntohl(igmp->igmp_group.s_addr), ifp, ifp->if_xname); case IGMP_LAZY_MEMBER: inm->inm_state = IGMP_LAZY_MEMBER; break; case IGMP_G_QUERY_PENDING_MEMBER: case IGMP_SG_QUERY_PENDING_MEMBER: case IGMP_LEAVING_MEMBER: break; } } out_locked: IN_MULTI_LIST_UNLOCK(); return (0); } int igmp_input(struct mbuf **mp, int *offp, int proto) { int iphlen; struct ifnet *ifp; struct igmp *igmp; struct ip *ip; struct mbuf *m; int igmplen; int minlen; int queryver; CTR3(KTR_IGMPV3, "%s: called w/mbuf (%p,%d)", __func__, *mp, *offp); m = *mp; ifp = m->m_pkthdr.rcvif; *mp = NULL; IGMPSTAT_INC(igps_rcv_total); ip = mtod(m, struct ip *); iphlen = *offp; igmplen = ntohs(ip->ip_len) - iphlen; /* * Validate lengths. */ if (igmplen < IGMP_MINLEN) { IGMPSTAT_INC(igps_rcv_tooshort); m_freem(m); return (IPPROTO_DONE); } /* * Always pullup to the minimum size for v1/v2 or v3 * to amortize calls to m_pullup(). */ minlen = iphlen; if (igmplen >= IGMP_V3_QUERY_MINLEN) minlen += IGMP_V3_QUERY_MINLEN; else minlen += IGMP_MINLEN; if ((!M_WRITABLE(m) || m->m_len < minlen) && (m = m_pullup(m, minlen)) == NULL) { IGMPSTAT_INC(igps_rcv_tooshort); return (IPPROTO_DONE); } ip = mtod(m, struct ip *); /* * Validate checksum. */ m->m_data += iphlen; m->m_len -= iphlen; igmp = mtod(m, struct igmp *); if (in_cksum(m, igmplen)) { IGMPSTAT_INC(igps_rcv_badsum); m_freem(m); return (IPPROTO_DONE); } m->m_data -= iphlen; m->m_len += iphlen; /* * IGMP control traffic is link-scope, and must have a TTL of 1. * DVMRP traffic (e.g. mrinfo, mtrace) is an exception; * probe packets may come from beyond the LAN. */ if (igmp->igmp_type != IGMP_DVMRP && ip->ip_ttl != 1) { IGMPSTAT_INC(igps_rcv_badttl); m_freem(m); return (IPPROTO_DONE); } switch (igmp->igmp_type) { case IGMP_HOST_MEMBERSHIP_QUERY: if (igmplen == IGMP_MINLEN) { if (igmp->igmp_code == 0) queryver = IGMP_VERSION_1; else queryver = IGMP_VERSION_2; } else if (igmplen >= IGMP_V3_QUERY_MINLEN) { queryver = IGMP_VERSION_3; } else { IGMPSTAT_INC(igps_rcv_tooshort); m_freem(m); return (IPPROTO_DONE); } switch (queryver) { case IGMP_VERSION_1: IGMPSTAT_INC(igps_rcv_v1v2_queries); if (!V_igmp_v1enable) break; if (igmp_input_v1_query(ifp, ip, igmp) != 0) { m_freem(m); return (IPPROTO_DONE); } break; case IGMP_VERSION_2: IGMPSTAT_INC(igps_rcv_v1v2_queries); if (!V_igmp_v2enable) break; if (igmp_input_v2_query(ifp, ip, igmp) != 0) { m_freem(m); return (IPPROTO_DONE); } break; case IGMP_VERSION_3: { struct igmpv3 *igmpv3; uint16_t igmpv3len; uint16_t nsrc; IGMPSTAT_INC(igps_rcv_v3_queries); igmpv3 = (struct igmpv3 *)igmp; /* * Validate length based on source count. */ nsrc = ntohs(igmpv3->igmp_numsrc); if (nsrc * sizeof(in_addr_t) > UINT16_MAX - iphlen - IGMP_V3_QUERY_MINLEN) { IGMPSTAT_INC(igps_rcv_tooshort); return (IPPROTO_DONE); } /* * m_pullup() may modify m, so pullup in * this scope. */ igmpv3len = iphlen + IGMP_V3_QUERY_MINLEN + sizeof(struct in_addr) * nsrc; if ((!M_WRITABLE(m) || m->m_len < igmpv3len) && (m = m_pullup(m, igmpv3len)) == NULL) { IGMPSTAT_INC(igps_rcv_tooshort); return (IPPROTO_DONE); } igmpv3 = (struct igmpv3 *)(mtod(m, uint8_t *) + iphlen); if (igmp_input_v3_query(ifp, ip, igmpv3) != 0) { m_freem(m); return (IPPROTO_DONE); } } break; } break; case IGMP_v1_HOST_MEMBERSHIP_REPORT: if (!V_igmp_v1enable) break; if (igmp_input_v1_report(ifp, ip, igmp) != 0) { m_freem(m); return (IPPROTO_DONE); } break; case IGMP_v2_HOST_MEMBERSHIP_REPORT: if (!V_igmp_v2enable) break; if (!ip_checkrouteralert(m)) IGMPSTAT_INC(igps_rcv_nora); if (igmp_input_v2_report(ifp, ip, igmp) != 0) { m_freem(m); return (IPPROTO_DONE); } break; case IGMP_v3_HOST_MEMBERSHIP_REPORT: /* * Hosts do not need to process IGMPv3 membership reports, * as report suppression is no longer required. */ if (!ip_checkrouteralert(m)) IGMPSTAT_INC(igps_rcv_nora); break; default: break; } /* * Pass all valid IGMP packets up to any process(es) listening on a * raw IGMP socket. */ *mp = m; return (rip_input(mp, offp, proto)); } /* * Fast timeout handler (global). * VIMAGE: Timeout handlers are expected to service all vimages. */ void igmp_fasttimo(void) { VNET_ITERATOR_DECL(vnet_iter); VNET_LIST_RLOCK_NOSLEEP(); VNET_FOREACH(vnet_iter) { CURVNET_SET(vnet_iter); igmp_fasttimo_vnet(); CURVNET_RESTORE(); } VNET_LIST_RUNLOCK_NOSLEEP(); } /* * Fast timeout handler (per-vnet). * Sends are shuffled off to a netisr to deal with Giant. * * VIMAGE: Assume caller has set up our curvnet. */ static void igmp_fasttimo_vnet(void) { struct mbufq scq; /* State-change packets */ struct mbufq qrq; /* Query response packets */ struct ifnet *ifp; struct igmp_ifsoftc *igi; struct ifmultiaddr *ifma, *next; struct in_multi *inm; struct in_multi_head inm_free_tmp; int loop, uri_fasthz; loop = 0; uri_fasthz = 0; /* * Quick check to see if any work needs to be done, in order to * minimize the overhead of fasttimo processing. * SMPng: XXX Unlocked reads. */ if (!V_current_state_timers_running && !V_interface_timers_running && !V_state_change_timers_running) return; SLIST_INIT(&inm_free_tmp); IN_MULTI_LIST_LOCK(); IGMP_LOCK(); /* * IGMPv3 General Query response timer processing. */ if (V_interface_timers_running) { CTR1(KTR_IGMPV3, "%s: interface timers running", __func__); V_interface_timers_running = 0; LIST_FOREACH(igi, &V_igi_head, igi_link) { if (igi->igi_v3_timer == 0) { /* Do nothing. */ } else if (--igi->igi_v3_timer == 0) { igmp_v3_dispatch_general_query(igi); } else { V_interface_timers_running = 1; } } } if (!V_current_state_timers_running && !V_state_change_timers_running) goto out_locked; V_current_state_timers_running = 0; V_state_change_timers_running = 0; CTR1(KTR_IGMPV3, "%s: state change timers running", __func__); /* * IGMPv1/v2/v3 host report and state-change timer processing. * Note: Processing a v3 group timer may remove a node. */ LIST_FOREACH(igi, &V_igi_head, igi_link) { ifp = igi->igi_ifp; if (igi->igi_version == IGMP_VERSION_3) { loop = (igi->igi_flags & IGIF_LOOPBACK) ? 1 : 0; uri_fasthz = IGMP_RANDOM_DELAY(igi->igi_uri * PR_FASTHZ); mbufq_init(&qrq, IGMP_MAX_G_GS_PACKETS); mbufq_init(&scq, IGMP_MAX_STATE_CHANGE_PACKETS); } IF_ADDR_WLOCK(ifp); restart: CK_STAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, next) { if (ifma->ifma_addr->sa_family != AF_INET || ifma->ifma_protospec == NULL) continue; inm = (struct in_multi *)ifma->ifma_protospec; switch (igi->igi_version) { case IGMP_VERSION_1: case IGMP_VERSION_2: igmp_v1v2_process_group_timer(inm, igi->igi_version); break; case IGMP_VERSION_3: igmp_v3_process_group_timers(&inm_free_tmp, &qrq, &scq, inm, uri_fasthz); break; } if (__predict_false(ifma_restart)) { ifma_restart = false; goto restart; } } IF_ADDR_WUNLOCK(ifp); if (igi->igi_version == IGMP_VERSION_3) { igmp_dispatch_queue(&qrq, 0, loop); igmp_dispatch_queue(&scq, 0, loop); /* * Free the in_multi reference(s) for this * IGMP lifecycle. */ inm_release_list_deferred(&inm_free_tmp); } } out_locked: IGMP_UNLOCK(); IN_MULTI_LIST_UNLOCK(); } /* * Update host report group timer for IGMPv1/v2. * Will update the global pending timer flags. */ static void igmp_v1v2_process_group_timer(struct in_multi *inm, const int version) { int report_timer_expired; IN_MULTI_LIST_LOCK_ASSERT(); IGMP_LOCK_ASSERT(); if (inm->inm_timer == 0) { report_timer_expired = 0; } else if (--inm->inm_timer == 0) { report_timer_expired = 1; } else { V_current_state_timers_running = 1; return; } switch (inm->inm_state) { case IGMP_NOT_MEMBER: case IGMP_SILENT_MEMBER: case IGMP_IDLE_MEMBER: case IGMP_LAZY_MEMBER: case IGMP_SLEEPING_MEMBER: case IGMP_AWAKENING_MEMBER: break; case IGMP_REPORTING_MEMBER: if (report_timer_expired) { inm->inm_state = IGMP_IDLE_MEMBER; (void)igmp_v1v2_queue_report(inm, (version == IGMP_VERSION_2) ? IGMP_v2_HOST_MEMBERSHIP_REPORT : IGMP_v1_HOST_MEMBERSHIP_REPORT); } break; case IGMP_G_QUERY_PENDING_MEMBER: case IGMP_SG_QUERY_PENDING_MEMBER: case IGMP_LEAVING_MEMBER: break; } } /* * Update a group's timers for IGMPv3. * Will update the global pending timer flags. * Note: Unlocked read from igi. */ static void igmp_v3_process_group_timers(struct in_multi_head *inmh, struct mbufq *qrq, struct mbufq *scq, struct in_multi *inm, const int uri_fasthz) { int query_response_timer_expired; int state_change_retransmit_timer_expired; IN_MULTI_LIST_LOCK_ASSERT(); IGMP_LOCK_ASSERT(); query_response_timer_expired = 0; state_change_retransmit_timer_expired = 0; /* * During a transition from v1/v2 compatibility mode back to v3, * a group record in REPORTING state may still have its group * timer active. This is a no-op in this function; it is easier * to deal with it here than to complicate the slow-timeout path. */ if (inm->inm_timer == 0) { query_response_timer_expired = 0; } else if (--inm->inm_timer == 0) { query_response_timer_expired = 1; } else { V_current_state_timers_running = 1; } if (inm->inm_sctimer == 0) { state_change_retransmit_timer_expired = 0; } else if (--inm->inm_sctimer == 0) { state_change_retransmit_timer_expired = 1; } else { V_state_change_timers_running = 1; } /* We are in fasttimo, so be quick about it. */ if (!state_change_retransmit_timer_expired && !query_response_timer_expired) return; switch (inm->inm_state) { case IGMP_NOT_MEMBER: case IGMP_SILENT_MEMBER: case IGMP_SLEEPING_MEMBER: case IGMP_LAZY_MEMBER: case IGMP_AWAKENING_MEMBER: case IGMP_IDLE_MEMBER: break; case IGMP_G_QUERY_PENDING_MEMBER: case IGMP_SG_QUERY_PENDING_MEMBER: /* * Respond to a previously pending Group-Specific * or Group-and-Source-Specific query by enqueueing * the appropriate Current-State report for * immediate transmission. */ if (query_response_timer_expired) { int retval __unused; retval = igmp_v3_enqueue_group_record(qrq, inm, 0, 1, (inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER)); CTR2(KTR_IGMPV3, "%s: enqueue record = %d", __func__, retval); inm->inm_state = IGMP_REPORTING_MEMBER; /* XXX Clear recorded sources for next time. */ inm_clear_recorded(inm); } /* FALLTHROUGH */ case IGMP_REPORTING_MEMBER: case IGMP_LEAVING_MEMBER: if (state_change_retransmit_timer_expired) { /* * State-change retransmission timer fired. * If there are any further pending retransmissions, * set the global pending state-change flag, and * reset the timer. */ if (--inm->inm_scrv > 0) { inm->inm_sctimer = uri_fasthz; V_state_change_timers_running = 1; } /* * Retransmit the previously computed state-change * report. If there are no further pending * retransmissions, the mbuf queue will be consumed. * Update T0 state to T1 as we have now sent * a state-change. */ (void)igmp_v3_merge_state_changes(inm, scq); inm_commit(inm); CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__, ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname); /* * If we are leaving the group for good, make sure * we release IGMP's reference to it. * This release must be deferred using a SLIST, * as we are called from a loop which traverses * the in_ifmultiaddr TAILQ. */ if (inm->inm_state == IGMP_LEAVING_MEMBER && inm->inm_scrv == 0) { inm->inm_state = IGMP_NOT_MEMBER; inm_rele_locked(inmh, inm); } } break; } } /* * Suppress a group's pending response to a group or source/group query. * * Do NOT suppress state changes. This leads to IGMPv3 inconsistency. * Do NOT update ST1/ST0 as this operation merely suppresses * the currently pending group record. * Do NOT suppress the response to a general query. It is possible but * it would require adding another state or flag. */ static void igmp_v3_suppress_group_record(struct in_multi *inm) { IN_MULTI_LIST_LOCK_ASSERT(); KASSERT(inm->inm_igi->igi_version == IGMP_VERSION_3, ("%s: not IGMPv3 mode on link", __func__)); if (inm->inm_state != IGMP_G_QUERY_PENDING_MEMBER || inm->inm_state != IGMP_SG_QUERY_PENDING_MEMBER) return; if (inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER) inm_clear_recorded(inm); inm->inm_timer = 0; inm->inm_state = IGMP_REPORTING_MEMBER; } /* * Switch to a different IGMP version on the given interface, * as per Section 7.2.1. */ static void igmp_set_version(struct igmp_ifsoftc *igi, const int version) { int old_version_timer; IGMP_LOCK_ASSERT(); CTR4(KTR_IGMPV3, "%s: switching to v%d on ifp %p(%s)", __func__, version, igi->igi_ifp, igi->igi_ifp->if_xname); if (version == IGMP_VERSION_1 || version == IGMP_VERSION_2) { /* * Compute the "Older Version Querier Present" timer as per * Section 8.12. */ old_version_timer = igi->igi_rv * igi->igi_qi + igi->igi_qri; old_version_timer *= PR_SLOWHZ; if (version == IGMP_VERSION_1) { igi->igi_v1_timer = old_version_timer; igi->igi_v2_timer = 0; } else if (version == IGMP_VERSION_2) { igi->igi_v1_timer = 0; igi->igi_v2_timer = old_version_timer; } } if (igi->igi_v1_timer == 0 && igi->igi_v2_timer > 0) { if (igi->igi_version != IGMP_VERSION_2) { igi->igi_version = IGMP_VERSION_2; igmp_v3_cancel_link_timers(igi); } } else if (igi->igi_v1_timer > 0) { if (igi->igi_version != IGMP_VERSION_1) { igi->igi_version = IGMP_VERSION_1; igmp_v3_cancel_link_timers(igi); } } } /* * Cancel pending IGMPv3 timers for the given link and all groups * joined on it; state-change, general-query, and group-query timers. * * Only ever called on a transition from v3 to Compatibility mode. Kill * the timers stone dead (this may be expensive for large N groups), they * will be restarted if Compatibility Mode deems that they must be due to * query processing. */ static void igmp_v3_cancel_link_timers(struct igmp_ifsoftc *igi) { struct ifmultiaddr *ifma, *ifmatmp; struct ifnet *ifp; struct in_multi *inm; struct in_multi_head inm_free_tmp; CTR3(KTR_IGMPV3, "%s: cancel v3 timers on ifp %p(%s)", __func__, igi->igi_ifp, igi->igi_ifp->if_xname); IN_MULTI_LIST_LOCK_ASSERT(); IGMP_LOCK_ASSERT(); NET_EPOCH_ASSERT(); SLIST_INIT(&inm_free_tmp); /* * Stop the v3 General Query Response on this link stone dead. * If fasttimo is woken up due to V_interface_timers_running, * the flag will be cleared if there are no pending link timers. */ igi->igi_v3_timer = 0; /* * Now clear the current-state and state-change report timers * for all memberships scoped to this link. */ ifp = igi->igi_ifp; IF_ADDR_WLOCK(ifp); CK_STAILQ_FOREACH_SAFE(ifma, &ifp->if_multiaddrs, ifma_link, ifmatmp) { if (ifma->ifma_addr->sa_family != AF_INET || ifma->ifma_protospec == NULL) continue; inm = (struct in_multi *)ifma->ifma_protospec; switch (inm->inm_state) { case IGMP_NOT_MEMBER: case IGMP_SILENT_MEMBER: case IGMP_IDLE_MEMBER: case IGMP_LAZY_MEMBER: case IGMP_SLEEPING_MEMBER: case IGMP_AWAKENING_MEMBER: /* * These states are either not relevant in v3 mode, * or are unreported. Do nothing. */ break; case IGMP_LEAVING_MEMBER: /* * If we are leaving the group and switching to * compatibility mode, we need to release the final * reference held for issuing the INCLUDE {}, and * transition to REPORTING to ensure the host leave * message is sent upstream to the old querier -- * transition to NOT would lose the leave and race. */ inm_rele_locked(&inm_free_tmp, inm); /* FALLTHROUGH */ case IGMP_G_QUERY_PENDING_MEMBER: case IGMP_SG_QUERY_PENDING_MEMBER: inm_clear_recorded(inm); /* FALLTHROUGH */ case IGMP_REPORTING_MEMBER: inm->inm_state = IGMP_REPORTING_MEMBER; break; } /* * Always clear state-change and group report timers. * Free any pending IGMPv3 state-change records. */ inm->inm_sctimer = 0; inm->inm_timer = 0; mbufq_drain(&inm->inm_scq); } IF_ADDR_WUNLOCK(ifp); inm_release_list_deferred(&inm_free_tmp); } /* * Update the Older Version Querier Present timers for a link. * See Section 7.2.1 of RFC 3376. */ static void igmp_v1v2_process_querier_timers(struct igmp_ifsoftc *igi) { IGMP_LOCK_ASSERT(); if (igi->igi_v1_timer == 0 && igi->igi_v2_timer == 0) { /* * IGMPv1 and IGMPv2 Querier Present timers expired. * * Revert to IGMPv3. */ if (igi->igi_version != IGMP_VERSION_3) { CTR5(KTR_IGMPV3, "%s: transition from v%d -> v%d on %p(%s)", __func__, igi->igi_version, IGMP_VERSION_3, igi->igi_ifp, igi->igi_ifp->if_xname); igi->igi_version = IGMP_VERSION_3; } } else if (igi->igi_v1_timer == 0 && igi->igi_v2_timer > 0) { /* * IGMPv1 Querier Present timer expired, * IGMPv2 Querier Present timer running. * If IGMPv2 was disabled since last timeout, * revert to IGMPv3. * If IGMPv2 is enabled, revert to IGMPv2. */ if (!V_igmp_v2enable) { CTR5(KTR_IGMPV3, "%s: transition from v%d -> v%d on %p(%s)", __func__, igi->igi_version, IGMP_VERSION_3, igi->igi_ifp, igi->igi_ifp->if_xname); igi->igi_v2_timer = 0; igi->igi_version = IGMP_VERSION_3; } else { --igi->igi_v2_timer; if (igi->igi_version != IGMP_VERSION_2) { CTR5(KTR_IGMPV3, "%s: transition from v%d -> v%d on %p(%s)", __func__, igi->igi_version, IGMP_VERSION_2, igi->igi_ifp, igi->igi_ifp->if_xname); igi->igi_version = IGMP_VERSION_2; igmp_v3_cancel_link_timers(igi); } } } else if (igi->igi_v1_timer > 0) { /* * IGMPv1 Querier Present timer running. * Stop IGMPv2 timer if running. * * If IGMPv1 was disabled since last timeout, * revert to IGMPv3. * If IGMPv1 is enabled, reset IGMPv2 timer if running. */ if (!V_igmp_v1enable) { CTR5(KTR_IGMPV3, "%s: transition from v%d -> v%d on %p(%s)", __func__, igi->igi_version, IGMP_VERSION_3, igi->igi_ifp, igi->igi_ifp->if_xname); igi->igi_v1_timer = 0; igi->igi_version = IGMP_VERSION_3; } else { --igi->igi_v1_timer; } if (igi->igi_v2_timer > 0) { CTR3(KTR_IGMPV3, "%s: cancel v2 timer on %p(%s)", __func__, igi->igi_ifp, igi->igi_ifp->if_xname); igi->igi_v2_timer = 0; } } } /* * Global slowtimo handler. * VIMAGE: Timeout handlers are expected to service all vimages. */ void igmp_slowtimo(void) { VNET_ITERATOR_DECL(vnet_iter); VNET_LIST_RLOCK_NOSLEEP(); VNET_FOREACH(vnet_iter) { CURVNET_SET(vnet_iter); igmp_slowtimo_vnet(); CURVNET_RESTORE(); } VNET_LIST_RUNLOCK_NOSLEEP(); } /* * Per-vnet slowtimo handler. */ static void igmp_slowtimo_vnet(void) { struct igmp_ifsoftc *igi; IGMP_LOCK(); LIST_FOREACH(igi, &V_igi_head, igi_link) { igmp_v1v2_process_querier_timers(igi); } IGMP_UNLOCK(); } /* * Dispatch an IGMPv1/v2 host report or leave message. * These are always small enough to fit inside a single mbuf. */ static int igmp_v1v2_queue_report(struct in_multi *inm, const int type) { struct epoch_tracker et; struct ifnet *ifp; struct igmp *igmp; struct ip *ip; struct mbuf *m; IN_MULTI_LIST_LOCK_ASSERT(); IGMP_LOCK_ASSERT(); ifp = inm->inm_ifp; m = m_gethdr(M_NOWAIT, MT_DATA); if (m == NULL) return (ENOMEM); M_ALIGN(m, sizeof(struct ip) + sizeof(struct igmp)); m->m_pkthdr.len = sizeof(struct ip) + sizeof(struct igmp); m->m_data += sizeof(struct ip); m->m_len = sizeof(struct igmp); igmp = mtod(m, struct igmp *); igmp->igmp_type = type; igmp->igmp_code = 0; igmp->igmp_group = inm->inm_addr; igmp->igmp_cksum = 0; igmp->igmp_cksum = in_cksum(m, sizeof(struct igmp)); m->m_data -= sizeof(struct ip); m->m_len += sizeof(struct ip); ip = mtod(m, struct ip *); ip->ip_tos = 0; ip->ip_len = htons(sizeof(struct ip) + sizeof(struct igmp)); ip->ip_off = 0; ip->ip_p = IPPROTO_IGMP; ip->ip_src.s_addr = INADDR_ANY; if (type == IGMP_HOST_LEAVE_MESSAGE) ip->ip_dst.s_addr = htonl(INADDR_ALLRTRS_GROUP); else ip->ip_dst = inm->inm_addr; igmp_save_context(m, ifp); m->m_flags |= M_IGMPV2; if (inm->inm_igi->igi_flags & IGIF_LOOPBACK) m->m_flags |= M_IGMP_LOOP; CTR2(KTR_IGMPV3, "%s: netisr_dispatch(NETISR_IGMP, %p)", __func__, m); NET_EPOCH_ENTER(et); netisr_dispatch(NETISR_IGMP, m); NET_EPOCH_EXIT(et); return (0); } /* * Process a state change from the upper layer for the given IPv4 group. * * Each socket holds a reference on the in_multi in its own ip_moptions. * The socket layer will have made the necessary updates to.the group * state, it is now up to IGMP to issue a state change report if there * has been any change between T0 (when the last state-change was issued) * and T1 (now). * * We use the IGMPv3 state machine at group level. The IGMP module * however makes the decision as to which IGMP protocol version to speak. * A state change *from* INCLUDE {} always means an initial join. * A state change *to* INCLUDE {} always means a final leave. * * FUTURE: If IGIF_V3LITE is enabled for this interface, then we can * save ourselves a bunch of work; any exclusive mode groups need not * compute source filter lists. * * VIMAGE: curvnet should have been set by caller, as this routine * is called from the socket option handlers. */ int igmp_change_state(struct in_multi *inm) { struct igmp_ifsoftc *igi; struct ifnet *ifp; int error; error = 0; IN_MULTI_LOCK_ASSERT(); /* * Try to detect if the upper layer just asked us to change state * for an interface which has now gone away. */ KASSERT(inm->inm_ifma != NULL, ("%s: no ifma", __func__)); ifp = inm->inm_ifma->ifma_ifp; /* * Sanity check that netinet's notion of ifp is the * same as net's. */ KASSERT(inm->inm_ifp == ifp, ("%s: bad ifp", __func__)); IGMP_LOCK(); igi = ((struct in_ifinfo *)ifp->if_afdata[AF_INET])->ii_igmp; KASSERT(igi != NULL, ("%s: no igmp_ifsoftc for ifp %p", __func__, ifp)); /* * If we detect a state transition to or from MCAST_UNDEFINED * for this group, then we are starting or finishing an IGMP * life cycle for this group. */ if (inm->inm_st[1].iss_fmode != inm->inm_st[0].iss_fmode) { CTR3(KTR_IGMPV3, "%s: inm transition %d -> %d", __func__, inm->inm_st[0].iss_fmode, inm->inm_st[1].iss_fmode); if (inm->inm_st[0].iss_fmode == MCAST_UNDEFINED) { CTR1(KTR_IGMPV3, "%s: initial join", __func__); error = igmp_initial_join(inm, igi); goto out_locked; } else if (inm->inm_st[1].iss_fmode == MCAST_UNDEFINED) { CTR1(KTR_IGMPV3, "%s: final leave", __func__); igmp_final_leave(inm, igi); goto out_locked; } } else { CTR1(KTR_IGMPV3, "%s: filter set change", __func__); } error = igmp_handle_state_change(inm, igi); out_locked: IGMP_UNLOCK(); return (error); } /* * Perform the initial join for an IGMP group. * * When joining a group: * If the group should have its IGMP traffic suppressed, do nothing. * IGMPv1 starts sending IGMPv1 host membership reports. * IGMPv2 starts sending IGMPv2 host membership reports. * IGMPv3 will schedule an IGMPv3 state-change report containing the * initial state of the membership. */ static int igmp_initial_join(struct in_multi *inm, struct igmp_ifsoftc *igi) { struct ifnet *ifp; struct mbufq *mq; int error, retval, syncstates; CTR4(KTR_IGMPV3, "%s: initial join 0x%08x on ifp %p(%s)", __func__, ntohl(inm->inm_addr.s_addr), inm->inm_ifp, inm->inm_ifp->if_xname); error = 0; syncstates = 1; ifp = inm->inm_ifp; IN_MULTI_LOCK_ASSERT(); IGMP_LOCK_ASSERT(); KASSERT(igi && igi->igi_ifp == ifp, ("%s: inconsistent ifp", __func__)); /* * Groups joined on loopback or marked as 'not reported', * e.g. 224.0.0.1, enter the IGMP_SILENT_MEMBER state and * are never reported in any IGMP protocol exchanges. * All other groups enter the appropriate IGMP state machine * for the version in use on this link. * A link marked as IGIF_SILENT causes IGMP to be completely * disabled for the link. */ if ((ifp->if_flags & IFF_LOOPBACK) || (igi->igi_flags & IGIF_SILENT) || !igmp_isgroupreported(inm->inm_addr)) { CTR1(KTR_IGMPV3, "%s: not kicking state machine for silent group", __func__); inm->inm_state = IGMP_SILENT_MEMBER; inm->inm_timer = 0; } else { /* * Deal with overlapping in_multi lifecycle. * If this group was LEAVING, then make sure * we drop the reference we picked up to keep the * group around for the final INCLUDE {} enqueue. */ if (igi->igi_version == IGMP_VERSION_3 && inm->inm_state == IGMP_LEAVING_MEMBER) { MPASS(inm->inm_refcount > 1); inm_rele_locked(NULL, inm); } inm->inm_state = IGMP_REPORTING_MEMBER; switch (igi->igi_version) { case IGMP_VERSION_1: case IGMP_VERSION_2: inm->inm_state = IGMP_IDLE_MEMBER; error = igmp_v1v2_queue_report(inm, (igi->igi_version == IGMP_VERSION_2) ? IGMP_v2_HOST_MEMBERSHIP_REPORT : IGMP_v1_HOST_MEMBERSHIP_REPORT); if (error == 0) { inm->inm_timer = IGMP_RANDOM_DELAY( IGMP_V1V2_MAX_RI * PR_FASTHZ); V_current_state_timers_running = 1; } break; case IGMP_VERSION_3: /* * Defer update of T0 to T1, until the first copy * of the state change has been transmitted. */ syncstates = 0; /* * Immediately enqueue a State-Change Report for * this interface, freeing any previous reports. * Don't kick the timers if there is nothing to do, * or if an error occurred. */ mq = &inm->inm_scq; mbufq_drain(mq); retval = igmp_v3_enqueue_group_record(mq, inm, 1, 0, 0); CTR2(KTR_IGMPV3, "%s: enqueue record = %d", __func__, retval); if (retval <= 0) { error = retval * -1; break; } /* * Schedule transmission of pending state-change * report up to RV times for this link. The timer * will fire at the next igmp_fasttimo (~200ms), * giving us an opportunity to merge the reports. */ if (igi->igi_flags & IGIF_LOOPBACK) { inm->inm_scrv = 1; } else { KASSERT(igi->igi_rv > 1, ("%s: invalid robustness %d", __func__, igi->igi_rv)); inm->inm_scrv = igi->igi_rv; } inm->inm_sctimer = 1; V_state_change_timers_running = 1; error = 0; break; } } /* * Only update the T0 state if state change is atomic, * i.e. we don't need to wait for a timer to fire before we * can consider the state change to have been communicated. */ if (syncstates) { inm_commit(inm); CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__, ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname); } return (error); } /* * Issue an intermediate state change during the IGMP life-cycle. */ static int igmp_handle_state_change(struct in_multi *inm, struct igmp_ifsoftc *igi) { struct ifnet *ifp; int retval; CTR4(KTR_IGMPV3, "%s: state change for 0x%08x on ifp %p(%s)", __func__, ntohl(inm->inm_addr.s_addr), inm->inm_ifp, inm->inm_ifp->if_xname); ifp = inm->inm_ifp; IN_MULTI_LIST_LOCK_ASSERT(); IGMP_LOCK_ASSERT(); KASSERT(igi && igi->igi_ifp == ifp, ("%s: inconsistent ifp", __func__)); if ((ifp->if_flags & IFF_LOOPBACK) || (igi->igi_flags & IGIF_SILENT) || !igmp_isgroupreported(inm->inm_addr) || (igi->igi_version != IGMP_VERSION_3)) { if (!igmp_isgroupreported(inm->inm_addr)) { CTR1(KTR_IGMPV3, "%s: not kicking state machine for silent group", __func__); } CTR1(KTR_IGMPV3, "%s: nothing to do", __func__); inm_commit(inm); CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__, ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname); return (0); } mbufq_drain(&inm->inm_scq); retval = igmp_v3_enqueue_group_record(&inm->inm_scq, inm, 1, 0, 0); CTR2(KTR_IGMPV3, "%s: enqueue record = %d", __func__, retval); if (retval <= 0) return (-retval); /* * If record(s) were enqueued, start the state-change * report timer for this group. */ inm->inm_scrv = ((igi->igi_flags & IGIF_LOOPBACK) ? 1 : igi->igi_rv); inm->inm_sctimer = 1; V_state_change_timers_running = 1; return (0); } /* * Perform the final leave for an IGMP group. * * When leaving a group: * IGMPv1 does nothing. * IGMPv2 sends a host leave message, if and only if we are the reporter. * IGMPv3 enqueues a state-change report containing a transition * to INCLUDE {} for immediate transmission. */ static void igmp_final_leave(struct in_multi *inm, struct igmp_ifsoftc *igi) { int syncstates; syncstates = 1; CTR4(KTR_IGMPV3, "%s: final leave 0x%08x on ifp %p(%s)", __func__, ntohl(inm->inm_addr.s_addr), inm->inm_ifp, inm->inm_ifp->if_xname); IN_MULTI_LIST_LOCK_ASSERT(); IGMP_LOCK_ASSERT(); switch (inm->inm_state) { case IGMP_NOT_MEMBER: case IGMP_SILENT_MEMBER: case IGMP_LEAVING_MEMBER: /* Already leaving or left; do nothing. */ CTR1(KTR_IGMPV3, "%s: not kicking state machine for silent group", __func__); break; case IGMP_REPORTING_MEMBER: case IGMP_IDLE_MEMBER: case IGMP_G_QUERY_PENDING_MEMBER: case IGMP_SG_QUERY_PENDING_MEMBER: if (igi->igi_version == IGMP_VERSION_2) { #ifdef INVARIANTS if (inm->inm_state == IGMP_G_QUERY_PENDING_MEMBER || inm->inm_state == IGMP_SG_QUERY_PENDING_MEMBER) panic("%s: IGMPv3 state reached, not IGMPv3 mode", __func__); #endif igmp_v1v2_queue_report(inm, IGMP_HOST_LEAVE_MESSAGE); inm->inm_state = IGMP_NOT_MEMBER; } else if (igi->igi_version == IGMP_VERSION_3) { /* * Stop group timer and all pending reports. * Immediately enqueue a state-change report * TO_IN {} to be sent on the next fast timeout, * giving us an opportunity to merge reports. */ mbufq_drain(&inm->inm_scq); inm->inm_timer = 0; if (igi->igi_flags & IGIF_LOOPBACK) { inm->inm_scrv = 1; } else { inm->inm_scrv = igi->igi_rv; } CTR4(KTR_IGMPV3, "%s: Leaving 0x%08x/%s with %d " "pending retransmissions.", __func__, ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname, inm->inm_scrv); if (inm->inm_scrv == 0) { inm->inm_state = IGMP_NOT_MEMBER; inm->inm_sctimer = 0; } else { int retval __unused; inm_acquire_locked(inm); retval = igmp_v3_enqueue_group_record( &inm->inm_scq, inm, 1, 0, 0); KASSERT(retval != 0, ("%s: enqueue record = %d", __func__, retval)); inm->inm_state = IGMP_LEAVING_MEMBER; inm->inm_sctimer = 1; V_state_change_timers_running = 1; syncstates = 0; } break; } break; case IGMP_LAZY_MEMBER: case IGMP_SLEEPING_MEMBER: case IGMP_AWAKENING_MEMBER: /* Our reports are suppressed; do nothing. */ break; } if (syncstates) { inm_commit(inm); CTR3(KTR_IGMPV3, "%s: T1 -> T0 for 0x%08x/%s", __func__, ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname); inm->inm_st[1].iss_fmode = MCAST_UNDEFINED; CTR3(KTR_IGMPV3, "%s: T1 now MCAST_UNDEFINED for 0x%08x/%s", __func__, ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname); } } /* * Enqueue an IGMPv3 group record to the given output queue. * * XXX This function could do with having the allocation code * split out, and the multiple-tree-walks coalesced into a single * routine as has been done in igmp_v3_enqueue_filter_change(). * * If is_state_change is zero, a current-state record is appended. * If is_state_change is non-zero, a state-change report is appended. * * If is_group_query is non-zero, an mbuf packet chain is allocated. * If is_group_query is zero, and if there is a packet with free space * at the tail of the queue, it will be appended to providing there * is enough free space. * Otherwise a new mbuf packet chain is allocated. * * If is_source_query is non-zero, each source is checked to see if * it was recorded for a Group-Source query, and will be omitted if * it is not both in-mode and recorded. * * The function will attempt to allocate leading space in the packet * for the IP/IGMP header to be prepended without fragmenting the chain. * * If successful the size of all data appended to the queue is returned, * otherwise an error code less than zero is returned, or zero if * no record(s) were appended. */ static int igmp_v3_enqueue_group_record(struct mbufq *mq, struct in_multi *inm, const int is_state_change, const int is_group_query, const int is_source_query) { struct igmp_grouprec ig; struct igmp_grouprec *pig; struct ifnet *ifp; struct ip_msource *ims, *nims; struct mbuf *m0, *m, *md; int is_filter_list_change; int minrec0len, m0srcs, msrcs, nbytes, off; int record_has_sources; int now; int type; in_addr_t naddr; uint8_t mode; IN_MULTI_LIST_LOCK_ASSERT(); ifp = inm->inm_ifp; is_filter_list_change = 0; m = NULL; m0 = NULL; m0srcs = 0; msrcs = 0; nbytes = 0; nims = NULL; record_has_sources = 1; pig = NULL; type = IGMP_DO_NOTHING; mode = inm->inm_st[1].iss_fmode; /* * If we did not transition out of ASM mode during t0->t1, * and there are no source nodes to process, we can skip * the generation of source records. */ if (inm->inm_st[0].iss_asm > 0 && inm->inm_st[1].iss_asm > 0 && inm->inm_nsrc == 0) record_has_sources = 0; if (is_state_change) { /* * Queue a state change record. * If the mode did not change, and there are non-ASM * listeners or source filters present, * we potentially need to issue two records for the group. * If we are transitioning to MCAST_UNDEFINED, we need * not send any sources. * If there are ASM listeners, and there was no filter * mode transition of any kind, do nothing. */ if (mode != inm->inm_st[0].iss_fmode) { if (mode == MCAST_EXCLUDE) { CTR1(KTR_IGMPV3, "%s: change to EXCLUDE", __func__); type = IGMP_CHANGE_TO_EXCLUDE_MODE; } else { CTR1(KTR_IGMPV3, "%s: change to INCLUDE", __func__); type = IGMP_CHANGE_TO_INCLUDE_MODE; if (mode == MCAST_UNDEFINED) record_has_sources = 0; } } else { if (record_has_sources) { is_filter_list_change = 1; } else { type = IGMP_DO_NOTHING; } } } else { /* * Queue a current state record. */ if (mode == MCAST_EXCLUDE) { type = IGMP_MODE_IS_EXCLUDE; } else if (mode == MCAST_INCLUDE) { type = IGMP_MODE_IS_INCLUDE; KASSERT(inm->inm_st[1].iss_asm == 0, ("%s: inm %p is INCLUDE but ASM count is %d", __func__, inm, inm->inm_st[1].iss_asm)); } } /* * Generate the filter list changes using a separate function. */ if (is_filter_list_change) return (igmp_v3_enqueue_filter_change(mq, inm)); if (type == IGMP_DO_NOTHING) { CTR3(KTR_IGMPV3, "%s: nothing to do for 0x%08x/%s", __func__, ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname); return (0); } /* * If any sources are present, we must be able to fit at least * one in the trailing space of the tail packet's mbuf, * ideally more. */ minrec0len = sizeof(struct igmp_grouprec); if (record_has_sources) minrec0len += sizeof(in_addr_t); CTR4(KTR_IGMPV3, "%s: queueing %s for 0x%08x/%s", __func__, igmp_rec_type_to_str(type), ntohl(inm->inm_addr.s_addr), inm->inm_ifp->if_xname); /* * Check if we have a packet in the tail of the queue for this * group into which the first group record for this group will fit. * Otherwise allocate a new packet. * Always allocate leading space for IP+RA_OPT+IGMP+REPORT. * Note: Group records for G/GSR query responses MUST be sent * in their own packet. */ m0 = mbufq_last(mq); if (!is_group_query && m0 != NULL && (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= IGMP_V3_REPORT_MAXRECS) && (m0->m_pkthdr.len + minrec0len) < (ifp->if_mtu - IGMP_LEADINGSPACE)) { m0srcs = (ifp->if_mtu - m0->m_pkthdr.len - sizeof(struct igmp_grouprec)) / sizeof(in_addr_t); m = m0; CTR1(KTR_IGMPV3, "%s: use existing packet", __func__); } else { if (mbufq_full(mq)) { CTR1(KTR_IGMPV3, "%s: outbound queue full", __func__); return (-ENOMEM); } m = NULL; m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE - sizeof(struct igmp_grouprec)) / sizeof(in_addr_t); if (!is_state_change && !is_group_query) { m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); if (m) m->m_data += IGMP_LEADINGSPACE; } if (m == NULL) { m = m_gethdr(M_NOWAIT, MT_DATA); if (m) M_ALIGN(m, IGMP_LEADINGSPACE); } if (m == NULL) return (-ENOMEM); igmp_save_context(m, ifp); CTR1(KTR_IGMPV3, "%s: allocated first packet", __func__); } /* * Append group record. * If we have sources, we don't know how many yet. */ ig.ig_type = type; ig.ig_datalen = 0; ig.ig_numsrc = 0; ig.ig_group = inm->inm_addr; if (!m_append(m, sizeof(struct igmp_grouprec), (void *)&ig)) { if (m != m0) m_freem(m); CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__); return (-ENOMEM); } nbytes += sizeof(struct igmp_grouprec); /* * Append as many sources as will fit in the first packet. * If we are appending to a new packet, the chain allocation * may potentially use clusters; use m_getptr() in this case. * If we are appending to an existing packet, we need to obtain * a pointer to the group record after m_append(), in case a new * mbuf was allocated. * Only append sources which are in-mode at t1. If we are * transitioning to MCAST_UNDEFINED state on the group, do not * include source entries. * Only report recorded sources in our filter set when responding * to a group-source query. */ if (record_has_sources) { if (m == m0) { md = m_last(m); pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) + md->m_len - nbytes); } else { md = m_getptr(m, 0, &off); pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) + off); } msrcs = 0; RB_FOREACH_SAFE(ims, ip_msource_tree, &inm->inm_srcs, nims) { CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__, ims->ims_haddr); now = ims_get_mode(inm, ims, 1); CTR2(KTR_IGMPV3, "%s: node is %d", __func__, now); if ((now != mode) || (now == mode && mode == MCAST_UNDEFINED)) { CTR1(KTR_IGMPV3, "%s: skip node", __func__); continue; } if (is_source_query && ims->ims_stp == 0) { CTR1(KTR_IGMPV3, "%s: skip unrecorded node", __func__); continue; } CTR1(KTR_IGMPV3, "%s: append node", __func__); naddr = htonl(ims->ims_haddr); if (!m_append(m, sizeof(in_addr_t), (void *)&naddr)) { if (m != m0) m_freem(m); CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__); return (-ENOMEM); } nbytes += sizeof(in_addr_t); ++msrcs; if (msrcs == m0srcs) break; } CTR2(KTR_IGMPV3, "%s: msrcs is %d this packet", __func__, msrcs); pig->ig_numsrc = htons(msrcs); nbytes += (msrcs * sizeof(in_addr_t)); } if (is_source_query && msrcs == 0) { CTR1(KTR_IGMPV3, "%s: no recorded sources to report", __func__); if (m != m0) m_freem(m); return (0); } /* * We are good to go with first packet. */ if (m != m0) { CTR1(KTR_IGMPV3, "%s: enqueueing first packet", __func__); m->m_pkthdr.PH_vt.vt_nrecs = 1; mbufq_enqueue(mq, m); } else m->m_pkthdr.PH_vt.vt_nrecs++; /* * No further work needed if no source list in packet(s). */ if (!record_has_sources) return (nbytes); /* * Whilst sources remain to be announced, we need to allocate * a new packet and fill out as many sources as will fit. * Always try for a cluster first. */ while (nims != NULL) { if (mbufq_full(mq)) { CTR1(KTR_IGMPV3, "%s: outbound queue full", __func__); return (-ENOMEM); } m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); if (m) m->m_data += IGMP_LEADINGSPACE; if (m == NULL) { m = m_gethdr(M_NOWAIT, MT_DATA); if (m) M_ALIGN(m, IGMP_LEADINGSPACE); } if (m == NULL) return (-ENOMEM); igmp_save_context(m, ifp); md = m_getptr(m, 0, &off); pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) + off); CTR1(KTR_IGMPV3, "%s: allocated next packet", __func__); if (!m_append(m, sizeof(struct igmp_grouprec), (void *)&ig)) { if (m != m0) m_freem(m); CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__); return (-ENOMEM); } m->m_pkthdr.PH_vt.vt_nrecs = 1; nbytes += sizeof(struct igmp_grouprec); m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE - sizeof(struct igmp_grouprec)) / sizeof(in_addr_t); msrcs = 0; RB_FOREACH_FROM(ims, ip_msource_tree, nims) { CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__, ims->ims_haddr); now = ims_get_mode(inm, ims, 1); if ((now != mode) || (now == mode && mode == MCAST_UNDEFINED)) { CTR1(KTR_IGMPV3, "%s: skip node", __func__); continue; } if (is_source_query && ims->ims_stp == 0) { CTR1(KTR_IGMPV3, "%s: skip unrecorded node", __func__); continue; } CTR1(KTR_IGMPV3, "%s: append node", __func__); naddr = htonl(ims->ims_haddr); if (!m_append(m, sizeof(in_addr_t), (void *)&naddr)) { if (m != m0) m_freem(m); CTR1(KTR_IGMPV3, "%s: m_append() failed.", __func__); return (-ENOMEM); } ++msrcs; if (msrcs == m0srcs) break; } pig->ig_numsrc = htons(msrcs); nbytes += (msrcs * sizeof(in_addr_t)); CTR1(KTR_IGMPV3, "%s: enqueueing next packet", __func__); mbufq_enqueue(mq, m); } return (nbytes); } /* * Type used to mark record pass completion. * We exploit the fact we can cast to this easily from the * current filter modes on each ip_msource node. */ typedef enum { REC_NONE = 0x00, /* MCAST_UNDEFINED */ REC_ALLOW = 0x01, /* MCAST_INCLUDE */ REC_BLOCK = 0x02, /* MCAST_EXCLUDE */ REC_FULL = REC_ALLOW | REC_BLOCK } rectype_t; /* * Enqueue an IGMPv3 filter list change to the given output queue. * * Source list filter state is held in an RB-tree. When the filter list * for a group is changed without changing its mode, we need to compute * the deltas between T0 and T1 for each source in the filter set, * and enqueue the appropriate ALLOW_NEW/BLOCK_OLD records. * * As we may potentially queue two record types, and the entire R-B tree * needs to be walked at once, we break this out into its own function * so we can generate a tightly packed queue of packets. * * XXX This could be written to only use one tree walk, although that makes * serializing into the mbuf chains a bit harder. For now we do two walks * which makes things easier on us, and it may or may not be harder on * the L2 cache. * * If successful the size of all data appended to the queue is returned, * otherwise an error code less than zero is returned, or zero if * no record(s) were appended. */ static int igmp_v3_enqueue_filter_change(struct mbufq *mq, struct in_multi *inm) { static const int MINRECLEN = sizeof(struct igmp_grouprec) + sizeof(in_addr_t); struct ifnet *ifp; struct igmp_grouprec ig; struct igmp_grouprec *pig; struct ip_msource *ims, *nims; struct mbuf *m, *m0, *md; in_addr_t naddr; int m0srcs, nbytes, npbytes, off, rsrcs, schanged; int nallow, nblock; uint8_t mode, now, then; rectype_t crt, drt, nrt; IN_MULTI_LIST_LOCK_ASSERT(); if (inm->inm_nsrc == 0 || (inm->inm_st[0].iss_asm > 0 && inm->inm_st[1].iss_asm > 0)) return (0); ifp = inm->inm_ifp; /* interface */ mode = inm->inm_st[1].iss_fmode; /* filter mode at t1 */ crt = REC_NONE; /* current group record type */ drt = REC_NONE; /* mask of completed group record types */ nrt = REC_NONE; /* record type for current node */ m0srcs = 0; /* # source which will fit in current mbuf chain */ nbytes = 0; /* # of bytes appended to group's state-change queue */ npbytes = 0; /* # of bytes appended this packet */ rsrcs = 0; /* # sources encoded in current record */ schanged = 0; /* # nodes encoded in overall filter change */ nallow = 0; /* # of source entries in ALLOW_NEW */ nblock = 0; /* # of source entries in BLOCK_OLD */ nims = NULL; /* next tree node pointer */ /* * For each possible filter record mode. * The first kind of source we encounter tells us which * is the first kind of record we start appending. * If a node transitioned to UNDEFINED at t1, its mode is treated * as the inverse of the group's filter mode. */ while (drt != REC_FULL) { do { m0 = mbufq_last(mq); if (m0 != NULL && (m0->m_pkthdr.PH_vt.vt_nrecs + 1 <= IGMP_V3_REPORT_MAXRECS) && (m0->m_pkthdr.len + MINRECLEN) < (ifp->if_mtu - IGMP_LEADINGSPACE)) { m = m0; m0srcs = (ifp->if_mtu - m0->m_pkthdr.len - sizeof(struct igmp_grouprec)) / sizeof(in_addr_t); CTR1(KTR_IGMPV3, "%s: use previous packet", __func__); } else { m = m_getcl(M_NOWAIT, MT_DATA, M_PKTHDR); if (m) m->m_data += IGMP_LEADINGSPACE; if (m == NULL) { m = m_gethdr(M_NOWAIT, MT_DATA); if (m) M_ALIGN(m, IGMP_LEADINGSPACE); } if (m == NULL) { CTR1(KTR_IGMPV3, "%s: m_get*() failed", __func__); return (-ENOMEM); } m->m_pkthdr.PH_vt.vt_nrecs = 0; igmp_save_context(m, ifp); m0srcs = (ifp->if_mtu - IGMP_LEADINGSPACE - sizeof(struct igmp_grouprec)) / sizeof(in_addr_t); npbytes = 0; CTR1(KTR_IGMPV3, "%s: allocated new packet", __func__); } /* * Append the IGMP group record header to the * current packet's data area. * Recalculate pointer to free space for next * group record, in case m_append() allocated * a new mbuf or cluster. */ memset(&ig, 0, sizeof(ig)); ig.ig_group = inm->inm_addr; if (!m_append(m, sizeof(ig), (void *)&ig)) { if (m != m0) m_freem(m); CTR1(KTR_IGMPV3, "%s: m_append() failed", __func__); return (-ENOMEM); } npbytes += sizeof(struct igmp_grouprec); if (m != m0) { /* new packet; offset in c hain */ md = m_getptr(m, npbytes - sizeof(struct igmp_grouprec), &off); pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) + off); } else { /* current packet; offset from last append */ md = m_last(m); pig = (struct igmp_grouprec *)(mtod(md, uint8_t *) + md->m_len - sizeof(struct igmp_grouprec)); } /* * Begin walking the tree for this record type * pass, or continue from where we left off * previously if we had to allocate a new packet. * Only report deltas in-mode at t1. * We need not report included sources as allowed * if we are in inclusive mode on the group, * however the converse is not true. */ rsrcs = 0; if (nims == NULL) nims = RB_MIN(ip_msource_tree, &inm->inm_srcs); RB_FOREACH_FROM(ims, ip_msource_tree, nims) { CTR2(KTR_IGMPV3, "%s: visit node 0x%08x", __func__, ims->ims_haddr); now = ims_get_mode(inm, ims, 1); then = ims_get_mode(inm, ims, 0); CTR3(KTR_IGMPV3, "%s: mode: t0 %d, t1 %d", __func__, then, now); if (now == then) { CTR1(KTR_IGMPV3, "%s: skip unchanged", __func__); continue; } if (mode == MCAST_EXCLUDE && now == MCAST_INCLUDE) { CTR1(KTR_IGMPV3, "%s: skip IN src on EX group", __func__); continue; } nrt = (rectype_t)now; if (nrt == REC_NONE) nrt = (rectype_t)(~mode & REC_FULL); if (schanged++ == 0) { crt = nrt; } else if (crt != nrt) continue; naddr = htonl(ims->ims_haddr); if (!m_append(m, sizeof(in_addr_t), (void *)&naddr)) { if (m != m0) m_freem(m); CTR1(KTR_IGMPV3, "%s: m_append() failed", __func__); return (-ENOMEM); } nallow += !!(crt == REC_ALLOW); nblock += !!(crt == REC_BLOCK); if (++rsrcs == m0srcs) break; } /* * If we did not append any tree nodes on this * pass, back out of allocations. */ if (rsrcs == 0) { npbytes -= sizeof(struct igmp_grouprec); if (m != m0) { CTR1(KTR_IGMPV3, "%s: m_free(m)", __func__); m_freem(m); } else { CTR1(KTR_IGMPV3, "%s: m_adj(m, -ig)", __func__); m_adj(m, -((int)sizeof( struct igmp_grouprec))); } continue; } npbytes += (rsrcs * sizeof(in_addr_t)); if (crt == REC_ALLOW) pig->ig_type = IGMP_ALLOW_NEW_SOURCES; else if (crt == REC_BLOCK) pig->ig_type = IGMP_BLOCK_OLD_SOURCES; pig->ig_numsrc = htons(rsrcs); /* * Count the new group record, and enqueue this * packet if it wasn't already queued. */ m->m_pkthdr.PH_vt.vt_nrecs++; if (m != m0) mbufq_enqueue(mq, m); nbytes += npbytes; } while (nims != NULL); drt |= crt; crt = (~crt & REC_FULL); } CTR3(KTR_IGMPV3, "%s: queued %d ALLOW_NEW, %d BLOCK_OLD", __func__, nallow, nblock); return (nbytes); } static int igmp_v3_merge_state_changes(struct in_multi *inm, struct mbufq *scq) { struct mbufq *gq; struct mbuf *m; /* pending state-change */ struct mbuf *m0; /* copy of pending state-change */ struct mbuf *mt; /* last state-change in packet */ int docopy, domerge; u_int recslen; docopy = 0; domerge = 0; recslen = 0; IN_MULTI_LIST_LOCK_ASSERT(); IGMP_LOCK_ASSERT(); /* * If there are further pending retransmissions, make a writable * copy of each queued state-change message before merging. */ if (inm->inm_scrv > 0) docopy = 1; gq = &inm->inm_scq; #ifdef KTR if (mbufq_first(gq) == NULL) { CTR2(KTR_IGMPV3, "%s: WARNING: queue for inm %p is empty", __func__, inm); } #endif m = mbufq_first(gq); while (m != NULL) { /* * Only merge the report into the current packet if * there is sufficient space to do so; an IGMPv3 report * packet may only contain 65,535 group records. * Always use a simple mbuf chain concatentation to do this, * as large state changes for single groups may have * allocated clusters. */ domerge = 0; mt = mbufq_last(scq); if (mt != NULL) { recslen = m_length(m, NULL); if ((mt->m_pkthdr.PH_vt.vt_nrecs + m->m_pkthdr.PH_vt.vt_nrecs <= IGMP_V3_REPORT_MAXRECS) && (mt->m_pkthdr.len + recslen <= (inm->inm_ifp->if_mtu - IGMP_LEADINGSPACE))) domerge = 1; } if (!domerge && mbufq_full(gq)) { CTR2(KTR_IGMPV3, "%s: outbound queue full, skipping whole packet %p", __func__, m); mt = m->m_nextpkt; if (!docopy) m_freem(m); m = mt; continue; } if (!docopy) { CTR2(KTR_IGMPV3, "%s: dequeueing %p", __func__, m); m0 = mbufq_dequeue(gq); m = m0->m_nextpkt; } else { CTR2(KTR_IGMPV3, "%s: copying %p", __func__, m); m0 = m_dup(m, M_NOWAIT); if (m0 == NULL) return (ENOMEM); m0->m_nextpkt = NULL; m = m->m_nextpkt; } if (!domerge) { CTR3(KTR_IGMPV3, "%s: queueing %p to scq %p)", __func__, m0, scq); mbufq_enqueue(scq, m0); } else { struct mbuf *mtl; /* last mbuf of packet mt */ CTR3(KTR_IGMPV3, "%s: merging %p with scq tail %p)", __func__, m0, mt); mtl = m_last(mt); m0->m_flags &= ~M_PKTHDR; mt->m_pkthdr.len += recslen; mt->m_pkthdr.PH_vt.vt_nrecs += m0->m_pkthdr.PH_vt.vt_nrecs; mtl->m_next = m0; } } return (0); } /* * Respond to a pending IGMPv3 General Query. */ static void igmp_v3_dispatch_general_query(struct igmp_ifsoftc *igi) { struct ifmultiaddr *ifma; struct ifnet *ifp; struct in_multi *inm; int retval __unused, loop; IN_MULTI_LIST_LOCK_ASSERT(); IGMP_LOCK_ASSERT(); NET_EPOCH_ASSERT(); KASSERT(igi->igi_version == IGMP_VERSION_3, ("%s: called when version %d", __func__, igi->igi_version)); /* * Check that there are some packets queued. If so, send them first. * For large number of groups the reply to general query can take * many packets, we should finish sending them before starting of * queuing the new reply. */ if (mbufq_len(&igi->igi_gq) != 0) goto send; ifp = igi->igi_ifp; CK_STAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) { if (ifma->ifma_addr->sa_family != AF_INET || ifma->ifma_protospec == NULL) continue; inm = (struct in_multi *)ifma->ifma_protospec; KASSERT(ifp == inm->inm_ifp, ("%s: inconsistent ifp", __func__)); switch (inm->inm_state) { case IGMP_NOT_MEMBER: case IGMP_SILENT_MEMBER: break; case IGMP_REPORTING_MEMBER: case IGMP_IDLE_MEMBER: case IGMP_LAZY_MEMBER: case IGMP_SLEEPING_MEMBER: case IGMP_AWAKENING_MEMBER: inm->inm_state = IGMP_REPORTING_MEMBER; retval = igmp_v3_enqueue_group_record(&igi->igi_gq, inm, 0, 0, 0); CTR2(KTR_IGMPV3, "%s: enqueue record = %d", __func__, retval); break; case IGMP_G_QUERY_PENDING_MEMBER: case IGMP_SG_QUERY_PENDING_MEMBER: case IGMP_LEAVING_MEMBER: break; } } send: loop = (igi->igi_flags & IGIF_LOOPBACK) ? 1 : 0; igmp_dispatch_queue(&igi->igi_gq, IGMP_MAX_RESPONSE_BURST, loop); /* * Slew transmission of bursts over 500ms intervals. */ if (mbufq_first(&igi->igi_gq) != NULL) { igi->igi_v3_timer = 1 + IGMP_RANDOM_DELAY( IGMP_RESPONSE_BURST_INTERVAL); V_interface_timers_running = 1; } } /* * Transmit the next pending IGMP message in the output queue. * * We get called from netisr_processqueue(). A mutex private to igmpoq * will be acquired and released around this routine. * * VIMAGE: Needs to store/restore vnet pointer on a per-mbuf-chain basis. * MRT: Nothing needs to be done, as IGMP traffic is always local to * a link and uses a link-scope multicast address. */ static void igmp_intr(struct mbuf *m) { struct ip_moptions imo; struct ifnet *ifp; struct mbuf *ipopts, *m0; int error; uint32_t ifindex; CTR2(KTR_IGMPV3, "%s: transmit %p", __func__, m); /* * Set VNET image pointer from enqueued mbuf chain * before doing anything else. Whilst we use interface * indexes to guard against interface detach, they are * unique to each VIMAGE and must be retrieved. */ CURVNET_SET((struct vnet *)(m->m_pkthdr.PH_loc.ptr)); ifindex = igmp_restore_context(m); /* * Check if the ifnet still exists. This limits the scope of * any race in the absence of a global ifp lock for low cost * (an array lookup). */ ifp = ifnet_byindex(ifindex); if (ifp == NULL) { CTR3(KTR_IGMPV3, "%s: dropped %p as ifindex %u went away.", __func__, m, ifindex); m_freem(m); IPSTAT_INC(ips_noroute); goto out; } ipopts = V_igmp_sendra ? m_raopt : NULL; imo.imo_multicast_ttl = 1; imo.imo_multicast_vif = -1; imo.imo_multicast_loop = (V_ip_mrouter != NULL); /* * If the user requested that IGMP traffic be explicitly * redirected to the loopback interface (e.g. they are running a * MANET interface and the routing protocol needs to see the * updates), handle this now. */ if (m->m_flags & M_IGMP_LOOP) imo.imo_multicast_ifp = V_loif; else imo.imo_multicast_ifp = ifp; if (m->m_flags & M_IGMPV2) { m0 = m; } else { m0 = igmp_v3_encap_report(ifp, m); if (m0 == NULL) { CTR2(KTR_IGMPV3, "%s: dropped %p", __func__, m); m_freem(m); IPSTAT_INC(ips_odropped); goto out; } } igmp_scrub_context(m0); m_clrprotoflags(m); m0->m_pkthdr.rcvif = V_loif; #ifdef MAC mac_netinet_igmp_send(ifp, m0); #endif error = ip_output(m0, ipopts, NULL, 0, &imo, NULL); if (error) { CTR3(KTR_IGMPV3, "%s: ip_output(%p) = %d", __func__, m0, error); goto out; } IGMPSTAT_INC(igps_snd_reports); out: /* * We must restore the existing vnet pointer before * continuing as we are run from netisr context. */ CURVNET_RESTORE(); } /* * Encapsulate an IGMPv3 report. * * The internal mbuf flag M_IGMPV3_HDR is used to indicate that the mbuf * chain has already had its IP/IGMPv3 header prepended. In this case * the function will not attempt to prepend; the lengths and checksums * will however be re-computed. * * Returns a pointer to the new mbuf chain head, or NULL if the * allocation failed. */ static struct mbuf * igmp_v3_encap_report(struct ifnet *ifp, struct mbuf *m) { struct rm_priotracker in_ifa_tracker; struct igmp_report *igmp; struct ip *ip; int hdrlen, igmpreclen; KASSERT((m->m_flags & M_PKTHDR), ("%s: mbuf chain %p is !M_PKTHDR", __func__, m)); igmpreclen = m_length(m, NULL); hdrlen = sizeof(struct ip) + sizeof(struct igmp_report); if (m->m_flags & M_IGMPV3_HDR) { igmpreclen -= hdrlen; } else { M_PREPEND(m, hdrlen, M_NOWAIT); if (m == NULL) return (NULL); m->m_flags |= M_IGMPV3_HDR; } CTR2(KTR_IGMPV3, "%s: igmpreclen is %d", __func__, igmpreclen); m->m_data += sizeof(struct ip); m->m_len -= sizeof(struct ip); igmp = mtod(m, struct igmp_report *); igmp->ir_type = IGMP_v3_HOST_MEMBERSHIP_REPORT; igmp->ir_rsv1 = 0; igmp->ir_rsv2 = 0; igmp->ir_numgrps = htons(m->m_pkthdr.PH_vt.vt_nrecs); igmp->ir_cksum = 0; igmp->ir_cksum = in_cksum(m, sizeof(struct igmp_report) + igmpreclen); m->m_pkthdr.PH_vt.vt_nrecs = 0; m->m_data -= sizeof(struct ip); m->m_len += sizeof(struct ip); ip = mtod(m, struct ip *); ip->ip_tos = IPTOS_PREC_INTERNETCONTROL; ip->ip_len = htons(hdrlen + igmpreclen); ip->ip_off = htons(IP_DF); ip->ip_p = IPPROTO_IGMP; ip->ip_sum = 0; ip->ip_src.s_addr = INADDR_ANY; if (m->m_flags & M_IGMP_LOOP) { struct in_ifaddr *ia; IFP_TO_IA(ifp, ia, &in_ifa_tracker); if (ia != NULL) ip->ip_src = ia->ia_addr.sin_addr; } ip->ip_dst.s_addr = htonl(INADDR_ALLRPTS_GROUP); return (m); } #ifdef KTR static char * igmp_rec_type_to_str(const int type) { switch (type) { case IGMP_CHANGE_TO_EXCLUDE_MODE: return "TO_EX"; break; case IGMP_CHANGE_TO_INCLUDE_MODE: return "TO_IN"; break; case IGMP_MODE_IS_EXCLUDE: return "MODE_EX"; break; case IGMP_MODE_IS_INCLUDE: return "MODE_IN"; break; case IGMP_ALLOW_NEW_SOURCES: return "ALLOW_NEW"; break; case IGMP_BLOCK_OLD_SOURCES: return "BLOCK_OLD"; break; default: break; } return "unknown"; } #endif #ifdef VIMAGE static void vnet_igmp_init(const void *unused __unused) { netisr_register_vnet(&igmp_nh); } VNET_SYSINIT(vnet_igmp_init, SI_SUB_PROTO_MC, SI_ORDER_ANY, vnet_igmp_init, NULL); static void vnet_igmp_uninit(const void *unused __unused) { /* This can happen when we shutdown the entire network stack. */ CTR1(KTR_IGMPV3, "%s: tearing down", __func__); netisr_unregister_vnet(&igmp_nh); } VNET_SYSUNINIT(vnet_igmp_uninit, SI_SUB_PROTO_MC, SI_ORDER_ANY, vnet_igmp_uninit, NULL); #endif #ifdef DDB DB_SHOW_COMMAND(igi_list, db_show_igi_list) { struct igmp_ifsoftc *igi, *tigi; LIST_HEAD(_igi_list, igmp_ifsoftc) *igi_head; if (!have_addr) { db_printf("usage: show igi_list \n"); return; } igi_head = (struct _igi_list *)addr; LIST_FOREACH_SAFE(igi, igi_head, igi_link, tigi) { db_printf("igmp_ifsoftc %p:\n", igi); db_printf(" ifp %p\n", igi->igi_ifp); db_printf(" version %u\n", igi->igi_version); db_printf(" v1_timer %u\n", igi->igi_v1_timer); db_printf(" v2_timer %u\n", igi->igi_v2_timer); db_printf(" v3_timer %u\n", igi->igi_v3_timer); db_printf(" flags %#x\n", igi->igi_flags); db_printf(" rv %u\n", igi->igi_rv); db_printf(" qi %u\n", igi->igi_qi); db_printf(" qri %u\n", igi->igi_qri); db_printf(" uri %u\n", igi->igi_uri); /* struct mbufq igi_gq; */ db_printf("\n"); } } #endif static int igmp_modevent(module_t mod, int type, void *unused __unused) { switch (type) { case MOD_LOAD: CTR1(KTR_IGMPV3, "%s: initializing", __func__); IGMP_LOCK_INIT(); m_raopt = igmp_ra_alloc(); netisr_register(&igmp_nh); break; case MOD_UNLOAD: CTR1(KTR_IGMPV3, "%s: tearing down", __func__); netisr_unregister(&igmp_nh); m_free(m_raopt); m_raopt = NULL; IGMP_LOCK_DESTROY(); break; default: return (EOPNOTSUPP); } return (0); } static moduledata_t igmp_mod = { "igmp", igmp_modevent, 0 }; DECLARE_MODULE(igmp, igmp_mod, SI_SUB_PROTO_MC, SI_ORDER_MIDDLE); diff --git a/sys/powerpc/powerpc/uma_machdep.c b/sys/powerpc/powerpc/uma_machdep.c index 23f998b88408..2aa475d976d2 100644 --- a/sys/powerpc/powerpc/uma_machdep.c +++ b/sys/powerpc/powerpc/uma_machdep.c @@ -1,113 +1,108 @@ /*- * Copyright (c) 2003 The FreeBSD Project * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include static int hw_uma_mdpages; SYSCTL_INT(_hw, OID_AUTO, uma_mdpages, CTLFLAG_RD, &hw_uma_mdpages, 0, "UMA MD pages in use"); void * uma_small_alloc(uma_zone_t zone, vm_size_t bytes, int domain, u_int8_t *flags, int wait) { void *va; vm_paddr_t pa; vm_page_t m; *flags = UMA_SLAB_PRIV; m = vm_page_alloc_domain(NULL, 0, domain, malloc2vm_flags(wait) | VM_ALLOC_WIRED | VM_ALLOC_NOOBJ); if (m == NULL) return (NULL); pa = VM_PAGE_TO_PHYS(m); - - /* On book-e sizeof(void *) < sizeof(vm_paddr_t) */ - if ((vm_offset_t)pa != pa) - return (NULL); - #ifdef __powerpc64__ if ((wait & M_NODUMP) == 0) dump_add_page(pa); #endif if (!hw_direct_map) { pmap_kenter(pa, pa); va = (void *)(vm_offset_t)pa; } else { va = (void *)(vm_offset_t)PHYS_TO_DMAP(pa); } if ((wait & M_ZERO) && (m->flags & PG_ZERO) == 0) bzero(va, PAGE_SIZE); atomic_add_int(&hw_uma_mdpages, 1); return (va); } void uma_small_free(void *mem, vm_size_t size, u_int8_t flags) { vm_page_t m; if (hw_direct_map) m = PHYS_TO_VM_PAGE(DMAP_TO_PHYS((vm_offset_t)mem)); else { m = PHYS_TO_VM_PAGE(pmap_kextract((vm_offset_t)mem)); pmap_kremove((vm_offset_t)mem); } KASSERT(m != NULL, ("Freeing UMA block at %p with no associated page", mem)); #ifdef __powerpc64__ dump_drop_page(VM_PAGE_TO_PHYS(m)); #endif vm_page_unwire_noq(m); vm_page_free(m); atomic_subtract_int(&hw_uma_mdpages, 1); } diff --git a/sys/sys/_smr.h b/sys/sys/_smr.h index 834d912783f5..272bf8c07077 100644 --- a/sys/sys/_smr.h +++ b/sys/sys/_smr.h @@ -1,38 +1,50 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2019, 2020 Jeffrey Roberson * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice unmodified, this list of conditions, and the following * disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. * * $FreeBSD$ * */ #ifndef _SYS__SMR_H_ #define _SYS__SMR_H_ typedef uint32_t smr_seq_t; typedef int32_t smr_delta_t; typedef struct smr *smr_t; +#define SMR_ENTERED(smr) \ + (curthread->td_critnest != 0 && zpcpu_get((smr))->c_seq != SMR_SEQ_INVALID) + +#define SMR_ASSERT_ENTERED(smr) \ + KASSERT(SMR_ENTERED(smr), ("Not in smr section")) + +#define SMR_ASSERT_NOT_ENTERED(smr) \ + KASSERT(!SMR_ENTERED(smr), ("In smr section.")); + +#define SMR_ASSERT(ex, fn) \ + KASSERT((ex), (fn ": Assertion " #ex " failed at %s:%d", __FILE__, __LINE__)) + #endif /* __SYS_SMR_H_ */ diff --git a/sys/sys/bufobj.h b/sys/sys/bufobj.h index b6ca9bdaac4c..e52af6cbba21 100644 --- a/sys/sys/bufobj.h +++ b/sys/sys/bufobj.h @@ -1,140 +1,141 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2004 Poul-Henning Kamp * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ /* * Architectural notes: * * bufobj is a new object which is what buffers hang from in the buffer * cache. * * This used to be vnodes, but we need non-vnode code to be able * to use the buffer cache as well, specifically geom classes like gbde, * raid3 and raid5. * * All vnodes will contain a bufobj initially, but down the road we may * want to only allocate bufobjs when they are needed. There could be a * large number of vnodes in the system which wouldn't need a bufobj during * their lifetime. * * The exact relationship to the vmobject is not determined at this point, * it may in fact be that we find them to be two sides of the same object * once things starts to crystalize. */ #ifndef _SYS_BUFOBJ_H_ #define _SYS_BUFOBJ_H_ #if defined(_KERNEL) || defined(_KVM_VNODE) #include #include #include #include struct bufobj; struct buf_ops; extern struct buf_ops buf_ops_bio; TAILQ_HEAD(buflists, buf); /* A Buffer list & trie */ struct bufv { struct buflists bv_hd; /* Sorted blocklist */ struct pctrie bv_root; /* Buf trie */ int bv_cnt; /* Number of buffers */ }; typedef void b_strategy_t(struct bufobj *, struct buf *); typedef int b_write_t(struct buf *); typedef int b_sync_t(struct bufobj *, int waitfor); typedef void b_bdflush_t(struct bufobj *, struct buf *); struct buf_ops { const char *bop_name; b_write_t *bop_write; b_strategy_t *bop_strategy; b_sync_t *bop_sync; b_bdflush_t *bop_bdflush; }; #define BO_STRATEGY(bo, bp) ((bo)->bo_ops->bop_strategy((bo), (bp))) #define BO_SYNC(bo, w) ((bo)->bo_ops->bop_sync((bo), (w))) #define BO_WRITE(bo, bp) ((bo)->bo_ops->bop_write((bp))) #define BO_BDFLUSH(bo, bp) ((bo)->bo_ops->bop_bdflush((bo), (bp))) /* * Locking notes: * 'S' is sync_mtx * 'v' is the vnode lock which embeds the bufobj. * '-' Constant and unchanging after initialization. */ struct bufobj { struct rwlock bo_lock; /* Lock which protects "i" things */ struct buf_ops *bo_ops; /* - Buffer operations */ struct vm_object *bo_object; /* v Place to store VM object */ LIST_ENTRY(bufobj) bo_synclist; /* S dirty vnode list */ void *bo_private; /* private pointer */ struct bufv bo_clean; /* i Clean buffers */ struct bufv bo_dirty; /* i Dirty buffers */ long bo_numoutput; /* i Writes in progress */ u_int bo_flag; /* i Flags */ int bo_domain; /* - Clean queue affinity */ int bo_bsize; /* - Block size for i/o */ }; /* * XXX BO_ONWORKLST could be replaced with a check for NULL list elements * in v_synclist. */ #define BO_ONWORKLST (1 << 0) /* On syncer work-list */ #define BO_WWAIT (1 << 1) /* Wait for output to complete */ #define BO_DEAD (1 << 2) /* Dead; only with INVARIANTS */ +#define BO_NOBUFS (1 << 3) /* No bufs allowed */ #define BO_LOCKPTR(bo) (&(bo)->bo_lock) #define BO_LOCK(bo) rw_wlock(BO_LOCKPTR((bo))) #define BO_UNLOCK(bo) rw_wunlock(BO_LOCKPTR((bo))) #define BO_RLOCK(bo) rw_rlock(BO_LOCKPTR((bo))) #define BO_RUNLOCK(bo) rw_runlock(BO_LOCKPTR((bo))) #define ASSERT_BO_WLOCKED(bo) rw_assert(BO_LOCKPTR((bo)), RA_WLOCKED) #define ASSERT_BO_LOCKED(bo) rw_assert(BO_LOCKPTR((bo)), RA_LOCKED) #define ASSERT_BO_UNLOCKED(bo) rw_assert(BO_LOCKPTR((bo)), RA_UNLOCKED) void bufobj_init(struct bufobj *bo, void *priv); void bufobj_wdrop(struct bufobj *bo); void bufobj_wref(struct bufobj *bo); void bufobj_wrefl(struct bufobj *bo); int bufobj_invalbuf(struct bufobj *bo, int flags, int slpflag, int slptimeo); int bufobj_wwait(struct bufobj *bo, int slpflag, int timeo); int bufsync(struct bufobj *bo, int waitfor); void bufbdflush(struct bufobj *bo, struct buf *bp); #endif /* defined(_KERNEL) || defined(_KVM_VNODE) */ #endif /* _SYS_BUFOBJ_H_ */ diff --git a/sys/sys/counter.h b/sys/sys/counter.h index 9960a2c39d8e..3e1b7828d5d4 100644 --- a/sys/sys/counter.h +++ b/sys/sys/counter.h @@ -1,78 +1,91 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2012 Gleb Smirnoff * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ #ifndef __SYS_COUNTER_H__ #define __SYS_COUNTER_H__ typedef uint64_t *counter_u64_t; #ifdef _KERNEL #include counter_u64_t counter_u64_alloc(int); void counter_u64_free(counter_u64_t); void counter_u64_zero(counter_u64_t); uint64_t counter_u64_fetch(counter_u64_t); #define COUNTER_ARRAY_ALLOC(a, n, wait) do { \ for (int _i = 0; _i < (n); _i++) \ (a)[_i] = counter_u64_alloc(wait); \ } while (0) #define COUNTER_ARRAY_FREE(a, n) do { \ for (int _i = 0; _i < (n); _i++) \ counter_u64_free((a)[_i]); \ } while (0) #define COUNTER_ARRAY_COPY(a, dstp, n) do { \ for (int _i = 0; _i < (n); _i++) \ ((uint64_t *)(dstp))[_i] = counter_u64_fetch((a)[_i]);\ } while (0) #define COUNTER_ARRAY_ZERO(a, n) do { \ for (int _i = 0; _i < (n); _i++) \ counter_u64_zero((a)[_i]); \ } while (0) /* * counter(9) based rate checking. */ struct counter_rate { counter_u64_t cr_rate; /* Events since last second */ volatile int cr_lock; /* Lock to clean the struct */ int cr_ticks; /* Ticks on last clean */ int cr_over; /* Over limit since cr_ticks? */ }; int64_t counter_ratecheck(struct counter_rate *, int64_t); +#define COUNTER_U64_SYSINIT(c) \ + SYSINIT(c##_counter_sysinit, SI_SUB_COUNTER, \ + SI_ORDER_ANY, counter_u64_sysinit, &c); \ + SYSUNINIT(c##_counter_sysuninit, SI_SUB_COUNTER, \ + SI_ORDER_ANY, counter_u64_sysuninit, &c) + +#define COUNTER_U64_DEFINE_EARLY(c) \ + counter_u64_t __read_mostly c = EARLY_COUNTER; \ + COUNTER_U64_SYSINIT(c) + +void counter_u64_sysinit(void *); +void counter_u64_sysuninit(void *); + #endif /* _KERNEL */ #endif /* ! __SYS_COUNTER_H__ */ diff --git a/sys/sys/kernel.h b/sys/sys/kernel.h index 839f26706f47..3b0e038b5c83 100644 --- a/sys/sys/kernel.h +++ b/sys/sys/kernel.h @@ -1,456 +1,457 @@ /*- * SPDX-License-Identifier: BSD-4-Clause * * Copyright (c) 1995 Terrence R. Lambert * All rights reserved. * * Copyright (c) 1990, 1993 * The Regents of the University of California. All rights reserved. * (c) UNIX System Laboratories, Inc. * All or some portions of this file are derived from material licensed * to the University of California by American Telephone and Telegraph * Co. or Unix System Laboratories, Inc. and are reproduced herein with * the permission of UNIX System Laboratories, Inc. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. All advertising materials mentioning features or use of this software * must display the following acknowledgement: * This product includes software developed by the University of * California, Berkeley and its contributors. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)kernel.h 8.3 (Berkeley) 1/21/94 * $FreeBSD$ */ #ifndef _SYS_KERNEL_H_ #define _SYS_KERNEL_H_ #include #ifdef _KERNEL /* for intrhook below */ #include /* for timestamping SYSINITs; other files may assume this is included here */ #include /* Global variables for the kernel. */ /* 1.1 */ extern char kernelname[MAXPATHLEN]; extern int tick; /* usec per tick (1000000 / hz) */ extern int hz; /* system clock's frequency */ extern int psratio; /* ratio: prof / stat */ extern int stathz; /* statistics clock's frequency */ extern int profhz; /* profiling clock's frequency */ extern int profprocs; /* number of process's profiling */ extern volatile int ticks; #endif /* _KERNEL */ /* * Enumerated types for known system startup interfaces. * * Startup occurs in ascending numeric order; the list entries are * sorted prior to attempting startup to guarantee order. Items * of the same level are arbitrated for order based on the 'order' * element. * * These numbers are arbitrary and are chosen ONLY for ordering; the * enumeration values are explicit rather than implicit to provide * for binary compatibility with inserted elements. * * The SI_SUB_LAST value must have the highest lexical value. */ enum sysinit_sub_id { SI_SUB_DUMMY = 0x0000000, /* not executed; for linker*/ SI_SUB_DONE = 0x0000001, /* processed*/ SI_SUB_TUNABLES = 0x0700000, /* establish tunable values */ SI_SUB_COPYRIGHT = 0x0800001, /* first use of console*/ - SI_SUB_VM = 0x1000000, /* virtual memory system init*/ + SI_SUB_VM = 0x1000000, /* virtual memory system init */ + SI_SUB_COUNTER = 0x1100000, /* counter(9) is initialized */ SI_SUB_KMEM = 0x1800000, /* kernel memory*/ SI_SUB_HYPERVISOR = 0x1A40000, /* * Hypervisor detection and * virtualization support * setup. */ SI_SUB_WITNESS = 0x1A80000, /* witness initialization */ SI_SUB_MTX_POOL_DYNAMIC = 0x1AC0000, /* dynamic mutex pool */ SI_SUB_LOCK = 0x1B00000, /* various locks */ SI_SUB_EVENTHANDLER = 0x1C00000, /* eventhandler init */ SI_SUB_VNET_PRELINK = 0x1E00000, /* vnet init before modules */ SI_SUB_KLD = 0x2000000, /* KLD and module setup */ SI_SUB_CPU = 0x2100000, /* CPU resource(s)*/ SI_SUB_RACCT = 0x2110000, /* resource accounting */ SI_SUB_KDTRACE = 0x2140000, /* Kernel dtrace hooks */ SI_SUB_RANDOM = 0x2160000, /* random number generator */ SI_SUB_MAC = 0x2180000, /* TrustedBSD MAC subsystem */ SI_SUB_MAC_POLICY = 0x21C0000, /* TrustedBSD MAC policies */ SI_SUB_MAC_LATE = 0x21D0000, /* TrustedBSD MAC subsystem */ SI_SUB_VNET = 0x21E0000, /* vnet 0 */ SI_SUB_INTRINSIC = 0x2200000, /* proc 0*/ SI_SUB_VM_CONF = 0x2300000, /* config VM, set limits*/ SI_SUB_DDB_SERVICES = 0x2380000, /* capture, scripting, etc. */ SI_SUB_RUN_QUEUE = 0x2400000, /* set up run queue*/ SI_SUB_KTRACE = 0x2480000, /* ktrace */ SI_SUB_OPENSOLARIS = 0x2490000, /* OpenSolaris compatibility */ SI_SUB_AUDIT = 0x24C0000, /* audit */ SI_SUB_CREATE_INIT = 0x2500000, /* create init process*/ SI_SUB_SCHED_IDLE = 0x2600000, /* required idle procs */ SI_SUB_MBUF = 0x2700000, /* mbuf subsystem */ SI_SUB_INTR = 0x2800000, /* interrupt threads */ SI_SUB_TASKQ = 0x2880000, /* task queues */ SI_SUB_EPOCH = 0x2888000, /* epoch subsystem */ #ifdef EARLY_AP_STARTUP SI_SUB_SMP = 0x2900000, /* start the APs*/ #endif SI_SUB_SOFTINTR = 0x2A00000, /* start soft interrupt thread */ SI_SUB_DEVFS = 0x2F00000, /* devfs ready for devices */ SI_SUB_INIT_IF = 0x3000000, /* prep for net interfaces */ SI_SUB_NETGRAPH = 0x3010000, /* Let Netgraph initialize */ SI_SUB_DTRACE = 0x3020000, /* DTrace subsystem */ SI_SUB_DTRACE_PROVIDER = 0x3048000, /* DTrace providers */ SI_SUB_DTRACE_ANON = 0x308C000, /* DTrace anon enabling */ SI_SUB_DRIVERS = 0x3100000, /* Let Drivers initialize */ SI_SUB_CONFIGURE = 0x3800000, /* Configure devices */ SI_SUB_VFS = 0x4000000, /* virtual filesystem*/ SI_SUB_CLOCKS = 0x4800000, /* real time and stat clocks*/ SI_SUB_SYSV_SHM = 0x6400000, /* System V shared memory*/ SI_SUB_SYSV_SEM = 0x6800000, /* System V semaphores*/ SI_SUB_SYSV_MSG = 0x6C00000, /* System V message queues*/ SI_SUB_P1003_1B = 0x6E00000, /* P1003.1B realtime */ SI_SUB_PSEUDO = 0x7000000, /* pseudo devices*/ SI_SUB_EXEC = 0x7400000, /* execve() handlers */ SI_SUB_PROTO_BEGIN = 0x8000000, /* VNET initialization */ SI_SUB_PROTO_PFIL = 0x8100000, /* Initialize pfil before FWs */ SI_SUB_PROTO_IF = 0x8400000, /* interfaces*/ SI_SUB_PROTO_DOMAININIT = 0x8600000, /* domain registration system */ SI_SUB_PROTO_MC = 0x8700000, /* Multicast */ SI_SUB_PROTO_DOMAIN = 0x8800000, /* domains (address families?)*/ SI_SUB_PROTO_FIREWALL = 0x8806000, /* Firewalls */ SI_SUB_PROTO_IFATTACHDOMAIN = 0x8808000,/* domain dependent data init */ SI_SUB_PROTO_END = 0x8ffffff, /* VNET helper functions */ SI_SUB_KPROF = 0x9000000, /* kernel profiling*/ SI_SUB_KICK_SCHEDULER = 0xa000000, /* start the timeout events*/ SI_SUB_INT_CONFIG_HOOKS = 0xa800000, /* Interrupts enabled config */ SI_SUB_ROOT_CONF = 0xb000000, /* Find root devices */ SI_SUB_INTRINSIC_POST = 0xd000000, /* proc 0 cleanup*/ SI_SUB_SYSCALLS = 0xd800000, /* register system calls */ SI_SUB_VNET_DONE = 0xdc00000, /* vnet registration complete */ SI_SUB_KTHREAD_INIT = 0xe000000, /* init process*/ SI_SUB_KTHREAD_PAGE = 0xe400000, /* pageout daemon*/ SI_SUB_KTHREAD_VM = 0xe800000, /* vm daemon*/ SI_SUB_KTHREAD_BUF = 0xea00000, /* buffer daemon*/ SI_SUB_KTHREAD_UPDATE = 0xec00000, /* update daemon*/ SI_SUB_KTHREAD_IDLE = 0xee00000, /* idle procs*/ #ifndef EARLY_AP_STARTUP SI_SUB_SMP = 0xf000000, /* start the APs*/ #endif SI_SUB_RACCTD = 0xf100000, /* start racctd*/ SI_SUB_LAST = 0xfffffff /* final initialization */ }; /* * Some enumerated orders; "ANY" sorts last. */ enum sysinit_elem_order { SI_ORDER_FIRST = 0x0000000, /* first*/ SI_ORDER_SECOND = 0x0000001, /* second*/ SI_ORDER_THIRD = 0x0000002, /* third*/ SI_ORDER_FOURTH = 0x0000003, /* fourth*/ SI_ORDER_FIFTH = 0x0000004, /* fifth*/ SI_ORDER_SIXTH = 0x0000005, /* sixth*/ SI_ORDER_SEVENTH = 0x0000006, /* seventh*/ SI_ORDER_EIGHTH = 0x0000007, /* eighth*/ SI_ORDER_MIDDLE = 0x1000000, /* somewhere in the middle */ SI_ORDER_ANY = 0xfffffff /* last*/ }; /* * A system initialization call instance * * At the moment there is one instance of sysinit. We probably do not * want two which is why this code is if'd out, but we definitely want * to discern SYSINIT's which take non-constant data pointers and * SYSINIT's which take constant data pointers, * * The C_* macros take functions expecting const void * arguments * while the non-C_* macros take functions expecting just void * arguments. * * With -Wcast-qual on, the compiler issues warnings: * - if we pass non-const data or functions taking non-const data * to a C_* macro. * * - if we pass const data to the normal macros * * However, no warning is issued if we pass a function taking const data * through a normal non-const macro. This is ok because the function is * saying it won't modify the data so we don't care whether the data is * modifiable or not. */ typedef void (*sysinit_nfunc_t)(void *); typedef void (*sysinit_cfunc_t)(const void *); struct sysinit { enum sysinit_sub_id subsystem; /* subsystem identifier*/ enum sysinit_elem_order order; /* init order within subsystem*/ sysinit_cfunc_t func; /* function */ const void *udata; /* multiplexer/argument */ }; /* * Default: no special processing * * The C_ version of SYSINIT is for data pointers to const * data ( and functions taking data pointers to const data ). * At the moment it is no different from SYSINIT and thus * still results in warnings. * * The casts are necessary to have the compiler produce the * correct warnings when -Wcast-qual is used. * */ #ifdef TSLOG struct sysinit_tslog { sysinit_cfunc_t func; const void * data; const char * name; }; static inline void sysinit_tslog_shim(const void * data) { const struct sysinit_tslog * x = data; TSRAW(curthread, TS_ENTER, "SYSINIT", x->name); (x->func)(x->data); TSRAW(curthread, TS_EXIT, "SYSINIT", x->name); } #define C_SYSINIT(uniquifier, subsystem, order, func, ident) \ static struct sysinit_tslog uniquifier ## _sys_init_tslog = { \ func, \ (ident), \ #uniquifier \ }; \ static struct sysinit uniquifier ## _sys_init = { \ subsystem, \ order, \ sysinit_tslog_shim, \ &uniquifier ## _sys_init_tslog \ }; \ DATA_WSET(sysinit_set,uniquifier ## _sys_init) #else #define C_SYSINIT(uniquifier, subsystem, order, func, ident) \ static struct sysinit uniquifier ## _sys_init = { \ subsystem, \ order, \ func, \ (ident) \ }; \ DATA_WSET(sysinit_set,uniquifier ## _sys_init) #endif #define SYSINIT(uniquifier, subsystem, order, func, ident) \ C_SYSINIT(uniquifier, subsystem, order, \ (sysinit_cfunc_t)(sysinit_nfunc_t)func, (void *)(ident)) /* * Called on module unload: no special processing */ #define C_SYSUNINIT(uniquifier, subsystem, order, func, ident) \ static struct sysinit uniquifier ## _sys_uninit = { \ subsystem, \ order, \ func, \ (ident) \ }; \ DATA_WSET(sysuninit_set,uniquifier ## _sys_uninit) #define SYSUNINIT(uniquifier, subsystem, order, func, ident) \ C_SYSUNINIT(uniquifier, subsystem, order, \ (sysinit_cfunc_t)(sysinit_nfunc_t)func, (void *)(ident)) void sysinit_add(struct sysinit **set, struct sysinit **set_end); /* * Infrastructure for tunable 'constants'. Value may be specified at compile * time or kernel load time. Rules relating tunables together can be placed * in a SYSINIT function at SI_SUB_TUNABLES with SI_ORDER_ANY. * * WARNING: developers should never use the reserved suffixes specified in * loader.conf(5) for any tunables or conflicts will result. */ /* * int * please avoid using for new tunables! */ extern void tunable_int_init(void *); struct tunable_int { const char *path; int *var; }; #define TUNABLE_INT(path, var) \ static struct tunable_int __CONCAT(__tunable_int_, __LINE__) = { \ (path), \ (var), \ }; \ SYSINIT(__CONCAT(__Tunable_init_, __LINE__), \ SI_SUB_TUNABLES, SI_ORDER_MIDDLE, tunable_int_init, \ &__CONCAT(__tunable_int_, __LINE__)) #define TUNABLE_INT_FETCH(path, var) getenv_int((path), (var)) /* * long */ extern void tunable_long_init(void *); struct tunable_long { const char *path; long *var; }; #define TUNABLE_LONG(path, var) \ static struct tunable_long __CONCAT(__tunable_long_, __LINE__) = { \ (path), \ (var), \ }; \ SYSINIT(__CONCAT(__Tunable_init_, __LINE__), \ SI_SUB_TUNABLES, SI_ORDER_MIDDLE, tunable_long_init,\ &__CONCAT(__tunable_long_, __LINE__)) #define TUNABLE_LONG_FETCH(path, var) getenv_long((path), (var)) /* * unsigned long */ extern void tunable_ulong_init(void *); struct tunable_ulong { const char *path; unsigned long *var; }; #define TUNABLE_ULONG(path, var) \ static struct tunable_ulong __CONCAT(__tunable_ulong_, __LINE__) = { \ (path), \ (var), \ }; \ SYSINIT(__CONCAT(__Tunable_init_, __LINE__), \ SI_SUB_TUNABLES, SI_ORDER_MIDDLE, tunable_ulong_init, \ &__CONCAT(__tunable_ulong_, __LINE__)) #define TUNABLE_ULONG_FETCH(path, var) getenv_ulong((path), (var)) /* * int64_t */ extern void tunable_int64_init(void *); struct tunable_int64 { const char *path; int64_t *var; }; #define TUNABLE_INT64(path, var) \ static struct tunable_int64 __CONCAT(__tunable_int64_, __LINE__) = { \ (path), \ (var), \ }; \ SYSINIT(__CONCAT(__Tunable_init_, __LINE__), \ SI_SUB_TUNABLES, SI_ORDER_MIDDLE, tunable_int64_init, \ &__CONCAT(__tunable_int64_, __LINE__)) #define TUNABLE_INT64_FETCH(path, var) getenv_int64((path), (var)) /* * uint64_t */ extern void tunable_uint64_init(void *); struct tunable_uint64 { const char *path; uint64_t *var; }; #define TUNABLE_UINT64(path, var) \ static struct tunable_uint64 __CONCAT(__tunable_uint64_, __LINE__) = { \ (path), \ (var), \ }; \ SYSINIT(__CONCAT(__Tunable_init_, __LINE__), \ SI_SUB_TUNABLES, SI_ORDER_MIDDLE, tunable_uint64_init, \ &__CONCAT(__tunable_uint64_, __LINE__)) #define TUNABLE_UINT64_FETCH(path, var) getenv_uint64((path), (var)) /* * quad */ extern void tunable_quad_init(void *); struct tunable_quad { const char *path; quad_t *var; }; #define TUNABLE_QUAD(path, var) \ static struct tunable_quad __CONCAT(__tunable_quad_, __LINE__) = { \ (path), \ (var), \ }; \ SYSINIT(__CONCAT(__Tunable_init_, __LINE__), \ SI_SUB_TUNABLES, SI_ORDER_MIDDLE, tunable_quad_init, \ &__CONCAT(__tunable_quad_, __LINE__)) #define TUNABLE_QUAD_FETCH(path, var) getenv_quad((path), (var)) extern void tunable_str_init(void *); struct tunable_str { const char *path; char *var; int size; }; #define TUNABLE_STR(path, var, size) \ static struct tunable_str __CONCAT(__tunable_str_, __LINE__) = { \ (path), \ (var), \ (size), \ }; \ SYSINIT(__CONCAT(__Tunable_init_, __LINE__), \ SI_SUB_TUNABLES, SI_ORDER_MIDDLE, tunable_str_init, \ &__CONCAT(__tunable_str_, __LINE__)) #define TUNABLE_STR_FETCH(path, var, size) \ getenv_string((path), (var), (size)) typedef void (*ich_func_t)(void *_arg); struct intr_config_hook { TAILQ_ENTRY(intr_config_hook) ich_links; ich_func_t ich_func; void *ich_arg; }; int config_intrhook_establish(struct intr_config_hook *hook); void config_intrhook_disestablish(struct intr_config_hook *hook); void config_intrhook_oneshot(ich_func_t _func, void *_arg); #endif /* !_SYS_KERNEL_H_*/ diff --git a/sys/sys/mount.h b/sys/sys/mount.h index 394c2f2f18cd..1941f2da2f09 100644 --- a/sys/sys/mount.h +++ b/sys/sys/mount.h @@ -1,1062 +1,1064 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1989, 1991, 1993 * The Regents of the University of California. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)mount.h 8.21 (Berkeley) 5/20/95 * $FreeBSD$ */ #ifndef _SYS_MOUNT_H_ #define _SYS_MOUNT_H_ #include #include #ifdef _KERNEL #include #include #include #include #include #endif /* * NOTE: When changing statfs structure, mount structure, MNT_* flags or * MNTK_* flags also update DDB show mount command in vfs_subr.c. */ typedef struct fsid { int32_t val[2]; } fsid_t; /* filesystem id type */ /* * File identifier. * These are unique per filesystem on a single machine. */ #define MAXFIDSZ 16 struct fid { u_short fid_len; /* length of data in bytes */ u_short fid_data0; /* force longword alignment */ char fid_data[MAXFIDSZ]; /* data (variable length) */ }; /* * filesystem statistics */ #define MFSNAMELEN 16 /* length of type name including null */ #define MNAMELEN 1024 /* size of on/from name bufs */ #define STATFS_VERSION 0x20140518 /* current version number */ struct statfs { uint32_t f_version; /* structure version number */ uint32_t f_type; /* type of filesystem */ uint64_t f_flags; /* copy of mount exported flags */ uint64_t f_bsize; /* filesystem fragment size */ uint64_t f_iosize; /* optimal transfer block size */ uint64_t f_blocks; /* total data blocks in filesystem */ uint64_t f_bfree; /* free blocks in filesystem */ int64_t f_bavail; /* free blocks avail to non-superuser */ uint64_t f_files; /* total file nodes in filesystem */ int64_t f_ffree; /* free nodes avail to non-superuser */ uint64_t f_syncwrites; /* count of sync writes since mount */ uint64_t f_asyncwrites; /* count of async writes since mount */ uint64_t f_syncreads; /* count of sync reads since mount */ uint64_t f_asyncreads; /* count of async reads since mount */ uint64_t f_spare[10]; /* unused spare */ uint32_t f_namemax; /* maximum filename length */ uid_t f_owner; /* user that mounted the filesystem */ fsid_t f_fsid; /* filesystem id */ char f_charspare[80]; /* spare string space */ char f_fstypename[MFSNAMELEN]; /* filesystem type name */ char f_mntfromname[MNAMELEN]; /* mounted filesystem */ char f_mntonname[MNAMELEN]; /* directory on which mounted */ }; #if defined(_WANT_FREEBSD11_STATFS) || defined(_KERNEL) #define FREEBSD11_STATFS_VERSION 0x20030518 /* current version number */ struct freebsd11_statfs { uint32_t f_version; /* structure version number */ uint32_t f_type; /* type of filesystem */ uint64_t f_flags; /* copy of mount exported flags */ uint64_t f_bsize; /* filesystem fragment size */ uint64_t f_iosize; /* optimal transfer block size */ uint64_t f_blocks; /* total data blocks in filesystem */ uint64_t f_bfree; /* free blocks in filesystem */ int64_t f_bavail; /* free blocks avail to non-superuser */ uint64_t f_files; /* total file nodes in filesystem */ int64_t f_ffree; /* free nodes avail to non-superuser */ uint64_t f_syncwrites; /* count of sync writes since mount */ uint64_t f_asyncwrites; /* count of async writes since mount */ uint64_t f_syncreads; /* count of sync reads since mount */ uint64_t f_asyncreads; /* count of async reads since mount */ uint64_t f_spare[10]; /* unused spare */ uint32_t f_namemax; /* maximum filename length */ uid_t f_owner; /* user that mounted the filesystem */ fsid_t f_fsid; /* filesystem id */ char f_charspare[80]; /* spare string space */ char f_fstypename[16]; /* filesystem type name */ char f_mntfromname[88]; /* mounted filesystem */ char f_mntonname[88]; /* directory on which mounted */ }; #endif /* _WANT_FREEBSD11_STATFS || _KERNEL */ #ifdef _KERNEL #define OMFSNAMELEN 16 /* length of fs type name, including null */ #define OMNAMELEN (88 - 2 * sizeof(long)) /* size of on/from name bufs */ /* XXX getfsstat.2 is out of date with write and read counter changes here. */ /* XXX statfs.2 is out of date with read counter changes here. */ struct ostatfs { long f_spare2; /* placeholder */ long f_bsize; /* fundamental filesystem block size */ long f_iosize; /* optimal transfer block size */ long f_blocks; /* total data blocks in filesystem */ long f_bfree; /* free blocks in fs */ long f_bavail; /* free blocks avail to non-superuser */ long f_files; /* total file nodes in filesystem */ long f_ffree; /* free file nodes in fs */ fsid_t f_fsid; /* filesystem id */ uid_t f_owner; /* user that mounted the filesystem */ int f_type; /* type of filesystem */ int f_flags; /* copy of mount exported flags */ long f_syncwrites; /* count of sync writes since mount */ long f_asyncwrites; /* count of async writes since mount */ char f_fstypename[OMFSNAMELEN]; /* fs type name */ char f_mntonname[OMNAMELEN]; /* directory on which mounted */ long f_syncreads; /* count of sync reads since mount */ long f_asyncreads; /* count of async reads since mount */ short f_spares1; /* unused spare */ char f_mntfromname[OMNAMELEN];/* mounted filesystem */ short f_spares2; /* unused spare */ /* * XXX on machines where longs are aligned to 8-byte boundaries, there * is an unnamed int32_t here. This spare was after the apparent end * of the struct until we bit off the read counters from f_mntonname. */ long f_spare[2]; /* unused spare */ }; TAILQ_HEAD(vnodelst, vnode); /* Mount options list */ TAILQ_HEAD(vfsoptlist, vfsopt); struct vfsopt { TAILQ_ENTRY(vfsopt) link; char *name; void *value; int len; int pos; int seen; }; /* * Structure per mounted filesystem. Each mounted filesystem has an * array of operations and an instance record. The filesystems are * put on a doubly linked list. * * Lock reference: * l - mnt_listmtx * m - mountlist_mtx * i - interlock * v - vnode freelist mutex * * Unmarked fields are considered stable as long as a ref is held. * */ struct mount { struct mtx mnt_mtx; /* mount structure interlock */ int mnt_gen; /* struct mount generation */ #define mnt_startzero mnt_list TAILQ_ENTRY(mount) mnt_list; /* (m) mount list */ struct vfsops *mnt_op; /* operations on fs */ struct vfsconf *mnt_vfc; /* configuration info */ struct vnode *mnt_vnodecovered; /* vnode we mounted on */ struct vnode *mnt_syncer; /* syncer vnode */ int mnt_ref; /* (i) Reference count */ struct vnodelst mnt_nvnodelist; /* (i) list of vnodes */ int mnt_nvnodelistsize; /* (i) # of vnodes */ int mnt_writeopcount; /* (i) write syscalls pending */ int mnt_kern_flag; /* (i) kernel only flags */ uint64_t mnt_flag; /* (i) flags shared with user */ struct vfsoptlist *mnt_opt; /* current mount options */ struct vfsoptlist *mnt_optnew; /* new options passed to fs */ int mnt_maxsymlinklen; /* max size of short symlink */ struct statfs mnt_stat; /* cache of filesystem stats */ struct ucred *mnt_cred; /* credentials of mounter */ void * mnt_data; /* private data */ time_t mnt_time; /* last time written*/ int mnt_iosize_max; /* max size for clusters, etc */ struct netexport *mnt_export; /* export list */ struct label *mnt_label; /* MAC label for the fs */ u_int mnt_hashseed; /* Random seed for vfs_hash */ int mnt_lockref; /* (i) Lock reference count */ int mnt_secondary_writes; /* (i) # of secondary writes */ int mnt_secondary_accwrites;/* (i) secondary wr. starts */ struct thread *mnt_susp_owner; /* (i) thread owning suspension */ #define mnt_endzero mnt_gjprovider char *mnt_gjprovider; /* gjournal provider name */ struct mtx mnt_listmtx; struct vnodelst mnt_lazyvnodelist; /* (l) list of lazy vnodes */ int mnt_lazyvnodelistsize; /* (l) # of lazy vnodes */ struct lock mnt_explock; /* vfs_export walkers lock */ TAILQ_ENTRY(mount) mnt_upper_link; /* (m) we in the all uppers */ TAILQ_HEAD(, mount) mnt_uppers; /* (m) upper mounts over us*/ int __aligned(CACHE_LINE_SIZE) mnt_vfs_ops;/* (i) pending vfs ops */ int *mnt_thread_in_ops_pcpu; int *mnt_ref_pcpu; int *mnt_lockref_pcpu; int *mnt_writeopcount_pcpu; struct vnode *mnt_rootvnode; }; /* * Definitions for MNT_VNODE_FOREACH_ALL. */ struct vnode *__mnt_vnode_next_all(struct vnode **mvp, struct mount *mp); struct vnode *__mnt_vnode_first_all(struct vnode **mvp, struct mount *mp); void __mnt_vnode_markerfree_all(struct vnode **mvp, struct mount *mp); #define MNT_VNODE_FOREACH_ALL(vp, mp, mvp) \ for (vp = __mnt_vnode_first_all(&(mvp), (mp)); \ (vp) != NULL; vp = __mnt_vnode_next_all(&(mvp), (mp))) #define MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp) \ do { \ MNT_ILOCK(mp); \ __mnt_vnode_markerfree_all(&(mvp), (mp)); \ /* MNT_IUNLOCK(mp); -- done in above function */ \ mtx_assert(MNT_MTX(mp), MA_NOTOWNED); \ } while (0) /* * Definitions for MNT_VNODE_FOREACH_LAZY. */ typedef int mnt_lazy_cb_t(struct vnode *, void *); struct vnode *__mnt_vnode_next_lazy(struct vnode **mvp, struct mount *mp, mnt_lazy_cb_t *cb, void *cbarg); struct vnode *__mnt_vnode_first_lazy(struct vnode **mvp, struct mount *mp, mnt_lazy_cb_t *cb, void *cbarg); void __mnt_vnode_markerfree_lazy(struct vnode **mvp, struct mount *mp); #define MNT_VNODE_FOREACH_LAZY(vp, mp, mvp, cb, cbarg) \ for (vp = __mnt_vnode_first_lazy(&(mvp), (mp), (cb), (cbarg)); \ (vp) != NULL; \ vp = __mnt_vnode_next_lazy(&(mvp), (mp), (cb), (cbarg))) #define MNT_VNODE_FOREACH_LAZY_ABORT(mp, mvp) \ __mnt_vnode_markerfree_lazy(&(mvp), (mp)) #define MNT_ILOCK(mp) mtx_lock(&(mp)->mnt_mtx) #define MNT_ITRYLOCK(mp) mtx_trylock(&(mp)->mnt_mtx) #define MNT_IUNLOCK(mp) mtx_unlock(&(mp)->mnt_mtx) #define MNT_MTX(mp) (&(mp)->mnt_mtx) #define MNT_REF(mp) do { \ mtx_assert(MNT_MTX(mp), MA_OWNED); \ mp->mnt_ref++; \ } while (0) #define MNT_REL(mp) do { \ mtx_assert(MNT_MTX(mp), MA_OWNED); \ (mp)->mnt_ref--; \ if ((mp)->mnt_vfs_ops && (mp)->mnt_ref < 0) \ vfs_dump_mount_counters(mp); \ if ((mp)->mnt_ref == 0 && (mp)->mnt_vfs_ops) \ wakeup((mp)); \ } while (0) #endif /* _KERNEL */ /* * User specifiable flags, stored in mnt_flag. */ #define MNT_RDONLY 0x0000000000000001ULL /* read only filesystem */ #define MNT_SYNCHRONOUS 0x0000000000000002ULL /* fs written synchronously */ #define MNT_NOEXEC 0x0000000000000004ULL /* can't exec from filesystem */ #define MNT_NOSUID 0x0000000000000008ULL /* don't honor setuid fs bits */ #define MNT_NFS4ACLS 0x0000000000000010ULL /* enable NFS version 4 ACLs */ #define MNT_UNION 0x0000000000000020ULL /* union with underlying fs */ #define MNT_ASYNC 0x0000000000000040ULL /* fs written asynchronously */ #define MNT_SUIDDIR 0x0000000000100000ULL /* special SUID dir handling */ #define MNT_SOFTDEP 0x0000000000200000ULL /* using soft updates */ #define MNT_NOSYMFOLLOW 0x0000000000400000ULL /* do not follow symlinks */ #define MNT_GJOURNAL 0x0000000002000000ULL /* GEOM journal support enabled */ #define MNT_MULTILABEL 0x0000000004000000ULL /* MAC support for objects */ #define MNT_ACLS 0x0000000008000000ULL /* ACL support enabled */ #define MNT_NOATIME 0x0000000010000000ULL /* dont update file access time */ #define MNT_NOCLUSTERR 0x0000000040000000ULL /* disable cluster read */ #define MNT_NOCLUSTERW 0x0000000080000000ULL /* disable cluster write */ #define MNT_SUJ 0x0000000100000000ULL /* using journaled soft updates */ #define MNT_AUTOMOUNTED 0x0000000200000000ULL /* mounted by automountd(8) */ #define MNT_UNTRUSTED 0x0000000800000000ULL /* filesys metadata untrusted */ /* * NFS export related mount flags. */ #define MNT_EXRDONLY 0x0000000000000080ULL /* exported read only */ #define MNT_EXPORTED 0x0000000000000100ULL /* filesystem is exported */ #define MNT_DEFEXPORTED 0x0000000000000200ULL /* exported to the world */ #define MNT_EXPORTANON 0x0000000000000400ULL /* anon uid mapping for all */ #define MNT_EXKERB 0x0000000000000800ULL /* exported with Kerberos */ #define MNT_EXPUBLIC 0x0000000020000000ULL /* public export (WebNFS) */ /* * Flags set by internal operations, * but visible to the user. * XXX some of these are not quite right.. (I've never seen the root flag set) */ #define MNT_LOCAL 0x0000000000001000ULL /* filesystem is stored locally */ #define MNT_QUOTA 0x0000000000002000ULL /* quotas are enabled on fs */ #define MNT_ROOTFS 0x0000000000004000ULL /* identifies the root fs */ #define MNT_USER 0x0000000000008000ULL /* mounted by a user */ #define MNT_IGNORE 0x0000000000800000ULL /* do not show entry in df */ #define MNT_VERIFIED 0x0000000400000000ULL /* filesystem is verified */ /* * Mask of flags that are visible to statfs(). * XXX I think that this could now become (~(MNT_CMDFLAGS)) * but the 'mount' program may need changing to handle this. */ #define MNT_VISFLAGMASK (MNT_RDONLY | MNT_SYNCHRONOUS | MNT_NOEXEC | \ MNT_NOSUID | MNT_UNION | MNT_SUJ | \ MNT_ASYNC | MNT_EXRDONLY | MNT_EXPORTED | \ MNT_DEFEXPORTED | MNT_EXPORTANON| MNT_EXKERB | \ MNT_LOCAL | MNT_USER | MNT_QUOTA | \ MNT_ROOTFS | MNT_NOATIME | MNT_NOCLUSTERR| \ MNT_NOCLUSTERW | MNT_SUIDDIR | MNT_SOFTDEP | \ MNT_IGNORE | MNT_EXPUBLIC | MNT_NOSYMFOLLOW | \ MNT_GJOURNAL | MNT_MULTILABEL | MNT_ACLS | \ MNT_NFS4ACLS | MNT_AUTOMOUNTED | MNT_VERIFIED | \ MNT_UNTRUSTED) /* Mask of flags that can be updated. */ #define MNT_UPDATEMASK (MNT_NOSUID | MNT_NOEXEC | \ MNT_SYNCHRONOUS | MNT_UNION | MNT_ASYNC | \ MNT_NOATIME | \ MNT_NOSYMFOLLOW | MNT_IGNORE | \ MNT_NOCLUSTERR | MNT_NOCLUSTERW | MNT_SUIDDIR | \ MNT_ACLS | MNT_USER | MNT_NFS4ACLS | \ MNT_AUTOMOUNTED | MNT_UNTRUSTED) /* * External filesystem command modifier flags. * Unmount can use the MNT_FORCE flag. * XXX: These are not STATES and really should be somewhere else. * XXX: MNT_BYFSID and MNT_NONBUSY collide with MNT_ACLS and MNT_MULTILABEL, * but because MNT_ACLS and MNT_MULTILABEL are only used for mount(2), * and MNT_BYFSID and MNT_NONBUSY are only used for unmount(2), * it's harmless. */ #define MNT_UPDATE 0x0000000000010000ULL /* not real mount, just update */ #define MNT_DELEXPORT 0x0000000000020000ULL /* delete export host lists */ #define MNT_RELOAD 0x0000000000040000ULL /* reload filesystem data */ #define MNT_FORCE 0x0000000000080000ULL /* force unmount or readonly */ #define MNT_SNAPSHOT 0x0000000001000000ULL /* snapshot the filesystem */ #define MNT_NONBUSY 0x0000000004000000ULL /* check vnode use counts. */ #define MNT_BYFSID 0x0000000008000000ULL /* specify filesystem by ID. */ #define MNT_NOCOVER 0x0000001000000000ULL /* Do not cover a mount point */ #define MNT_EMPTYDIR 0x0000002000000000ULL /* Only mount on empty dir */ #define MNT_CMDFLAGS (MNT_UPDATE | MNT_DELEXPORT | MNT_RELOAD | \ MNT_FORCE | MNT_SNAPSHOT | MNT_NONBUSY | \ MNT_BYFSID | MNT_NOCOVER | MNT_EMPTYDIR) /* * Internal filesystem control flags stored in mnt_kern_flag. * * MNTK_UNMOUNT locks the mount entry so that name lookup cannot * proceed past the mount point. This keeps the subtree stable during * mounts and unmounts. When non-forced unmount flushes all vnodes * from the mp queue, the MNTK_UNMOUNT flag prevents insmntque() from * queueing new vnodes. * * MNTK_UNMOUNTF permits filesystems to detect a forced unmount while * dounmount() is still waiting to lock the mountpoint. This allows * the filesystem to cancel operations that might otherwise deadlock * with the unmount attempt (used by NFS). */ #define MNTK_UNMOUNTF 0x00000001 /* forced unmount in progress */ #define MNTK_ASYNC 0x00000002 /* filtered async flag */ #define MNTK_SOFTDEP 0x00000004 /* async disabled by softdep */ #define MNTK_NOMSYNC 0x00000008 /* don't do msync */ #define MNTK_DRAINING 0x00000010 /* lock draining is happening */ #define MNTK_REFEXPIRE 0x00000020 /* refcount expiring is happening */ #define MNTK_EXTENDED_SHARED 0x00000040 /* Allow shared locking for more ops */ #define MNTK_SHARED_WRITES 0x00000080 /* Allow shared locking for writes */ #define MNTK_NO_IOPF 0x00000100 /* Disallow page faults during reads and writes. Filesystem shall properly handle i/o state on EFAULT. */ #define MNTK_VGONE_UPPER 0x00000200 #define MNTK_VGONE_WAITER 0x00000400 #define MNTK_LOOKUP_EXCL_DOTDOT 0x00000800 #define MNTK_MARKER 0x00001000 #define MNTK_UNMAPPED_BUFS 0x00002000 #define MNTK_USES_BCACHE 0x00004000 /* FS uses the buffer cache. */ #define MNTK_TEXT_REFS 0x00008000 /* Keep use ref for text */ #define MNTK_VMSETSIZE_BUG 0x00010000 #define MNTK_UNIONFS 0x00020000 /* A hack for F_ISUNIONSTACK */ #define MNTK_NOASYNC 0x00800000 /* disable async */ #define MNTK_UNMOUNT 0x01000000 /* unmount in progress */ #define MNTK_MWAIT 0x02000000 /* waiting for unmount to finish */ #define MNTK_SUSPEND 0x08000000 /* request write suspension */ #define MNTK_SUSPEND2 0x04000000 /* block secondary writes */ #define MNTK_SUSPENDED 0x10000000 /* write operations are suspended */ #define MNTK_NULL_NOCACHE 0x20000000 /* auto disable cache for nullfs mounts over this fs */ #define MNTK_LOOKUP_SHARED 0x40000000 /* FS supports shared lock lookups */ #define MNTK_NOKNOTE 0x80000000 /* Don't send KNOTEs from VOP hooks */ #ifdef _KERNEL static inline int MNT_SHARED_WRITES(struct mount *mp) { return (mp != NULL && (mp->mnt_kern_flag & MNTK_SHARED_WRITES) != 0); } static inline int MNT_EXTENDED_SHARED(struct mount *mp) { return (mp != NULL && (mp->mnt_kern_flag & MNTK_EXTENDED_SHARED) != 0); } #endif /* * Sysctl CTL_VFS definitions. * * Second level identifier specifies which filesystem. Second level * identifier VFS_VFSCONF returns information about all filesystems. * Second level identifier VFS_GENERIC is non-terminal. */ #define VFS_VFSCONF 0 /* get configured filesystems */ #define VFS_GENERIC 0 /* generic filesystem information */ /* * Third level identifiers for VFS_GENERIC are given below; third * level identifiers for specific filesystems are given in their * mount specific header files. */ #define VFS_MAXTYPENUM 1 /* int: highest defined filesystem type */ #define VFS_CONF 2 /* struct: vfsconf for filesystem given as next argument */ /* * Flags for various system call interfaces. * * waitfor flags to vfs_sync() and getfsstat() */ #define MNT_WAIT 1 /* synchronously wait for I/O to complete */ #define MNT_NOWAIT 2 /* start all I/O, but do not wait for it */ #define MNT_LAZY 3 /* push data not written by filesystem syncer */ #define MNT_SUSPEND 4 /* Suspend file system after sync */ /* * Generic file handle */ struct fhandle { fsid_t fh_fsid; /* Filesystem id of mount point */ struct fid fh_fid; /* Filesys specific id */ }; typedef struct fhandle fhandle_t; /* * Old export arguments without security flavor list */ struct oexport_args { int ex_flags; /* export related flags */ uid_t ex_root; /* mapping for root uid */ struct xucred ex_anon; /* mapping for anonymous user */ struct sockaddr *ex_addr; /* net address to which exported */ u_char ex_addrlen; /* and the net address length */ struct sockaddr *ex_mask; /* mask of valid bits in saddr */ u_char ex_masklen; /* and the smask length */ char *ex_indexfile; /* index file for WebNFS URLs */ }; /* * Export arguments for local filesystem mount calls. */ #define MAXSECFLAVORS 5 struct export_args { int ex_flags; /* export related flags */ uid_t ex_root; /* mapping for root uid */ struct xucred ex_anon; /* mapping for anonymous user */ struct sockaddr *ex_addr; /* net address to which exported */ u_char ex_addrlen; /* and the net address length */ struct sockaddr *ex_mask; /* mask of valid bits in saddr */ u_char ex_masklen; /* and the smask length */ char *ex_indexfile; /* index file for WebNFS URLs */ int ex_numsecflavors; /* security flavor count */ int ex_secflavors[MAXSECFLAVORS]; /* list of security flavors */ }; /* * Structure holding information for a publicly exported filesystem * (WebNFS). Currently the specs allow just for one such filesystem. */ struct nfs_public { int np_valid; /* Do we hold valid information */ fhandle_t np_handle; /* Filehandle for pub fs (internal) */ struct mount *np_mount; /* Mountpoint of exported fs */ char *np_index; /* Index file */ }; /* * Filesystem configuration information. One of these exists for each * type of filesystem supported by the kernel. These are searched at * mount time to identify the requested filesystem. * * XXX: Never change the first two arguments! */ struct vfsconf { u_int vfc_version; /* ABI version number */ char vfc_name[MFSNAMELEN]; /* filesystem type name */ struct vfsops *vfc_vfsops; /* filesystem operations vector */ struct vfsops *vfc_vfsops_sd; /* ... signal-deferred */ int vfc_typenum; /* historic filesystem type number */ int vfc_refcount; /* number mounted of this type */ int vfc_flags; /* permanent flags */ int vfc_prison_flag; /* prison allow.mount.* flag */ struct vfsoptdecl *vfc_opts; /* mount options */ TAILQ_ENTRY(vfsconf) vfc_list; /* list of vfscons */ }; /* Userland version of the struct vfsconf. */ struct xvfsconf { struct vfsops *vfc_vfsops; /* filesystem operations vector */ char vfc_name[MFSNAMELEN]; /* filesystem type name */ int vfc_typenum; /* historic filesystem type number */ int vfc_refcount; /* number mounted of this type */ int vfc_flags; /* permanent flags */ struct vfsconf *vfc_next; /* next in list */ }; #ifndef BURN_BRIDGES struct ovfsconf { void *vfc_vfsops; char vfc_name[32]; int vfc_index; int vfc_refcount; int vfc_flags; }; #endif /* * NB: these flags refer to IMPLEMENTATION properties, not properties of * any actual mounts; i.e., it does not make sense to change the flags. */ #define VFCF_STATIC 0x00010000 /* statically compiled into kernel */ #define VFCF_NETWORK 0x00020000 /* may get data over the network */ #define VFCF_READONLY 0x00040000 /* writes are not implemented */ #define VFCF_SYNTHETIC 0x00080000 /* data does not represent real files */ #define VFCF_LOOPBACK 0x00100000 /* aliases some other mounted FS */ #define VFCF_UNICODE 0x00200000 /* stores file names as Unicode */ #define VFCF_JAIL 0x00400000 /* can be mounted from within a jail */ #define VFCF_DELEGADMIN 0x00800000 /* supports delegated administration */ #define VFCF_SBDRY 0x01000000 /* Stop at Boundary: defer stop requests to kernel->user (AST) transition */ typedef uint32_t fsctlop_t; struct vfsidctl { int vc_vers; /* should be VFSIDCTL_VERS1 (below) */ fsid_t vc_fsid; /* fsid to operate on */ char vc_fstypename[MFSNAMELEN]; /* type of fs 'nfs' or '*' */ fsctlop_t vc_op; /* operation VFS_CTL_* (below) */ void *vc_ptr; /* pointer to data structure */ size_t vc_len; /* sizeof said structure */ u_int32_t vc_spare[12]; /* spare (must be zero) */ }; /* vfsidctl API version. */ #define VFS_CTL_VERS1 0x01 /* * New style VFS sysctls, do not reuse/conflict with the namespace for * private sysctls. * All "global" sysctl ops have the 33rd bit set: * 0x...1.... * Private sysctl ops should have the 33rd bit unset. */ #define VFS_CTL_QUERY 0x00010001 /* anything wrong? (vfsquery) */ #define VFS_CTL_TIMEO 0x00010002 /* set timeout for vfs notification */ #define VFS_CTL_NOLOCKS 0x00010003 /* disable file locking */ struct vfsquery { u_int32_t vq_flags; u_int32_t vq_spare[31]; }; /* vfsquery flags */ #define VQ_NOTRESP 0x0001 /* server down */ #define VQ_NEEDAUTH 0x0002 /* server bad auth */ #define VQ_LOWDISK 0x0004 /* we're low on space */ #define VQ_MOUNT 0x0008 /* new filesystem arrived */ #define VQ_UNMOUNT 0x0010 /* filesystem has left */ #define VQ_DEAD 0x0020 /* filesystem is dead, needs force unmount */ #define VQ_ASSIST 0x0040 /* filesystem needs assistance from external program */ #define VQ_NOTRESPLOCK 0x0080 /* server lockd down */ #define VQ_FLAG0100 0x0100 /* placeholder */ #define VQ_FLAG0200 0x0200 /* placeholder */ #define VQ_FLAG0400 0x0400 /* placeholder */ #define VQ_FLAG0800 0x0800 /* placeholder */ #define VQ_FLAG1000 0x1000 /* placeholder */ #define VQ_FLAG2000 0x2000 /* placeholder */ #define VQ_FLAG4000 0x4000 /* placeholder */ #define VQ_FLAG8000 0x8000 /* placeholder */ #ifdef _KERNEL /* Point a sysctl request at a vfsidctl's data. */ #define VCTLTOREQ(vc, req) \ do { \ (req)->newptr = (vc)->vc_ptr; \ (req)->newlen = (vc)->vc_len; \ (req)->newidx = 0; \ } while (0) #endif struct iovec; struct uio; #ifdef _KERNEL /* * vfs_busy specific flags and mask. */ #define MBF_NOWAIT 0x01 #define MBF_MNTLSTLOCK 0x02 #define MBF_MASK (MBF_NOWAIT | MBF_MNTLSTLOCK) #ifdef MALLOC_DECLARE MALLOC_DECLARE(M_MOUNT); MALLOC_DECLARE(M_STATFS); #endif extern int maxvfsconf; /* highest defined filesystem type */ TAILQ_HEAD(vfsconfhead, vfsconf); extern struct vfsconfhead vfsconf; /* * Operations supported on mounted filesystem. */ struct mount_args; struct nameidata; struct sysctl_req; struct mntarg; /* * N.B., vfs_cmount is the ancient vfsop invoked by the old mount(2) syscall. * The new way is vfs_mount. * * vfs_cmount implementations typically translate arguments from their * respective old per-FS structures into the key-value list supported by * nmount(2), then use kernel_mount(9) to mimic nmount(2) from kernelspace. * * Filesystems with mounters that use nmount(2) do not need to and should not * implement vfs_cmount. Hopefully a future cleanup can remove vfs_cmount and * mount(2) entirely. */ typedef int vfs_cmount_t(struct mntarg *ma, void *data, uint64_t flags); typedef int vfs_unmount_t(struct mount *mp, int mntflags); typedef int vfs_root_t(struct mount *mp, int flags, struct vnode **vpp); typedef int vfs_quotactl_t(struct mount *mp, int cmds, uid_t uid, void *arg); typedef int vfs_statfs_t(struct mount *mp, struct statfs *sbp); typedef int vfs_sync_t(struct mount *mp, int waitfor); typedef int vfs_vget_t(struct mount *mp, ino_t ino, int flags, struct vnode **vpp); typedef int vfs_fhtovp_t(struct mount *mp, struct fid *fhp, int flags, struct vnode **vpp); typedef int vfs_checkexp_t(struct mount *mp, struct sockaddr *nam, int *extflagsp, struct ucred **credanonp, int *numsecflavors, int **secflavors); typedef int vfs_init_t(struct vfsconf *); typedef int vfs_uninit_t(struct vfsconf *); typedef int vfs_extattrctl_t(struct mount *mp, int cmd, struct vnode *filename_vp, int attrnamespace, const char *attrname); typedef int vfs_mount_t(struct mount *mp); typedef int vfs_sysctl_t(struct mount *mp, fsctlop_t op, struct sysctl_req *req); typedef void vfs_susp_clean_t(struct mount *mp); typedef void vfs_notify_lowervp_t(struct mount *mp, struct vnode *lowervp); typedef void vfs_purge_t(struct mount *mp); struct vfsops { vfs_mount_t *vfs_mount; vfs_cmount_t *vfs_cmount; vfs_unmount_t *vfs_unmount; vfs_root_t *vfs_root; vfs_root_t *vfs_cachedroot; vfs_quotactl_t *vfs_quotactl; vfs_statfs_t *vfs_statfs; vfs_sync_t *vfs_sync; vfs_vget_t *vfs_vget; vfs_fhtovp_t *vfs_fhtovp; vfs_checkexp_t *vfs_checkexp; vfs_init_t *vfs_init; vfs_uninit_t *vfs_uninit; vfs_extattrctl_t *vfs_extattrctl; vfs_sysctl_t *vfs_sysctl; vfs_susp_clean_t *vfs_susp_clean; vfs_notify_lowervp_t *vfs_reclaim_lowervp; vfs_notify_lowervp_t *vfs_unlink_lowervp; vfs_purge_t *vfs_purge; vfs_mount_t *vfs_spare[6]; /* spares for ABI compat */ }; vfs_statfs_t __vfs_statfs; #define VFS_MOUNT(MP) ({ \ int _rc; \ \ TSRAW(curthread, TS_ENTER, "VFS_MOUNT", (MP)->mnt_vfc->vfc_name);\ _rc = (*(MP)->mnt_op->vfs_mount)(MP); \ TSRAW(curthread, TS_EXIT, "VFS_MOUNT", (MP)->mnt_vfc->vfc_name);\ _rc; }) #define VFS_UNMOUNT(MP, FORCE) ({ \ int _rc; \ \ _rc = (*(MP)->mnt_op->vfs_unmount)(MP, FORCE); \ _rc; }) #define VFS_ROOT(MP, FLAGS, VPP) ({ \ int _rc; \ \ _rc = (*(MP)->mnt_op->vfs_root)(MP, FLAGS, VPP); \ _rc; }) #define VFS_CACHEDROOT(MP, FLAGS, VPP) ({ \ int _rc; \ \ _rc = (*(MP)->mnt_op->vfs_cachedroot)(MP, FLAGS, VPP); \ _rc; }) #define VFS_QUOTACTL(MP, C, U, A) ({ \ int _rc; \ \ _rc = (*(MP)->mnt_op->vfs_quotactl)(MP, C, U, A); \ _rc; }) #define VFS_STATFS(MP, SBP) ({ \ int _rc; \ \ _rc = __vfs_statfs((MP), (SBP)); \ _rc; }) #define VFS_SYNC(MP, WAIT) ({ \ int _rc; \ \ _rc = (*(MP)->mnt_op->vfs_sync)(MP, WAIT); \ _rc; }) #define VFS_VGET(MP, INO, FLAGS, VPP) ({ \ int _rc; \ \ _rc = (*(MP)->mnt_op->vfs_vget)(MP, INO, FLAGS, VPP); \ _rc; }) #define VFS_FHTOVP(MP, FIDP, FLAGS, VPP) ({ \ int _rc; \ \ _rc = (*(MP)->mnt_op->vfs_fhtovp)(MP, FIDP, FLAGS, VPP); \ _rc; }) #define VFS_CHECKEXP(MP, NAM, EXFLG, CRED, NUMSEC, SEC) ({ \ int _rc; \ \ _rc = (*(MP)->mnt_op->vfs_checkexp)(MP, NAM, EXFLG, CRED, NUMSEC,\ SEC); \ _rc; }) #define VFS_EXTATTRCTL(MP, C, FN, NS, N) ({ \ int _rc; \ \ _rc = (*(MP)->mnt_op->vfs_extattrctl)(MP, C, FN, NS, N); \ _rc; }) #define VFS_SYSCTL(MP, OP, REQ) ({ \ int _rc; \ \ _rc = (*(MP)->mnt_op->vfs_sysctl)(MP, OP, REQ); \ _rc; }) #define VFS_SUSP_CLEAN(MP) do { \ if (*(MP)->mnt_op->vfs_susp_clean != NULL) { \ (*(MP)->mnt_op->vfs_susp_clean)(MP); \ } \ } while (0) #define VFS_RECLAIM_LOWERVP(MP, VP) do { \ if (*(MP)->mnt_op->vfs_reclaim_lowervp != NULL) { \ (*(MP)->mnt_op->vfs_reclaim_lowervp)((MP), (VP)); \ } \ } while (0) #define VFS_UNLINK_LOWERVP(MP, VP) do { \ if (*(MP)->mnt_op->vfs_unlink_lowervp != NULL) { \ (*(MP)->mnt_op->vfs_unlink_lowervp)((MP), (VP)); \ } \ } while (0) #define VFS_PURGE(MP) do { \ if (*(MP)->mnt_op->vfs_purge != NULL) { \ (*(MP)->mnt_op->vfs_purge)(MP); \ } \ } while (0) #define VFS_KNOTE_LOCKED(vp, hint) do \ { \ if (((vp)->v_vflag & VV_NOKNOTE) == 0) \ VN_KNOTE((vp), (hint), KNF_LISTLOCKED); \ } while (0) #define VFS_KNOTE_UNLOCKED(vp, hint) do \ { \ if (((vp)->v_vflag & VV_NOKNOTE) == 0) \ VN_KNOTE((vp), (hint), 0); \ } while (0) #define VFS_NOTIFY_UPPER_RECLAIM 1 #define VFS_NOTIFY_UPPER_UNLINK 2 #include /* * Version numbers. */ #define VFS_VERSION_00 0x19660120 #define VFS_VERSION_01 0x20121030 #define VFS_VERSION_02 0x20180504 #define VFS_VERSION VFS_VERSION_02 #define VFS_SET(vfsops, fsname, flags) \ static struct vfsconf fsname ## _vfsconf = { \ .vfc_version = VFS_VERSION, \ .vfc_name = #fsname, \ .vfc_vfsops = &vfsops, \ .vfc_typenum = -1, \ .vfc_flags = flags, \ }; \ static moduledata_t fsname ## _mod = { \ #fsname, \ vfs_modevent, \ & fsname ## _vfsconf \ }; \ DECLARE_MODULE(fsname, fsname ## _mod, SI_SUB_VFS, SI_ORDER_MIDDLE) /* * exported vnode operations */ int dounmount(struct mount *, int, struct thread *); int kernel_mount(struct mntarg *ma, uint64_t flags); int kernel_vmount(int flags, ...); struct mntarg *mount_arg(struct mntarg *ma, const char *name, const void *val, int len); struct mntarg *mount_argb(struct mntarg *ma, int flag, const char *name); struct mntarg *mount_argf(struct mntarg *ma, const char *name, const char *fmt, ...); struct mntarg *mount_argsu(struct mntarg *ma, const char *name, const void *val, int len); void statfs_scale_blocks(struct statfs *sf, long max_size); struct vfsconf *vfs_byname(const char *); struct vfsconf *vfs_byname_kld(const char *, struct thread *td, int *); void vfs_mount_destroy(struct mount *); void vfs_event_signal(fsid_t *, u_int32_t, intptr_t); void vfs_freeopts(struct vfsoptlist *opts); void vfs_deleteopt(struct vfsoptlist *opts, const char *name); int vfs_buildopts(struct uio *auio, struct vfsoptlist **options); int vfs_flagopt(struct vfsoptlist *opts, const char *name, uint64_t *w, uint64_t val); int vfs_getopt(struct vfsoptlist *, const char *, void **, int *); int vfs_getopt_pos(struct vfsoptlist *opts, const char *name); int vfs_getopt_size(struct vfsoptlist *opts, const char *name, off_t *value); char *vfs_getopts(struct vfsoptlist *, const char *, int *error); int vfs_copyopt(struct vfsoptlist *, const char *, void *, int); int vfs_filteropt(struct vfsoptlist *, const char **legal); void vfs_opterror(struct vfsoptlist *opts, const char *fmt, ...); int vfs_scanopt(struct vfsoptlist *opts, const char *name, const char *fmt, ...); int vfs_setopt(struct vfsoptlist *opts, const char *name, void *value, int len); int vfs_setopt_part(struct vfsoptlist *opts, const char *name, void *value, int len); int vfs_setopts(struct vfsoptlist *opts, const char *name, const char *value); int vfs_setpublicfs /* set publicly exported fs */ (struct mount *, struct netexport *, struct export_args *); void vfs_periodic(struct mount *, int); int vfs_busy(struct mount *, int); int vfs_export /* process mount export info */ (struct mount *, struct export_args *); void vfs_allocate_syncvnode(struct mount *); void vfs_deallocate_syncvnode(struct mount *); int vfs_donmount(struct thread *td, uint64_t fsflags, struct uio *fsoptions); void vfs_getnewfsid(struct mount *); struct cdev *vfs_getrootfsid(struct mount *); struct mount *vfs_getvfs(fsid_t *); /* return vfs given fsid */ struct mount *vfs_busyfs(fsid_t *); int vfs_modevent(module_t, int, void *); void vfs_mount_error(struct mount *, const char *, ...); void vfs_mountroot(void); /* mount our root filesystem */ void vfs_mountedfrom(struct mount *, const char *from); void vfs_notify_upper(struct vnode *, int); void vfs_oexport_conv(const struct oexport_args *oexp, struct export_args *exp); void vfs_ref(struct mount *); void vfs_rel(struct mount *); struct mount *vfs_mount_alloc(struct vnode *, struct vfsconf *, const char *, struct ucred *); int vfs_suser(struct mount *, struct thread *); void vfs_unbusy(struct mount *); void vfs_unmountall(void); extern TAILQ_HEAD(mntlist, mount) mountlist; /* mounted filesystem list */ extern struct mtx mountlist_mtx; extern struct nfs_public nfs_pub; extern struct sx vfsconf_sx; #define vfsconf_lock() sx_xlock(&vfsconf_sx) #define vfsconf_unlock() sx_xunlock(&vfsconf_sx) #define vfsconf_slock() sx_slock(&vfsconf_sx) #define vfsconf_sunlock() sx_sunlock(&vfsconf_sx) +struct vnode *mntfs_allocvp(struct mount *, struct vnode *); +void mntfs_freevp(struct vnode *); /* * Declarations for these vfs default operations are located in * kern/vfs_default.c. They will be automatically used to replace * null entries in VFS ops tables when registering a new filesystem * type in the global table. */ vfs_root_t vfs_stdroot; vfs_quotactl_t vfs_stdquotactl; vfs_statfs_t vfs_stdstatfs; vfs_sync_t vfs_stdsync; vfs_sync_t vfs_stdnosync; vfs_vget_t vfs_stdvget; vfs_fhtovp_t vfs_stdfhtovp; vfs_checkexp_t vfs_stdcheckexp; vfs_init_t vfs_stdinit; vfs_uninit_t vfs_stduninit; vfs_extattrctl_t vfs_stdextattrctl; vfs_sysctl_t vfs_stdsysctl; void syncer_suspend(void); void syncer_resume(void); struct vnode *vfs_cache_root_clear(struct mount *); void vfs_cache_root_set(struct mount *, struct vnode *); void vfs_op_barrier_wait(struct mount *); void vfs_op_enter(struct mount *); void vfs_op_exit_locked(struct mount *); void vfs_op_exit(struct mount *); #ifdef DIAGNOSTIC void vfs_assert_mount_counters(struct mount *); void vfs_dump_mount_counters(struct mount *); #else #define vfs_assert_mount_counters(mp) do { } while (0) #define vfs_dump_mount_counters(mp) do { } while (0) #endif enum mount_counter { MNT_COUNT_REF, MNT_COUNT_LOCKREF, MNT_COUNT_WRITEOPCOUNT }; int vfs_mount_fetch_counter(struct mount *, enum mount_counter); /* * Code transitioning mnt_vfs_ops to > 0 issues IPIs until it observes * all CPUs not executing code enclosed by mnt_thread_in_ops_pcpu. * * This provides an invariant that by the time the last CPU is observed not * executing, everyone else entering will see the counter > 0 and exit. * * Note there is no barrier between vfs_ops and the rest of the code in the * section. It is not necessary as the writer has to wait for everyone to drain * before making any changes or only make changes safe while the section is * executed. */ #define vfs_op_thread_entered(mp) ({ \ MPASS(curthread->td_critnest > 0); \ *zpcpu_get(mp->mnt_thread_in_ops_pcpu) == 1; \ }) #define vfs_op_thread_enter(mp) ({ \ bool _retval = true; \ critical_enter(); \ MPASS(!vfs_op_thread_entered(mp)); \ zpcpu_set_protected(mp->mnt_thread_in_ops_pcpu, 1); \ __compiler_membar(); \ if (__predict_false(mp->mnt_vfs_ops > 0)) { \ vfs_op_thread_exit(mp); \ _retval = false; \ } \ _retval; \ }) #define vfs_op_thread_exit(mp) do { \ MPASS(vfs_op_thread_entered(mp)); \ __compiler_membar(); \ zpcpu_set_protected(mp->mnt_thread_in_ops_pcpu, 0); \ critical_exit(); \ } while (0) #define vfs_mp_count_add_pcpu(mp, count, val) do { \ MPASS(vfs_op_thread_entered(mp)); \ zpcpu_add_protected(mp->mnt_##count##_pcpu, val); \ } while (0) #define vfs_mp_count_sub_pcpu(mp, count, val) do { \ MPASS(vfs_op_thread_entered(mp)); \ zpcpu_sub_protected(mp->mnt_##count##_pcpu, val); \ } while (0) #else /* !_KERNEL */ #include struct stat; __BEGIN_DECLS int fhlink(struct fhandle *, const char *); int fhlinkat(struct fhandle *, int, const char *); int fhopen(const struct fhandle *, int); int fhreadlink(struct fhandle *, char *, size_t); int fhstat(const struct fhandle *, struct stat *); int fhstatfs(const struct fhandle *, struct statfs *); int fstatfs(int, struct statfs *); int getfh(const char *, fhandle_t *); int getfhat(int, char *, struct fhandle *, int); int getfsstat(struct statfs *, long, int); int getmntinfo(struct statfs **, int); int lgetfh(const char *, fhandle_t *); int mount(const char *, const char *, int, void *); int nmount(struct iovec *, unsigned int, int); int statfs(const char *, struct statfs *); int unmount(const char *, int); /* C library stuff */ int getvfsbyname(const char *, struct xvfsconf *); __END_DECLS #endif /* _KERNEL */ #endif /* !_SYS_MOUNT_H_ */ diff --git a/sys/sys/smr.h b/sys/sys/smr.h index 9218deadbc85..c110be9a66c2 100644 --- a/sys/sys/smr.h +++ b/sys/sys/smr.h @@ -1,366 +1,261 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2019, 2020 Jeffrey Roberson * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice unmodified, this list of conditions, and the following * disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. * * $FreeBSD$ * */ #ifndef _SYS_SMR_H_ #define _SYS_SMR_H_ #include /* * Safe memory reclamation. See subr_smr.c for a description of the - * algorithm. + * algorithm, and smr_types.h for macros to define and access SMR-protected + * data structures. * * Readers synchronize with smr_enter()/exit() and writers may either * free directly to a SMR UMA zone or use smr_synchronize or wait. */ /* * Modular arithmetic for comparing sequence numbers that have * potentially wrapped. Copied from tcp_seq.h. */ #define SMR_SEQ_LT(a, b) ((smr_delta_t)((a)-(b)) < 0) #define SMR_SEQ_LEQ(a, b) ((smr_delta_t)((a)-(b)) <= 0) #define SMR_SEQ_GT(a, b) ((smr_delta_t)((a)-(b)) > 0) #define SMR_SEQ_GEQ(a, b) ((smr_delta_t)((a)-(b)) >= 0) #define SMR_SEQ_DELTA(a, b) ((smr_delta_t)((a)-(b))) #define SMR_SEQ_MIN(a, b) (SMR_SEQ_LT((a), (b)) ? (a) : (b)) #define SMR_SEQ_MAX(a, b) (SMR_SEQ_GT((a), (b)) ? (a) : (b)) #define SMR_SEQ_INVALID 0 /* Shared SMR state. */ union s_wr { struct { smr_seq_t seq; /* Current write sequence #. */ int ticks; /* tick of last update (LAZY) */ }; uint64_t _pair; }; struct smr_shared { const char *s_name; /* Name for debugging/reporting. */ union s_wr s_wr; /* Write sequence */ smr_seq_t s_rd_seq; /* Minimum observed read sequence. */ }; typedef struct smr_shared *smr_shared_t; /* Per-cpu SMR state. */ struct smr { smr_seq_t c_seq; /* Current observed sequence. */ smr_shared_t c_shared; /* Shared SMR state. */ int c_deferred; /* Deferred advance counter. */ int c_limit; /* Deferred advance limit. */ int c_flags; /* SMR Configuration */ }; #define SMR_LAZY 0x0001 /* Higher latency write, fast read. */ #define SMR_DEFERRED 0x0002 /* Aggregate updates to wr_seq. */ -#define SMR_ENTERED(smr) \ - (curthread->td_critnest != 0 && zpcpu_get((smr))->c_seq != SMR_SEQ_INVALID) - -#define SMR_ASSERT_ENTERED(smr) \ - KASSERT(SMR_ENTERED(smr), ("Not in smr section")) - -#define SMR_ASSERT_NOT_ENTERED(smr) \ - KASSERT(!SMR_ENTERED(smr), ("In smr section.")); - -#define SMR_ASSERT(ex, fn) \ - KASSERT((ex), (fn ": Assertion " #ex " failed at %s:%d", __FILE__, __LINE__)) - -/* - * SMR Accessors are meant to provide safe access to SMR protected - * pointers and prevent misuse and accidental access. - * - * Accessors are grouped by type: - * entered - Use while in a read section (between smr_enter/smr_exit()) - * serialized - Use while holding a lock that serializes writers. Updates - * are synchronized with readers via included barriers. - * unserialized - Use after the memory is out of scope and not visible to - * readers. - * - * All acceses include a parameter for an assert to verify the required - * synchronization. For example, a writer might use: - * - * smr_serialized_store(pointer, value, mtx_owned(&writelock)); - * - * These are only enabled in INVARIANTS kernels. - */ - -/* Type restricting pointer access to force smr accessors. */ -#define SMR_TYPE_DECLARE(smrtype, type) \ -typedef struct { \ - type __ptr; /* Do not access directly */ \ -} smrtype - -/* - * Read from an SMR protected pointer while in a read section. - */ -#define smr_entered_load(p, smr) ({ \ - SMR_ASSERT(SMR_ENTERED((smr)), "smr_entered_load"); \ - (__typeof((p)->__ptr))atomic_load_acq_ptr((uintptr_t *)&(p)->__ptr); \ -}) - -/* - * Read from an SMR protected pointer while serialized by an - * external mechanism. 'ex' should contain an assert that the - * external mechanism is held. i.e. mtx_owned() - */ -#define smr_serialized_load(p, ex) ({ \ - SMR_ASSERT(ex, "smr_serialized_load"); \ - (__typeof((p)->__ptr))atomic_load_ptr(&(p)->__ptr); \ -}) - -/* - * Store 'v' to an SMR protected pointer while serialized by an - * external mechanism. 'ex' should contain an assert that the - * external mechanism is held. i.e. mtx_owned() - * - * Writers that are serialized with mutual exclusion or on a single - * thread should use smr_serialized_store() rather than swap. - */ -#define smr_serialized_store(p, v, ex) do { \ - SMR_ASSERT(ex, "smr_serialized_store"); \ - __typeof((p)->__ptr) _v = (v); \ - atomic_store_rel_ptr((uintptr_t *)&(p)->__ptr, (uintptr_t)_v); \ -} while (0) - -/* - * swap 'v' with an SMR protected pointer and return the old value - * while serialized by an external mechanism. 'ex' should contain - * an assert that the external mechanism is provided. i.e. mtx_owned() - * - * Swap permits multiple writers to update a pointer concurrently. - */ -#define smr_serialized_swap(p, v, ex) ({ \ - SMR_ASSERT(ex, "smr_serialized_swap"); \ - __typeof((p)->__ptr) _v = (v); \ - /* Release barrier guarantees contents are visible to reader */ \ - atomic_thread_fence_rel(); \ - (__typeof((p)->__ptr))atomic_swap_ptr( \ - (uintptr_t *)&(p)->__ptr, (uintptr_t)_v); \ -}) - -/* - * Read from an SMR protected pointer when no serialization is required - * such as in the destructor callback or when the caller guarantees other - * synchronization. - */ -#define smr_unserialized_load(p, ex) ({ \ - SMR_ASSERT(ex, "smr_unserialized_load"); \ - (__typeof((p)->__ptr))atomic_load_ptr(&(p)->__ptr); \ -}) - -/* - * Store to an SMR protected pointer when no serialiation is required - * such as in the destructor callback or when the caller guarantees other - * synchronization. - */ -#define smr_unserialized_store(p, v, ex) do { \ - SMR_ASSERT(ex, "smr_unserialized_store"); \ - __typeof((p)->__ptr) _v = (v); \ - atomic_store_ptr((uintptr_t *)&(p)->__ptr, (uintptr_t)_v); \ -} while (0) - /* * Return the current write sequence number. This is not the same as the * current goal which may be in the future. */ static inline smr_seq_t smr_shared_current(smr_shared_t s) { return (atomic_load_int(&s->s_wr.seq)); } static inline smr_seq_t smr_current(smr_t smr) { return (smr_shared_current(zpcpu_get(smr)->c_shared)); } /* * Enter a read section. */ static inline void smr_enter(smr_t smr) { critical_enter(); smr = zpcpu_get(smr); KASSERT((smr->c_flags & SMR_LAZY) == 0, ("smr_enter(%s) lazy smr.", smr->c_shared->s_name)); KASSERT(smr->c_seq == 0, ("smr_enter(%s) does not support recursion.", smr->c_shared->s_name)); /* * Store the current observed write sequence number in our * per-cpu state so that it can be queried via smr_poll(). * Frees that are newer than this stored value will be * deferred until we call smr_exit(). * * An acquire barrier is used to synchronize with smr_exit() * and smr_poll(). * * It is possible that a long delay between loading the wr_seq * and storing the c_seq could create a situation where the * rd_seq advances beyond our stored c_seq. In this situation * only the observed wr_seq is stale, the fence still orders * the load. See smr_poll() for details on how this condition * is detected and handled there. */ /* This is an add because we do not have atomic_store_acq_int */ atomic_add_acq_int(&smr->c_seq, smr_shared_current(smr->c_shared)); } /* * Exit a read section. */ static inline void smr_exit(smr_t smr) { smr = zpcpu_get(smr); CRITICAL_ASSERT(curthread); KASSERT((smr->c_flags & SMR_LAZY) == 0, ("smr_exit(%s) lazy smr.", smr->c_shared->s_name)); KASSERT(smr->c_seq != SMR_SEQ_INVALID, ("smr_exit(%s) not in a smr section.", smr->c_shared->s_name)); /* * Clear the recorded sequence number. This allows poll() to * detect CPUs not in read sections. * * Use release semantics to retire any stores before the sequence * number is cleared. */ atomic_store_rel_int(&smr->c_seq, SMR_SEQ_INVALID); critical_exit(); } /* * Enter a lazy smr section. This is used for read-mostly state that * can tolerate a high free latency. */ static inline void smr_lazy_enter(smr_t smr) { critical_enter(); smr = zpcpu_get(smr); KASSERT((smr->c_flags & SMR_LAZY) != 0, ("smr_lazy_enter(%s) non-lazy smr.", smr->c_shared->s_name)); KASSERT(smr->c_seq == 0, ("smr_lazy_enter(%s) does not support recursion.", smr->c_shared->s_name)); /* * This needs no serialization. If an interrupt occurs before we * assign sr_seq to c_seq any speculative loads will be discarded. * If we assign a stale wr_seq value due to interrupt we use the * same algorithm that renders smr_enter() safe. */ atomic_store_int(&smr->c_seq, smr_shared_current(smr->c_shared)); } /* * Exit a lazy smr section. This is used for read-mostly state that * can tolerate a high free latency. */ static inline void smr_lazy_exit(smr_t smr) { smr = zpcpu_get(smr); CRITICAL_ASSERT(curthread); KASSERT((smr->c_flags & SMR_LAZY) != 0, ("smr_lazy_enter(%s) non-lazy smr.", smr->c_shared->s_name)); KASSERT(smr->c_seq != SMR_SEQ_INVALID, ("smr_lazy_exit(%s) not in a smr section.", smr->c_shared->s_name)); /* * All loads/stores must be retired before the sequence becomes * visible. The fence compiles away on amd64. Another * alternative would be to omit the fence but store the exit * time and wait 1 tick longer. */ atomic_thread_fence_rel(); atomic_store_int(&smr->c_seq, SMR_SEQ_INVALID); critical_exit(); } /* * Advances the write sequence number. Returns the sequence number * required to ensure that all modifications are visible to readers. */ smr_seq_t smr_advance(smr_t smr); /* * Returns true if a goal sequence has been reached. If * wait is true this will busy loop until success. */ bool smr_poll(smr_t smr, smr_seq_t goal, bool wait); /* Create a new SMR context. */ smr_t smr_create(const char *name, int limit, int flags); /* Destroy the context. */ void smr_destroy(smr_t smr); /* * Blocking wait for all readers to observe 'goal'. */ static inline bool smr_wait(smr_t smr, smr_seq_t goal) { return (smr_poll(smr, goal, true)); } /* * Synchronize advances the write sequence and returns when all * readers have observed it. * * If your application can cache a sequence number returned from * smr_advance() and poll or wait at a later time there will * be less chance of busy looping while waiting for readers. */ static inline void smr_synchronize(smr_t smr) { smr_wait(smr, smr_advance(smr)); } /* Only at startup. */ void smr_init(void); #endif /* _SYS_SMR_H_ */ diff --git a/sys/sys/smr_types.h b/sys/sys/smr_types.h new file mode 100644 index 000000000000..3497f21af846 --- /dev/null +++ b/sys/sys/smr_types.h @@ -0,0 +1,138 @@ +/*- + * SPDX-License-Identifier: BSD-2-Clause-FreeBSD + * + * Copyright (c) 2019, 2020 Jeffrey Roberson + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice unmodified, this list of conditions, and the following + * disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR + * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES + * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. + * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, + * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT + * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, + * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY + * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF + * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + * + * $FreeBSD$ + */ + +#ifndef _SYS_SMR_TYPES_H_ +#define _SYS_SMR_TYPES_H_ + +#include + +/* + * SMR Accessors are meant to provide safe access to SMR protected + * pointers and prevent misuse and accidental access. + * + * Accessors are grouped by type: + * entered - Use while in a read section (between smr_enter/smr_exit()) + * serialized - Use while holding a lock that serializes writers. Updates + * are synchronized with readers via included barriers. + * unserialized - Use after the memory is out of scope and not visible to + * readers. + * + * All acceses include a parameter for an assert to verify the required + * synchronization. For example, a writer might use: + * + * smr_serialized_store(pointer, value, mtx_owned(&writelock)); + * + * These are only enabled in INVARIANTS kernels. + */ + +/* Type restricting pointer access to force smr accessors. */ +#define SMR_POINTER(type) \ +struct { \ + type __ptr; /* Do not access directly */ \ +} + +/* + * Read from an SMR protected pointer while in a read section. + */ +#define smr_entered_load(p, smr) ({ \ + SMR_ASSERT(SMR_ENTERED((smr)), "smr_entered_load"); \ + (__typeof((p)->__ptr))atomic_load_acq_ptr((uintptr_t *)&(p)->__ptr); \ +}) + +/* + * Read from an SMR protected pointer while serialized by an + * external mechanism. 'ex' should contain an assert that the + * external mechanism is held. i.e. mtx_owned() + */ +#define smr_serialized_load(p, ex) ({ \ + SMR_ASSERT(ex, "smr_serialized_load"); \ + (__typeof((p)->__ptr))atomic_load_ptr(&(p)->__ptr); \ +}) + +/* + * Store 'v' to an SMR protected pointer while serialized by an + * external mechanism. 'ex' should contain an assert that the + * external mechanism is held. i.e. mtx_owned() + * + * Writers that are serialized with mutual exclusion or on a single + * thread should use smr_serialized_store() rather than swap. + */ +#define smr_serialized_store(p, v, ex) do { \ + SMR_ASSERT(ex, "smr_serialized_store"); \ + __typeof((p)->__ptr) _v = (v); \ + atomic_store_rel_ptr((uintptr_t *)&(p)->__ptr, (uintptr_t)_v); \ +} while (0) + +/* + * swap 'v' with an SMR protected pointer and return the old value + * while serialized by an external mechanism. 'ex' should contain + * an assert that the external mechanism is provided. i.e. mtx_owned() + * + * Swap permits multiple writers to update a pointer concurrently. + */ +#define smr_serialized_swap(p, v, ex) ({ \ + SMR_ASSERT(ex, "smr_serialized_swap"); \ + __typeof((p)->__ptr) _v = (v); \ + /* Release barrier guarantees contents are visible to reader */ \ + atomic_thread_fence_rel(); \ + (__typeof((p)->__ptr))atomic_swap_ptr( \ + (uintptr_t *)&(p)->__ptr, (uintptr_t)_v); \ +}) + +/* + * Read from an SMR protected pointer when no serialization is required + * such as in the destructor callback or when the caller guarantees other + * synchronization. + */ +#define smr_unserialized_load(p, ex) ({ \ + SMR_ASSERT(ex, "smr_unserialized_load"); \ + (__typeof((p)->__ptr))atomic_load_ptr(&(p)->__ptr); \ +}) + +/* + * Store to an SMR protected pointer when no serialiation is required + * such as in the destructor callback or when the caller guarantees other + * synchronization. + */ +#define smr_unserialized_store(p, v, ex) do { \ + SMR_ASSERT(ex, "smr_unserialized_store"); \ + __typeof((p)->__ptr) _v = (v); \ + atomic_store_ptr((uintptr_t *)&(p)->__ptr, (uintptr_t)_v); \ +} while (0) + +#ifndef _KERNEL + +/* + * Load an SMR protected pointer when accessing kernel data structures through + * libkvm. + */ +#define smr_kvm_load(p) ((p)->__ptr) + +#endif /* !_KERNEL */ +#endif /* !_SYS_SMR_TYPES_H_ */ diff --git a/sys/ufs/ffs/ffs_alloc.c b/sys/ufs/ffs/ffs_alloc.c index fb270b11d912..0b8068db0da7 100644 --- a/sys/ufs/ffs/ffs_alloc.c +++ b/sys/ufs/ffs/ffs_alloc.c @@ -1,3661 +1,3664 @@ /*- * SPDX-License-Identifier: (BSD-2-Clause-FreeBSD AND BSD-3-Clause) * * Copyright (c) 2002 Networks Associates Technology, Inc. * All rights reserved. * * This software was developed for the FreeBSD Project by Marshall * Kirk McKusick and Network Associates Laboratories, the Security * Research Division of Network Associates, Inc. under DARPA/SPAWAR * contract N66001-01-C-8035 ("CBOSS"), as part of the DARPA CHATS * research program * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * Copyright (c) 1982, 1986, 1989, 1993 * The Regents of the University of California. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)ffs_alloc.c 8.18 (Berkeley) 5/26/95 */ #include __FBSDID("$FreeBSD$"); #include "opt_quota.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include typedef ufs2_daddr_t allocfcn_t(struct inode *ip, u_int cg, ufs2_daddr_t bpref, int size, int rsize); static ufs2_daddr_t ffs_alloccg(struct inode *, u_int, ufs2_daddr_t, int, int); static ufs2_daddr_t ffs_alloccgblk(struct inode *, struct buf *, ufs2_daddr_t, int); static void ffs_blkfree_cg(struct ufsmount *, struct fs *, struct vnode *, ufs2_daddr_t, long, ino_t, struct workhead *); #ifdef INVARIANTS static int ffs_checkblk(struct inode *, ufs2_daddr_t, long); #endif static ufs2_daddr_t ffs_clusteralloc(struct inode *, u_int, ufs2_daddr_t, int); static ino_t ffs_dirpref(struct inode *); static ufs2_daddr_t ffs_fragextend(struct inode *, u_int, ufs2_daddr_t, int, int); static ufs2_daddr_t ffs_hashalloc (struct inode *, u_int, ufs2_daddr_t, int, int, allocfcn_t *); static ufs2_daddr_t ffs_nodealloccg(struct inode *, u_int, ufs2_daddr_t, int, int); static ufs1_daddr_t ffs_mapsearch(struct fs *, struct cg *, ufs2_daddr_t, int); static int ffs_reallocblks_ufs1(struct vop_reallocblks_args *); static int ffs_reallocblks_ufs2(struct vop_reallocblks_args *); static void ffs_ckhash_cg(struct buf *); /* * Allocate a block in the filesystem. * * The size of the requested block is given, which must be some * multiple of fs_fsize and <= fs_bsize. * A preference may be optionally specified. If a preference is given * the following hierarchy is used to allocate a block: * 1) allocate the requested block. * 2) allocate a rotationally optimal block in the same cylinder. * 3) allocate a block in the same cylinder group. * 4) quadradically rehash into other cylinder groups, until an * available block is located. * If no block preference is given the following hierarchy is used * to allocate a block: * 1) allocate a block in the cylinder group that contains the * inode for the file. * 2) quadradically rehash into other cylinder groups, until an * available block is located. */ int ffs_alloc(ip, lbn, bpref, size, flags, cred, bnp) struct inode *ip; ufs2_daddr_t lbn, bpref; int size, flags; struct ucred *cred; ufs2_daddr_t *bnp; { struct fs *fs; struct ufsmount *ump; ufs2_daddr_t bno; u_int cg, reclaimed; int64_t delta; #ifdef QUOTA int error; #endif *bnp = 0; ump = ITOUMP(ip); fs = ump->um_fs; mtx_assert(UFS_MTX(ump), MA_OWNED); #ifdef INVARIANTS if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) { printf("dev = %s, bsize = %ld, size = %d, fs = %s\n", devtoname(ump->um_dev), (long)fs->fs_bsize, size, fs->fs_fsmnt); panic("ffs_alloc: bad size"); } if (cred == NOCRED) panic("ffs_alloc: missing credential"); #endif /* INVARIANTS */ reclaimed = 0; retry: #ifdef QUOTA UFS_UNLOCK(ump); error = chkdq(ip, btodb(size), cred, 0); if (error) return (error); UFS_LOCK(ump); #endif if (size == fs->fs_bsize && fs->fs_cstotal.cs_nbfree == 0) goto nospace; if (priv_check_cred(cred, PRIV_VFS_BLOCKRESERVE) && freespace(fs, fs->fs_minfree) - numfrags(fs, size) < 0) goto nospace; if (bpref >= fs->fs_size) bpref = 0; if (bpref == 0) cg = ino_to_cg(fs, ip->i_number); else cg = dtog(fs, bpref); bno = ffs_hashalloc(ip, cg, bpref, size, size, ffs_alloccg); if (bno > 0) { delta = btodb(size); DIP_SET(ip, i_blocks, DIP(ip, i_blocks) + delta); if (flags & IO_EXT) UFS_INODE_SET_FLAG(ip, IN_CHANGE); else UFS_INODE_SET_FLAG(ip, IN_CHANGE | IN_UPDATE); *bnp = bno; return (0); } nospace: #ifdef QUOTA UFS_UNLOCK(ump); /* * Restore user's disk quota because allocation failed. */ (void) chkdq(ip, -btodb(size), cred, FORCE); UFS_LOCK(ump); #endif if (reclaimed == 0 && (flags & IO_BUFLOCKED) == 0) { reclaimed = 1; softdep_request_cleanup(fs, ITOV(ip), cred, FLUSH_BLOCKS_WAIT); goto retry; } if (reclaimed > 0 && ppsratecheck(&ump->um_last_fullmsg, &ump->um_secs_fullmsg, 1)) { UFS_UNLOCK(ump); ffs_fserr(fs, ip->i_number, "filesystem full"); uprintf("\n%s: write failed, filesystem is full\n", fs->fs_fsmnt); } else { UFS_UNLOCK(ump); } return (ENOSPC); } /* * Reallocate a fragment to a bigger size * * The number and size of the old block is given, and a preference * and new size is also specified. The allocator attempts to extend * the original block. Failing that, the regular block allocator is * invoked to get an appropriate block. */ int ffs_realloccg(ip, lbprev, bprev, bpref, osize, nsize, flags, cred, bpp) struct inode *ip; ufs2_daddr_t lbprev; ufs2_daddr_t bprev; ufs2_daddr_t bpref; int osize, nsize, flags; struct ucred *cred; struct buf **bpp; { struct vnode *vp; struct fs *fs; struct buf *bp; struct ufsmount *ump; u_int cg, request, reclaimed; int error, gbflags; ufs2_daddr_t bno; int64_t delta; vp = ITOV(ip); ump = ITOUMP(ip); fs = ump->um_fs; bp = NULL; gbflags = (flags & BA_UNMAPPED) != 0 ? GB_UNMAPPED : 0; mtx_assert(UFS_MTX(ump), MA_OWNED); #ifdef INVARIANTS if (vp->v_mount->mnt_kern_flag & MNTK_SUSPENDED) panic("ffs_realloccg: allocation on suspended filesystem"); if ((u_int)osize > fs->fs_bsize || fragoff(fs, osize) != 0 || (u_int)nsize > fs->fs_bsize || fragoff(fs, nsize) != 0) { printf( "dev = %s, bsize = %ld, osize = %d, nsize = %d, fs = %s\n", devtoname(ump->um_dev), (long)fs->fs_bsize, osize, nsize, fs->fs_fsmnt); panic("ffs_realloccg: bad size"); } if (cred == NOCRED) panic("ffs_realloccg: missing credential"); #endif /* INVARIANTS */ reclaimed = 0; retry: if (priv_check_cred(cred, PRIV_VFS_BLOCKRESERVE) && freespace(fs, fs->fs_minfree) - numfrags(fs, nsize - osize) < 0) { goto nospace; } if (bprev == 0) { printf("dev = %s, bsize = %ld, bprev = %jd, fs = %s\n", devtoname(ump->um_dev), (long)fs->fs_bsize, (intmax_t)bprev, fs->fs_fsmnt); panic("ffs_realloccg: bad bprev"); } UFS_UNLOCK(ump); /* * Allocate the extra space in the buffer. */ error = bread_gb(vp, lbprev, osize, NOCRED, gbflags, &bp); if (error) { return (error); } if (bp->b_blkno == bp->b_lblkno) { if (lbprev >= UFS_NDADDR) panic("ffs_realloccg: lbprev out of range"); bp->b_blkno = fsbtodb(fs, bprev); } #ifdef QUOTA error = chkdq(ip, btodb(nsize - osize), cred, 0); if (error) { brelse(bp); return (error); } #endif /* * Check for extension in the existing location. */ *bpp = NULL; cg = dtog(fs, bprev); UFS_LOCK(ump); bno = ffs_fragextend(ip, cg, bprev, osize, nsize); if (bno) { if (bp->b_blkno != fsbtodb(fs, bno)) panic("ffs_realloccg: bad blockno"); delta = btodb(nsize - osize); DIP_SET(ip, i_blocks, DIP(ip, i_blocks) + delta); if (flags & IO_EXT) UFS_INODE_SET_FLAG(ip, IN_CHANGE); else UFS_INODE_SET_FLAG(ip, IN_CHANGE | IN_UPDATE); allocbuf(bp, nsize); bp->b_flags |= B_DONE; vfs_bio_bzero_buf(bp, osize, nsize - osize); if ((bp->b_flags & (B_MALLOC | B_VMIO)) == B_VMIO) vfs_bio_set_valid(bp, osize, nsize - osize); *bpp = bp; return (0); } /* * Allocate a new disk location. */ if (bpref >= fs->fs_size) bpref = 0; switch ((int)fs->fs_optim) { case FS_OPTSPACE: /* * Allocate an exact sized fragment. Although this makes * best use of space, we will waste time relocating it if * the file continues to grow. If the fragmentation is * less than half of the minimum free reserve, we choose * to begin optimizing for time. */ request = nsize; if (fs->fs_minfree <= 5 || fs->fs_cstotal.cs_nffree > (off_t)fs->fs_dsize * fs->fs_minfree / (2 * 100)) break; log(LOG_NOTICE, "%s: optimization changed from SPACE to TIME\n", fs->fs_fsmnt); fs->fs_optim = FS_OPTTIME; break; case FS_OPTTIME: /* * At this point we have discovered a file that is trying to * grow a small fragment to a larger fragment. To save time, * we allocate a full sized block, then free the unused portion. * If the file continues to grow, the `ffs_fragextend' call * above will be able to grow it in place without further * copying. If aberrant programs cause disk fragmentation to * grow within 2% of the free reserve, we choose to begin * optimizing for space. */ request = fs->fs_bsize; if (fs->fs_cstotal.cs_nffree < (off_t)fs->fs_dsize * (fs->fs_minfree - 2) / 100) break; log(LOG_NOTICE, "%s: optimization changed from TIME to SPACE\n", fs->fs_fsmnt); fs->fs_optim = FS_OPTSPACE; break; default: printf("dev = %s, optim = %ld, fs = %s\n", devtoname(ump->um_dev), (long)fs->fs_optim, fs->fs_fsmnt); panic("ffs_realloccg: bad optim"); /* NOTREACHED */ } bno = ffs_hashalloc(ip, cg, bpref, request, nsize, ffs_alloccg); if (bno > 0) { bp->b_blkno = fsbtodb(fs, bno); if (!DOINGSOFTDEP(vp)) /* * The usual case is that a smaller fragment that * was just allocated has been replaced with a bigger * fragment or a full-size block. If it is marked as * B_DELWRI, the current contents have not been written * to disk. It is possible that the block was written * earlier, but very uncommon. If the block has never * been written, there is no need to send a BIO_DELETE * for it when it is freed. The gain from avoiding the * TRIMs for the common case of unwritten blocks far * exceeds the cost of the write amplification for the * uncommon case of failing to send a TRIM for a block * that had been written. */ ffs_blkfree(ump, fs, ump->um_devvp, bprev, (long)osize, ip->i_number, vp->v_type, NULL, (bp->b_flags & B_DELWRI) != 0 ? NOTRIM_KEY : SINGLETON_KEY); delta = btodb(nsize - osize); DIP_SET(ip, i_blocks, DIP(ip, i_blocks) + delta); if (flags & IO_EXT) UFS_INODE_SET_FLAG(ip, IN_CHANGE); else UFS_INODE_SET_FLAG(ip, IN_CHANGE | IN_UPDATE); allocbuf(bp, nsize); bp->b_flags |= B_DONE; vfs_bio_bzero_buf(bp, osize, nsize - osize); if ((bp->b_flags & (B_MALLOC | B_VMIO)) == B_VMIO) vfs_bio_set_valid(bp, osize, nsize - osize); *bpp = bp; return (0); } #ifdef QUOTA UFS_UNLOCK(ump); /* * Restore user's disk quota because allocation failed. */ (void) chkdq(ip, -btodb(nsize - osize), cred, FORCE); UFS_LOCK(ump); #endif nospace: /* * no space available */ if (reclaimed == 0 && (flags & IO_BUFLOCKED) == 0) { reclaimed = 1; UFS_UNLOCK(ump); if (bp) { brelse(bp); bp = NULL; } UFS_LOCK(ump); softdep_request_cleanup(fs, vp, cred, FLUSH_BLOCKS_WAIT); goto retry; } if (reclaimed > 0 && ppsratecheck(&ump->um_last_fullmsg, &ump->um_secs_fullmsg, 1)) { UFS_UNLOCK(ump); ffs_fserr(fs, ip->i_number, "filesystem full"); uprintf("\n%s: write failed, filesystem is full\n", fs->fs_fsmnt); } else { UFS_UNLOCK(ump); } if (bp) brelse(bp); return (ENOSPC); } /* * Reallocate a sequence of blocks into a contiguous sequence of blocks. * * The vnode and an array of buffer pointers for a range of sequential * logical blocks to be made contiguous is given. The allocator attempts * to find a range of sequential blocks starting as close as possible * from the end of the allocation for the logical block immediately * preceding the current range. If successful, the physical block numbers * in the buffer pointers and in the inode are changed to reflect the new * allocation. If unsuccessful, the allocation is left unchanged. The * success in doing the reallocation is returned. Note that the error * return is not reflected back to the user. Rather the previous block * allocation will be used. */ SYSCTL_NODE(_vfs, OID_AUTO, ffs, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "FFS filesystem"); static int doasyncfree = 1; SYSCTL_INT(_vfs_ffs, OID_AUTO, doasyncfree, CTLFLAG_RW, &doasyncfree, 0, "do not force synchronous writes when blocks are reallocated"); static int doreallocblks = 1; SYSCTL_INT(_vfs_ffs, OID_AUTO, doreallocblks, CTLFLAG_RW, &doreallocblks, 0, "enable block reallocation"); static int dotrimcons = 1; SYSCTL_INT(_vfs_ffs, OID_AUTO, dotrimcons, CTLFLAG_RWTUN, &dotrimcons, 0, "enable BIO_DELETE / TRIM consolidation"); static int maxclustersearch = 10; SYSCTL_INT(_vfs_ffs, OID_AUTO, maxclustersearch, CTLFLAG_RW, &maxclustersearch, 0, "max number of cylinder group to search for contigous blocks"); #ifdef DIAGNOSTIC static int prtrealloc = 0; SYSCTL_INT(_debug, OID_AUTO, ffs_prtrealloc, CTLFLAG_RW, &prtrealloc, 0, "print out FFS filesystem block reallocation operations"); #endif int ffs_reallocblks(ap) struct vop_reallocblks_args /* { struct vnode *a_vp; struct cluster_save *a_buflist; } */ *ap; { struct ufsmount *ump; /* * We used to skip reallocating the blocks of a file into a * contiguous sequence if the underlying flash device requested * BIO_DELETE notifications, because devices that benefit from * BIO_DELETE also benefit from not moving the data. However, * the destination for the data is usually moved before the data * is written to the initially allocated location, so we rarely * suffer the penalty of extra writes. With the addition of the * consolidation of contiguous blocks into single BIO_DELETE * operations, having fewer but larger contiguous blocks reduces * the number of (slow and expensive) BIO_DELETE operations. So * when doing BIO_DELETE consolidation, we do block reallocation. * * Skip if reallocblks has been disabled globally. */ ump = ap->a_vp->v_mount->mnt_data; if ((((ump->um_flags) & UM_CANDELETE) != 0 && dotrimcons == 0) || doreallocblks == 0) return (ENOSPC); /* * We can't wait in softdep prealloc as it may fsync and recurse * here. Instead we simply fail to reallocate blocks if this * rare condition arises. */ if (DOINGSOFTDEP(ap->a_vp)) if (softdep_prealloc(ap->a_vp, MNT_NOWAIT) != 0) return (ENOSPC); if (ump->um_fstype == UFS1) return (ffs_reallocblks_ufs1(ap)); return (ffs_reallocblks_ufs2(ap)); } static int ffs_reallocblks_ufs1(ap) struct vop_reallocblks_args /* { struct vnode *a_vp; struct cluster_save *a_buflist; } */ *ap; { struct fs *fs; struct inode *ip; struct vnode *vp; struct buf *sbp, *ebp, *bp; ufs1_daddr_t *bap, *sbap, *ebap; struct cluster_save *buflist; struct ufsmount *ump; ufs_lbn_t start_lbn, end_lbn; ufs1_daddr_t soff, newblk, blkno; ufs2_daddr_t pref; struct indir start_ap[UFS_NIADDR + 1], end_ap[UFS_NIADDR + 1], *idp; int i, cg, len, start_lvl, end_lvl, ssize; vp = ap->a_vp; ip = VTOI(vp); ump = ITOUMP(ip); fs = ump->um_fs; /* * If we are not tracking block clusters or if we have less than 4% * free blocks left, then do not attempt to cluster. Running with * less than 5% free block reserve is not recommended and those that * choose to do so do not expect to have good file layout. */ if (fs->fs_contigsumsize <= 0 || freespace(fs, 4) < 0) return (ENOSPC); buflist = ap->a_buflist; len = buflist->bs_nchildren; start_lbn = buflist->bs_children[0]->b_lblkno; end_lbn = start_lbn + len - 1; #ifdef INVARIANTS for (i = 0; i < len; i++) if (!ffs_checkblk(ip, dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize)) panic("ffs_reallocblks: unallocated block 1"); for (i = 1; i < len; i++) if (buflist->bs_children[i]->b_lblkno != start_lbn + i) panic("ffs_reallocblks: non-logical cluster"); blkno = buflist->bs_children[0]->b_blkno; ssize = fsbtodb(fs, fs->fs_frag); for (i = 1; i < len - 1; i++) if (buflist->bs_children[i]->b_blkno != blkno + (i * ssize)) panic("ffs_reallocblks: non-physical cluster %d", i); #endif /* * If the cluster crosses the boundary for the first indirect * block, leave space for the indirect block. Indirect blocks * are initially laid out in a position after the last direct * block. Block reallocation would usually destroy locality by * moving the indirect block out of the way to make room for * data blocks if we didn't compensate here. We should also do * this for other indirect block boundaries, but it is only * important for the first one. */ if (start_lbn < UFS_NDADDR && end_lbn >= UFS_NDADDR) return (ENOSPC); /* * If the latest allocation is in a new cylinder group, assume that * the filesystem has decided to move and do not force it back to * the previous cylinder group. */ if (dtog(fs, dbtofsb(fs, buflist->bs_children[0]->b_blkno)) != dtog(fs, dbtofsb(fs, buflist->bs_children[len - 1]->b_blkno))) return (ENOSPC); if (ufs_getlbns(vp, start_lbn, start_ap, &start_lvl) || ufs_getlbns(vp, end_lbn, end_ap, &end_lvl)) return (ENOSPC); /* * Get the starting offset and block map for the first block. */ if (start_lvl == 0) { sbap = &ip->i_din1->di_db[0]; soff = start_lbn; } else { idp = &start_ap[start_lvl - 1]; if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &sbp)) { brelse(sbp); return (ENOSPC); } sbap = (ufs1_daddr_t *)sbp->b_data; soff = idp->in_off; } /* * If the block range spans two block maps, get the second map. */ ebap = NULL; if (end_lvl == 0 || (idp = &end_ap[end_lvl - 1])->in_off + 1 >= len) { ssize = len; } else { #ifdef INVARIANTS if (start_lvl > 0 && start_ap[start_lvl - 1].in_lbn == idp->in_lbn) panic("ffs_reallocblk: start == end"); #endif ssize = len - (idp->in_off + 1); if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &ebp)) goto fail; ebap = (ufs1_daddr_t *)ebp->b_data; } /* * Find the preferred location for the cluster. If we have not * previously failed at this endeavor, then follow our standard * preference calculation. If we have failed at it, then pick up * where we last ended our search. */ UFS_LOCK(ump); if (ip->i_nextclustercg == -1) pref = ffs_blkpref_ufs1(ip, start_lbn, soff, sbap); else pref = cgdata(fs, ip->i_nextclustercg); /* * Search the block map looking for an allocation of the desired size. * To avoid wasting too much time, we limit the number of cylinder * groups that we will search. */ cg = dtog(fs, pref); for (i = min(maxclustersearch, fs->fs_ncg); i > 0; i--) { if ((newblk = ffs_clusteralloc(ip, cg, pref, len)) != 0) break; cg += 1; if (cg >= fs->fs_ncg) cg = 0; } /* * If we have failed in our search, record where we gave up for * next time. Otherwise, fall back to our usual search citerion. */ if (newblk == 0) { ip->i_nextclustercg = cg; UFS_UNLOCK(ump); goto fail; } ip->i_nextclustercg = -1; /* * We have found a new contiguous block. * * First we have to replace the old block pointers with the new * block pointers in the inode and indirect blocks associated * with the file. */ #ifdef DIAGNOSTIC if (prtrealloc) printf("realloc: ino %ju, lbns %jd-%jd\n\told:", (uintmax_t)ip->i_number, (intmax_t)start_lbn, (intmax_t)end_lbn); #endif blkno = newblk; for (bap = &sbap[soff], i = 0; i < len; i++, blkno += fs->fs_frag) { if (i == ssize) { bap = ebap; soff = -i; } #ifdef INVARIANTS if (!ffs_checkblk(ip, dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize)) panic("ffs_reallocblks: unallocated block 2"); if (dbtofsb(fs, buflist->bs_children[i]->b_blkno) != *bap) panic("ffs_reallocblks: alloc mismatch"); #endif #ifdef DIAGNOSTIC if (prtrealloc) printf(" %d,", *bap); #endif if (DOINGSOFTDEP(vp)) { if (sbap == &ip->i_din1->di_db[0] && i < ssize) softdep_setup_allocdirect(ip, start_lbn + i, blkno, *bap, fs->fs_bsize, fs->fs_bsize, buflist->bs_children[i]); else softdep_setup_allocindir_page(ip, start_lbn + i, i < ssize ? sbp : ebp, soff + i, blkno, *bap, buflist->bs_children[i]); } *bap++ = blkno; } /* * Next we must write out the modified inode and indirect blocks. * For strict correctness, the writes should be synchronous since * the old block values may have been written to disk. In practise * they are almost never written, but if we are concerned about * strict correctness, the `doasyncfree' flag should be set to zero. * * The test on `doasyncfree' should be changed to test a flag * that shows whether the associated buffers and inodes have * been written. The flag should be set when the cluster is * started and cleared whenever the buffer or inode is flushed. * We can then check below to see if it is set, and do the * synchronous write only when it has been cleared. */ if (sbap != &ip->i_din1->di_db[0]) { if (doasyncfree) bdwrite(sbp); else bwrite(sbp); } else { UFS_INODE_SET_FLAG(ip, IN_CHANGE | IN_UPDATE); if (!doasyncfree) ffs_update(vp, 1); } if (ssize < len) { if (doasyncfree) bdwrite(ebp); else bwrite(ebp); } /* * Last, free the old blocks and assign the new blocks to the buffers. */ #ifdef DIAGNOSTIC if (prtrealloc) printf("\n\tnew:"); #endif for (blkno = newblk, i = 0; i < len; i++, blkno += fs->fs_frag) { bp = buflist->bs_children[i]; if (!DOINGSOFTDEP(vp)) /* * The usual case is that a set of N-contiguous blocks * that was just allocated has been replaced with a * set of N+1-contiguous blocks. If they are marked as * B_DELWRI, the current contents have not been written * to disk. It is possible that the blocks were written * earlier, but very uncommon. If the blocks have never * been written, there is no need to send a BIO_DELETE * for them when they are freed. The gain from avoiding * the TRIMs for the common case of unwritten blocks * far exceeds the cost of the write amplification for * the uncommon case of failing to send a TRIM for the * blocks that had been written. */ ffs_blkfree(ump, fs, ump->um_devvp, dbtofsb(fs, bp->b_blkno), fs->fs_bsize, ip->i_number, vp->v_type, NULL, (bp->b_flags & B_DELWRI) != 0 ? NOTRIM_KEY : SINGLETON_KEY); bp->b_blkno = fsbtodb(fs, blkno); #ifdef INVARIANTS if (!ffs_checkblk(ip, dbtofsb(fs, bp->b_blkno), fs->fs_bsize)) panic("ffs_reallocblks: unallocated block 3"); #endif #ifdef DIAGNOSTIC if (prtrealloc) printf(" %d,", blkno); #endif } #ifdef DIAGNOSTIC if (prtrealloc) { prtrealloc--; printf("\n"); } #endif return (0); fail: if (ssize < len) brelse(ebp); if (sbap != &ip->i_din1->di_db[0]) brelse(sbp); return (ENOSPC); } static int ffs_reallocblks_ufs2(ap) struct vop_reallocblks_args /* { struct vnode *a_vp; struct cluster_save *a_buflist; } */ *ap; { struct fs *fs; struct inode *ip; struct vnode *vp; struct buf *sbp, *ebp, *bp; ufs2_daddr_t *bap, *sbap, *ebap; struct cluster_save *buflist; struct ufsmount *ump; ufs_lbn_t start_lbn, end_lbn; ufs2_daddr_t soff, newblk, blkno, pref; struct indir start_ap[UFS_NIADDR + 1], end_ap[UFS_NIADDR + 1], *idp; int i, cg, len, start_lvl, end_lvl, ssize; vp = ap->a_vp; ip = VTOI(vp); ump = ITOUMP(ip); fs = ump->um_fs; /* * If we are not tracking block clusters or if we have less than 4% * free blocks left, then do not attempt to cluster. Running with * less than 5% free block reserve is not recommended and those that * choose to do so do not expect to have good file layout. */ if (fs->fs_contigsumsize <= 0 || freespace(fs, 4) < 0) return (ENOSPC); buflist = ap->a_buflist; len = buflist->bs_nchildren; start_lbn = buflist->bs_children[0]->b_lblkno; end_lbn = start_lbn + len - 1; #ifdef INVARIANTS for (i = 0; i < len; i++) if (!ffs_checkblk(ip, dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize)) panic("ffs_reallocblks: unallocated block 1"); for (i = 1; i < len; i++) if (buflist->bs_children[i]->b_lblkno != start_lbn + i) panic("ffs_reallocblks: non-logical cluster"); blkno = buflist->bs_children[0]->b_blkno; ssize = fsbtodb(fs, fs->fs_frag); for (i = 1; i < len - 1; i++) if (buflist->bs_children[i]->b_blkno != blkno + (i * ssize)) panic("ffs_reallocblks: non-physical cluster %d", i); #endif /* * If the cluster crosses the boundary for the first indirect * block, do not move anything in it. Indirect blocks are * usually initially laid out in a position between the data * blocks. Block reallocation would usually destroy locality by * moving the indirect block out of the way to make room for * data blocks if we didn't compensate here. We should also do * this for other indirect block boundaries, but it is only * important for the first one. */ if (start_lbn < UFS_NDADDR && end_lbn >= UFS_NDADDR) return (ENOSPC); /* * If the latest allocation is in a new cylinder group, assume that * the filesystem has decided to move and do not force it back to * the previous cylinder group. */ if (dtog(fs, dbtofsb(fs, buflist->bs_children[0]->b_blkno)) != dtog(fs, dbtofsb(fs, buflist->bs_children[len - 1]->b_blkno))) return (ENOSPC); if (ufs_getlbns(vp, start_lbn, start_ap, &start_lvl) || ufs_getlbns(vp, end_lbn, end_ap, &end_lvl)) return (ENOSPC); /* * Get the starting offset and block map for the first block. */ if (start_lvl == 0) { sbap = &ip->i_din2->di_db[0]; soff = start_lbn; } else { idp = &start_ap[start_lvl - 1]; if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &sbp)) { brelse(sbp); return (ENOSPC); } sbap = (ufs2_daddr_t *)sbp->b_data; soff = idp->in_off; } /* * If the block range spans two block maps, get the second map. */ ebap = NULL; if (end_lvl == 0 || (idp = &end_ap[end_lvl - 1])->in_off + 1 >= len) { ssize = len; } else { #ifdef INVARIANTS if (start_lvl > 0 && start_ap[start_lvl - 1].in_lbn == idp->in_lbn) panic("ffs_reallocblk: start == end"); #endif ssize = len - (idp->in_off + 1); if (bread(vp, idp->in_lbn, (int)fs->fs_bsize, NOCRED, &ebp)) goto fail; ebap = (ufs2_daddr_t *)ebp->b_data; } /* * Find the preferred location for the cluster. If we have not * previously failed at this endeavor, then follow our standard * preference calculation. If we have failed at it, then pick up * where we last ended our search. */ UFS_LOCK(ump); if (ip->i_nextclustercg == -1) pref = ffs_blkpref_ufs2(ip, start_lbn, soff, sbap); else pref = cgdata(fs, ip->i_nextclustercg); /* * Search the block map looking for an allocation of the desired size. * To avoid wasting too much time, we limit the number of cylinder * groups that we will search. */ cg = dtog(fs, pref); for (i = min(maxclustersearch, fs->fs_ncg); i > 0; i--) { if ((newblk = ffs_clusteralloc(ip, cg, pref, len)) != 0) break; cg += 1; if (cg >= fs->fs_ncg) cg = 0; } /* * If we have failed in our search, record where we gave up for * next time. Otherwise, fall back to our usual search citerion. */ if (newblk == 0) { ip->i_nextclustercg = cg; UFS_UNLOCK(ump); goto fail; } ip->i_nextclustercg = -1; /* * We have found a new contiguous block. * * First we have to replace the old block pointers with the new * block pointers in the inode and indirect blocks associated * with the file. */ #ifdef DIAGNOSTIC if (prtrealloc) printf("realloc: ino %ju, lbns %jd-%jd\n\told:", (uintmax_t)ip->i_number, (intmax_t)start_lbn, (intmax_t)end_lbn); #endif blkno = newblk; for (bap = &sbap[soff], i = 0; i < len; i++, blkno += fs->fs_frag) { if (i == ssize) { bap = ebap; soff = -i; } #ifdef INVARIANTS if (!ffs_checkblk(ip, dbtofsb(fs, buflist->bs_children[i]->b_blkno), fs->fs_bsize)) panic("ffs_reallocblks: unallocated block 2"); if (dbtofsb(fs, buflist->bs_children[i]->b_blkno) != *bap) panic("ffs_reallocblks: alloc mismatch"); #endif #ifdef DIAGNOSTIC if (prtrealloc) printf(" %jd,", (intmax_t)*bap); #endif if (DOINGSOFTDEP(vp)) { if (sbap == &ip->i_din2->di_db[0] && i < ssize) softdep_setup_allocdirect(ip, start_lbn + i, blkno, *bap, fs->fs_bsize, fs->fs_bsize, buflist->bs_children[i]); else softdep_setup_allocindir_page(ip, start_lbn + i, i < ssize ? sbp : ebp, soff + i, blkno, *bap, buflist->bs_children[i]); } *bap++ = blkno; } /* * Next we must write out the modified inode and indirect blocks. * For strict correctness, the writes should be synchronous since * the old block values may have been written to disk. In practise * they are almost never written, but if we are concerned about * strict correctness, the `doasyncfree' flag should be set to zero. * * The test on `doasyncfree' should be changed to test a flag * that shows whether the associated buffers and inodes have * been written. The flag should be set when the cluster is * started and cleared whenever the buffer or inode is flushed. * We can then check below to see if it is set, and do the * synchronous write only when it has been cleared. */ if (sbap != &ip->i_din2->di_db[0]) { if (doasyncfree) bdwrite(sbp); else bwrite(sbp); } else { UFS_INODE_SET_FLAG(ip, IN_CHANGE | IN_UPDATE); if (!doasyncfree) ffs_update(vp, 1); } if (ssize < len) { if (doasyncfree) bdwrite(ebp); else bwrite(ebp); } /* * Last, free the old blocks and assign the new blocks to the buffers. */ #ifdef DIAGNOSTIC if (prtrealloc) printf("\n\tnew:"); #endif for (blkno = newblk, i = 0; i < len; i++, blkno += fs->fs_frag) { bp = buflist->bs_children[i]; if (!DOINGSOFTDEP(vp)) /* * The usual case is that a set of N-contiguous blocks * that was just allocated has been replaced with a * set of N+1-contiguous blocks. If they are marked as * B_DELWRI, the current contents have not been written * to disk. It is possible that the blocks were written * earlier, but very uncommon. If the blocks have never * been written, there is no need to send a BIO_DELETE * for them when they are freed. The gain from avoiding * the TRIMs for the common case of unwritten blocks * far exceeds the cost of the write amplification for * the uncommon case of failing to send a TRIM for the * blocks that had been written. */ ffs_blkfree(ump, fs, ump->um_devvp, dbtofsb(fs, bp->b_blkno), fs->fs_bsize, ip->i_number, vp->v_type, NULL, (bp->b_flags & B_DELWRI) != 0 ? NOTRIM_KEY : SINGLETON_KEY); bp->b_blkno = fsbtodb(fs, blkno); #ifdef INVARIANTS if (!ffs_checkblk(ip, dbtofsb(fs, bp->b_blkno), fs->fs_bsize)) panic("ffs_reallocblks: unallocated block 3"); #endif #ifdef DIAGNOSTIC if (prtrealloc) printf(" %jd,", (intmax_t)blkno); #endif } #ifdef DIAGNOSTIC if (prtrealloc) { prtrealloc--; printf("\n"); } #endif return (0); fail: if (ssize < len) brelse(ebp); if (sbap != &ip->i_din2->di_db[0]) brelse(sbp); return (ENOSPC); } /* * Allocate an inode in the filesystem. * * If allocating a directory, use ffs_dirpref to select the inode. * If allocating in a directory, the following hierarchy is followed: * 1) allocate the preferred inode. * 2) allocate an inode in the same cylinder group. * 3) quadradically rehash into other cylinder groups, until an * available inode is located. * If no inode preference is given the following hierarchy is used * to allocate an inode: * 1) allocate an inode in cylinder group 0. * 2) quadradically rehash into other cylinder groups, until an * available inode is located. */ int ffs_valloc(pvp, mode, cred, vpp) struct vnode *pvp; int mode; struct ucred *cred; struct vnode **vpp; { struct inode *pip; struct fs *fs; struct inode *ip; struct timespec ts; struct ufsmount *ump; ino_t ino, ipref; u_int cg; int error, error1, reclaimed; *vpp = NULL; pip = VTOI(pvp); ump = ITOUMP(pip); fs = ump->um_fs; UFS_LOCK(ump); reclaimed = 0; retry: if (fs->fs_cstotal.cs_nifree == 0) goto noinodes; if ((mode & IFMT) == IFDIR) ipref = ffs_dirpref(pip); else ipref = pip->i_number; if (ipref >= fs->fs_ncg * fs->fs_ipg) ipref = 0; cg = ino_to_cg(fs, ipref); /* * Track number of dirs created one after another * in a same cg without intervening by files. */ if ((mode & IFMT) == IFDIR) { if (fs->fs_contigdirs[cg] < 255) fs->fs_contigdirs[cg]++; } else { if (fs->fs_contigdirs[cg] > 0) fs->fs_contigdirs[cg]--; } ino = (ino_t)ffs_hashalloc(pip, cg, ipref, mode, 0, (allocfcn_t *)ffs_nodealloccg); if (ino == 0) goto noinodes; /* * Get rid of the cached old vnode, force allocation of a new vnode * for this inode. */ error = ffs_vgetf(pvp->v_mount, ino, LK_EXCLUSIVE, vpp, FFSV_REPLACE); if (error) { error1 = ffs_vgetf(pvp->v_mount, ino, LK_EXCLUSIVE, vpp, FFSV_FORCEINSMQ | FFSV_REPLACE); ffs_vfree(pvp, ino, mode); if (error1 == 0) { ip = VTOI(*vpp); if (ip->i_mode) goto dup_alloc; UFS_INODE_SET_FLAG(ip, IN_MODIFIED); vput(*vpp); } return (error); } ip = VTOI(*vpp); if (ip->i_mode) { dup_alloc: printf("mode = 0%o, inum = %ju, fs = %s\n", ip->i_mode, (uintmax_t)ip->i_number, fs->fs_fsmnt); panic("ffs_valloc: dup alloc"); } if (DIP(ip, i_blocks) && (fs->fs_flags & FS_UNCLEAN) == 0) { /* XXX */ printf("free inode %s/%lu had %ld blocks\n", fs->fs_fsmnt, (u_long)ino, (long)DIP(ip, i_blocks)); DIP_SET(ip, i_blocks, 0); } ip->i_flags = 0; DIP_SET(ip, i_flags, 0); /* * Set up a new generation number for this inode. */ while (ip->i_gen == 0 || ++ip->i_gen == 0) ip->i_gen = arc4random(); DIP_SET(ip, i_gen, ip->i_gen); if (fs->fs_magic == FS_UFS2_MAGIC) { vfs_timestamp(&ts); ip->i_din2->di_birthtime = ts.tv_sec; ip->i_din2->di_birthnsec = ts.tv_nsec; } ip->i_flag = 0; (*vpp)->v_vflag = 0; (*vpp)->v_type = VNON; if (fs->fs_magic == FS_UFS2_MAGIC) { (*vpp)->v_op = &ffs_vnodeops2; UFS_INODE_SET_FLAG(ip, IN_UFS2); } else { (*vpp)->v_op = &ffs_vnodeops1; } return (0); noinodes: if (reclaimed == 0) { reclaimed = 1; softdep_request_cleanup(fs, pvp, cred, FLUSH_INODES_WAIT); goto retry; } if (ppsratecheck(&ump->um_last_fullmsg, &ump->um_secs_fullmsg, 1)) { UFS_UNLOCK(ump); ffs_fserr(fs, pip->i_number, "out of inodes"); uprintf("\n%s: create/symlink failed, no inodes free\n", fs->fs_fsmnt); } else { UFS_UNLOCK(ump); } return (ENOSPC); } /* * Find a cylinder group to place a directory. * * The policy implemented by this algorithm is to allocate a * directory inode in the same cylinder group as its parent * directory, but also to reserve space for its files inodes * and data. Restrict the number of directories which may be * allocated one after another in the same cylinder group * without intervening allocation of files. * * If we allocate a first level directory then force allocation * in another cylinder group. */ static ino_t ffs_dirpref(pip) struct inode *pip; { struct fs *fs; int cg, prefcg, dirsize, cgsize; u_int avgifree, avgbfree, avgndir, curdirsize; u_int minifree, minbfree, maxndir; u_int mincg, minndir; u_int maxcontigdirs; mtx_assert(UFS_MTX(ITOUMP(pip)), MA_OWNED); fs = ITOFS(pip); avgifree = fs->fs_cstotal.cs_nifree / fs->fs_ncg; avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg; avgndir = fs->fs_cstotal.cs_ndir / fs->fs_ncg; /* * Force allocation in another cg if creating a first level dir. */ ASSERT_VOP_LOCKED(ITOV(pip), "ffs_dirpref"); if (ITOV(pip)->v_vflag & VV_ROOT) { prefcg = arc4random() % fs->fs_ncg; mincg = prefcg; minndir = fs->fs_ipg; for (cg = prefcg; cg < fs->fs_ncg; cg++) if (fs->fs_cs(fs, cg).cs_ndir < minndir && fs->fs_cs(fs, cg).cs_nifree >= avgifree && fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { mincg = cg; minndir = fs->fs_cs(fs, cg).cs_ndir; } for (cg = 0; cg < prefcg; cg++) if (fs->fs_cs(fs, cg).cs_ndir < minndir && fs->fs_cs(fs, cg).cs_nifree >= avgifree && fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { mincg = cg; minndir = fs->fs_cs(fs, cg).cs_ndir; } return ((ino_t)(fs->fs_ipg * mincg)); } /* * Count various limits which used for * optimal allocation of a directory inode. */ maxndir = min(avgndir + fs->fs_ipg / 16, fs->fs_ipg); minifree = avgifree - avgifree / 4; if (minifree < 1) minifree = 1; minbfree = avgbfree - avgbfree / 4; if (minbfree < 1) minbfree = 1; cgsize = fs->fs_fsize * fs->fs_fpg; dirsize = fs->fs_avgfilesize * fs->fs_avgfpdir; curdirsize = avgndir ? (cgsize - avgbfree * fs->fs_bsize) / avgndir : 0; if (dirsize < curdirsize) dirsize = curdirsize; if (dirsize <= 0) maxcontigdirs = 0; /* dirsize overflowed */ else maxcontigdirs = min((avgbfree * fs->fs_bsize) / dirsize, 255); if (fs->fs_avgfpdir > 0) maxcontigdirs = min(maxcontigdirs, fs->fs_ipg / fs->fs_avgfpdir); if (maxcontigdirs == 0) maxcontigdirs = 1; /* * Limit number of dirs in one cg and reserve space for * regular files, but only if we have no deficit in * inodes or space. * * We are trying to find a suitable cylinder group nearby * our preferred cylinder group to place a new directory. * We scan from our preferred cylinder group forward looking * for a cylinder group that meets our criterion. If we get * to the final cylinder group and do not find anything, * we start scanning forwards from the beginning of the * filesystem. While it might seem sensible to start scanning * backwards or even to alternate looking forward and backward, * this approach fails badly when the filesystem is nearly full. * Specifically, we first search all the areas that have no space * and finally try the one preceding that. We repeat this on * every request and in the case of the final block end up * searching the entire filesystem. By jumping to the front * of the filesystem, our future forward searches always look * in new cylinder groups so finds every possible block after * one pass over the filesystem. */ prefcg = ino_to_cg(fs, pip->i_number); for (cg = prefcg; cg < fs->fs_ncg; cg++) if (fs->fs_cs(fs, cg).cs_ndir < maxndir && fs->fs_cs(fs, cg).cs_nifree >= minifree && fs->fs_cs(fs, cg).cs_nbfree >= minbfree) { if (fs->fs_contigdirs[cg] < maxcontigdirs) return ((ino_t)(fs->fs_ipg * cg)); } for (cg = 0; cg < prefcg; cg++) if (fs->fs_cs(fs, cg).cs_ndir < maxndir && fs->fs_cs(fs, cg).cs_nifree >= minifree && fs->fs_cs(fs, cg).cs_nbfree >= minbfree) { if (fs->fs_contigdirs[cg] < maxcontigdirs) return ((ino_t)(fs->fs_ipg * cg)); } /* * This is a backstop when we have deficit in space. */ for (cg = prefcg; cg < fs->fs_ncg; cg++) if (fs->fs_cs(fs, cg).cs_nifree >= avgifree) return ((ino_t)(fs->fs_ipg * cg)); for (cg = 0; cg < prefcg; cg++) if (fs->fs_cs(fs, cg).cs_nifree >= avgifree) break; return ((ino_t)(fs->fs_ipg * cg)); } /* * Select the desired position for the next block in a file. The file is * logically divided into sections. The first section is composed of the * direct blocks and the next fs_maxbpg blocks. Each additional section * contains fs_maxbpg blocks. * * If no blocks have been allocated in the first section, the policy is to * request a block in the same cylinder group as the inode that describes * the file. The first indirect is allocated immediately following the last * direct block and the data blocks for the first indirect immediately * follow it. * * If no blocks have been allocated in any other section, the indirect * block(s) are allocated in the same cylinder group as its inode in an * area reserved immediately following the inode blocks. The policy for * the data blocks is to place them in a cylinder group with a greater than * average number of free blocks. An appropriate cylinder group is found * by using a rotor that sweeps the cylinder groups. When a new group of * blocks is needed, the sweep begins in the cylinder group following the * cylinder group from which the previous allocation was made. The sweep * continues until a cylinder group with greater than the average number * of free blocks is found. If the allocation is for the first block in an * indirect block or the previous block is a hole, then the information on * the previous allocation is unavailable; here a best guess is made based * on the logical block number being allocated. * * If a section is already partially allocated, the policy is to * allocate blocks contiguously within the section if possible. */ ufs2_daddr_t ffs_blkpref_ufs1(ip, lbn, indx, bap) struct inode *ip; ufs_lbn_t lbn; int indx; ufs1_daddr_t *bap; { struct fs *fs; u_int cg, inocg; u_int avgbfree, startcg; ufs2_daddr_t pref, prevbn; KASSERT(indx <= 0 || bap != NULL, ("need non-NULL bap")); mtx_assert(UFS_MTX(ITOUMP(ip)), MA_OWNED); fs = ITOFS(ip); /* * Allocation of indirect blocks is indicated by passing negative * values in indx: -1 for single indirect, -2 for double indirect, * -3 for triple indirect. As noted below, we attempt to allocate * the first indirect inline with the file data. For all later * indirect blocks, the data is often allocated in other cylinder * groups. However to speed random file access and to speed up * fsck, the filesystem reserves the first fs_metaspace blocks * (typically half of fs_minfree) of the data area of each cylinder * group to hold these later indirect blocks. */ inocg = ino_to_cg(fs, ip->i_number); if (indx < 0) { /* * Our preference for indirect blocks is the zone at the * beginning of the inode's cylinder group data area that * we try to reserve for indirect blocks. */ pref = cgmeta(fs, inocg); /* * If we are allocating the first indirect block, try to * place it immediately following the last direct block. */ if (indx == -1 && lbn < UFS_NDADDR + NINDIR(fs) && ip->i_din1->di_db[UFS_NDADDR - 1] != 0) pref = ip->i_din1->di_db[UFS_NDADDR - 1] + fs->fs_frag; return (pref); } /* * If we are allocating the first data block in the first indirect * block and the indirect has been allocated in the data block area, * try to place it immediately following the indirect block. */ if (lbn == UFS_NDADDR) { pref = ip->i_din1->di_ib[0]; if (pref != 0 && pref >= cgdata(fs, inocg) && pref < cgbase(fs, inocg + 1)) return (pref + fs->fs_frag); } /* * If we are at the beginning of a file, or we have already allocated * the maximum number of blocks per cylinder group, or we do not * have a block allocated immediately preceding us, then we need * to decide where to start allocating new blocks. */ if (indx == 0) { prevbn = 0; } else { prevbn = bap[indx - 1]; if (UFS_CHECK_BLKNO(ITOVFS(ip), ip->i_number, prevbn, fs->fs_bsize) != 0) prevbn = 0; } if (indx % fs->fs_maxbpg == 0 || prevbn == 0) { /* * If we are allocating a directory data block, we want * to place it in the metadata area. */ if ((ip->i_mode & IFMT) == IFDIR) return (cgmeta(fs, inocg)); /* * Until we fill all the direct and all the first indirect's * blocks, we try to allocate in the data area of the inode's * cylinder group. */ if (lbn < UFS_NDADDR + NINDIR(fs)) return (cgdata(fs, inocg)); /* * Find a cylinder with greater than average number of * unused data blocks. */ if (indx == 0 || prevbn == 0) startcg = inocg + lbn / fs->fs_maxbpg; else startcg = dtog(fs, prevbn) + 1; startcg %= fs->fs_ncg; avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg; for (cg = startcg; cg < fs->fs_ncg; cg++) if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { fs->fs_cgrotor = cg; return (cgdata(fs, cg)); } for (cg = 0; cg <= startcg; cg++) if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { fs->fs_cgrotor = cg; return (cgdata(fs, cg)); } return (0); } /* * Otherwise, we just always try to lay things out contiguously. */ return (prevbn + fs->fs_frag); } /* * Same as above, but for UFS2 */ ufs2_daddr_t ffs_blkpref_ufs2(ip, lbn, indx, bap) struct inode *ip; ufs_lbn_t lbn; int indx; ufs2_daddr_t *bap; { struct fs *fs; u_int cg, inocg; u_int avgbfree, startcg; ufs2_daddr_t pref, prevbn; KASSERT(indx <= 0 || bap != NULL, ("need non-NULL bap")); mtx_assert(UFS_MTX(ITOUMP(ip)), MA_OWNED); fs = ITOFS(ip); /* * Allocation of indirect blocks is indicated by passing negative * values in indx: -1 for single indirect, -2 for double indirect, * -3 for triple indirect. As noted below, we attempt to allocate * the first indirect inline with the file data. For all later * indirect blocks, the data is often allocated in other cylinder * groups. However to speed random file access and to speed up * fsck, the filesystem reserves the first fs_metaspace blocks * (typically half of fs_minfree) of the data area of each cylinder * group to hold these later indirect blocks. */ inocg = ino_to_cg(fs, ip->i_number); if (indx < 0) { /* * Our preference for indirect blocks is the zone at the * beginning of the inode's cylinder group data area that * we try to reserve for indirect blocks. */ pref = cgmeta(fs, inocg); /* * If we are allocating the first indirect block, try to * place it immediately following the last direct block. */ if (indx == -1 && lbn < UFS_NDADDR + NINDIR(fs) && ip->i_din2->di_db[UFS_NDADDR - 1] != 0) pref = ip->i_din2->di_db[UFS_NDADDR - 1] + fs->fs_frag; return (pref); } /* * If we are allocating the first data block in the first indirect * block and the indirect has been allocated in the data block area, * try to place it immediately following the indirect block. */ if (lbn == UFS_NDADDR) { pref = ip->i_din2->di_ib[0]; if (pref != 0 && pref >= cgdata(fs, inocg) && pref < cgbase(fs, inocg + 1)) return (pref + fs->fs_frag); } /* * If we are at the beginning of a file, or we have already allocated * the maximum number of blocks per cylinder group, or we do not * have a block allocated immediately preceding us, then we need * to decide where to start allocating new blocks. */ if (indx == 0) { prevbn = 0; } else { prevbn = bap[indx - 1]; if (UFS_CHECK_BLKNO(ITOVFS(ip), ip->i_number, prevbn, fs->fs_bsize) != 0) prevbn = 0; } if (indx % fs->fs_maxbpg == 0 || prevbn == 0) { /* * If we are allocating a directory data block, we want * to place it in the metadata area. */ if ((ip->i_mode & IFMT) == IFDIR) return (cgmeta(fs, inocg)); /* * Until we fill all the direct and all the first indirect's * blocks, we try to allocate in the data area of the inode's * cylinder group. */ if (lbn < UFS_NDADDR + NINDIR(fs)) return (cgdata(fs, inocg)); /* * Find a cylinder with greater than average number of * unused data blocks. */ if (indx == 0 || prevbn == 0) startcg = inocg + lbn / fs->fs_maxbpg; else startcg = dtog(fs, prevbn) + 1; startcg %= fs->fs_ncg; avgbfree = fs->fs_cstotal.cs_nbfree / fs->fs_ncg; for (cg = startcg; cg < fs->fs_ncg; cg++) if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { fs->fs_cgrotor = cg; return (cgdata(fs, cg)); } for (cg = 0; cg <= startcg; cg++) if (fs->fs_cs(fs, cg).cs_nbfree >= avgbfree) { fs->fs_cgrotor = cg; return (cgdata(fs, cg)); } return (0); } /* * Otherwise, we just always try to lay things out contiguously. */ return (prevbn + fs->fs_frag); } /* * Implement the cylinder overflow algorithm. * * The policy implemented by this algorithm is: * 1) allocate the block in its requested cylinder group. * 2) quadradically rehash on the cylinder group number. * 3) brute force search for a free block. * * Must be called with the UFS lock held. Will release the lock on success * and return with it held on failure. */ /*VARARGS5*/ static ufs2_daddr_t ffs_hashalloc(ip, cg, pref, size, rsize, allocator) struct inode *ip; u_int cg; ufs2_daddr_t pref; int size; /* Search size for data blocks, mode for inodes */ int rsize; /* Real allocated size. */ allocfcn_t *allocator; { struct fs *fs; ufs2_daddr_t result; u_int i, icg = cg; mtx_assert(UFS_MTX(ITOUMP(ip)), MA_OWNED); #ifdef INVARIANTS if (ITOV(ip)->v_mount->mnt_kern_flag & MNTK_SUSPENDED) panic("ffs_hashalloc: allocation on suspended filesystem"); #endif fs = ITOFS(ip); /* * 1: preferred cylinder group */ result = (*allocator)(ip, cg, pref, size, rsize); if (result) return (result); /* * 2: quadratic rehash */ for (i = 1; i < fs->fs_ncg; i *= 2) { cg += i; if (cg >= fs->fs_ncg) cg -= fs->fs_ncg; result = (*allocator)(ip, cg, 0, size, rsize); if (result) return (result); } /* * 3: brute force search * Note that we start at i == 2, since 0 was checked initially, * and 1 is always checked in the quadratic rehash. */ cg = (icg + 2) % fs->fs_ncg; for (i = 2; i < fs->fs_ncg; i++) { result = (*allocator)(ip, cg, 0, size, rsize); if (result) return (result); cg++; if (cg == fs->fs_ncg) cg = 0; } return (0); } /* * Determine whether a fragment can be extended. * * Check to see if the necessary fragments are available, and * if they are, allocate them. */ static ufs2_daddr_t ffs_fragextend(ip, cg, bprev, osize, nsize) struct inode *ip; u_int cg; ufs2_daddr_t bprev; int osize, nsize; { struct fs *fs; struct cg *cgp; struct buf *bp; struct ufsmount *ump; int nffree; long bno; int frags, bbase; int i, error; u_int8_t *blksfree; ump = ITOUMP(ip); fs = ump->um_fs; if (fs->fs_cs(fs, cg).cs_nffree < numfrags(fs, nsize - osize)) return (0); frags = numfrags(fs, nsize); bbase = fragnum(fs, bprev); if (bbase > fragnum(fs, (bprev + frags - 1))) { /* cannot extend across a block boundary */ return (0); } UFS_UNLOCK(ump); if ((error = ffs_getcg(fs, ump->um_devvp, cg, 0, &bp, &cgp)) != 0) goto fail; bno = dtogd(fs, bprev); blksfree = cg_blksfree(cgp); for (i = numfrags(fs, osize); i < frags; i++) if (isclr(blksfree, bno + i)) goto fail; /* * the current fragment can be extended * deduct the count on fragment being extended into * increase the count on the remaining fragment (if any) * allocate the extended piece */ for (i = frags; i < fs->fs_frag - bbase; i++) if (isclr(blksfree, bno + i)) break; cgp->cg_frsum[i - numfrags(fs, osize)]--; if (i != frags) cgp->cg_frsum[i - frags]++; for (i = numfrags(fs, osize), nffree = 0; i < frags; i++) { clrbit(blksfree, bno + i); cgp->cg_cs.cs_nffree--; nffree++; } UFS_LOCK(ump); fs->fs_cstotal.cs_nffree -= nffree; fs->fs_cs(fs, cg).cs_nffree -= nffree; fs->fs_fmod = 1; ACTIVECLEAR(fs, cg); UFS_UNLOCK(ump); if (DOINGSOFTDEP(ITOV(ip))) softdep_setup_blkmapdep(bp, UFSTOVFS(ump), bprev, frags, numfrags(fs, osize)); bdwrite(bp); return (bprev); fail: brelse(bp); UFS_LOCK(ump); return (0); } /* * Determine whether a block can be allocated. * * Check to see if a block of the appropriate size is available, * and if it is, allocate it. */ static ufs2_daddr_t ffs_alloccg(ip, cg, bpref, size, rsize) struct inode *ip; u_int cg; ufs2_daddr_t bpref; int size; int rsize; { struct fs *fs; struct cg *cgp; struct buf *bp; struct ufsmount *ump; ufs1_daddr_t bno; ufs2_daddr_t blkno; int i, allocsiz, error, frags; u_int8_t *blksfree; ump = ITOUMP(ip); fs = ump->um_fs; if (fs->fs_cs(fs, cg).cs_nbfree == 0 && size == fs->fs_bsize) return (0); UFS_UNLOCK(ump); if ((error = ffs_getcg(fs, ump->um_devvp, cg, 0, &bp, &cgp)) != 0 || (cgp->cg_cs.cs_nbfree == 0 && size == fs->fs_bsize)) goto fail; if (size == fs->fs_bsize) { UFS_LOCK(ump); blkno = ffs_alloccgblk(ip, bp, bpref, rsize); ACTIVECLEAR(fs, cg); UFS_UNLOCK(ump); bdwrite(bp); return (blkno); } /* * check to see if any fragments are already available * allocsiz is the size which will be allocated, hacking * it down to a smaller size if necessary */ blksfree = cg_blksfree(cgp); frags = numfrags(fs, size); for (allocsiz = frags; allocsiz < fs->fs_frag; allocsiz++) if (cgp->cg_frsum[allocsiz] != 0) break; if (allocsiz == fs->fs_frag) { /* * no fragments were available, so a block will be * allocated, and hacked up */ if (cgp->cg_cs.cs_nbfree == 0) goto fail; UFS_LOCK(ump); blkno = ffs_alloccgblk(ip, bp, bpref, rsize); ACTIVECLEAR(fs, cg); UFS_UNLOCK(ump); bdwrite(bp); return (blkno); } KASSERT(size == rsize, ("ffs_alloccg: size(%d) != rsize(%d)", size, rsize)); bno = ffs_mapsearch(fs, cgp, bpref, allocsiz); if (bno < 0) goto fail; for (i = 0; i < frags; i++) clrbit(blksfree, bno + i); cgp->cg_cs.cs_nffree -= frags; cgp->cg_frsum[allocsiz]--; if (frags != allocsiz) cgp->cg_frsum[allocsiz - frags]++; UFS_LOCK(ump); fs->fs_cstotal.cs_nffree -= frags; fs->fs_cs(fs, cg).cs_nffree -= frags; fs->fs_fmod = 1; blkno = cgbase(fs, cg) + bno; ACTIVECLEAR(fs, cg); UFS_UNLOCK(ump); if (DOINGSOFTDEP(ITOV(ip))) softdep_setup_blkmapdep(bp, UFSTOVFS(ump), blkno, frags, 0); bdwrite(bp); return (blkno); fail: brelse(bp); UFS_LOCK(ump); return (0); } /* * Allocate a block in a cylinder group. * * This algorithm implements the following policy: * 1) allocate the requested block. * 2) allocate a rotationally optimal block in the same cylinder. * 3) allocate the next available block on the block rotor for the * specified cylinder group. * Note that this routine only allocates fs_bsize blocks; these * blocks may be fragmented by the routine that allocates them. */ static ufs2_daddr_t ffs_alloccgblk(ip, bp, bpref, size) struct inode *ip; struct buf *bp; ufs2_daddr_t bpref; int size; { struct fs *fs; struct cg *cgp; struct ufsmount *ump; ufs1_daddr_t bno; ufs2_daddr_t blkno; u_int8_t *blksfree; int i, cgbpref; ump = ITOUMP(ip); fs = ump->um_fs; mtx_assert(UFS_MTX(ump), MA_OWNED); cgp = (struct cg *)bp->b_data; blksfree = cg_blksfree(cgp); if (bpref == 0) { bpref = cgbase(fs, cgp->cg_cgx) + cgp->cg_rotor + fs->fs_frag; } else if ((cgbpref = dtog(fs, bpref)) != cgp->cg_cgx) { /* map bpref to correct zone in this cg */ if (bpref < cgdata(fs, cgbpref)) bpref = cgmeta(fs, cgp->cg_cgx); else bpref = cgdata(fs, cgp->cg_cgx); } /* * if the requested block is available, use it */ bno = dtogd(fs, blknum(fs, bpref)); if (ffs_isblock(fs, blksfree, fragstoblks(fs, bno))) goto gotit; /* * Take the next available block in this cylinder group. */ bno = ffs_mapsearch(fs, cgp, bpref, (int)fs->fs_frag); if (bno < 0) return (0); /* Update cg_rotor only if allocated from the data zone */ if (bno >= dtogd(fs, cgdata(fs, cgp->cg_cgx))) cgp->cg_rotor = bno; gotit: blkno = fragstoblks(fs, bno); ffs_clrblock(fs, blksfree, (long)blkno); ffs_clusteracct(fs, cgp, blkno, -1); cgp->cg_cs.cs_nbfree--; fs->fs_cstotal.cs_nbfree--; fs->fs_cs(fs, cgp->cg_cgx).cs_nbfree--; fs->fs_fmod = 1; blkno = cgbase(fs, cgp->cg_cgx) + bno; /* * If the caller didn't want the whole block free the frags here. */ size = numfrags(fs, size); if (size != fs->fs_frag) { bno = dtogd(fs, blkno); for (i = size; i < fs->fs_frag; i++) setbit(blksfree, bno + i); i = fs->fs_frag - size; cgp->cg_cs.cs_nffree += i; fs->fs_cstotal.cs_nffree += i; fs->fs_cs(fs, cgp->cg_cgx).cs_nffree += i; fs->fs_fmod = 1; cgp->cg_frsum[i]++; } /* XXX Fixme. */ UFS_UNLOCK(ump); if (DOINGSOFTDEP(ITOV(ip))) softdep_setup_blkmapdep(bp, UFSTOVFS(ump), blkno, size, 0); UFS_LOCK(ump); return (blkno); } /* * Determine whether a cluster can be allocated. * * We do not currently check for optimal rotational layout if there * are multiple choices in the same cylinder group. Instead we just * take the first one that we find following bpref. */ static ufs2_daddr_t ffs_clusteralloc(ip, cg, bpref, len) struct inode *ip; u_int cg; ufs2_daddr_t bpref; int len; { struct fs *fs; struct cg *cgp; struct buf *bp; struct ufsmount *ump; int i, run, bit, map, got, error; ufs2_daddr_t bno; u_char *mapp; int32_t *lp; u_int8_t *blksfree; ump = ITOUMP(ip); fs = ump->um_fs; if (fs->fs_maxcluster[cg] < len) return (0); UFS_UNLOCK(ump); if ((error = ffs_getcg(fs, ump->um_devvp, cg, 0, &bp, &cgp)) != 0) { UFS_LOCK(ump); return (0); } /* * Check to see if a cluster of the needed size (or bigger) is * available in this cylinder group. */ lp = &cg_clustersum(cgp)[len]; for (i = len; i <= fs->fs_contigsumsize; i++) if (*lp++ > 0) break; if (i > fs->fs_contigsumsize) { /* * This is the first time looking for a cluster in this * cylinder group. Update the cluster summary information * to reflect the true maximum sized cluster so that * future cluster allocation requests can avoid reading * the cylinder group map only to find no clusters. */ lp = &cg_clustersum(cgp)[len - 1]; for (i = len - 1; i > 0; i--) if (*lp-- > 0) break; UFS_LOCK(ump); fs->fs_maxcluster[cg] = i; brelse(bp); return (0); } /* * Search the cluster map to find a big enough cluster. * We take the first one that we find, even if it is larger * than we need as we prefer to get one close to the previous * block allocation. We do not search before the current * preference point as we do not want to allocate a block * that is allocated before the previous one (as we will * then have to wait for another pass of the elevator * algorithm before it will be read). We prefer to fail and * be recalled to try an allocation in the next cylinder group. */ if (dtog(fs, bpref) != cg) bpref = cgdata(fs, cg); else bpref = blknum(fs, bpref); bpref = fragstoblks(fs, dtogd(fs, bpref)); mapp = &cg_clustersfree(cgp)[bpref / NBBY]; map = *mapp++; bit = 1 << (bpref % NBBY); for (run = 0, got = bpref; got < cgp->cg_nclusterblks; got++) { if ((map & bit) == 0) { run = 0; } else { run++; if (run == len) break; } if ((got & (NBBY - 1)) != (NBBY - 1)) { bit <<= 1; } else { map = *mapp++; bit = 1; } } if (got >= cgp->cg_nclusterblks) { UFS_LOCK(ump); brelse(bp); return (0); } /* * Allocate the cluster that we have found. */ blksfree = cg_blksfree(cgp); for (i = 1; i <= len; i++) if (!ffs_isblock(fs, blksfree, got - run + i)) panic("ffs_clusteralloc: map mismatch"); bno = cgbase(fs, cg) + blkstofrags(fs, got - run + 1); if (dtog(fs, bno) != cg) panic("ffs_clusteralloc: allocated out of group"); len = blkstofrags(fs, len); UFS_LOCK(ump); for (i = 0; i < len; i += fs->fs_frag) if (ffs_alloccgblk(ip, bp, bno + i, fs->fs_bsize) != bno + i) panic("ffs_clusteralloc: lost block"); ACTIVECLEAR(fs, cg); UFS_UNLOCK(ump); bdwrite(bp); return (bno); } static inline struct buf * getinobuf(struct inode *ip, u_int cg, u_int32_t cginoblk, int gbflags) { struct fs *fs; fs = ITOFS(ip); return (getblk(ITODEVVP(ip), fsbtodb(fs, ino_to_fsba(fs, cg * fs->fs_ipg + cginoblk)), (int)fs->fs_bsize, 0, 0, gbflags)); } /* * Synchronous inode initialization is needed only when barrier writes do not * work as advertised, and will impose a heavy cost on file creation in a newly * created filesystem. */ static int doasyncinodeinit = 1; SYSCTL_INT(_vfs_ffs, OID_AUTO, doasyncinodeinit, CTLFLAG_RWTUN, &doasyncinodeinit, 0, "Perform inode block initialization using asynchronous writes"); /* * Determine whether an inode can be allocated. * * Check to see if an inode is available, and if it is, * allocate it using the following policy: * 1) allocate the requested inode. * 2) allocate the next available inode after the requested * inode in the specified cylinder group. */ static ufs2_daddr_t ffs_nodealloccg(ip, cg, ipref, mode, unused) struct inode *ip; u_int cg; ufs2_daddr_t ipref; int mode; int unused; { struct fs *fs; struct cg *cgp; struct buf *bp, *ibp; struct ufsmount *ump; u_int8_t *inosused, *loc; struct ufs2_dinode *dp2; int error, start, len, i; u_int32_t old_initediblk; ump = ITOUMP(ip); fs = ump->um_fs; check_nifree: if (fs->fs_cs(fs, cg).cs_nifree == 0) return (0); UFS_UNLOCK(ump); if ((error = ffs_getcg(fs, ump->um_devvp, cg, 0, &bp, &cgp)) != 0) { UFS_LOCK(ump); return (0); } restart: if (cgp->cg_cs.cs_nifree == 0) { brelse(bp); UFS_LOCK(ump); return (0); } inosused = cg_inosused(cgp); if (ipref) { ipref %= fs->fs_ipg; if (isclr(inosused, ipref)) goto gotit; } start = cgp->cg_irotor / NBBY; len = howmany(fs->fs_ipg - cgp->cg_irotor, NBBY); loc = memcchr(&inosused[start], 0xff, len); if (loc == NULL) { len = start + 1; start = 0; loc = memcchr(&inosused[start], 0xff, len); if (loc == NULL) { printf("cg = %d, irotor = %ld, fs = %s\n", cg, (long)cgp->cg_irotor, fs->fs_fsmnt); panic("ffs_nodealloccg: map corrupted"); /* NOTREACHED */ } } ipref = (loc - inosused) * NBBY + ffs(~*loc) - 1; gotit: /* * Check to see if we need to initialize more inodes. */ if (fs->fs_magic == FS_UFS2_MAGIC && ipref + INOPB(fs) > cgp->cg_initediblk && cgp->cg_initediblk < cgp->cg_niblk) { old_initediblk = cgp->cg_initediblk; /* * Free the cylinder group lock before writing the * initialized inode block. Entering the * babarrierwrite() with the cylinder group lock * causes lock order violation between the lock and * snaplk. * * Another thread can decide to initialize the same * inode block, but whichever thread first gets the * cylinder group lock after writing the newly * allocated inode block will update it and the other * will realize that it has lost and leave the * cylinder group unchanged. */ ibp = getinobuf(ip, cg, old_initediblk, GB_LOCK_NOWAIT); brelse(bp); if (ibp == NULL) { /* * The inode block buffer is already owned by * another thread, which must initialize it. * Wait on the buffer to allow another thread * to finish the updates, with dropped cg * buffer lock, then retry. */ ibp = getinobuf(ip, cg, old_initediblk, 0); brelse(ibp); UFS_LOCK(ump); goto check_nifree; } bzero(ibp->b_data, (int)fs->fs_bsize); dp2 = (struct ufs2_dinode *)(ibp->b_data); for (i = 0; i < INOPB(fs); i++) { while (dp2->di_gen == 0) dp2->di_gen = arc4random(); dp2++; } /* * Rather than adding a soft updates dependency to ensure * that the new inode block is written before it is claimed * by the cylinder group map, we just do a barrier write * here. The barrier write will ensure that the inode block * gets written before the updated cylinder group map can be * written. The barrier write should only slow down bulk * loading of newly created filesystems. */ if (doasyncinodeinit) babarrierwrite(ibp); else bwrite(ibp); /* * After the inode block is written, try to update the * cg initediblk pointer. If another thread beat us * to it, then leave it unchanged as the other thread * has already set it correctly. */ error = ffs_getcg(fs, ump->um_devvp, cg, 0, &bp, &cgp); UFS_LOCK(ump); ACTIVECLEAR(fs, cg); UFS_UNLOCK(ump); if (error != 0) return (error); if (cgp->cg_initediblk == old_initediblk) cgp->cg_initediblk += INOPB(fs); goto restart; } cgp->cg_irotor = ipref; UFS_LOCK(ump); ACTIVECLEAR(fs, cg); setbit(inosused, ipref); cgp->cg_cs.cs_nifree--; fs->fs_cstotal.cs_nifree--; fs->fs_cs(fs, cg).cs_nifree--; fs->fs_fmod = 1; if ((mode & IFMT) == IFDIR) { cgp->cg_cs.cs_ndir++; fs->fs_cstotal.cs_ndir++; fs->fs_cs(fs, cg).cs_ndir++; } UFS_UNLOCK(ump); if (DOINGSOFTDEP(ITOV(ip))) softdep_setup_inomapdep(bp, ip, cg * fs->fs_ipg + ipref, mode); bdwrite(bp); return ((ino_t)(cg * fs->fs_ipg + ipref)); } /* * Free a block or fragment. * * The specified block or fragment is placed back in the * free map. If a fragment is deallocated, a possible * block reassembly is checked. */ static void ffs_blkfree_cg(ump, fs, devvp, bno, size, inum, dephd) struct ufsmount *ump; struct fs *fs; struct vnode *devvp; ufs2_daddr_t bno; long size; ino_t inum; struct workhead *dephd; { struct mount *mp; struct cg *cgp; struct buf *bp; ufs1_daddr_t fragno, cgbno; int i, blk, frags, bbase, error; u_int cg; u_int8_t *blksfree; struct cdev *dev; cg = dtog(fs, bno); if (devvp->v_type == VREG) { /* devvp is a snapshot */ MPASS(devvp->v_mount->mnt_data == ump); dev = ump->um_devvp->v_rdev; } else if (devvp->v_type == VCHR) { /* devvp is a normal disk device */ dev = devvp->v_rdev; ASSERT_VOP_LOCKED(devvp, "ffs_blkfree_cg"); } else return; #ifdef INVARIANTS if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0 || fragnum(fs, bno) + numfrags(fs, size) > fs->fs_frag) { printf("dev=%s, bno = %jd, bsize = %ld, size = %ld, fs = %s\n", devtoname(dev), (intmax_t)bno, (long)fs->fs_bsize, size, fs->fs_fsmnt); panic("ffs_blkfree_cg: bad size"); } #endif if ((u_int)bno >= fs->fs_size) { printf("bad block %jd, ino %lu\n", (intmax_t)bno, (u_long)inum); ffs_fserr(fs, inum, "bad block"); return; } if ((error = ffs_getcg(fs, devvp, cg, 0, &bp, &cgp)) != 0) return; cgbno = dtogd(fs, bno); blksfree = cg_blksfree(cgp); UFS_LOCK(ump); if (size == fs->fs_bsize) { fragno = fragstoblks(fs, cgbno); if (!ffs_isfreeblock(fs, blksfree, fragno)) { if (devvp->v_type == VREG) { UFS_UNLOCK(ump); /* devvp is a snapshot */ brelse(bp); return; } printf("dev = %s, block = %jd, fs = %s\n", devtoname(dev), (intmax_t)bno, fs->fs_fsmnt); panic("ffs_blkfree_cg: freeing free block"); } ffs_setblock(fs, blksfree, fragno); ffs_clusteracct(fs, cgp, fragno, 1); cgp->cg_cs.cs_nbfree++; fs->fs_cstotal.cs_nbfree++; fs->fs_cs(fs, cg).cs_nbfree++; } else { bbase = cgbno - fragnum(fs, cgbno); /* * decrement the counts associated with the old frags */ blk = blkmap(fs, blksfree, bbase); ffs_fragacct(fs, blk, cgp->cg_frsum, -1); /* * deallocate the fragment */ frags = numfrags(fs, size); for (i = 0; i < frags; i++) { if (isset(blksfree, cgbno + i)) { printf("dev = %s, block = %jd, fs = %s\n", devtoname(dev), (intmax_t)(bno + i), fs->fs_fsmnt); panic("ffs_blkfree_cg: freeing free frag"); } setbit(blksfree, cgbno + i); } cgp->cg_cs.cs_nffree += i; fs->fs_cstotal.cs_nffree += i; fs->fs_cs(fs, cg).cs_nffree += i; /* * add back in counts associated with the new frags */ blk = blkmap(fs, blksfree, bbase); ffs_fragacct(fs, blk, cgp->cg_frsum, 1); /* * if a complete block has been reassembled, account for it */ fragno = fragstoblks(fs, bbase); if (ffs_isblock(fs, blksfree, fragno)) { cgp->cg_cs.cs_nffree -= fs->fs_frag; fs->fs_cstotal.cs_nffree -= fs->fs_frag; fs->fs_cs(fs, cg).cs_nffree -= fs->fs_frag; ffs_clusteracct(fs, cgp, fragno, 1); cgp->cg_cs.cs_nbfree++; fs->fs_cstotal.cs_nbfree++; fs->fs_cs(fs, cg).cs_nbfree++; } } fs->fs_fmod = 1; ACTIVECLEAR(fs, cg); UFS_UNLOCK(ump); mp = UFSTOVFS(ump); if (MOUNTEDSOFTDEP(mp) && devvp->v_type == VCHR) softdep_setup_blkfree(UFSTOVFS(ump), bp, bno, numfrags(fs, size), dephd); bdwrite(bp); } /* * Structures and routines associated with trim management. * * The following requests are passed to trim_lookup to indicate * the actions that should be taken. */ #define NEW 1 /* if found, error else allocate and hash it */ #define OLD 2 /* if not found, error, else return it */ #define REPLACE 3 /* if not found, error else unhash and reallocate it */ #define DONE 4 /* if not found, error else unhash and return it */ #define SINGLE 5 /* don't look up, just allocate it and don't hash it */ MALLOC_DEFINE(M_TRIM, "ufs_trim", "UFS trim structures"); #define TRIMLIST_HASH(ump, key) \ (&(ump)->um_trimhash[(key) & (ump)->um_trimlisthashsize]) /* * These structures describe each of the block free requests aggregated * together to make up a trim request. */ struct trim_blkreq { TAILQ_ENTRY(trim_blkreq) blkreqlist; ufs2_daddr_t bno; long size; struct workhead *pdephd; struct workhead dephd; }; /* * Description of a trim request. */ struct ffs_blkfree_trim_params { TAILQ_HEAD(, trim_blkreq) blklist; LIST_ENTRY(ffs_blkfree_trim_params) hashlist; struct task task; struct ufsmount *ump; struct vnode *devvp; ino_t inum; ufs2_daddr_t bno; long size; long key; }; static void ffs_blkfree_trim_completed(struct buf *); static void ffs_blkfree_trim_task(void *ctx, int pending __unused); static struct ffs_blkfree_trim_params *trim_lookup(struct ufsmount *, struct vnode *, ufs2_daddr_t, long, ino_t, u_long, int); static void ffs_blkfree_sendtrim(struct ffs_blkfree_trim_params *); /* * Called on trim completion to start a task to free the associated block(s). */ static void ffs_blkfree_trim_completed(bp) struct buf *bp; { struct ffs_blkfree_trim_params *tp; tp = bp->b_fsprivate1; free(bp, M_TRIM); TASK_INIT(&tp->task, 0, ffs_blkfree_trim_task, tp); taskqueue_enqueue(tp->ump->um_trim_tq, &tp->task); } /* * Trim completion task that free associated block(s). */ static void ffs_blkfree_trim_task(ctx, pending) void *ctx; int pending; { struct ffs_blkfree_trim_params *tp; struct trim_blkreq *blkelm; struct ufsmount *ump; tp = ctx; ump = tp->ump; while ((blkelm = TAILQ_FIRST(&tp->blklist)) != NULL) { ffs_blkfree_cg(ump, ump->um_fs, tp->devvp, blkelm->bno, blkelm->size, tp->inum, blkelm->pdephd); TAILQ_REMOVE(&tp->blklist, blkelm, blkreqlist); free(blkelm, M_TRIM); } vn_finished_secondary_write(UFSTOVFS(ump)); UFS_LOCK(ump); ump->um_trim_inflight -= 1; ump->um_trim_inflight_blks -= numfrags(ump->um_fs, tp->size); UFS_UNLOCK(ump); free(tp, M_TRIM); } /* * Lookup a trim request by inode number. * Allocate if requested (NEW, REPLACE, SINGLE). */ static struct ffs_blkfree_trim_params * trim_lookup(ump, devvp, bno, size, inum, key, alloctype) struct ufsmount *ump; struct vnode *devvp; ufs2_daddr_t bno; long size; ino_t inum; u_long key; int alloctype; { struct trimlist_hashhead *tphashhead; struct ffs_blkfree_trim_params *tp, *ntp; ntp = malloc(sizeof(struct ffs_blkfree_trim_params), M_TRIM, M_WAITOK); if (alloctype != SINGLE) { KASSERT(key >= FIRST_VALID_KEY, ("trim_lookup: invalid key")); UFS_LOCK(ump); tphashhead = TRIMLIST_HASH(ump, key); LIST_FOREACH(tp, tphashhead, hashlist) if (key == tp->key) break; } switch (alloctype) { case NEW: KASSERT(tp == NULL, ("trim_lookup: found trim")); break; case OLD: KASSERT(tp != NULL, ("trim_lookup: missing call to ffs_blkrelease_start()")); UFS_UNLOCK(ump); free(ntp, M_TRIM); return (tp); case REPLACE: KASSERT(tp != NULL, ("trim_lookup: missing REPLACE trim")); LIST_REMOVE(tp, hashlist); /* tp will be freed by caller */ break; case DONE: KASSERT(tp != NULL, ("trim_lookup: missing DONE trim")); LIST_REMOVE(tp, hashlist); UFS_UNLOCK(ump); free(ntp, M_TRIM); return (tp); } TAILQ_INIT(&ntp->blklist); ntp->ump = ump; ntp->devvp = devvp; ntp->bno = bno; ntp->size = size; ntp->inum = inum; ntp->key = key; if (alloctype != SINGLE) { LIST_INSERT_HEAD(tphashhead, ntp, hashlist); UFS_UNLOCK(ump); } return (ntp); } /* * Dispatch a trim request. */ static void ffs_blkfree_sendtrim(tp) struct ffs_blkfree_trim_params *tp; { struct ufsmount *ump; struct mount *mp; struct buf *bp; /* * Postpone the set of the free bit in the cg bitmap until the * BIO_DELETE is completed. Otherwise, due to disk queue * reordering, TRIM might be issued after we reuse the block * and write some new data into it. */ ump = tp->ump; bp = malloc(sizeof(*bp), M_TRIM, M_WAITOK | M_ZERO); bp->b_iocmd = BIO_DELETE; bp->b_iooffset = dbtob(fsbtodb(ump->um_fs, tp->bno)); bp->b_iodone = ffs_blkfree_trim_completed; bp->b_bcount = tp->size; bp->b_fsprivate1 = tp; UFS_LOCK(ump); ump->um_trim_total += 1; ump->um_trim_inflight += 1; ump->um_trim_inflight_blks += numfrags(ump->um_fs, tp->size); ump->um_trim_total_blks += numfrags(ump->um_fs, tp->size); UFS_UNLOCK(ump); mp = UFSTOVFS(ump); vn_start_secondary_write(NULL, &mp, 0); g_vfs_strategy(ump->um_bo, bp); } /* * Allocate a new key to use to identify a range of blocks. */ u_long ffs_blkrelease_start(ump, devvp, inum) struct ufsmount *ump; struct vnode *devvp; ino_t inum; { static u_long masterkey; u_long key; if (((ump->um_flags & UM_CANDELETE) == 0) || dotrimcons == 0) return (SINGLETON_KEY); do { key = atomic_fetchadd_long(&masterkey, 1); } while (key < FIRST_VALID_KEY); (void) trim_lookup(ump, devvp, 0, 0, inum, key, NEW); return (key); } /* * Deallocate a key that has been used to identify a range of blocks. */ void ffs_blkrelease_finish(ump, key) struct ufsmount *ump; u_long key; { struct ffs_blkfree_trim_params *tp; if (((ump->um_flags & UM_CANDELETE) == 0) || dotrimcons == 0) return; /* * If the vfs.ffs.dotrimcons sysctl option is enabled while * a file deletion is active, specifically after a call * to ffs_blkrelease_start() but before the call to * ffs_blkrelease_finish(), ffs_blkrelease_start() will * have handed out SINGLETON_KEY rather than starting a * collection sequence. Thus if we get a SINGLETON_KEY * passed to ffs_blkrelease_finish(), we just return rather * than trying to finish the nonexistent sequence. */ if (key == SINGLETON_KEY) { #ifdef INVARIANTS printf("%s: vfs.ffs.dotrimcons enabled on active filesystem\n", ump->um_mountp->mnt_stat.f_mntonname); #endif return; } /* * We are done with sending blocks using this key. Look up the key * using the DONE alloctype (in tp) to request that it be unhashed * as we will not be adding to it. If the key has never been used, * tp->size will be zero, so we can just free tp. Otherwise the call * to ffs_blkfree_sendtrim(tp) causes the block range described by * tp to be issued (and then tp to be freed). */ tp = trim_lookup(ump, NULL, 0, 0, 0, key, DONE); if (tp->size == 0) free(tp, M_TRIM); else ffs_blkfree_sendtrim(tp); } /* * Setup to free a block or fragment. * * Check for snapshots that might want to claim the block. * If trims are requested, prepare a trim request. Attempt to * aggregate consecutive blocks into a single trim request. */ void ffs_blkfree(ump, fs, devvp, bno, size, inum, vtype, dephd, key) struct ufsmount *ump; struct fs *fs; struct vnode *devvp; ufs2_daddr_t bno; long size; ino_t inum; enum vtype vtype; struct workhead *dephd; u_long key; { struct ffs_blkfree_trim_params *tp, *ntp; struct trim_blkreq *blkelm; /* * Check to see if a snapshot wants to claim the block. * Check that devvp is a normal disk device, not a snapshot, * it has a snapshot(s) associated with it, and one of the * snapshots wants to claim the block. */ if (devvp->v_type == VCHR && (devvp->v_vflag & VV_COPYONWRITE) && ffs_snapblkfree(fs, devvp, bno, size, inum, vtype, dephd)) { return; } /* * Nothing to delay if TRIM is not required for this block or TRIM * is disabled or the operation is performed on a snapshot. */ if (key == NOTRIM_KEY || ((ump->um_flags & UM_CANDELETE) == 0) || devvp->v_type == VREG) { ffs_blkfree_cg(ump, fs, devvp, bno, size, inum, dephd); return; } blkelm = malloc(sizeof(struct trim_blkreq), M_TRIM, M_WAITOK); blkelm->bno = bno; blkelm->size = size; if (dephd == NULL) { blkelm->pdephd = NULL; } else { LIST_INIT(&blkelm->dephd); LIST_SWAP(dephd, &blkelm->dephd, worklist, wk_list); blkelm->pdephd = &blkelm->dephd; } if (key == SINGLETON_KEY) { /* * Just a single non-contiguous piece. Use the SINGLE * alloctype to return a trim request that will not be * hashed for future lookup. */ tp = trim_lookup(ump, devvp, bno, size, inum, key, SINGLE); TAILQ_INSERT_HEAD(&tp->blklist, blkelm, blkreqlist); ffs_blkfree_sendtrim(tp); return; } /* * The callers of this function are not tracking whether or not * the blocks are contiguous. They are just saying that they * are freeing a set of blocks. It is this code that determines * the pieces of that range that are actually contiguous. * * Calling ffs_blkrelease_start() will have created an entry * that we will use. */ tp = trim_lookup(ump, devvp, bno, size, inum, key, OLD); if (tp->size == 0) { /* * First block of a potential range, set block and size * for the trim block. */ tp->bno = bno; tp->size = size; TAILQ_INSERT_HEAD(&tp->blklist, blkelm, blkreqlist); return; } /* * If this block is a continuation of the range (either * follows at the end or preceeds in the front) then we * add it to the front or back of the list and return. * * If it is not a continuation of the trim that we were * building, using the REPLACE alloctype, we request that * the old trim request (still in tp) be unhashed and a * new range started (in ntp). The ffs_blkfree_sendtrim(tp) * call causes the block range described by tp to be issued * (and then tp to be freed). */ if (bno + numfrags(fs, size) == tp->bno) { TAILQ_INSERT_HEAD(&tp->blklist, blkelm, blkreqlist); tp->bno = bno; tp->size += size; return; } else if (bno == tp->bno + numfrags(fs, tp->size)) { TAILQ_INSERT_TAIL(&tp->blklist, blkelm, blkreqlist); tp->size += size; return; } ntp = trim_lookup(ump, devvp, bno, size, inum, key, REPLACE); TAILQ_INSERT_HEAD(&ntp->blklist, blkelm, blkreqlist); ffs_blkfree_sendtrim(tp); } #ifdef INVARIANTS /* * Verify allocation of a block or fragment. Returns true if block or * fragment is allocated, false if it is free. */ static int ffs_checkblk(ip, bno, size) struct inode *ip; ufs2_daddr_t bno; long size; { struct fs *fs; struct cg *cgp; struct buf *bp; ufs1_daddr_t cgbno; int i, error, frags, free; u_int8_t *blksfree; fs = ITOFS(ip); if ((u_int)size > fs->fs_bsize || fragoff(fs, size) != 0) { printf("bsize = %ld, size = %ld, fs = %s\n", (long)fs->fs_bsize, size, fs->fs_fsmnt); panic("ffs_checkblk: bad size"); } if ((u_int)bno >= fs->fs_size) panic("ffs_checkblk: bad block %jd", (intmax_t)bno); error = ffs_getcg(fs, ITODEVVP(ip), dtog(fs, bno), 0, &bp, &cgp); if (error) panic("ffs_checkblk: cylinder group read failed"); blksfree = cg_blksfree(cgp); cgbno = dtogd(fs, bno); if (size == fs->fs_bsize) { free = ffs_isblock(fs, blksfree, fragstoblks(fs, cgbno)); } else { frags = numfrags(fs, size); for (free = 0, i = 0; i < frags; i++) if (isset(blksfree, cgbno + i)) free++; if (free != 0 && free != frags) panic("ffs_checkblk: partially free fragment"); } brelse(bp); return (!free); } #endif /* INVARIANTS */ /* * Free an inode. */ int ffs_vfree(pvp, ino, mode) struct vnode *pvp; ino_t ino; int mode; { struct ufsmount *ump; if (DOINGSOFTDEP(pvp)) { softdep_freefile(pvp, ino, mode); return (0); } ump = VFSTOUFS(pvp->v_mount); return (ffs_freefile(ump, ump->um_fs, ump->um_devvp, ino, mode, NULL)); } /* * Do the actual free operation. * The specified inode is placed back in the free map. */ int ffs_freefile(ump, fs, devvp, ino, mode, wkhd) struct ufsmount *ump; struct fs *fs; struct vnode *devvp; ino_t ino; int mode; struct workhead *wkhd; { struct cg *cgp; struct buf *bp; int error; u_int cg; u_int8_t *inosused; struct cdev *dev; ino_t cgino; cg = ino_to_cg(fs, ino); if (devvp->v_type == VREG) { /* devvp is a snapshot */ MPASS(devvp->v_mount->mnt_data == ump); dev = ump->um_devvp->v_rdev; } else if (devvp->v_type == VCHR) { /* devvp is a normal disk device */ dev = devvp->v_rdev; } else { bp = NULL; return (0); } if (ino >= fs->fs_ipg * fs->fs_ncg) panic("ffs_freefile: range: dev = %s, ino = %ju, fs = %s", devtoname(dev), (uintmax_t)ino, fs->fs_fsmnt); if ((error = ffs_getcg(fs, devvp, cg, 0, &bp, &cgp)) != 0) return (error); inosused = cg_inosused(cgp); cgino = ino % fs->fs_ipg; if (isclr(inosused, cgino)) { printf("dev = %s, ino = %ju, fs = %s\n", devtoname(dev), (uintmax_t)ino, fs->fs_fsmnt); if (fs->fs_ronly == 0) panic("ffs_freefile: freeing free inode"); } clrbit(inosused, cgino); if (cgino < cgp->cg_irotor) cgp->cg_irotor = cgino; cgp->cg_cs.cs_nifree++; UFS_LOCK(ump); fs->fs_cstotal.cs_nifree++; fs->fs_cs(fs, cg).cs_nifree++; if ((mode & IFMT) == IFDIR) { cgp->cg_cs.cs_ndir--; fs->fs_cstotal.cs_ndir--; fs->fs_cs(fs, cg).cs_ndir--; } fs->fs_fmod = 1; ACTIVECLEAR(fs, cg); UFS_UNLOCK(ump); if (MOUNTEDSOFTDEP(UFSTOVFS(ump)) && devvp->v_type == VCHR) softdep_setup_inofree(UFSTOVFS(ump), bp, ino, wkhd); bdwrite(bp); return (0); } /* * Check to see if a file is free. * Used to check for allocated files in snapshots. */ int ffs_checkfreefile(fs, devvp, ino) struct fs *fs; struct vnode *devvp; ino_t ino; { struct cg *cgp; struct buf *bp; int ret, error; u_int cg; u_int8_t *inosused; cg = ino_to_cg(fs, ino); if ((devvp->v_type != VREG) && (devvp->v_type != VCHR)) return (1); if (ino >= fs->fs_ipg * fs->fs_ncg) return (1); if ((error = ffs_getcg(fs, devvp, cg, 0, &bp, &cgp)) != 0) return (1); inosused = cg_inosused(cgp); ino %= fs->fs_ipg; ret = isclr(inosused, ino); brelse(bp); return (ret); } /* * Find a block of the specified size in the specified cylinder group. * * It is a panic if a request is made to find a block if none are * available. */ static ufs1_daddr_t ffs_mapsearch(fs, cgp, bpref, allocsiz) struct fs *fs; struct cg *cgp; ufs2_daddr_t bpref; int allocsiz; { ufs1_daddr_t bno; int start, len, loc, i; int blk, field, subfield, pos; u_int8_t *blksfree; /* * find the fragment by searching through the free block * map for an appropriate bit pattern */ if (bpref) start = dtogd(fs, bpref) / NBBY; else start = cgp->cg_frotor / NBBY; blksfree = cg_blksfree(cgp); len = howmany(fs->fs_fpg, NBBY) - start; loc = scanc((u_int)len, (u_char *)&blksfree[start], fragtbl[fs->fs_frag], (u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY)))); if (loc == 0) { len = start + 1; start = 0; loc = scanc((u_int)len, (u_char *)&blksfree[0], fragtbl[fs->fs_frag], (u_char)(1 << (allocsiz - 1 + (fs->fs_frag % NBBY)))); if (loc == 0) { printf("start = %d, len = %d, fs = %s\n", start, len, fs->fs_fsmnt); panic("ffs_alloccg: map corrupted"); /* NOTREACHED */ } } bno = (start + len - loc) * NBBY; cgp->cg_frotor = bno; /* * found the byte in the map * sift through the bits to find the selected frag */ for (i = bno + NBBY; bno < i; bno += fs->fs_frag) { blk = blkmap(fs, blksfree, bno); blk <<= 1; field = around[allocsiz]; subfield = inside[allocsiz]; for (pos = 0; pos <= fs->fs_frag - allocsiz; pos++) { if ((blk & field) == subfield) return (bno + pos); field <<= 1; subfield <<= 1; } } printf("bno = %lu, fs = %s\n", (u_long)bno, fs->fs_fsmnt); panic("ffs_alloccg: block not in map"); return (-1); } static const struct statfs * ffs_getmntstat(struct vnode *devvp) { if (devvp->v_type == VCHR) return (&devvp->v_rdev->si_mountpt->mnt_stat); return (ffs_getmntstat(VFSTOUFS(devvp->v_mount)->um_devvp)); } /* * Fetch and verify a cylinder group. */ int ffs_getcg(fs, devvp, cg, flags, bpp, cgpp) struct fs *fs; struct vnode *devvp; u_int cg; int flags; struct buf **bpp; struct cg **cgpp; { struct buf *bp; struct cg *cgp; const struct statfs *sfs; daddr_t blkno; int error; *bpp = NULL; *cgpp = NULL; if ((fs->fs_metackhash & CK_CYLGRP) != 0) flags |= GB_CKHASH; if (devvp->v_type == VREG) blkno = fragstoblks(fs, cgtod(fs, cg)); else blkno = fsbtodb(fs, cgtod(fs, cg)); error = breadn_flags(devvp, blkno, blkno, (int)fs->fs_cgsize, NULL, NULL, 0, NOCRED, flags, ffs_ckhash_cg, &bp); if (error != 0) return (error); cgp = (struct cg *)bp->b_data; if ((fs->fs_metackhash & CK_CYLGRP) != 0 && (bp->b_flags & B_CKHASH) != 0 && cgp->cg_ckhash != bp->b_ckhash) { sfs = ffs_getmntstat(devvp); printf("UFS %s%s (%s) cylinder checksum failed: cg %u, cgp: " "0x%x != bp: 0x%jx\n", devvp->v_type == VCHR ? "" : "snapshot of ", sfs->f_mntfromname, sfs->f_mntonname, cg, cgp->cg_ckhash, (uintmax_t)bp->b_ckhash); bp->b_flags &= ~B_CKHASH; bp->b_flags |= B_INVAL | B_NOCACHE; brelse(bp); return (EIO); } if (!cg_chkmagic(cgp) || cgp->cg_cgx != cg) { sfs = ffs_getmntstat(devvp); printf("UFS %s%s (%s)", devvp->v_type == VCHR ? "" : "snapshot of ", sfs->f_mntfromname, sfs->f_mntonname); if (!cg_chkmagic(cgp)) printf(" cg %u: bad magic number 0x%x should be 0x%x\n", cg, cgp->cg_magic, CG_MAGIC); else printf(": wrong cylinder group cg %u != cgx %u\n", cg, cgp->cg_cgx); bp->b_flags &= ~B_CKHASH; bp->b_flags |= B_INVAL | B_NOCACHE; brelse(bp); return (EIO); } bp->b_flags &= ~B_CKHASH; bp->b_xflags |= BX_BKGRDWRITE; /* * If we are using check hashes on the cylinder group then we want * to limit changing the cylinder group time to when we are actually * going to write it to disk so that its check hash remains correct * in memory. If the CK_CYLGRP flag is set the time is updated in * ffs_bufwrite() as the buffer is queued for writing. Otherwise we * update the time here as we have done historically. */ if ((fs->fs_metackhash & CK_CYLGRP) != 0) bp->b_xflags |= BX_CYLGRP; else cgp->cg_old_time = cgp->cg_time = time_second; *bpp = bp; *cgpp = cgp; return (0); } static void ffs_ckhash_cg(bp) struct buf *bp; { uint32_t ckhash; struct cg *cgp; cgp = (struct cg *)bp->b_data; ckhash = cgp->cg_ckhash; cgp->cg_ckhash = 0; bp->b_ckhash = calculate_crc32c(~0L, bp->b_data, bp->b_bcount); cgp->cg_ckhash = ckhash; } /* * Fserr prints the name of a filesystem with an error diagnostic. * * The form of the error message is: * fs: error message */ void ffs_fserr(fs, inum, cp) struct fs *fs; ino_t inum; char *cp; { struct thread *td = curthread; /* XXX */ struct proc *p = td->td_proc; log(LOG_ERR, "pid %d (%s), uid %d inumber %ju on %s: %s\n", p->p_pid, p->p_comm, td->td_ucred->cr_uid, (uintmax_t)inum, fs->fs_fsmnt, cp); } /* * This function provides the capability for the fsck program to * update an active filesystem. Fourteen operations are provided: * * adjrefcnt(inode, amt) - adjusts the reference count on the * specified inode by the specified amount. Under normal * operation the count should always go down. Decrementing * the count to zero will cause the inode to be freed. * adjblkcnt(inode, amt) - adjust the number of blocks used by the * inode by the specified amount. * adjsize(inode, size) - set the size of the inode to the * specified size. * adjndir, adjbfree, adjifree, adjffree, adjnumclusters(amt) - * adjust the superblock summary. * freedirs(inode, count) - directory inodes [inode..inode + count - 1] * are marked as free. Inodes should never have to be marked * as in use. * freefiles(inode, count) - file inodes [inode..inode + count - 1] * are marked as free. Inodes should never have to be marked * as in use. * freeblks(blockno, size) - blocks [blockno..blockno + size - 1] * are marked as free. Blocks should never have to be marked * as in use. * setflags(flags, set/clear) - the fs_flags field has the specified * flags set (second parameter +1) or cleared (second parameter -1). * setcwd(dirinode) - set the current directory to dirinode in the * filesystem associated with the snapshot. * setdotdot(oldvalue, newvalue) - Verify that the inode number for ".." * in the current directory is oldvalue then change it to newvalue. * unlink(nameptr, oldvalue) - Verify that the inode number associated * with nameptr in the current directory is oldvalue then unlink it. * * The following functions may only be used on a quiescent filesystem * by the soft updates journal. They are not safe to be run on an active * filesystem. * * setinode(inode, dip) - the specified disk inode is replaced with the * contents pointed to by dip. * setbufoutput(fd, flags) - output associated with the specified file * descriptor (which must reference the character device supporting * the filesystem) switches from using physio to running through the * buffer cache when flags is set to 1. The descriptor reverts to * physio for output when flags is set to zero. */ static int sysctl_ffs_fsck(SYSCTL_HANDLER_ARGS); SYSCTL_PROC(_vfs_ffs, FFS_ADJ_REFCNT, adjrefcnt, CTLFLAG_WR | CTLTYPE_STRUCT | CTLFLAG_NEEDGIANT, 0, 0, sysctl_ffs_fsck, "S,fsck", "Adjust Inode Reference Count"); static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_BLKCNT, adjblkcnt, CTLFLAG_WR | CTLFLAG_NEEDGIANT, sysctl_ffs_fsck, "Adjust Inode Used Blocks Count"); static SYSCTL_NODE(_vfs_ffs, FFS_SET_SIZE, setsize, CTLFLAG_WR | CTLFLAG_NEEDGIANT, sysctl_ffs_fsck, "Set the inode size"); static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_NDIR, adjndir, CTLFLAG_WR | CTLFLAG_NEEDGIANT, sysctl_ffs_fsck, "Adjust number of directories"); static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_NBFREE, adjnbfree, CTLFLAG_WR | CTLFLAG_NEEDGIANT, sysctl_ffs_fsck, "Adjust number of free blocks"); static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_NIFREE, adjnifree, CTLFLAG_WR | CTLFLAG_NEEDGIANT, sysctl_ffs_fsck, "Adjust number of free inodes"); static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_NFFREE, adjnffree, CTLFLAG_WR | CTLFLAG_NEEDGIANT, sysctl_ffs_fsck, "Adjust number of free frags"); static SYSCTL_NODE(_vfs_ffs, FFS_ADJ_NUMCLUSTERS, adjnumclusters, CTLFLAG_WR | CTLFLAG_NEEDGIANT, sysctl_ffs_fsck, "Adjust number of free clusters"); static SYSCTL_NODE(_vfs_ffs, FFS_DIR_FREE, freedirs, CTLFLAG_WR | CTLFLAG_NEEDGIANT, sysctl_ffs_fsck, "Free Range of Directory Inodes"); static SYSCTL_NODE(_vfs_ffs, FFS_FILE_FREE, freefiles, CTLFLAG_WR | CTLFLAG_NEEDGIANT, sysctl_ffs_fsck, "Free Range of File Inodes"); static SYSCTL_NODE(_vfs_ffs, FFS_BLK_FREE, freeblks, CTLFLAG_WR | CTLFLAG_NEEDGIANT, sysctl_ffs_fsck, "Free Range of Blocks"); static SYSCTL_NODE(_vfs_ffs, FFS_SET_FLAGS, setflags, CTLFLAG_WR | CTLFLAG_NEEDGIANT, sysctl_ffs_fsck, "Change Filesystem Flags"); static SYSCTL_NODE(_vfs_ffs, FFS_SET_CWD, setcwd, CTLFLAG_WR | CTLFLAG_NEEDGIANT, sysctl_ffs_fsck, "Set Current Working Directory"); static SYSCTL_NODE(_vfs_ffs, FFS_SET_DOTDOT, setdotdot, CTLFLAG_WR | CTLFLAG_NEEDGIANT, sysctl_ffs_fsck, "Change Value of .. Entry"); static SYSCTL_NODE(_vfs_ffs, FFS_UNLINK, unlink, CTLFLAG_WR | CTLFLAG_NEEDGIANT, sysctl_ffs_fsck, "Unlink a Duplicate Name"); static SYSCTL_NODE(_vfs_ffs, FFS_SET_INODE, setinode, CTLFLAG_WR | CTLFLAG_NEEDGIANT, sysctl_ffs_fsck, "Update an On-Disk Inode"); static SYSCTL_NODE(_vfs_ffs, FFS_SET_BUFOUTPUT, setbufoutput, CTLFLAG_WR | CTLFLAG_NEEDGIANT, sysctl_ffs_fsck, "Set Buffered Writing for Descriptor"); #ifdef DIAGNOSTIC static int fsckcmds = 0; SYSCTL_INT(_debug, OID_AUTO, ffs_fsckcmds, CTLFLAG_RW, &fsckcmds, 0, "print out fsck_ffs-based filesystem update commands"); #endif /* DIAGNOSTIC */ static int buffered_write(struct file *, struct uio *, struct ucred *, int, struct thread *); static int sysctl_ffs_fsck(SYSCTL_HANDLER_ARGS) { struct thread *td = curthread; struct fsck_cmd cmd; struct ufsmount *ump; struct vnode *vp, *dvp, *fdvp; struct inode *ip, *dp; struct mount *mp; struct fs *fs; struct pwd *pwd; ufs2_daddr_t blkno; long blkcnt, blksize; u_long key; struct file *fp, *vfp; cap_rights_t rights; int filetype, error; static struct fileops *origops, bufferedops; if (req->newlen > sizeof cmd) return (EBADRPC); if ((error = SYSCTL_IN(req, &cmd, sizeof cmd)) != 0) return (error); if (cmd.version != FFS_CMD_VERSION) return (ERPCMISMATCH); if ((error = getvnode(td, cmd.handle, cap_rights_init(&rights, CAP_FSCK), &fp)) != 0) return (error); vp = fp->f_data; if (vp->v_type != VREG && vp->v_type != VDIR) { fdrop(fp, td); return (EINVAL); } vn_start_write(vp, &mp, V_WAIT); if (mp == NULL || strncmp(mp->mnt_stat.f_fstypename, "ufs", MFSNAMELEN)) { vn_finished_write(mp); fdrop(fp, td); return (EINVAL); } ump = VFSTOUFS(mp); if ((mp->mnt_flag & MNT_RDONLY) && ump->um_fsckpid != td->td_proc->p_pid) { vn_finished_write(mp); fdrop(fp, td); return (EROFS); } fs = ump->um_fs; filetype = IFREG; switch (oidp->oid_number) { case FFS_SET_FLAGS: #ifdef DIAGNOSTIC if (fsckcmds) printf("%s: %s flags\n", mp->mnt_stat.f_mntonname, cmd.size > 0 ? "set" : "clear"); #endif /* DIAGNOSTIC */ if (cmd.size > 0) fs->fs_flags |= (long)cmd.value; else fs->fs_flags &= ~(long)cmd.value; break; case FFS_ADJ_REFCNT: #ifdef DIAGNOSTIC if (fsckcmds) { printf("%s: adjust inode %jd link count by %jd\n", mp->mnt_stat.f_mntonname, (intmax_t)cmd.value, (intmax_t)cmd.size); } #endif /* DIAGNOSTIC */ if ((error = ffs_vget(mp, (ino_t)cmd.value, LK_EXCLUSIVE, &vp))) break; ip = VTOI(vp); ip->i_nlink += cmd.size; DIP_SET(ip, i_nlink, ip->i_nlink); ip->i_effnlink += cmd.size; UFS_INODE_SET_FLAG(ip, IN_CHANGE | IN_MODIFIED); error = ffs_update(vp, 1); if (DOINGSOFTDEP(vp)) softdep_change_linkcnt(ip); vput(vp); break; case FFS_ADJ_BLKCNT: #ifdef DIAGNOSTIC if (fsckcmds) { printf("%s: adjust inode %jd block count by %jd\n", mp->mnt_stat.f_mntonname, (intmax_t)cmd.value, (intmax_t)cmd.size); } #endif /* DIAGNOSTIC */ if ((error = ffs_vget(mp, (ino_t)cmd.value, LK_EXCLUSIVE, &vp))) break; ip = VTOI(vp); DIP_SET(ip, i_blocks, DIP(ip, i_blocks) + cmd.size); UFS_INODE_SET_FLAG(ip, IN_CHANGE | IN_MODIFIED); error = ffs_update(vp, 1); vput(vp); break; case FFS_SET_SIZE: #ifdef DIAGNOSTIC if (fsckcmds) { printf("%s: set inode %jd size to %jd\n", mp->mnt_stat.f_mntonname, (intmax_t)cmd.value, (intmax_t)cmd.size); } #endif /* DIAGNOSTIC */ if ((error = ffs_vget(mp, (ino_t)cmd.value, LK_EXCLUSIVE, &vp))) break; ip = VTOI(vp); DIP_SET(ip, i_size, cmd.size); UFS_INODE_SET_FLAG(ip, IN_CHANGE | IN_MODIFIED); error = ffs_update(vp, 1); vput(vp); break; case FFS_DIR_FREE: filetype = IFDIR; /* fall through */ case FFS_FILE_FREE: #ifdef DIAGNOSTIC if (fsckcmds) { if (cmd.size == 1) printf("%s: free %s inode %ju\n", mp->mnt_stat.f_mntonname, filetype == IFDIR ? "directory" : "file", (uintmax_t)cmd.value); else printf("%s: free %s inodes %ju-%ju\n", mp->mnt_stat.f_mntonname, filetype == IFDIR ? "directory" : "file", (uintmax_t)cmd.value, (uintmax_t)(cmd.value + cmd.size - 1)); } #endif /* DIAGNOSTIC */ while (cmd.size > 0) { if ((error = ffs_freefile(ump, fs, ump->um_devvp, cmd.value, filetype, NULL))) break; cmd.size -= 1; cmd.value += 1; } break; case FFS_BLK_FREE: #ifdef DIAGNOSTIC if (fsckcmds) { if (cmd.size == 1) printf("%s: free block %jd\n", mp->mnt_stat.f_mntonname, (intmax_t)cmd.value); else printf("%s: free blocks %jd-%jd\n", mp->mnt_stat.f_mntonname, (intmax_t)cmd.value, (intmax_t)cmd.value + cmd.size - 1); } #endif /* DIAGNOSTIC */ blkno = cmd.value; blkcnt = cmd.size; blksize = fs->fs_frag - (blkno % fs->fs_frag); key = ffs_blkrelease_start(ump, ump->um_devvp, UFS_ROOTINO); while (blkcnt > 0) { if (blkcnt < blksize) blksize = blkcnt; ffs_blkfree(ump, fs, ump->um_devvp, blkno, blksize * fs->fs_fsize, UFS_ROOTINO, VDIR, NULL, key); blkno += blksize; blkcnt -= blksize; blksize = fs->fs_frag; } ffs_blkrelease_finish(ump, key); break; /* * Adjust superblock summaries. fsck(8) is expected to * submit deltas when necessary. */ case FFS_ADJ_NDIR: #ifdef DIAGNOSTIC if (fsckcmds) { printf("%s: adjust number of directories by %jd\n", mp->mnt_stat.f_mntonname, (intmax_t)cmd.value); } #endif /* DIAGNOSTIC */ fs->fs_cstotal.cs_ndir += cmd.value; break; case FFS_ADJ_NBFREE: #ifdef DIAGNOSTIC if (fsckcmds) { printf("%s: adjust number of free blocks by %+jd\n", mp->mnt_stat.f_mntonname, (intmax_t)cmd.value); } #endif /* DIAGNOSTIC */ fs->fs_cstotal.cs_nbfree += cmd.value; break; case FFS_ADJ_NIFREE: #ifdef DIAGNOSTIC if (fsckcmds) { printf("%s: adjust number of free inodes by %+jd\n", mp->mnt_stat.f_mntonname, (intmax_t)cmd.value); } #endif /* DIAGNOSTIC */ fs->fs_cstotal.cs_nifree += cmd.value; break; case FFS_ADJ_NFFREE: #ifdef DIAGNOSTIC if (fsckcmds) { printf("%s: adjust number of free frags by %+jd\n", mp->mnt_stat.f_mntonname, (intmax_t)cmd.value); } #endif /* DIAGNOSTIC */ fs->fs_cstotal.cs_nffree += cmd.value; break; case FFS_ADJ_NUMCLUSTERS: #ifdef DIAGNOSTIC if (fsckcmds) { printf("%s: adjust number of free clusters by %+jd\n", mp->mnt_stat.f_mntonname, (intmax_t)cmd.value); } #endif /* DIAGNOSTIC */ fs->fs_cstotal.cs_numclusters += cmd.value; break; case FFS_SET_CWD: #ifdef DIAGNOSTIC if (fsckcmds) { printf("%s: set current directory to inode %jd\n", mp->mnt_stat.f_mntonname, (intmax_t)cmd.value); } #endif /* DIAGNOSTIC */ if ((error = ffs_vget(mp, (ino_t)cmd.value, LK_SHARED, &vp))) break; AUDIT_ARG_VNODE1(vp); if ((error = change_dir(vp, td)) != 0) { vput(vp); break; } VOP_UNLOCK(vp); pwd_chdir(td, vp); break; case FFS_SET_DOTDOT: #ifdef DIAGNOSTIC if (fsckcmds) { printf("%s: change .. in cwd from %jd to %jd\n", mp->mnt_stat.f_mntonname, (intmax_t)cmd.value, (intmax_t)cmd.size); } #endif /* DIAGNOSTIC */ /* * First we have to get and lock the parent directory * to which ".." points. */ error = ffs_vget(mp, (ino_t)cmd.value, LK_EXCLUSIVE, &fdvp); if (error) break; /* * Now we get and lock the child directory containing "..". */ pwd = pwd_hold(td); dvp = pwd->pwd_cdir; if ((error = vget(dvp, LK_EXCLUSIVE, td)) != 0) { vput(fdvp); pwd_drop(pwd); break; } dp = VTOI(dvp); dp->i_offset = 12; /* XXX mastertemplate.dot_reclen */ error = ufs_dirrewrite(dp, VTOI(fdvp), (ino_t)cmd.size, DT_DIR, 0); cache_purge(fdvp); cache_purge(dvp); vput(dvp); vput(fdvp); pwd_drop(pwd); break; case FFS_UNLINK: #ifdef DIAGNOSTIC if (fsckcmds) { char buf[32]; if (copyinstr((char *)(intptr_t)cmd.value, buf,32,NULL)) strncpy(buf, "Name_too_long", 32); printf("%s: unlink %s (inode %jd)\n", mp->mnt_stat.f_mntonname, buf, (intmax_t)cmd.size); } #endif /* DIAGNOSTIC */ /* * kern_funlinkat will do its own start/finish writes and * they do not nest, so drop ours here. Setting mp == NULL * indicates that vn_finished_write is not needed down below. */ vn_finished_write(mp); mp = NULL; error = kern_funlinkat(td, AT_FDCWD, (char *)(intptr_t)cmd.value, FD_NONE, UIO_USERSPACE, 0, (ino_t)cmd.size); break; case FFS_SET_INODE: if (ump->um_fsckpid != td->td_proc->p_pid) { error = EPERM; break; } #ifdef DIAGNOSTIC if (fsckcmds) { printf("%s: update inode %jd\n", mp->mnt_stat.f_mntonname, (intmax_t)cmd.value); } #endif /* DIAGNOSTIC */ if ((error = ffs_vget(mp, (ino_t)cmd.value, LK_EXCLUSIVE, &vp))) break; AUDIT_ARG_VNODE1(vp); ip = VTOI(vp); if (I_IS_UFS1(ip)) error = copyin((void *)(intptr_t)cmd.size, ip->i_din1, sizeof(struct ufs1_dinode)); else error = copyin((void *)(intptr_t)cmd.size, ip->i_din2, sizeof(struct ufs2_dinode)); if (error) { vput(vp); break; } UFS_INODE_SET_FLAG(ip, IN_CHANGE | IN_MODIFIED); error = ffs_update(vp, 1); vput(vp); break; case FFS_SET_BUFOUTPUT: if (ump->um_fsckpid != td->td_proc->p_pid) { error = EPERM; break; } if (ITOUMP(VTOI(vp)) != ump) { error = EINVAL; break; } #ifdef DIAGNOSTIC if (fsckcmds) { printf("%s: %s buffered output for descriptor %jd\n", mp->mnt_stat.f_mntonname, cmd.size == 1 ? "enable" : "disable", (intmax_t)cmd.value); } #endif /* DIAGNOSTIC */ if ((error = getvnode(td, cmd.value, cap_rights_init(&rights, CAP_FSCK), &vfp)) != 0) break; if (vfp->f_vnode->v_type != VCHR) { fdrop(vfp, td); error = EINVAL; break; } if (origops == NULL) { origops = vfp->f_ops; bcopy((void *)origops, (void *)&bufferedops, sizeof(bufferedops)); bufferedops.fo_write = buffered_write; } if (cmd.size == 1) atomic_store_rel_ptr((volatile uintptr_t *)&vfp->f_ops, (uintptr_t)&bufferedops); else atomic_store_rel_ptr((volatile uintptr_t *)&vfp->f_ops, (uintptr_t)origops); fdrop(vfp, td); break; default: #ifdef DIAGNOSTIC if (fsckcmds) { printf("Invalid request %d from fsck\n", oidp->oid_number); } #endif /* DIAGNOSTIC */ error = EINVAL; break; } fdrop(fp, td); vn_finished_write(mp); return (error); } /* * Function to switch a descriptor to use the buffer cache to stage * its I/O. This is needed so that writes to the filesystem device * will give snapshots a chance to copy modified blocks for which it * needs to retain copies. */ static int buffered_write(fp, uio, active_cred, flags, td) struct file *fp; struct uio *uio; struct ucred *active_cred; int flags; struct thread *td; { struct vnode *devvp, *vp; struct inode *ip; struct buf *bp; struct fs *fs; + struct ufsmount *ump; struct filedesc *fdp; int error; daddr_t lbn; /* * The devvp is associated with the /dev filesystem. To discover * the filesystem with which the device is associated, we depend * on the application setting the current directory to a location * within the filesystem being written. Yes, this is an ugly hack. */ devvp = fp->f_vnode; if (!vn_isdisk(devvp, NULL)) return (EINVAL); fdp = td->td_proc->p_fd; FILEDESC_SLOCK(fdp); vp = fdp->fd_pwd->pwd_cdir; vref(vp); FILEDESC_SUNLOCK(fdp); vn_lock(vp, LK_SHARED | LK_RETRY); /* * Check that the current directory vnode indeed belongs to * UFS before trying to dereference UFS-specific v_data fields. */ if (vp->v_op != &ffs_vnodeops1 && vp->v_op != &ffs_vnodeops2) { vput(vp); return (EINVAL); } ip = VTOI(vp); - if (ITODEVVP(ip) != devvp) { + ump = ip->i_ump; + if (ump->um_odevvp != devvp) { vput(vp); return (EINVAL); } + devvp = ump->um_devvp; fs = ITOFS(ip); vput(vp); foffset_lock_uio(fp, uio, flags); vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); #ifdef DIAGNOSTIC if (fsckcmds) { printf("%s: buffered write for block %jd\n", fs->fs_fsmnt, (intmax_t)btodb(uio->uio_offset)); } #endif /* DIAGNOSTIC */ /* * All I/O must be contained within a filesystem block, start on * a fragment boundary, and be a multiple of fragments in length. */ if (uio->uio_resid > fs->fs_bsize - (uio->uio_offset % fs->fs_bsize) || fragoff(fs, uio->uio_offset) != 0 || fragoff(fs, uio->uio_resid) != 0) { error = EINVAL; goto out; } lbn = numfrags(fs, uio->uio_offset); bp = getblk(devvp, lbn, uio->uio_resid, 0, 0, 0); bp->b_flags |= B_RELBUF; if ((error = uiomove((char *)bp->b_data, uio->uio_resid, uio)) != 0) { brelse(bp); goto out; } error = bwrite(bp); out: VOP_UNLOCK(devvp); foffset_unlock_uio(fp, uio, flags | FOF_NEXTOFF); return (error); } diff --git a/sys/ufs/ffs/ffs_vfsops.c b/sys/ufs/ffs/ffs_vfsops.c index 5051c0a11eb8..1ae7fdc6184f 100644 --- a/sys/ufs/ffs/ffs_vfsops.c +++ b/sys/ufs/ffs/ffs_vfsops.c @@ -1,2408 +1,2434 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1989, 1991, 1993, 1994 * The Regents of the University of California. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)ffs_vfsops.c 8.31 (Berkeley) 5/20/95 */ #include __FBSDID("$FreeBSD$"); #include "opt_quota.h" #include "opt_ufs.h" #include "opt_ffs.h" #include "opt_ddb.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include static uma_zone_t uma_inode, uma_ufs1, uma_ufs2; static int ffs_mountfs(struct vnode *, struct mount *, struct thread *); static void ffs_oldfscompat_read(struct fs *, struct ufsmount *, ufs2_daddr_t); static void ffs_ifree(struct ufsmount *ump, struct inode *ip); static int ffs_sync_lazy(struct mount *mp); static int ffs_use_bread(void *devfd, off_t loc, void **bufp, int size); static int ffs_use_bwrite(void *devfd, off_t loc, void *buf, int size); static vfs_init_t ffs_init; static vfs_uninit_t ffs_uninit; static vfs_extattrctl_t ffs_extattrctl; static vfs_cmount_t ffs_cmount; static vfs_unmount_t ffs_unmount; static vfs_mount_t ffs_mount; static vfs_statfs_t ffs_statfs; static vfs_fhtovp_t ffs_fhtovp; static vfs_sync_t ffs_sync; static struct vfsops ufs_vfsops = { .vfs_extattrctl = ffs_extattrctl, .vfs_fhtovp = ffs_fhtovp, .vfs_init = ffs_init, .vfs_mount = ffs_mount, .vfs_cmount = ffs_cmount, .vfs_quotactl = ufs_quotactl, .vfs_root = vfs_cache_root, .vfs_cachedroot = ufs_root, .vfs_statfs = ffs_statfs, .vfs_sync = ffs_sync, .vfs_uninit = ffs_uninit, .vfs_unmount = ffs_unmount, .vfs_vget = ffs_vget, .vfs_susp_clean = process_deferred_inactive, }; VFS_SET(ufs_vfsops, ufs, 0); MODULE_VERSION(ufs, 1); static b_strategy_t ffs_geom_strategy; static b_write_t ffs_bufwrite; static struct buf_ops ffs_ops = { .bop_name = "FFS", .bop_write = ffs_bufwrite, .bop_strategy = ffs_geom_strategy, .bop_sync = bufsync, #ifdef NO_FFS_SNAPSHOT .bop_bdflush = bufbdflush, #else .bop_bdflush = ffs_bdflush, #endif }; /* * Note that userquota and groupquota options are not currently used * by UFS/FFS code and generally mount(8) does not pass those options * from userland, but they can be passed by loader(8) via * vfs.root.mountfrom.options. */ static const char *ffs_opts[] = { "acls", "async", "noatime", "noclusterr", "noclusterw", "noexec", "export", "force", "from", "groupquota", "multilabel", "nfsv4acls", "fsckpid", "snapshot", "nosuid", "suiddir", "nosymfollow", "sync", "union", "userquota", "untrusted", NULL }; static int ffs_mount(struct mount *mp) { - struct vnode *devvp; + struct vnode *devvp, *odevvp; struct thread *td; struct ufsmount *ump = NULL; struct fs *fs; pid_t fsckpid = 0; int error, error1, flags; uint64_t mntorflags, saved_mnt_flag; accmode_t accmode; struct nameidata ndp; char *fspec; td = curthread; if (vfs_filteropt(mp->mnt_optnew, ffs_opts)) return (EINVAL); if (uma_inode == NULL) { uma_inode = uma_zcreate("FFS inode", sizeof(struct inode), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); uma_ufs1 = uma_zcreate("FFS1 dinode", sizeof(struct ufs1_dinode), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); uma_ufs2 = uma_zcreate("FFS2 dinode", sizeof(struct ufs2_dinode), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0); } vfs_deleteopt(mp->mnt_optnew, "groupquota"); vfs_deleteopt(mp->mnt_optnew, "userquota"); fspec = vfs_getopts(mp->mnt_optnew, "from", &error); if (error) return (error); mntorflags = 0; if (vfs_getopt(mp->mnt_optnew, "untrusted", NULL, NULL) == 0) mntorflags |= MNT_UNTRUSTED; if (vfs_getopt(mp->mnt_optnew, "acls", NULL, NULL) == 0) mntorflags |= MNT_ACLS; if (vfs_getopt(mp->mnt_optnew, "snapshot", NULL, NULL) == 0) { mntorflags |= MNT_SNAPSHOT; /* * Once we have set the MNT_SNAPSHOT flag, do not * persist "snapshot" in the options list. */ vfs_deleteopt(mp->mnt_optnew, "snapshot"); vfs_deleteopt(mp->mnt_opt, "snapshot"); } if (vfs_getopt(mp->mnt_optnew, "fsckpid", NULL, NULL) == 0 && vfs_scanopt(mp->mnt_optnew, "fsckpid", "%d", &fsckpid) == 1) { /* * Once we have set the restricted PID, do not * persist "fsckpid" in the options list. */ vfs_deleteopt(mp->mnt_optnew, "fsckpid"); vfs_deleteopt(mp->mnt_opt, "fsckpid"); if (mp->mnt_flag & MNT_UPDATE) { if (VFSTOUFS(mp)->um_fs->fs_ronly == 0 && vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0) == 0) { vfs_mount_error(mp, "Checker enable: Must be read-only"); return (EINVAL); } } else if (vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0) == 0) { vfs_mount_error(mp, "Checker enable: Must be read-only"); return (EINVAL); } /* Set to -1 if we are done */ if (fsckpid == 0) fsckpid = -1; } if (vfs_getopt(mp->mnt_optnew, "nfsv4acls", NULL, NULL) == 0) { if (mntorflags & MNT_ACLS) { vfs_mount_error(mp, "\"acls\" and \"nfsv4acls\" options " "are mutually exclusive"); return (EINVAL); } mntorflags |= MNT_NFS4ACLS; } MNT_ILOCK(mp); mp->mnt_flag |= mntorflags; MNT_IUNLOCK(mp); /* * If updating, check whether changing from read-only to * read/write; if there is no device name, that's all we do. */ if (mp->mnt_flag & MNT_UPDATE) { ump = VFSTOUFS(mp); fs = ump->um_fs; + odevvp = ump->um_odevvp; devvp = ump->um_devvp; if (fsckpid == -1 && ump->um_fsckpid > 0) { if ((error = ffs_flushfiles(mp, WRITECLOSE, td)) != 0 || (error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) return (error); g_topology_lock(); /* * Return to normal read-only mode. */ error = g_access(ump->um_cp, 0, -1, 0); g_topology_unlock(); ump->um_fsckpid = 0; } if (fs->fs_ronly == 0 && vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) { /* * Flush any dirty data and suspend filesystem. */ if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0) return (error); error = vfs_write_suspend_umnt(mp); if (error != 0) return (error); /* * Check for and optionally get rid of files open * for writing. */ flags = WRITECLOSE; if (mp->mnt_flag & MNT_FORCE) flags |= FORCECLOSE; if (MOUNTEDSOFTDEP(mp)) { error = softdep_flushfiles(mp, flags, td); } else { error = ffs_flushfiles(mp, flags, td); } if (error) { vfs_write_resume(mp, 0); return (error); } if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) { printf("WARNING: %s Update error: blocks %jd " "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks, fs->fs_pendinginodes); fs->fs_pendingblocks = 0; fs->fs_pendinginodes = 0; } if ((fs->fs_flags & (FS_UNCLEAN | FS_NEEDSFSCK)) == 0) fs->fs_clean = 1; if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) { fs->fs_ronly = 0; fs->fs_clean = 0; vfs_write_resume(mp, 0); return (error); } if (MOUNTEDSOFTDEP(mp)) softdep_unmount(mp); g_topology_lock(); /* * Drop our write and exclusive access. */ g_access(ump->um_cp, 0, -1, -1); g_topology_unlock(); fs->fs_ronly = 1; MNT_ILOCK(mp); mp->mnt_flag |= MNT_RDONLY; MNT_IUNLOCK(mp); /* * Allow the writers to note that filesystem * is ro now. */ vfs_write_resume(mp, 0); } if ((mp->mnt_flag & MNT_RELOAD) && (error = ffs_reload(mp, td, 0)) != 0) return (error); if (fs->fs_ronly && !vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) { /* * If we are running a checker, do not allow upgrade. */ if (ump->um_fsckpid > 0) { vfs_mount_error(mp, "Active checker, cannot upgrade to write"); return (EINVAL); } /* * If upgrade to read-write by non-root, then verify * that user has necessary permissions on the device. */ - vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); - error = VOP_ACCESS(devvp, VREAD | VWRITE, + vn_lock(odevvp, LK_EXCLUSIVE | LK_RETRY); + error = VOP_ACCESS(odevvp, VREAD | VWRITE, td->td_ucred, td); if (error) error = priv_check(td, PRIV_VFS_MOUNT_PERM); + VOP_UNLOCK(odevvp); if (error) { - VOP_UNLOCK(devvp); return (error); } - VOP_UNLOCK(devvp); fs->fs_flags &= ~FS_UNCLEAN; if (fs->fs_clean == 0) { fs->fs_flags |= FS_UNCLEAN; if ((mp->mnt_flag & MNT_FORCE) || ((fs->fs_flags & (FS_SUJ | FS_NEEDSFSCK)) == 0 && (fs->fs_flags & FS_DOSOFTDEP))) { printf("WARNING: %s was not properly " "dismounted\n", fs->fs_fsmnt); } else { vfs_mount_error(mp, "R/W mount of %s denied. %s.%s", fs->fs_fsmnt, "Filesystem is not clean - run fsck", (fs->fs_flags & FS_SUJ) == 0 ? "" : " Forced mount will invalidate" " journal contents"); return (EPERM); } } g_topology_lock(); /* * Request exclusive write access. */ error = g_access(ump->um_cp, 0, 1, 1); g_topology_unlock(); if (error) return (error); if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0) return (error); error = vfs_write_suspend_umnt(mp); if (error != 0) return (error); fs->fs_ronly = 0; MNT_ILOCK(mp); saved_mnt_flag = MNT_RDONLY; if (MOUNTEDSOFTDEP(mp) && (mp->mnt_flag & MNT_ASYNC) != 0) saved_mnt_flag |= MNT_ASYNC; mp->mnt_flag &= ~saved_mnt_flag; MNT_IUNLOCK(mp); fs->fs_mtime = time_second; /* check to see if we need to start softdep */ if ((fs->fs_flags & FS_DOSOFTDEP) && (error = softdep_mount(devvp, mp, fs, td->td_ucred))){ fs->fs_ronly = 1; MNT_ILOCK(mp); mp->mnt_flag |= saved_mnt_flag; MNT_IUNLOCK(mp); vfs_write_resume(mp, 0); return (error); } fs->fs_clean = 0; if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) { fs->fs_ronly = 1; MNT_ILOCK(mp); mp->mnt_flag |= saved_mnt_flag; MNT_IUNLOCK(mp); vfs_write_resume(mp, 0); return (error); } if (fs->fs_snapinum[0] != 0) ffs_snapshot_mount(mp); vfs_write_resume(mp, 0); } /* * Soft updates is incompatible with "async", * so if we are doing softupdates stop the user * from setting the async flag in an update. * Softdep_mount() clears it in an initial mount * or ro->rw remount. */ if (MOUNTEDSOFTDEP(mp)) { /* XXX: Reset too late ? */ MNT_ILOCK(mp); mp->mnt_flag &= ~MNT_ASYNC; MNT_IUNLOCK(mp); } /* * Keep MNT_ACLS flag if it is stored in superblock. */ if ((fs->fs_flags & FS_ACLS) != 0) { /* XXX: Set too late ? */ MNT_ILOCK(mp); mp->mnt_flag |= MNT_ACLS; MNT_IUNLOCK(mp); } if ((fs->fs_flags & FS_NFS4ACLS) != 0) { /* XXX: Set too late ? */ MNT_ILOCK(mp); mp->mnt_flag |= MNT_NFS4ACLS; MNT_IUNLOCK(mp); } /* * If this is a request from fsck to clean up the filesystem, * then allow the specified pid to proceed. */ if (fsckpid > 0) { if (ump->um_fsckpid != 0) { vfs_mount_error(mp, "Active checker already running on %s", fs->fs_fsmnt); return (EINVAL); } KASSERT(MOUNTEDSOFTDEP(mp) == 0, ("soft updates enabled on read-only file system")); g_topology_lock(); /* * Request write access. */ error = g_access(ump->um_cp, 0, 1, 0); g_topology_unlock(); if (error) { vfs_mount_error(mp, "Checker activation failed on %s", fs->fs_fsmnt); return (error); } ump->um_fsckpid = fsckpid; if (fs->fs_snapinum[0] != 0) ffs_snapshot_mount(mp); fs->fs_mtime = time_second; fs->fs_fmod = 1; fs->fs_clean = 0; (void) ffs_sbupdate(ump, MNT_WAIT, 0); } /* * If this is a snapshot request, take the snapshot. */ if (mp->mnt_flag & MNT_SNAPSHOT) return (ffs_snapshot(mp, fspec)); /* * Must not call namei() while owning busy ref. */ vfs_unbusy(mp); } /* * Not an update, or updating the name: look up the name * and verify that it refers to a sensible disk device. */ NDINIT(&ndp, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fspec, td); error = namei(&ndp); if ((mp->mnt_flag & MNT_UPDATE) != 0) { /* * Unmount does not start if MNT_UPDATE is set. Mount * update busies mp before setting MNT_UPDATE. We * must be able to retain our busy ref succesfully, * without sleep. */ error1 = vfs_busy(mp, MBF_NOWAIT); MPASS(error1 == 0); } if (error != 0) return (error); NDFREE(&ndp, NDF_ONLY_PNBUF); devvp = ndp.ni_vp; if (!vn_isdisk(devvp, &error)) { vput(devvp); return (error); } /* * If mount by non-root, then verify that user has necessary * permissions on the device. */ accmode = VREAD; if ((mp->mnt_flag & MNT_RDONLY) == 0) accmode |= VWRITE; error = VOP_ACCESS(devvp, accmode, td->td_ucred, td); if (error) error = priv_check(td, PRIV_VFS_MOUNT_PERM); if (error) { vput(devvp); return (error); } if (mp->mnt_flag & MNT_UPDATE) { /* * Update only * * If it's not the same vnode, or at least the same device * then it's not correct. */ if (devvp->v_rdev != ump->um_devvp->v_rdev) error = EINVAL; /* needs translation */ vput(devvp); if (error) return (error); } else { /* * New mount * * We need the name for the mount point (also used for * "last mounted on") copied in. If an error occurs, * the mount point is discarded by the upper level code. * Note that vfs_mount_alloc() populates f_mntonname for us. */ if ((error = ffs_mountfs(devvp, mp, td)) != 0) { vrele(devvp); return (error); } if (fsckpid > 0) { KASSERT(MOUNTEDSOFTDEP(mp) == 0, ("soft updates enabled on read-only file system")); ump = VFSTOUFS(mp); fs = ump->um_fs; g_topology_lock(); /* * Request write access. */ error = g_access(ump->um_cp, 0, 1, 0); g_topology_unlock(); if (error) { printf("WARNING: %s: Checker activation " "failed\n", fs->fs_fsmnt); } else { ump->um_fsckpid = fsckpid; if (fs->fs_snapinum[0] != 0) ffs_snapshot_mount(mp); fs->fs_mtime = time_second; fs->fs_clean = 0; (void) ffs_sbupdate(ump, MNT_WAIT, 0); } } } vfs_mountedfrom(mp, fspec); return (0); } /* * Compatibility with old mount system call. */ static int ffs_cmount(struct mntarg *ma, void *data, uint64_t flags) { struct ufs_args args; struct export_args exp; int error; if (data == NULL) return (EINVAL); error = copyin(data, &args, sizeof args); if (error) return (error); vfs_oexport_conv(&args.export, &exp); ma = mount_argsu(ma, "from", args.fspec, MAXPATHLEN); ma = mount_arg(ma, "export", &exp, sizeof(exp)); error = kernel_mount(ma, flags); return (error); } /* * Reload all incore data for a filesystem (used after running fsck on * the root filesystem and finding things to fix). If the 'force' flag * is 0, the filesystem must be mounted read-only. * * Things to do to update the mount: * 1) invalidate all cached meta-data. * 2) re-read superblock from disk. * 3) re-read summary information from disk. * 4) invalidate all inactive vnodes. * 5) clear MNTK_SUSPEND2 and MNTK_SUSPENDED flags, allowing secondary * writers, if requested. * 6) invalidate all cached file data. * 7) re-read inode data for all active vnodes. */ int ffs_reload(struct mount *mp, struct thread *td, int flags) { struct vnode *vp, *mvp, *devvp; struct inode *ip; void *space; struct buf *bp; struct fs *fs, *newfs; struct ufsmount *ump; ufs2_daddr_t sblockloc; int i, blks, error; u_long size; int32_t *lp; ump = VFSTOUFS(mp); MNT_ILOCK(mp); if ((mp->mnt_flag & MNT_RDONLY) == 0 && (flags & FFSR_FORCE) == 0) { MNT_IUNLOCK(mp); return (EINVAL); } MNT_IUNLOCK(mp); /* * Step 1: invalidate all cached meta-data. */ devvp = VFSTOUFS(mp)->um_devvp; vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); if (vinvalbuf(devvp, 0, 0, 0) != 0) panic("ffs_reload: dirty1"); VOP_UNLOCK(devvp); /* * Step 2: re-read superblock from disk. */ fs = VFSTOUFS(mp)->um_fs; if ((error = bread(devvp, btodb(fs->fs_sblockloc), fs->fs_sbsize, NOCRED, &bp)) != 0) return (error); newfs = (struct fs *)bp->b_data; if ((newfs->fs_magic != FS_UFS1_MAGIC && newfs->fs_magic != FS_UFS2_MAGIC) || newfs->fs_bsize > MAXBSIZE || newfs->fs_bsize < sizeof(struct fs)) { brelse(bp); return (EIO); /* XXX needs translation */ } /* * Copy pointer fields back into superblock before copying in XXX * new superblock. These should really be in the ufsmount. XXX * Note that important parameters (eg fs_ncg) are unchanged. */ newfs->fs_csp = fs->fs_csp; newfs->fs_maxcluster = fs->fs_maxcluster; newfs->fs_contigdirs = fs->fs_contigdirs; newfs->fs_active = fs->fs_active; newfs->fs_ronly = fs->fs_ronly; sblockloc = fs->fs_sblockloc; bcopy(newfs, fs, (u_int)fs->fs_sbsize); brelse(bp); mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen; ffs_oldfscompat_read(fs, VFSTOUFS(mp), sblockloc); UFS_LOCK(ump); if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) { printf("WARNING: %s: reload pending error: blocks %jd " "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks, fs->fs_pendinginodes); fs->fs_pendingblocks = 0; fs->fs_pendinginodes = 0; } UFS_UNLOCK(ump); /* * Step 3: re-read summary information from disk. */ size = fs->fs_cssize; blks = howmany(size, fs->fs_fsize); if (fs->fs_contigsumsize > 0) size += fs->fs_ncg * sizeof(int32_t); size += fs->fs_ncg * sizeof(u_int8_t); free(fs->fs_csp, M_UFSMNT); space = malloc(size, M_UFSMNT, M_WAITOK); fs->fs_csp = space; for (i = 0; i < blks; i += fs->fs_frag) { size = fs->fs_bsize; if (i + fs->fs_frag > blks) size = (blks - i) * fs->fs_fsize; error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size, NOCRED, &bp); if (error) return (error); bcopy(bp->b_data, space, (u_int)size); space = (char *)space + size; brelse(bp); } /* * We no longer know anything about clusters per cylinder group. */ if (fs->fs_contigsumsize > 0) { fs->fs_maxcluster = lp = space; for (i = 0; i < fs->fs_ncg; i++) *lp++ = fs->fs_contigsumsize; space = lp; } size = fs->fs_ncg * sizeof(u_int8_t); fs->fs_contigdirs = (u_int8_t *)space; bzero(fs->fs_contigdirs, size); if ((flags & FFSR_UNSUSPEND) != 0) { MNT_ILOCK(mp); mp->mnt_kern_flag &= ~(MNTK_SUSPENDED | MNTK_SUSPEND2); wakeup(&mp->mnt_flag); MNT_IUNLOCK(mp); } loop: MNT_VNODE_FOREACH_ALL(vp, mp, mvp) { /* * Skip syncer vnode. */ if (vp->v_type == VNON) { VI_UNLOCK(vp); continue; } /* * Step 4: invalidate all cached file data. */ if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td)) { MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); goto loop; } if (vinvalbuf(vp, 0, 0, 0)) panic("ffs_reload: dirty2"); /* * Step 5: re-read inode data for all active vnodes. */ ip = VTOI(vp); error = bread(devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)), (int)fs->fs_bsize, NOCRED, &bp); if (error) { vput(vp); MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); return (error); } if ((error = ffs_load_inode(bp, ip, fs, ip->i_number)) != 0) { brelse(bp); vput(vp); MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); return (error); } ip->i_effnlink = ip->i_nlink; brelse(bp); vput(vp); } return (0); } /* * Common code for mount and mountroot */ static int -ffs_mountfs(devvp, mp, td) - struct vnode *devvp; +ffs_mountfs(odevvp, mp, td) + struct vnode *odevvp; struct mount *mp; struct thread *td; { struct ufsmount *ump; struct fs *fs; struct cdev *dev; int error, i, len, ronly; struct ucred *cred; struct g_consumer *cp; struct mount *nmp; + struct vnode *devvp; int candelete, canspeedup; off_t loc; fs = NULL; ump = NULL; cred = td ? td->td_ucred : NOCRED; ronly = (mp->mnt_flag & MNT_RDONLY) != 0; + devvp = mntfs_allocvp(mp, odevvp); + VOP_UNLOCK(odevvp); KASSERT(devvp->v_type == VCHR, ("reclaimed devvp")); dev = devvp->v_rdev; if (atomic_cmpset_acq_ptr((uintptr_t *)&dev->si_mountpt, 0, (uintptr_t)mp) == 0) { - VOP_UNLOCK(devvp); + mntfs_freevp(devvp); return (EBUSY); } g_topology_lock(); error = g_vfs_open(devvp, &cp, "ffs", ronly ? 0 : 1); g_topology_unlock(); if (error != 0) { atomic_store_rel_ptr((uintptr_t *)&dev->si_mountpt, 0); - VOP_UNLOCK(devvp); + mntfs_freevp(devvp); return (error); } dev_ref(dev); devvp->v_bufobj.bo_ops = &ffs_ops; - VOP_UNLOCK(devvp); + BO_LOCK(&odevvp->v_bufobj); + odevvp->v_bufobj.bo_flag |= BO_NOBUFS; + BO_UNLOCK(&odevvp->v_bufobj); if (dev->si_iosize_max != 0) mp->mnt_iosize_max = dev->si_iosize_max; if (mp->mnt_iosize_max > MAXPHYS) mp->mnt_iosize_max = MAXPHYS; if ((SBLOCKSIZE % cp->provider->sectorsize) != 0) { error = EINVAL; vfs_mount_error(mp, "Invalid sectorsize %d for superblock size %d", cp->provider->sectorsize, SBLOCKSIZE); goto out; } /* fetch the superblock and summary information */ loc = STDSB; if ((mp->mnt_flag & MNT_ROOTFS) != 0) loc = STDSB_NOHASHFAIL; if ((error = ffs_sbget(devvp, &fs, loc, M_UFSMNT, ffs_use_bread)) != 0) goto out; /* none of these types of check-hashes are maintained by this kernel */ fs->fs_metackhash &= ~(CK_INDIR | CK_DIR); /* no support for any undefined flags */ fs->fs_flags &= FS_SUPPORTED; fs->fs_flags &= ~FS_UNCLEAN; if (fs->fs_clean == 0) { fs->fs_flags |= FS_UNCLEAN; if (ronly || (mp->mnt_flag & MNT_FORCE) || ((fs->fs_flags & (FS_SUJ | FS_NEEDSFSCK)) == 0 && (fs->fs_flags & FS_DOSOFTDEP))) { printf("WARNING: %s was not properly dismounted\n", fs->fs_fsmnt); } else { vfs_mount_error(mp, "R/W mount of %s denied. %s%s", fs->fs_fsmnt, "Filesystem is not clean - run fsck.", (fs->fs_flags & FS_SUJ) == 0 ? "" : " Forced mount will invalidate journal contents"); error = EPERM; goto out; } if ((fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) && (mp->mnt_flag & MNT_FORCE)) { printf("WARNING: %s: lost blocks %jd files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks, fs->fs_pendinginodes); fs->fs_pendingblocks = 0; fs->fs_pendinginodes = 0; } } if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) { printf("WARNING: %s: mount pending error: blocks %jd " "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks, fs->fs_pendinginodes); fs->fs_pendingblocks = 0; fs->fs_pendinginodes = 0; } if ((fs->fs_flags & FS_GJOURNAL) != 0) { #ifdef UFS_GJOURNAL /* * Get journal provider name. */ len = 1024; mp->mnt_gjprovider = malloc((u_long)len, M_UFSMNT, M_WAITOK); if (g_io_getattr("GJOURNAL::provider", cp, &len, mp->mnt_gjprovider) == 0) { mp->mnt_gjprovider = realloc(mp->mnt_gjprovider, len, M_UFSMNT, M_WAITOK); MNT_ILOCK(mp); mp->mnt_flag |= MNT_GJOURNAL; MNT_IUNLOCK(mp); } else { printf("WARNING: %s: GJOURNAL flag on fs " "but no gjournal provider below\n", mp->mnt_stat.f_mntonname); free(mp->mnt_gjprovider, M_UFSMNT); mp->mnt_gjprovider = NULL; } #else printf("WARNING: %s: GJOURNAL flag on fs but no " "UFS_GJOURNAL support\n", mp->mnt_stat.f_mntonname); #endif } else { mp->mnt_gjprovider = NULL; } ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK | M_ZERO); ump->um_cp = cp; ump->um_bo = &devvp->v_bufobj; ump->um_fs = fs; if (fs->fs_magic == FS_UFS1_MAGIC) { ump->um_fstype = UFS1; ump->um_balloc = ffs_balloc_ufs1; } else { ump->um_fstype = UFS2; ump->um_balloc = ffs_balloc_ufs2; } ump->um_blkatoff = ffs_blkatoff; ump->um_truncate = ffs_truncate; ump->um_update = ffs_update; ump->um_valloc = ffs_valloc; ump->um_vfree = ffs_vfree; ump->um_ifree = ffs_ifree; ump->um_rdonly = ffs_rdonly; ump->um_snapgone = ffs_snapgone; if ((mp->mnt_flag & MNT_UNTRUSTED) != 0) ump->um_check_blkno = ffs_check_blkno; else ump->um_check_blkno = NULL; mtx_init(UFS_MTX(ump), "FFS", "FFS Lock", MTX_DEF); ffs_oldfscompat_read(fs, ump, fs->fs_sblockloc); fs->fs_ronly = ronly; fs->fs_active = NULL; mp->mnt_data = ump; mp->mnt_stat.f_fsid.val[0] = fs->fs_id[0]; mp->mnt_stat.f_fsid.val[1] = fs->fs_id[1]; nmp = NULL; if (fs->fs_id[0] == 0 || fs->fs_id[1] == 0 || (nmp = vfs_getvfs(&mp->mnt_stat.f_fsid))) { if (nmp) vfs_rel(nmp); vfs_getnewfsid(mp); } mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen; MNT_ILOCK(mp); mp->mnt_flag |= MNT_LOCAL; MNT_IUNLOCK(mp); if ((fs->fs_flags & FS_MULTILABEL) != 0) { #ifdef MAC MNT_ILOCK(mp); mp->mnt_flag |= MNT_MULTILABEL; MNT_IUNLOCK(mp); #else printf("WARNING: %s: multilabel flag on fs but " "no MAC support\n", mp->mnt_stat.f_mntonname); #endif } if ((fs->fs_flags & FS_ACLS) != 0) { #ifdef UFS_ACL MNT_ILOCK(mp); if (mp->mnt_flag & MNT_NFS4ACLS) printf("WARNING: %s: ACLs flag on fs conflicts with " "\"nfsv4acls\" mount option; option ignored\n", mp->mnt_stat.f_mntonname); mp->mnt_flag &= ~MNT_NFS4ACLS; mp->mnt_flag |= MNT_ACLS; MNT_IUNLOCK(mp); #else printf("WARNING: %s: ACLs flag on fs but no ACLs support\n", mp->mnt_stat.f_mntonname); #endif } if ((fs->fs_flags & FS_NFS4ACLS) != 0) { #ifdef UFS_ACL MNT_ILOCK(mp); if (mp->mnt_flag & MNT_ACLS) printf("WARNING: %s: NFSv4 ACLs flag on fs conflicts " "with \"acls\" mount option; option ignored\n", mp->mnt_stat.f_mntonname); mp->mnt_flag &= ~MNT_ACLS; mp->mnt_flag |= MNT_NFS4ACLS; MNT_IUNLOCK(mp); #else printf("WARNING: %s: NFSv4 ACLs flag on fs but no " "ACLs support\n", mp->mnt_stat.f_mntonname); #endif } if ((fs->fs_flags & FS_TRIM) != 0) { len = sizeof(int); if (g_io_getattr("GEOM::candelete", cp, &len, &candelete) == 0) { if (candelete) ump->um_flags |= UM_CANDELETE; else printf("WARNING: %s: TRIM flag on fs but disk " "does not support TRIM\n", mp->mnt_stat.f_mntonname); } else { printf("WARNING: %s: TRIM flag on fs but disk does " "not confirm that it supports TRIM\n", mp->mnt_stat.f_mntonname); } if (((ump->um_flags) & UM_CANDELETE) != 0) { ump->um_trim_tq = taskqueue_create("trim", M_WAITOK, taskqueue_thread_enqueue, &ump->um_trim_tq); taskqueue_start_threads(&ump->um_trim_tq, 1, PVFS, "%s trim", mp->mnt_stat.f_mntonname); ump->um_trimhash = hashinit(MAXTRIMIO, M_TRIM, &ump->um_trimlisthashsize); } } len = sizeof(int); if (g_io_getattr("GEOM::canspeedup", cp, &len, &canspeedup) == 0) { if (canspeedup) ump->um_flags |= UM_CANSPEEDUP; } ump->um_mountp = mp; ump->um_dev = dev; ump->um_devvp = devvp; + ump->um_odevvp = odevvp; ump->um_nindir = fs->fs_nindir; ump->um_bptrtodb = fs->fs_fsbtodb; ump->um_seqinc = fs->fs_frag; for (i = 0; i < MAXQUOTAS; i++) ump->um_quotas[i] = NULLVP; #ifdef UFS_EXTATTR ufs_extattr_uepm_init(&ump->um_extattr); #endif /* * Set FS local "last mounted on" information (NULL pad) */ bzero(fs->fs_fsmnt, MAXMNTLEN); strlcpy(fs->fs_fsmnt, mp->mnt_stat.f_mntonname, MAXMNTLEN); mp->mnt_stat.f_iosize = fs->fs_bsize; if (mp->mnt_flag & MNT_ROOTFS) { /* * Root mount; update timestamp in mount structure. * this will be used by the common root mount code * to update the system clock. */ mp->mnt_time = fs->fs_time; } if (ronly == 0) { fs->fs_mtime = time_second; if ((fs->fs_flags & FS_DOSOFTDEP) && (error = softdep_mount(devvp, mp, fs, cred)) != 0) { ffs_flushfiles(mp, FORCECLOSE, td); goto out; } if (fs->fs_snapinum[0] != 0) ffs_snapshot_mount(mp); fs->fs_fmod = 1; fs->fs_clean = 0; (void) ffs_sbupdate(ump, MNT_WAIT, 0); } /* * Initialize filesystem state information in mount struct. */ MNT_ILOCK(mp); mp->mnt_kern_flag |= MNTK_LOOKUP_SHARED | MNTK_EXTENDED_SHARED | MNTK_NO_IOPF | MNTK_UNMAPPED_BUFS | MNTK_USES_BCACHE; MNT_IUNLOCK(mp); #ifdef UFS_EXTATTR #ifdef UFS_EXTATTR_AUTOSTART /* * * Auto-starting does the following: * - check for /.attribute in the fs, and extattr_start if so * - for each file in .attribute, enable that file with * an attribute of the same name. * Not clear how to report errors -- probably eat them. * This would all happen while the filesystem was busy/not * available, so would effectively be "atomic". */ (void) ufs_extattr_autostart(mp, td); #endif /* !UFS_EXTATTR_AUTOSTART */ #endif /* !UFS_EXTATTR */ return (0); out: if (fs != NULL) { free(fs->fs_csp, M_UFSMNT); free(fs, M_UFSMNT); } if (cp != NULL) { g_topology_lock(); g_vfs_close(cp); g_topology_unlock(); } if (ump) { mtx_destroy(UFS_MTX(ump)); if (mp->mnt_gjprovider != NULL) { free(mp->mnt_gjprovider, M_UFSMNT); mp->mnt_gjprovider = NULL; } free(ump, M_UFSMNT); mp->mnt_data = NULL; } + BO_LOCK(&odevvp->v_bufobj); + odevvp->v_bufobj.bo_flag &= ~BO_NOBUFS; + BO_UNLOCK(&odevvp->v_bufobj); atomic_store_rel_ptr((uintptr_t *)&dev->si_mountpt, 0); + mntfs_freevp(devvp); dev_rel(dev); return (error); } /* * A read function for use by filesystem-layer routines. */ static int ffs_use_bread(void *devfd, off_t loc, void **bufp, int size) { struct buf *bp; int error; KASSERT(*bufp == NULL, ("ffs_use_bread: non-NULL *bufp %p\n", *bufp)); *bufp = malloc(size, M_UFSMNT, M_WAITOK); if ((error = bread((struct vnode *)devfd, btodb(loc), size, NOCRED, &bp)) != 0) return (error); bcopy(bp->b_data, *bufp, size); bp->b_flags |= B_INVAL | B_NOCACHE; brelse(bp); return (0); } #include static int bigcgs = 0; SYSCTL_INT(_debug, OID_AUTO, bigcgs, CTLFLAG_RW, &bigcgs, 0, ""); /* * Sanity checks for loading old filesystem superblocks. * See ffs_oldfscompat_write below for unwound actions. * * XXX - Parts get retired eventually. * Unfortunately new bits get added. */ static void ffs_oldfscompat_read(fs, ump, sblockloc) struct fs *fs; struct ufsmount *ump; ufs2_daddr_t sblockloc; { off_t maxfilesize; /* * If not yet done, update fs_flags location and value of fs_sblockloc. */ if ((fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) { fs->fs_flags = fs->fs_old_flags; fs->fs_old_flags |= FS_FLAGS_UPDATED; fs->fs_sblockloc = sblockloc; } /* * If not yet done, update UFS1 superblock with new wider fields. */ if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_maxbsize != fs->fs_bsize) { fs->fs_maxbsize = fs->fs_bsize; fs->fs_time = fs->fs_old_time; fs->fs_size = fs->fs_old_size; fs->fs_dsize = fs->fs_old_dsize; fs->fs_csaddr = fs->fs_old_csaddr; fs->fs_cstotal.cs_ndir = fs->fs_old_cstotal.cs_ndir; fs->fs_cstotal.cs_nbfree = fs->fs_old_cstotal.cs_nbfree; fs->fs_cstotal.cs_nifree = fs->fs_old_cstotal.cs_nifree; fs->fs_cstotal.cs_nffree = fs->fs_old_cstotal.cs_nffree; } if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_old_inodefmt < FS_44INODEFMT) { fs->fs_maxfilesize = ((uint64_t)1 << 31) - 1; fs->fs_qbmask = ~fs->fs_bmask; fs->fs_qfmask = ~fs->fs_fmask; } if (fs->fs_magic == FS_UFS1_MAGIC) { ump->um_savedmaxfilesize = fs->fs_maxfilesize; maxfilesize = (uint64_t)0x80000000 * fs->fs_bsize - 1; if (fs->fs_maxfilesize > maxfilesize) fs->fs_maxfilesize = maxfilesize; } /* Compatibility for old filesystems */ if (fs->fs_avgfilesize <= 0) fs->fs_avgfilesize = AVFILESIZ; if (fs->fs_avgfpdir <= 0) fs->fs_avgfpdir = AFPDIR; if (bigcgs) { fs->fs_save_cgsize = fs->fs_cgsize; fs->fs_cgsize = fs->fs_bsize; } } /* * Unwinding superblock updates for old filesystems. * See ffs_oldfscompat_read above for details. * * XXX - Parts get retired eventually. * Unfortunately new bits get added. */ void ffs_oldfscompat_write(fs, ump) struct fs *fs; struct ufsmount *ump; { /* * Copy back UFS2 updated fields that UFS1 inspects. */ if (fs->fs_magic == FS_UFS1_MAGIC) { fs->fs_old_time = fs->fs_time; fs->fs_old_cstotal.cs_ndir = fs->fs_cstotal.cs_ndir; fs->fs_old_cstotal.cs_nbfree = fs->fs_cstotal.cs_nbfree; fs->fs_old_cstotal.cs_nifree = fs->fs_cstotal.cs_nifree; fs->fs_old_cstotal.cs_nffree = fs->fs_cstotal.cs_nffree; fs->fs_maxfilesize = ump->um_savedmaxfilesize; } if (bigcgs) { fs->fs_cgsize = fs->fs_save_cgsize; fs->fs_save_cgsize = 0; } } /* * unmount system call */ static int ffs_unmount(mp, mntflags) struct mount *mp; int mntflags; { struct thread *td; struct ufsmount *ump = VFSTOUFS(mp); struct fs *fs; int error, flags, susp; #ifdef UFS_EXTATTR int e_restart; #endif flags = 0; td = curthread; fs = ump->um_fs; susp = 0; if (mntflags & MNT_FORCE) { flags |= FORCECLOSE; susp = fs->fs_ronly == 0; } #ifdef UFS_EXTATTR if ((error = ufs_extattr_stop(mp, td))) { if (error != EOPNOTSUPP) printf("WARNING: unmount %s: ufs_extattr_stop " "returned errno %d\n", mp->mnt_stat.f_mntonname, error); e_restart = 0; } else { ufs_extattr_uepm_destroy(&ump->um_extattr); e_restart = 1; } #endif if (susp) { error = vfs_write_suspend_umnt(mp); if (error != 0) goto fail1; } if (MOUNTEDSOFTDEP(mp)) error = softdep_flushfiles(mp, flags, td); else error = ffs_flushfiles(mp, flags, td); if (error != 0 && error != ENXIO) goto fail; UFS_LOCK(ump); if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) { printf("WARNING: unmount %s: pending error: blocks %jd " "files %d\n", fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks, fs->fs_pendinginodes); fs->fs_pendingblocks = 0; fs->fs_pendinginodes = 0; } UFS_UNLOCK(ump); if (MOUNTEDSOFTDEP(mp)) softdep_unmount(mp); if (fs->fs_ronly == 0 || ump->um_fsckpid > 0) { fs->fs_clean = fs->fs_flags & (FS_UNCLEAN|FS_NEEDSFSCK) ? 0 : 1; error = ffs_sbupdate(ump, MNT_WAIT, 0); if (error && error != ENXIO) { fs->fs_clean = 0; goto fail; } } if (susp) vfs_write_resume(mp, VR_START_WRITE); if (ump->um_trim_tq != NULL) { while (ump->um_trim_inflight != 0) pause("ufsutr", hz); taskqueue_drain_all(ump->um_trim_tq); taskqueue_free(ump->um_trim_tq); free (ump->um_trimhash, M_TRIM); } g_topology_lock(); if (ump->um_fsckpid > 0) { /* * Return to normal read-only mode. */ error = g_access(ump->um_cp, 0, -1, 0); ump->um_fsckpid = 0; } g_vfs_close(ump->um_cp); g_topology_unlock(); + BO_LOCK(&ump->um_odevvp->v_bufobj); + ump->um_odevvp->v_bufobj.bo_flag &= ~BO_NOBUFS; + BO_UNLOCK(&ump->um_odevvp->v_bufobj); atomic_store_rel_ptr((uintptr_t *)&ump->um_dev->si_mountpt, 0); - vrele(ump->um_devvp); + mntfs_freevp(ump->um_devvp); + vrele(ump->um_odevvp); dev_rel(ump->um_dev); mtx_destroy(UFS_MTX(ump)); if (mp->mnt_gjprovider != NULL) { free(mp->mnt_gjprovider, M_UFSMNT); mp->mnt_gjprovider = NULL; } free(fs->fs_csp, M_UFSMNT); free(fs, M_UFSMNT); free(ump, M_UFSMNT); mp->mnt_data = NULL; MNT_ILOCK(mp); mp->mnt_flag &= ~MNT_LOCAL; MNT_IUNLOCK(mp); if (td->td_su == mp) { td->td_su = NULL; vfs_rel(mp); } return (error); fail: if (susp) vfs_write_resume(mp, VR_START_WRITE); fail1: #ifdef UFS_EXTATTR if (e_restart) { ufs_extattr_uepm_init(&ump->um_extattr); #ifdef UFS_EXTATTR_AUTOSTART (void) ufs_extattr_autostart(mp, td); #endif } #endif return (error); } /* * Flush out all the files in a filesystem. */ int ffs_flushfiles(mp, flags, td) struct mount *mp; int flags; struct thread *td; { struct ufsmount *ump; int qerror, error; ump = VFSTOUFS(mp); qerror = 0; #ifdef QUOTA if (mp->mnt_flag & MNT_QUOTA) { int i; error = vflush(mp, 0, SKIPSYSTEM|flags, td); if (error) return (error); for (i = 0; i < MAXQUOTAS; i++) { error = quotaoff(td, mp, i); if (error != 0) { if ((flags & EARLYFLUSH) == 0) return (error); else qerror = error; } } /* * Here we fall through to vflush again to ensure that * we have gotten rid of all the system vnodes, unless * quotas must not be closed. */ } #endif ASSERT_VOP_LOCKED(ump->um_devvp, "ffs_flushfiles"); if (ump->um_devvp->v_vflag & VV_COPYONWRITE) { if ((error = vflush(mp, 0, SKIPSYSTEM | flags, td)) != 0) return (error); ffs_snapshot_unmount(mp); flags |= FORCECLOSE; /* * Here we fall through to vflush again to ensure * that we have gotten rid of all the system vnodes. */ } /* * Do not close system files if quotas were not closed, to be * able to sync the remaining dquots. The freeblks softupdate * workitems might hold a reference on a dquot, preventing * quotaoff() from completing. Next round of * softdep_flushworklist() iteration should process the * blockers, allowing the next run of quotaoff() to finally * flush held dquots. * * Otherwise, flush all the files. */ if (qerror == 0 && (error = vflush(mp, 0, flags, td)) != 0) return (error); /* * Flush filesystem metadata. */ vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY); error = VOP_FSYNC(ump->um_devvp, MNT_WAIT, td); VOP_UNLOCK(ump->um_devvp); return (error); } /* * Get filesystem statistics. */ static int ffs_statfs(mp, sbp) struct mount *mp; struct statfs *sbp; { struct ufsmount *ump; struct fs *fs; ump = VFSTOUFS(mp); fs = ump->um_fs; if (fs->fs_magic != FS_UFS1_MAGIC && fs->fs_magic != FS_UFS2_MAGIC) panic("ffs_statfs"); sbp->f_version = STATFS_VERSION; sbp->f_bsize = fs->fs_fsize; sbp->f_iosize = fs->fs_bsize; sbp->f_blocks = fs->fs_dsize; UFS_LOCK(ump); sbp->f_bfree = fs->fs_cstotal.cs_nbfree * fs->fs_frag + fs->fs_cstotal.cs_nffree + dbtofsb(fs, fs->fs_pendingblocks); sbp->f_bavail = freespace(fs, fs->fs_minfree) + dbtofsb(fs, fs->fs_pendingblocks); sbp->f_files = fs->fs_ncg * fs->fs_ipg - UFS_ROOTINO; sbp->f_ffree = fs->fs_cstotal.cs_nifree + fs->fs_pendinginodes; UFS_UNLOCK(ump); sbp->f_namemax = UFS_MAXNAMLEN; return (0); } static bool sync_doupdate(struct inode *ip) { return ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) != 0); } static int ffs_sync_lazy_filter(struct vnode *vp, void *arg __unused) { struct inode *ip; /* * Flags are safe to access because ->v_data invalidation * is held off by listmtx. */ if (vp->v_type == VNON) return (false); ip = VTOI(vp); if (!sync_doupdate(ip) && (vp->v_iflag & VI_OWEINACT) == 0) return (false); return (true); } /* * For a lazy sync, we only care about access times, quotas and the * superblock. Other filesystem changes are already converted to * cylinder group blocks or inode blocks updates and are written to * disk by syncer. */ static int ffs_sync_lazy(mp) struct mount *mp; { struct vnode *mvp, *vp; struct inode *ip; struct thread *td; int allerror, error; allerror = 0; td = curthread; if ((mp->mnt_flag & MNT_NOATIME) != 0) { #ifdef QUOTA qsync(mp); #endif goto sbupdate; } MNT_VNODE_FOREACH_LAZY(vp, mp, mvp, ffs_sync_lazy_filter, NULL) { if (vp->v_type == VNON) { VI_UNLOCK(vp); continue; } ip = VTOI(vp); /* * The IN_ACCESS flag is converted to IN_MODIFIED by * ufs_close() and ufs_getattr() by the calls to * ufs_itimes_locked(), without subsequent UFS_UPDATE(). * Test also all the other timestamp flags too, to pick up * any other cases that could be missed. */ if (!sync_doupdate(ip) && (vp->v_iflag & VI_OWEINACT) == 0) { VI_UNLOCK(vp); continue; } if ((error = vget(vp, LK_EXCLUSIVE | LK_NOWAIT | LK_INTERLOCK, td)) != 0) continue; #ifdef QUOTA qsyncvp(vp); #endif if (sync_doupdate(ip)) error = ffs_update(vp, 0); if (error != 0) allerror = error; vput(vp); } sbupdate: if (VFSTOUFS(mp)->um_fs->fs_fmod != 0 && (error = ffs_sbupdate(VFSTOUFS(mp), MNT_LAZY, 0)) != 0) allerror = error; return (allerror); } /* * Go through the disk queues to initiate sandbagged IO; * go through the inodes to write those that have been modified; * initiate the writing of the super block if it has been modified. * * Note: we are always called with the filesystem marked busy using * vfs_busy(). */ static int ffs_sync(mp, waitfor) struct mount *mp; int waitfor; { struct vnode *mvp, *vp, *devvp; struct thread *td; struct inode *ip; struct ufsmount *ump = VFSTOUFS(mp); struct fs *fs; int error, count, lockreq, allerror = 0; int suspend; int suspended; int secondary_writes; int secondary_accwrites; int softdep_deps; int softdep_accdeps; struct bufobj *bo; suspend = 0; suspended = 0; td = curthread; fs = ump->um_fs; if (fs->fs_fmod != 0 && fs->fs_ronly != 0 && ump->um_fsckpid == 0) panic("%s: ffs_sync: modification on read-only filesystem", fs->fs_fsmnt); if (waitfor == MNT_LAZY) { if (!rebooting) return (ffs_sync_lazy(mp)); waitfor = MNT_NOWAIT; } /* * Write back each (modified) inode. */ lockreq = LK_EXCLUSIVE | LK_NOWAIT; if (waitfor == MNT_SUSPEND) { suspend = 1; waitfor = MNT_WAIT; } if (waitfor == MNT_WAIT) lockreq = LK_EXCLUSIVE; lockreq |= LK_INTERLOCK | LK_SLEEPFAIL; loop: /* Grab snapshot of secondary write counts */ MNT_ILOCK(mp); secondary_writes = mp->mnt_secondary_writes; secondary_accwrites = mp->mnt_secondary_accwrites; MNT_IUNLOCK(mp); /* Grab snapshot of softdep dependency counts */ softdep_get_depcounts(mp, &softdep_deps, &softdep_accdeps); MNT_VNODE_FOREACH_ALL(vp, mp, mvp) { /* * Depend on the vnode interlock to keep things stable enough * for a quick test. Since there might be hundreds of * thousands of vnodes, we cannot afford even a subroutine * call unless there's a good chance that we have work to do. */ if (vp->v_type == VNON) { VI_UNLOCK(vp); continue; } ip = VTOI(vp); if ((ip->i_flag & (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 && vp->v_bufobj.bo_dirty.bv_cnt == 0) { VI_UNLOCK(vp); continue; } if ((error = vget(vp, lockreq, td)) != 0) { if (error == ENOENT || error == ENOLCK) { MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp); goto loop; } continue; } #ifdef QUOTA qsyncvp(vp); #endif if ((error = ffs_syncvnode(vp, waitfor, 0)) != 0) allerror = error; vput(vp); } /* * Force stale filesystem control information to be flushed. */ if (waitfor == MNT_WAIT || rebooting) { if ((error = softdep_flushworklist(ump->um_mountp, &count, td))) allerror = error; /* Flushed work items may create new vnodes to clean */ if (allerror == 0 && count) goto loop; } devvp = ump->um_devvp; bo = &devvp->v_bufobj; BO_LOCK(bo); if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0) { BO_UNLOCK(bo); vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); error = VOP_FSYNC(devvp, waitfor, td); VOP_UNLOCK(devvp); if (MOUNTEDSOFTDEP(mp) && (error == 0 || error == EAGAIN)) error = ffs_sbupdate(ump, waitfor, 0); if (error != 0) allerror = error; if (allerror == 0 && waitfor == MNT_WAIT) goto loop; } else if (suspend != 0) { if (softdep_check_suspend(mp, devvp, softdep_deps, softdep_accdeps, secondary_writes, secondary_accwrites) != 0) { MNT_IUNLOCK(mp); goto loop; /* More work needed */ } mtx_assert(MNT_MTX(mp), MA_OWNED); mp->mnt_kern_flag |= MNTK_SUSPEND2 | MNTK_SUSPENDED; MNT_IUNLOCK(mp); suspended = 1; } else BO_UNLOCK(bo); /* * Write back modified superblock. */ if (fs->fs_fmod != 0 && (error = ffs_sbupdate(ump, waitfor, suspended)) != 0) allerror = error; return (allerror); } int ffs_vget(mp, ino, flags, vpp) struct mount *mp; ino_t ino; int flags; struct vnode **vpp; { return (ffs_vgetf(mp, ino, flags, vpp, 0)); } int ffs_vgetf(mp, ino, flags, vpp, ffs_flags) struct mount *mp; ino_t ino; int flags; struct vnode **vpp; int ffs_flags; { struct fs *fs; struct inode *ip; struct ufsmount *ump; struct buf *bp; struct vnode *vp; int error; MPASS((ffs_flags & FFSV_REPLACE) == 0 || (flags & LK_EXCLUSIVE) != 0); error = vfs_hash_get(mp, ino, flags, curthread, vpp, NULL, NULL); if (error != 0) return (error); if (*vpp != NULL) { if ((ffs_flags & FFSV_REPLACE) == 0) return (0); vgone(*vpp); vput(*vpp); } /* * We must promote to an exclusive lock for vnode creation. This * can happen if lookup is passed LOCKSHARED. */ if ((flags & LK_TYPE_MASK) == LK_SHARED) { flags &= ~LK_TYPE_MASK; flags |= LK_EXCLUSIVE; } /* * We do not lock vnode creation as it is believed to be too * expensive for such rare case as simultaneous creation of vnode * for same ino by different processes. We just allow them to race * and check later to decide who wins. Let the race begin! */ ump = VFSTOUFS(mp); fs = ump->um_fs; ip = uma_zalloc(uma_inode, M_WAITOK | M_ZERO); /* Allocate a new vnode/inode. */ error = getnewvnode("ufs", mp, fs->fs_magic == FS_UFS1_MAGIC ? &ffs_vnodeops1 : &ffs_vnodeops2, &vp); if (error) { *vpp = NULL; uma_zfree(uma_inode, ip); return (error); } /* * FFS supports recursive locking. */ lockmgr(vp->v_vnlock, LK_EXCLUSIVE, NULL); VN_LOCK_AREC(vp); vp->v_data = ip; vp->v_bufobj.bo_bsize = fs->fs_bsize; ip->i_vnode = vp; ip->i_ump = ump; ip->i_number = ino; ip->i_ea_refs = 0; ip->i_nextclustercg = -1; ip->i_flag = fs->fs_magic == FS_UFS1_MAGIC ? 0 : IN_UFS2; ip->i_mode = 0; /* ensure error cases below throw away vnode */ #ifdef QUOTA { int i; for (i = 0; i < MAXQUOTAS; i++) ip->i_dquot[i] = NODQUOT; } #endif if (ffs_flags & FFSV_FORCEINSMQ) vp->v_vflag |= VV_FORCEINSMQ; error = insmntque(vp, mp); if (error != 0) { uma_zfree(uma_inode, ip); *vpp = NULL; return (error); } vp->v_vflag &= ~VV_FORCEINSMQ; error = vfs_hash_insert(vp, ino, flags, curthread, vpp, NULL, NULL); if (error != 0) return (error); if (*vpp != NULL) { /* * Calls from ffs_valloc() (i.e. FFSV_REPLACE set) * operate on empty inode, which must not be found by * other threads until fully filled. Vnode for empty * inode must be not re-inserted on the hash by other * thread, after removal by us at the beginning. */ MPASS((ffs_flags & FFSV_REPLACE) == 0); return (0); } /* Read in the disk contents for the inode, copy into the inode. */ error = bread(ump->um_devvp, fsbtodb(fs, ino_to_fsba(fs, ino)), (int)fs->fs_bsize, NOCRED, &bp); if (error) { /* * The inode does not contain anything useful, so it would * be misleading to leave it on its hash chain. With mode * still zero, it will be unlinked and returned to the free * list by vput(). */ vgone(vp); vput(vp); *vpp = NULL; return (error); } if (I_IS_UFS1(ip)) ip->i_din1 = uma_zalloc(uma_ufs1, M_WAITOK); else ip->i_din2 = uma_zalloc(uma_ufs2, M_WAITOK); if ((error = ffs_load_inode(bp, ip, fs, ino)) != 0) { bqrelse(bp); vgone(vp); vput(vp); *vpp = NULL; return (error); } if (DOINGSOFTDEP(vp)) softdep_load_inodeblock(ip); else ip->i_effnlink = ip->i_nlink; bqrelse(bp); /* * Initialize the vnode from the inode, check for aliases. * Note that the underlying vnode may have changed. */ error = ufs_vinit(mp, I_IS_UFS1(ip) ? &ffs_fifoops1 : &ffs_fifoops2, &vp); if (error) { vgone(vp); vput(vp); *vpp = NULL; return (error); } /* * Finish inode initialization. */ if (vp->v_type != VFIFO) { /* FFS supports shared locking for all files except fifos. */ VN_LOCK_ASHARE(vp); } /* * Set up a generation number for this inode if it does not * already have one. This should only happen on old filesystems. */ if (ip->i_gen == 0) { while (ip->i_gen == 0) ip->i_gen = arc4random(); if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) { UFS_INODE_SET_FLAG(ip, IN_MODIFIED); DIP_SET(ip, i_gen, ip->i_gen); } } #ifdef MAC if ((mp->mnt_flag & MNT_MULTILABEL) && ip->i_mode) { /* * If this vnode is already allocated, and we're running * multi-label, attempt to perform a label association * from the extended attributes on the inode. */ error = mac_vnode_associate_extattr(mp, vp); if (error) { /* ufs_inactive will release ip->i_devvp ref. */ vgone(vp); vput(vp); *vpp = NULL; return (error); } } #endif *vpp = vp; return (0); } /* * File handle to vnode * * Have to be really careful about stale file handles: * - check that the inode number is valid * - for UFS2 check that the inode number is initialized * - call ffs_vget() to get the locked inode * - check for an unallocated inode (i_mode == 0) * - check that the given client host has export rights and return * those rights via. exflagsp and credanonp */ static int ffs_fhtovp(mp, fhp, flags, vpp) struct mount *mp; struct fid *fhp; int flags; struct vnode **vpp; { struct ufid *ufhp; struct ufsmount *ump; struct fs *fs; struct cg *cgp; struct buf *bp; ino_t ino; u_int cg; int error; ufhp = (struct ufid *)fhp; ino = ufhp->ufid_ino; ump = VFSTOUFS(mp); fs = ump->um_fs; if (ino < UFS_ROOTINO || ino >= fs->fs_ncg * fs->fs_ipg) return (ESTALE); /* * Need to check if inode is initialized because UFS2 does lazy * initialization and nfs_fhtovp can offer arbitrary inode numbers. */ if (fs->fs_magic != FS_UFS2_MAGIC) return (ufs_fhtovp(mp, ufhp, flags, vpp)); cg = ino_to_cg(fs, ino); if ((error = ffs_getcg(fs, ump->um_devvp, cg, 0, &bp, &cgp)) != 0) return (error); if (ino >= cg * fs->fs_ipg + cgp->cg_initediblk) { brelse(bp); return (ESTALE); } brelse(bp); return (ufs_fhtovp(mp, ufhp, flags, vpp)); } /* * Initialize the filesystem. */ static int ffs_init(vfsp) struct vfsconf *vfsp; { ffs_susp_initialize(); softdep_initialize(); return (ufs_init(vfsp)); } /* * Undo the work of ffs_init(). */ static int ffs_uninit(vfsp) struct vfsconf *vfsp; { int ret; ret = ufs_uninit(vfsp); softdep_uninitialize(); ffs_susp_uninitialize(); return (ret); } /* * Structure used to pass information from ffs_sbupdate to its * helper routine ffs_use_bwrite. */ struct devfd { struct ufsmount *ump; struct buf *sbbp; int waitfor; int suspended; int error; }; /* * Write a superblock and associated information back to disk. */ int ffs_sbupdate(ump, waitfor, suspended) struct ufsmount *ump; int waitfor; int suspended; { struct fs *fs; struct buf *sbbp; struct devfd devfd; fs = ump->um_fs; if (fs->fs_ronly == 1 && (ump->um_mountp->mnt_flag & (MNT_RDONLY | MNT_UPDATE)) != (MNT_RDONLY | MNT_UPDATE) && ump->um_fsckpid == 0) panic("ffs_sbupdate: write read-only filesystem"); /* * We use the superblock's buf to serialize calls to ffs_sbupdate(). */ sbbp = getblk(ump->um_devvp, btodb(fs->fs_sblockloc), (int)fs->fs_sbsize, 0, 0, 0); /* * Initialize info needed for write function. */ devfd.ump = ump; devfd.sbbp = sbbp; devfd.waitfor = waitfor; devfd.suspended = suspended; devfd.error = 0; return (ffs_sbput(&devfd, fs, fs->fs_sblockloc, ffs_use_bwrite)); } /* * Write function for use by filesystem-layer routines. */ static int ffs_use_bwrite(void *devfd, off_t loc, void *buf, int size) { struct devfd *devfdp; struct ufsmount *ump; struct buf *bp; struct fs *fs; int error; devfdp = devfd; ump = devfdp->ump; fs = ump->um_fs; /* * Writing the superblock summary information. */ if (loc != fs->fs_sblockloc) { bp = getblk(ump->um_devvp, btodb(loc), size, 0, 0, 0); bcopy(buf, bp->b_data, (u_int)size); if (devfdp->suspended) bp->b_flags |= B_VALIDSUSPWRT; if (devfdp->waitfor != MNT_WAIT) bawrite(bp); else if ((error = bwrite(bp)) != 0) devfdp->error = error; return (0); } /* * Writing the superblock itself. We need to do special checks for it. */ bp = devfdp->sbbp; if (devfdp->error != 0) { brelse(bp); return (devfdp->error); } if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_sblockloc != SBLOCK_UFS1 && (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) { printf("WARNING: %s: correcting fs_sblockloc from %jd to %d\n", fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS1); fs->fs_sblockloc = SBLOCK_UFS1; } if (fs->fs_magic == FS_UFS2_MAGIC && fs->fs_sblockloc != SBLOCK_UFS2 && (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) { printf("WARNING: %s: correcting fs_sblockloc from %jd to %d\n", fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS2); fs->fs_sblockloc = SBLOCK_UFS2; } if (MOUNTEDSOFTDEP(ump->um_mountp)) softdep_setup_sbupdate(ump, (struct fs *)bp->b_data, bp); bcopy((caddr_t)fs, bp->b_data, (u_int)fs->fs_sbsize); fs = (struct fs *)bp->b_data; ffs_oldfscompat_write(fs, ump); /* * Because we may have made changes to the superblock, we need to * recompute its check-hash. */ fs->fs_ckhash = ffs_calc_sbhash(fs); if (devfdp->suspended) bp->b_flags |= B_VALIDSUSPWRT; if (devfdp->waitfor != MNT_WAIT) bawrite(bp); else if ((error = bwrite(bp)) != 0) devfdp->error = error; return (devfdp->error); } static int ffs_extattrctl(struct mount *mp, int cmd, struct vnode *filename_vp, int attrnamespace, const char *attrname) { #ifdef UFS_EXTATTR return (ufs_extattrctl(mp, cmd, filename_vp, attrnamespace, attrname)); #else return (vfs_stdextattrctl(mp, cmd, filename_vp, attrnamespace, attrname)); #endif } static void ffs_ifree(struct ufsmount *ump, struct inode *ip) { if (ump->um_fstype == UFS1 && ip->i_din1 != NULL) uma_zfree(uma_ufs1, ip->i_din1); else if (ip->i_din2 != NULL) uma_zfree(uma_ufs2, ip->i_din2); uma_zfree(uma_inode, ip); } static int dobkgrdwrite = 1; SYSCTL_INT(_debug, OID_AUTO, dobkgrdwrite, CTLFLAG_RW, &dobkgrdwrite, 0, "Do background writes (honoring the BV_BKGRDWRITE flag)?"); /* * Complete a background write started from bwrite. */ static void ffs_backgroundwritedone(struct buf *bp) { struct bufobj *bufobj; struct buf *origbp; /* * Find the original buffer that we are writing. */ bufobj = bp->b_bufobj; BO_LOCK(bufobj); if ((origbp = gbincore(bp->b_bufobj, bp->b_lblkno)) == NULL) panic("backgroundwritedone: lost buffer"); /* * We should mark the cylinder group buffer origbp as * dirty, to not loose the failed write. */ if ((bp->b_ioflags & BIO_ERROR) != 0) origbp->b_vflags |= BV_BKGRDERR; BO_UNLOCK(bufobj); /* * Process dependencies then return any unfinished ones. */ if (!LIST_EMPTY(&bp->b_dep) && (bp->b_ioflags & BIO_ERROR) == 0) buf_complete(bp); #ifdef SOFTUPDATES if (!LIST_EMPTY(&bp->b_dep)) softdep_move_dependencies(bp, origbp); #endif /* * This buffer is marked B_NOCACHE so when it is released * by biodone it will be tossed. */ bp->b_flags |= B_NOCACHE; bp->b_flags &= ~B_CACHE; pbrelvp(bp); /* * Prevent brelse() from trying to keep and re-dirtying bp on * errors. It causes b_bufobj dereference in * bdirty()/reassignbuf(), and b_bufobj was cleared in * pbrelvp() above. */ if ((bp->b_ioflags & BIO_ERROR) != 0) bp->b_flags |= B_INVAL; bufdone(bp); BO_LOCK(bufobj); /* * Clear the BV_BKGRDINPROG flag in the original buffer * and awaken it if it is waiting for the write to complete. * If BV_BKGRDINPROG is not set in the original buffer it must * have been released and re-instantiated - which is not legal. */ KASSERT((origbp->b_vflags & BV_BKGRDINPROG), ("backgroundwritedone: lost buffer2")); origbp->b_vflags &= ~BV_BKGRDINPROG; if (origbp->b_vflags & BV_BKGRDWAIT) { origbp->b_vflags &= ~BV_BKGRDWAIT; wakeup(&origbp->b_xflags); } BO_UNLOCK(bufobj); } /* * Write, release buffer on completion. (Done by iodone * if async). Do not bother writing anything if the buffer * is invalid. * * Note that we set B_CACHE here, indicating that buffer is * fully valid and thus cacheable. This is true even of NFS * now so we set it generally. This could be set either here * or in biodone() since the I/O is synchronous. We put it * here. */ static int ffs_bufwrite(struct buf *bp) { struct buf *newbp; struct cg *cgp; CTR3(KTR_BUF, "bufwrite(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags); if (bp->b_flags & B_INVAL) { brelse(bp); return (0); } if (!BUF_ISLOCKED(bp)) panic("bufwrite: buffer is not busy???"); /* * If a background write is already in progress, delay * writing this block if it is asynchronous. Otherwise * wait for the background write to complete. */ BO_LOCK(bp->b_bufobj); if (bp->b_vflags & BV_BKGRDINPROG) { if (bp->b_flags & B_ASYNC) { BO_UNLOCK(bp->b_bufobj); bdwrite(bp); return (0); } bp->b_vflags |= BV_BKGRDWAIT; msleep(&bp->b_xflags, BO_LOCKPTR(bp->b_bufobj), PRIBIO, "bwrbg", 0); if (bp->b_vflags & BV_BKGRDINPROG) panic("bufwrite: still writing"); } bp->b_vflags &= ~BV_BKGRDERR; BO_UNLOCK(bp->b_bufobj); /* * If this buffer is marked for background writing and we * do not have to wait for it, make a copy and write the * copy so as to leave this buffer ready for further use. * * This optimization eats a lot of memory. If we have a page * or buffer shortfall we can't do it. */ if (dobkgrdwrite && (bp->b_xflags & BX_BKGRDWRITE) && (bp->b_flags & B_ASYNC) && !vm_page_count_severe() && !buf_dirty_count_severe()) { KASSERT(bp->b_iodone == NULL, ("bufwrite: needs chained iodone (%p)", bp->b_iodone)); /* get a new block */ newbp = geteblk(bp->b_bufsize, GB_NOWAIT_BD); if (newbp == NULL) goto normal_write; KASSERT(buf_mapped(bp), ("Unmapped cg")); memcpy(newbp->b_data, bp->b_data, bp->b_bufsize); BO_LOCK(bp->b_bufobj); bp->b_vflags |= BV_BKGRDINPROG; BO_UNLOCK(bp->b_bufobj); newbp->b_xflags |= (bp->b_xflags & BX_FSPRIV) | BX_BKGRDMARKER; newbp->b_lblkno = bp->b_lblkno; newbp->b_blkno = bp->b_blkno; newbp->b_offset = bp->b_offset; newbp->b_iodone = ffs_backgroundwritedone; newbp->b_flags |= B_ASYNC; newbp->b_flags &= ~B_INVAL; pbgetvp(bp->b_vp, newbp); #ifdef SOFTUPDATES /* * Move over the dependencies. If there are rollbacks, * leave the parent buffer dirtied as it will need to * be written again. */ if (LIST_EMPTY(&bp->b_dep) || softdep_move_dependencies(bp, newbp) == 0) bundirty(bp); #else bundirty(bp); #endif /* * Initiate write on the copy, release the original. The * BKGRDINPROG flag prevents it from going away until * the background write completes. We have to recalculate * its check hash in case the buffer gets freed and then * reconstituted from the buffer cache during a later read. */ if ((bp->b_xflags & BX_CYLGRP) != 0) { cgp = (struct cg *)bp->b_data; cgp->cg_ckhash = 0; cgp->cg_ckhash = calculate_crc32c(~0L, bp->b_data, bp->b_bcount); } bqrelse(bp); bp = newbp; } else /* Mark the buffer clean */ bundirty(bp); /* Let the normal bufwrite do the rest for us */ normal_write: /* * If we are writing a cylinder group, update its time. */ if ((bp->b_xflags & BX_CYLGRP) != 0) { cgp = (struct cg *)bp->b_data; cgp->cg_old_time = cgp->cg_time = time_second; } return (bufwrite(bp)); } static void ffs_geom_strategy(struct bufobj *bo, struct buf *bp) { struct vnode *vp; struct buf *tbp; int error, nocopy; + /* + * This is the bufobj strategy for the private VCHR vnodes + * used by FFS to access the underlying storage device. + * We override the default bufobj strategy and thus bypass + * VOP_STRATEGY() for these vnodes. + */ vp = bo2vnode(bo); + KASSERT(bp->b_vp == NULL || bp->b_vp->v_type != VCHR || + bp->b_vp->v_rdev == NULL || + bp->b_vp->v_rdev->si_mountpt == NULL || + VFSTOUFS(bp->b_vp->v_rdev->si_mountpt) == NULL || + vp == VFSTOUFS(bp->b_vp->v_rdev->si_mountpt)->um_devvp, + ("ffs_geom_strategy() with wrong vp")); if (bp->b_iocmd == BIO_WRITE) { if ((bp->b_flags & B_VALIDSUSPWRT) == 0 && bp->b_vp != NULL && bp->b_vp->v_mount != NULL && (bp->b_vp->v_mount->mnt_kern_flag & MNTK_SUSPENDED) != 0) panic("ffs_geom_strategy: bad I/O"); nocopy = bp->b_flags & B_NOCOPY; bp->b_flags &= ~(B_VALIDSUSPWRT | B_NOCOPY); if ((vp->v_vflag & VV_COPYONWRITE) && nocopy == 0 && vp->v_rdev->si_snapdata != NULL) { if ((bp->b_flags & B_CLUSTER) != 0) { runningbufwakeup(bp); TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head, b_cluster.cluster_entry) { error = ffs_copyonwrite(vp, tbp); if (error != 0 && error != EOPNOTSUPP) { bp->b_error = error; bp->b_ioflags |= BIO_ERROR; bufdone(bp); return; } } bp->b_runningbufspace = bp->b_bufsize; atomic_add_long(&runningbufspace, bp->b_runningbufspace); } else { error = ffs_copyonwrite(vp, bp); if (error != 0 && error != EOPNOTSUPP) { bp->b_error = error; bp->b_ioflags |= BIO_ERROR; bufdone(bp); return; } } } #ifdef SOFTUPDATES if ((bp->b_flags & B_CLUSTER) != 0) { TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head, b_cluster.cluster_entry) { if (!LIST_EMPTY(&tbp->b_dep)) buf_start(tbp); } } else { if (!LIST_EMPTY(&bp->b_dep)) buf_start(bp); } #endif /* * Check for metadata that needs check-hashes and update them. */ switch (bp->b_xflags & BX_FSPRIV) { case BX_CYLGRP: ((struct cg *)bp->b_data)->cg_ckhash = 0; ((struct cg *)bp->b_data)->cg_ckhash = calculate_crc32c(~0L, bp->b_data, bp->b_bcount); break; case BX_SUPERBLOCK: case BX_INODE: case BX_INDIR: case BX_DIR: printf("Check-hash write is unimplemented!!!\n"); break; case 0: break; default: printf("multiple buffer types 0x%b\n", (u_int)(bp->b_xflags & BX_FSPRIV), PRINT_UFS_BUF_XFLAGS); break; } } g_vfs_strategy(bo, bp); } int ffs_own_mount(const struct mount *mp) { if (mp->mnt_op == &ufs_vfsops) return (1); return (0); } #ifdef DDB #ifdef SOFTUPDATES /* defined in ffs_softdep.c */ extern void db_print_ffs(struct ufsmount *ump); DB_SHOW_COMMAND(ffs, db_show_ffs) { struct mount *mp; struct ufsmount *ump; if (have_addr) { ump = VFSTOUFS((struct mount *)addr); db_print_ffs(ump); return; } TAILQ_FOREACH(mp, &mountlist, mnt_list) { if (!strcmp(mp->mnt_stat.f_fstypename, ufs_vfsconf.vfc_name)) db_print_ffs(VFSTOUFS(mp)); } } #endif /* SOFTUPDATES */ #endif /* DDB */ diff --git a/sys/ufs/ufs/ufsmount.h b/sys/ufs/ufs/ufsmount.h index 493571442d67..2d669338f98c 100644 --- a/sys/ufs/ufs/ufsmount.h +++ b/sys/ufs/ufs/ufsmount.h @@ -1,186 +1,187 @@ /*- * SPDX-License-Identifier: BSD-3-Clause * * Copyright (c) 1982, 1986, 1989, 1993 * The Regents of the University of California. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * @(#)ufsmount.h 8.6 (Berkeley) 3/30/95 * $FreeBSD$ */ #ifndef _UFS_UFS_UFSMOUNT_H_ #define _UFS_UFS_UFSMOUNT_H_ /* * Arguments to mount UFS-based filesystems */ struct ufs_args { char *fspec; /* block special device to mount */ struct oexport_args export; /* network export information */ }; #ifdef _KERNEL #ifdef MALLOC_DECLARE MALLOC_DECLARE(M_UFSMNT); MALLOC_DECLARE(M_TRIM); #endif struct buf; struct inode; struct nameidata; struct taskqueue; struct timeval; struct ucred; struct uio; struct vnode; struct ufs_extattr_per_mount; struct jblocks; struct inodedep; TAILQ_HEAD(inodedeplst, inodedep); LIST_HEAD(bmsafemaphd, bmsafemap); LIST_HEAD(trimlist_hashhead, ffs_blkfree_trim_params); /* * This structure describes the UFS specific mount structure data. * The function operators are used to support different versions of * UFS (UFS1, UFS2, etc). * * Lock reference: * c - set at allocation then constant until freed * i - ufsmount interlock (UFS_LOCK / UFS_UNLOCK) * q - associated quota file is locked * r - ref to parent mount structure is held (vfs_busy / vfs_unbusy) * u - managed by user process fsck_ufs */ struct ufsmount { struct mount *um_mountp; /* (r) filesystem vfs struct */ struct cdev *um_dev; /* (r) device mounted */ struct g_consumer *um_cp; /* (r) GEOM access point */ struct bufobj *um_bo; /* (r) Buffer cache object */ - struct vnode *um_devvp; /* (r) blk dev mounted vnode */ + struct vnode *um_odevvp; /* (r) devfs dev vnode */ + struct vnode *um_devvp; /* (r) mntfs private vnode */ u_long um_fstype; /* (c) type of filesystem */ struct fs *um_fs; /* (r) pointer to superblock */ struct ufs_extattr_per_mount um_extattr; /* (c) extended attrs */ u_long um_nindir; /* (c) indirect ptrs per blk */ u_long um_bptrtodb; /* (c) indir disk block ptr */ u_long um_seqinc; /* (c) inc between seq blocks */ struct mtx um_lock; /* (c) Protects ufsmount & fs */ pid_t um_fsckpid; /* (u) PID can do fsck sysctl */ struct mount_softdeps *um_softdep; /* (c) softdep mgmt structure */ struct vnode *um_quotas[MAXQUOTAS]; /* (q) pointer to quota files */ struct ucred *um_cred[MAXQUOTAS]; /* (q) quota file access cred */ time_t um_btime[MAXQUOTAS]; /* (q) block quota time limit */ time_t um_itime[MAXQUOTAS]; /* (q) inode quota time limit */ char um_qflags[MAXQUOTAS]; /* (i) quota specific flags */ int64_t um_savedmaxfilesize; /* (c) track maxfilesize */ u_int um_flags; /* (i) filesystem flags */ struct timeval um_last_fullmsg; /* (i) last full msg time */ int um_secs_fullmsg; /* (i) seconds since full msg */ struct timeval um_last_integritymsg; /* (i) last integrity msg */ int um_secs_integritymsg; /* (i) secs since integ msg */ u_int um_trim_inflight; /* (i) outstanding trim count */ u_int um_trim_inflight_blks; /* (i) outstanding trim blks */ u_long um_trim_total; /* (i) total trim count */ u_long um_trim_total_blks; /* (i) total trim block count */ struct taskqueue *um_trim_tq; /* (c) trim request queue */ struct trimlist_hashhead *um_trimhash; /* (i) trimlist hash table */ u_long um_trimlisthashsize; /* (i) trim hash table size-1 */ /* (c) - below function ptrs */ int (*um_balloc)(struct vnode *, off_t, int, struct ucred *, int, struct buf **); int (*um_blkatoff)(struct vnode *, off_t, char **, struct buf **); int (*um_truncate)(struct vnode *, off_t, int, struct ucred *); int (*um_update)(struct vnode *, int); int (*um_valloc)(struct vnode *, int, struct ucred *, struct vnode **); int (*um_vfree)(struct vnode *, ino_t, int); void (*um_ifree)(struct ufsmount *, struct inode *); int (*um_rdonly)(struct inode *); void (*um_snapgone)(struct inode *); int (*um_check_blkno)(struct mount *, ino_t, daddr_t, int); }; /* * filesystem flags */ #define UM_CANDELETE 0x00000001 /* devvp supports TRIM */ #define UM_WRITESUSPENDED 0x00000002 /* suspension in progress */ #define UM_CANSPEEDUP 0x00000004 /* devvp supports SPEEDUP */ /* * function prototypes */ #define UFS_BALLOC(aa, bb, cc, dd, ee, ff) \ VFSTOUFS((aa)->v_mount)->um_balloc(aa, bb, cc, dd, ee, ff) #define UFS_BLKATOFF(aa, bb, cc, dd) \ VFSTOUFS((aa)->v_mount)->um_blkatoff(aa, bb, cc, dd) #define UFS_TRUNCATE(aa, bb, cc, dd) \ VFSTOUFS((aa)->v_mount)->um_truncate(aa, bb, cc, dd) #define UFS_UPDATE(aa, bb) VFSTOUFS((aa)->v_mount)->um_update(aa, bb) #define UFS_VALLOC(aa, bb, cc, dd) \ VFSTOUFS((aa)->v_mount)->um_valloc(aa, bb, cc, dd) #define UFS_VFREE(aa, bb, cc) VFSTOUFS((aa)->v_mount)->um_vfree(aa, bb, cc) #define UFS_IFREE(aa, bb) ((aa)->um_ifree(aa, bb)) #define UFS_RDONLY(aa) (ITOUMP(aa)->um_rdonly(aa)) #define UFS_SNAPGONE(aa) (ITOUMP(aa)->um_snapgone(aa)) #define UFS_CHECK_BLKNO(aa, bb, cc, dd) \ (VFSTOUFS(aa)->um_check_blkno == NULL ? 0 : \ VFSTOUFS(aa)->um_check_blkno(aa, bb, cc, dd)) #define UFS_LOCK(aa) mtx_lock(&(aa)->um_lock) #define UFS_UNLOCK(aa) mtx_unlock(&(aa)->um_lock) #define UFS_MTX(aa) (&(aa)->um_lock) /* * Filesystem types */ #define UFS1 1 #define UFS2 2 /* * Flags describing the state of quotas. */ #define QTF_OPENING 0x01 /* Q_QUOTAON in progress */ #define QTF_CLOSING 0x02 /* Q_QUOTAOFF in progress */ #define QTF_64BIT 0x04 /* 64-bit quota file */ /* Convert mount ptr to ufsmount ptr. */ #define VFSTOUFS(mp) ((struct ufsmount *)((mp)->mnt_data)) #define UFSTOVFS(ump) (ump)->um_mountp /* * Macros to access filesystem parameters in the ufsmount structure. * Used by ufs_bmap. */ #define MNINDIR(ump) ((ump)->um_nindir) #define blkptrtodb(ump, b) ((b) << (ump)->um_bptrtodb) #define is_sequential(ump, a, b) ((b) == (a) + ump->um_seqinc) #endif /* _KERNEL */ #endif diff --git a/sys/vm/swap_pager.c b/sys/vm/swap_pager.c index 312c50add406..eae12d5a3953 100644 --- a/sys/vm/swap_pager.c +++ b/sys/vm/swap_pager.c @@ -1,3090 +1,3081 @@ /*- * SPDX-License-Identifier: BSD-4-Clause * * Copyright (c) 1998 Matthew Dillon, * Copyright (c) 1994 John S. Dyson * Copyright (c) 1990 University of Utah. * Copyright (c) 1982, 1986, 1989, 1993 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed to Berkeley by * the Systems Programming Group of the University of Utah Computer * Science Department. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. All advertising materials mentioning features or use of this software * must display the following acknowledgement: * This product includes software developed by the University of * California, Berkeley and its contributors. * 4. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * New Swap System * Matthew Dillon * * Radix Bitmap 'blists'. * * - The new swapper uses the new radix bitmap code. This should scale * to arbitrarily small or arbitrarily large swap spaces and an almost * arbitrary degree of fragmentation. * * Features: * * - on the fly reallocation of swap during putpages. The new system * does not try to keep previously allocated swap blocks for dirty * pages. * * - on the fly deallocation of swap * * - No more garbage collection required. Unnecessarily allocated swap * blocks only exist for dirty vm_page_t's now and these are already * cycled (in a high-load system) by the pager. We also do on-the-fly * removal of invalidated swap blocks when a page is destroyed * or renamed. * * from: Utah $Hdr: swap_pager.c 1.4 91/04/30$ * * @(#)swap_pager.c 8.9 (Berkeley) 3/21/94 * @(#)vm_swap.c 8.5 (Berkeley) 2/17/94 */ #include __FBSDID("$FreeBSD$"); #include "opt_vm.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* * MAX_PAGEOUT_CLUSTER must be a power of 2 between 1 and 64. * The 64-page limit is due to the radix code (kern/subr_blist.c). */ #ifndef MAX_PAGEOUT_CLUSTER #define MAX_PAGEOUT_CLUSTER 32 #endif #if !defined(SWB_NPAGES) #define SWB_NPAGES MAX_PAGEOUT_CLUSTER #endif #define SWAP_META_PAGES PCTRIE_COUNT /* * A swblk structure maps each page index within a * SWAP_META_PAGES-aligned and sized range to the address of an * on-disk swap block (or SWAPBLK_NONE). The collection of these * mappings for an entire vm object is implemented as a pc-trie. */ struct swblk { vm_pindex_t p; daddr_t d[SWAP_META_PAGES]; }; static MALLOC_DEFINE(M_VMPGDATA, "vm_pgdata", "swap pager private data"); static struct mtx sw_dev_mtx; static TAILQ_HEAD(, swdevt) swtailq = TAILQ_HEAD_INITIALIZER(swtailq); static struct swdevt *swdevhd; /* Allocate from here next */ static int nswapdev; /* Number of swap devices */ int swap_pager_avail; static struct sx swdev_syscall_lock; /* serialize swap(on|off) */ static u_long swap_reserved; static u_long swap_total; static int sysctl_page_shift(SYSCTL_HANDLER_ARGS); static SYSCTL_NODE(_vm_stats, OID_AUTO, swap, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "VM swap stats"); SYSCTL_PROC(_vm, OID_AUTO, swap_reserved, CTLTYPE_U64 | CTLFLAG_RD | CTLFLAG_MPSAFE, &swap_reserved, 0, sysctl_page_shift, "A", "Amount of swap storage needed to back all allocated anonymous memory."); SYSCTL_PROC(_vm, OID_AUTO, swap_total, CTLTYPE_U64 | CTLFLAG_RD | CTLFLAG_MPSAFE, &swap_total, 0, sysctl_page_shift, "A", "Total amount of available swap storage."); static int overcommit = 0; SYSCTL_INT(_vm, VM_OVERCOMMIT, overcommit, CTLFLAG_RW, &overcommit, 0, "Configure virtual memory overcommit behavior. See tuning(7) " "for details."); static unsigned long swzone; SYSCTL_ULONG(_vm, OID_AUTO, swzone, CTLFLAG_RD, &swzone, 0, "Actual size of swap metadata zone"); static unsigned long swap_maxpages; SYSCTL_ULONG(_vm, OID_AUTO, swap_maxpages, CTLFLAG_RD, &swap_maxpages, 0, "Maximum amount of swap supported"); -static counter_u64_t swap_free_deferred; +static COUNTER_U64_DEFINE_EARLY(swap_free_deferred); SYSCTL_COUNTER_U64(_vm_stats_swap, OID_AUTO, free_deferred, CTLFLAG_RD, &swap_free_deferred, "Number of pages that deferred freeing swap space"); -static counter_u64_t swap_free_completed; +static COUNTER_U64_DEFINE_EARLY(swap_free_completed); SYSCTL_COUNTER_U64(_vm_stats_swap, OID_AUTO, free_completed, CTLFLAG_RD, &swap_free_completed, "Number of deferred frees completed"); /* bits from overcommit */ #define SWAP_RESERVE_FORCE_ON (1 << 0) #define SWAP_RESERVE_RLIMIT_ON (1 << 1) #define SWAP_RESERVE_ALLOW_NONWIRED (1 << 2) static int sysctl_page_shift(SYSCTL_HANDLER_ARGS) { uint64_t newval; u_long value = *(u_long *)arg1; newval = ((uint64_t)value) << PAGE_SHIFT; return (sysctl_handle_64(oidp, &newval, 0, req)); } int swap_reserve(vm_ooffset_t incr) { return (swap_reserve_by_cred(incr, curthread->td_ucred)); } int swap_reserve_by_cred(vm_ooffset_t incr, struct ucred *cred) { u_long r, s, prev, pincr; int res, error; static int curfail; static struct timeval lastfail; struct uidinfo *uip; uip = cred->cr_ruidinfo; KASSERT((incr & PAGE_MASK) == 0, ("%s: incr: %ju & PAGE_MASK", __func__, (uintmax_t)incr)); #ifdef RACCT if (racct_enable) { PROC_LOCK(curproc); error = racct_add(curproc, RACCT_SWAP, incr); PROC_UNLOCK(curproc); if (error != 0) return (0); } #endif pincr = atop(incr); res = 0; prev = atomic_fetchadd_long(&swap_reserved, pincr); r = prev + pincr; if (overcommit & SWAP_RESERVE_ALLOW_NONWIRED) { s = vm_cnt.v_page_count - vm_cnt.v_free_reserved - vm_wire_count(); } else s = 0; s += swap_total; if ((overcommit & SWAP_RESERVE_FORCE_ON) == 0 || r <= s || (error = priv_check(curthread, PRIV_VM_SWAP_NOQUOTA)) == 0) { res = 1; } else { prev = atomic_fetchadd_long(&swap_reserved, -pincr); if (prev < pincr) panic("swap_reserved < incr on overcommit fail"); } if (res) { prev = atomic_fetchadd_long(&uip->ui_vmsize, pincr); if ((overcommit & SWAP_RESERVE_RLIMIT_ON) != 0 && prev + pincr > lim_cur(curthread, RLIMIT_SWAP) && priv_check(curthread, PRIV_VM_SWAP_NORLIMIT)) { res = 0; prev = atomic_fetchadd_long(&uip->ui_vmsize, -pincr); if (prev < pincr) panic("uip->ui_vmsize < incr on overcommit fail"); } } if (!res && ppsratecheck(&lastfail, &curfail, 1)) { printf("uid %d, pid %d: swap reservation for %jd bytes failed\n", uip->ui_uid, curproc->p_pid, incr); } #ifdef RACCT if (racct_enable && !res) { PROC_LOCK(curproc); racct_sub(curproc, RACCT_SWAP, incr); PROC_UNLOCK(curproc); } #endif return (res); } void swap_reserve_force(vm_ooffset_t incr) { struct uidinfo *uip; u_long pincr; KASSERT((incr & PAGE_MASK) == 0, ("%s: incr: %ju & PAGE_MASK", __func__, (uintmax_t)incr)); PROC_LOCK(curproc); #ifdef RACCT if (racct_enable) racct_add_force(curproc, RACCT_SWAP, incr); #endif pincr = atop(incr); atomic_add_long(&swap_reserved, pincr); uip = curproc->p_ucred->cr_ruidinfo; atomic_add_long(&uip->ui_vmsize, pincr); PROC_UNLOCK(curproc); } void swap_release(vm_ooffset_t decr) { struct ucred *cred; PROC_LOCK(curproc); cred = curproc->p_ucred; swap_release_by_cred(decr, cred); PROC_UNLOCK(curproc); } void swap_release_by_cred(vm_ooffset_t decr, struct ucred *cred) { u_long prev, pdecr; struct uidinfo *uip; uip = cred->cr_ruidinfo; KASSERT((decr & PAGE_MASK) == 0, ("%s: decr: %ju & PAGE_MASK", __func__, (uintmax_t)decr)); pdecr = atop(decr); prev = atomic_fetchadd_long(&swap_reserved, -pdecr); if (prev < pdecr) panic("swap_reserved < decr"); prev = atomic_fetchadd_long(&uip->ui_vmsize, -pdecr); if (prev < pdecr) printf("negative vmsize for uid = %d\n", uip->ui_uid); #ifdef RACCT if (racct_enable) racct_sub_cred(cred, RACCT_SWAP, decr); #endif } static int swap_pager_full = 2; /* swap space exhaustion (task killing) */ static int swap_pager_almost_full = 1; /* swap space exhaustion (w/hysteresis)*/ static struct mtx swbuf_mtx; /* to sync nsw_wcount_async */ static int nsw_wcount_async; /* limit async write buffers */ static int nsw_wcount_async_max;/* assigned maximum */ static int nsw_cluster_max; /* maximum VOP I/O allowed */ static int sysctl_swap_async_max(SYSCTL_HANDLER_ARGS); SYSCTL_PROC(_vm, OID_AUTO, swap_async_max, CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, 0, sysctl_swap_async_max, "I", "Maximum running async swap ops"); static int sysctl_swap_fragmentation(SYSCTL_HANDLER_ARGS); SYSCTL_PROC(_vm, OID_AUTO, swap_fragmentation, CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0, sysctl_swap_fragmentation, "A", "Swap Fragmentation Info"); static struct sx sw_alloc_sx; /* * "named" and "unnamed" anon region objects. Try to reduce the overhead * of searching a named list by hashing it just a little. */ #define NOBJLISTS 8 #define NOBJLIST(handle) \ (&swap_pager_object_list[((int)(intptr_t)handle >> 4) & (NOBJLISTS-1)]) static struct pagerlst swap_pager_object_list[NOBJLISTS]; static uma_zone_t swwbuf_zone; static uma_zone_t swrbuf_zone; static uma_zone_t swblk_zone; static uma_zone_t swpctrie_zone; /* * pagerops for OBJT_SWAP - "swap pager". Some ops are also global procedure * calls hooked from other parts of the VM system and do not appear here. * (see vm/swap_pager.h). */ static vm_object_t swap_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot, vm_ooffset_t offset, struct ucred *); static void swap_pager_dealloc(vm_object_t object); static int swap_pager_getpages(vm_object_t, vm_page_t *, int, int *, int *); static int swap_pager_getpages_async(vm_object_t, vm_page_t *, int, int *, int *, pgo_getpages_iodone_t, void *); static void swap_pager_putpages(vm_object_t, vm_page_t *, int, boolean_t, int *); static boolean_t swap_pager_haspage(vm_object_t object, vm_pindex_t pindex, int *before, int *after); static void swap_pager_init(void); static void swap_pager_unswapped(vm_page_t); static void swap_pager_swapoff(struct swdevt *sp); static void swap_pager_update_writecount(vm_object_t object, vm_offset_t start, vm_offset_t end); static void swap_pager_release_writecount(vm_object_t object, vm_offset_t start, vm_offset_t end); struct pagerops swappagerops = { .pgo_init = swap_pager_init, /* early system initialization of pager */ .pgo_alloc = swap_pager_alloc, /* allocate an OBJT_SWAP object */ .pgo_dealloc = swap_pager_dealloc, /* deallocate an OBJT_SWAP object */ .pgo_getpages = swap_pager_getpages, /* pagein */ .pgo_getpages_async = swap_pager_getpages_async, /* pagein (async) */ .pgo_putpages = swap_pager_putpages, /* pageout */ .pgo_haspage = swap_pager_haspage, /* get backing store status for page */ .pgo_pageunswapped = swap_pager_unswapped, /* remove swap related to page */ .pgo_update_writecount = swap_pager_update_writecount, .pgo_release_writecount = swap_pager_release_writecount, }; /* * swap_*() routines are externally accessible. swp_*() routines are * internal. */ static int nswap_lowat = 128; /* in pages, swap_pager_almost_full warn */ static int nswap_hiwat = 512; /* in pages, swap_pager_almost_full warn */ SYSCTL_INT(_vm, OID_AUTO, dmmax, CTLFLAG_RD, &nsw_cluster_max, 0, "Maximum size of a swap block in pages"); static void swp_sizecheck(void); static void swp_pager_async_iodone(struct buf *bp); static bool swp_pager_swblk_empty(struct swblk *sb, int start, int limit); static void swp_pager_free_empty_swblk(vm_object_t, struct swblk *sb); static int swapongeom(struct vnode *); static int swaponvp(struct thread *, struct vnode *, u_long); static int swapoff_one(struct swdevt *sp, struct ucred *cred); /* * Swap bitmap functions */ static void swp_pager_freeswapspace(daddr_t blk, daddr_t npages); static daddr_t swp_pager_getswapspace(int *npages); /* * Metadata functions */ static daddr_t swp_pager_meta_build(vm_object_t, vm_pindex_t, daddr_t); static void swp_pager_meta_free(vm_object_t, vm_pindex_t, vm_pindex_t); static void swp_pager_meta_transfer(vm_object_t src, vm_object_t dst, vm_pindex_t pindex, vm_pindex_t count); static void swp_pager_meta_free_all(vm_object_t); static daddr_t swp_pager_meta_lookup(vm_object_t, vm_pindex_t); static void swp_pager_init_freerange(daddr_t *start, daddr_t *num) { *start = SWAPBLK_NONE; *num = 0; } static void swp_pager_update_freerange(daddr_t *start, daddr_t *num, daddr_t addr) { if (*start + *num == addr) { (*num)++; } else { swp_pager_freeswapspace(*start, *num); *start = addr; *num = 1; } } static void * swblk_trie_alloc(struct pctrie *ptree) { return (uma_zalloc(swpctrie_zone, M_NOWAIT | (curproc == pageproc ? M_USE_RESERVE : 0))); } static void swblk_trie_free(struct pctrie *ptree, void *node) { uma_zfree(swpctrie_zone, node); } PCTRIE_DEFINE(SWAP, swblk, p, swblk_trie_alloc, swblk_trie_free); /* * SWP_SIZECHECK() - update swap_pager_full indication * * update the swap_pager_almost_full indication and warn when we are * about to run out of swap space, using lowat/hiwat hysteresis. * * Clear swap_pager_full ( task killing ) indication when lowat is met. * * No restrictions on call * This routine may not block. */ static void swp_sizecheck(void) { if (swap_pager_avail < nswap_lowat) { if (swap_pager_almost_full == 0) { printf("swap_pager: out of swap space\n"); swap_pager_almost_full = 1; } } else { swap_pager_full = 0; if (swap_pager_avail > nswap_hiwat) swap_pager_almost_full = 0; } } /* * SWAP_PAGER_INIT() - initialize the swap pager! * * Expected to be started from system init. NOTE: This code is run * before much else so be careful what you depend on. Most of the VM * system has yet to be initialized at this point. */ static void swap_pager_init(void) { /* * Initialize object lists */ int i; for (i = 0; i < NOBJLISTS; ++i) TAILQ_INIT(&swap_pager_object_list[i]); mtx_init(&sw_dev_mtx, "swapdev", NULL, MTX_DEF); sx_init(&sw_alloc_sx, "swspsx"); sx_init(&swdev_syscall_lock, "swsysc"); } -static void -swap_pager_counters(void) -{ - - swap_free_deferred = counter_u64_alloc(M_WAITOK); - swap_free_completed = counter_u64_alloc(M_WAITOK); -} -SYSINIT(swap_counters, SI_SUB_CPU, SI_ORDER_ANY, swap_pager_counters, NULL); - /* * SWAP_PAGER_SWAP_INIT() - swap pager initialization from pageout process * * Expected to be started from pageout process once, prior to entering * its main loop. */ void swap_pager_swap_init(void) { unsigned long n, n2; /* * Number of in-transit swap bp operations. Don't * exhaust the pbufs completely. Make sure we * initialize workable values (0 will work for hysteresis * but it isn't very efficient). * * The nsw_cluster_max is constrained by the bp->b_pages[] * array, which has MAXPHYS / PAGE_SIZE entries, and our locally * defined MAX_PAGEOUT_CLUSTER. Also be aware that swap ops are * constrained by the swap device interleave stripe size. * * Currently we hardwire nsw_wcount_async to 4. This limit is * designed to prevent other I/O from having high latencies due to * our pageout I/O. The value 4 works well for one or two active swap * devices but is probably a little low if you have more. Even so, * a higher value would probably generate only a limited improvement * with three or four active swap devices since the system does not * typically have to pageout at extreme bandwidths. We will want * at least 2 per swap devices, and 4 is a pretty good value if you * have one NFS swap device due to the command/ack latency over NFS. * So it all works out pretty well. */ nsw_cluster_max = min(MAXPHYS / PAGE_SIZE, MAX_PAGEOUT_CLUSTER); nsw_wcount_async = 4; nsw_wcount_async_max = nsw_wcount_async; mtx_init(&swbuf_mtx, "async swbuf mutex", NULL, MTX_DEF); swwbuf_zone = pbuf_zsecond_create("swwbuf", nswbuf / 4); swrbuf_zone = pbuf_zsecond_create("swrbuf", nswbuf / 2); /* * Initialize our zone, taking the user's requested size or * estimating the number we need based on the number of pages * in the system. */ n = maxswzone != 0 ? maxswzone / sizeof(struct swblk) : vm_cnt.v_page_count / 2; swpctrie_zone = uma_zcreate("swpctrie", pctrie_node_size(), NULL, NULL, pctrie_zone_init, NULL, UMA_ALIGN_PTR, 0); if (swpctrie_zone == NULL) panic("failed to create swap pctrie zone."); swblk_zone = uma_zcreate("swblk", sizeof(struct swblk), NULL, NULL, NULL, NULL, _Alignof(struct swblk) - 1, 0); if (swblk_zone == NULL) panic("failed to create swap blk zone."); n2 = n; do { if (uma_zone_reserve_kva(swblk_zone, n)) break; /* * if the allocation failed, try a zone two thirds the * size of the previous attempt. */ n -= ((n + 2) / 3); } while (n > 0); /* * Often uma_zone_reserve_kva() cannot reserve exactly the * requested size. Account for the difference when * calculating swap_maxpages. */ n = uma_zone_get_max(swblk_zone); if (n < n2) printf("Swap blk zone entries changed from %lu to %lu.\n", n2, n); /* absolute maximum we can handle assuming 100% efficiency */ swap_maxpages = n * SWAP_META_PAGES; swzone = n * sizeof(struct swblk); if (!uma_zone_reserve_kva(swpctrie_zone, n)) printf("Cannot reserve swap pctrie zone, " "reduce kern.maxswzone.\n"); } static vm_object_t swap_pager_alloc_init(void *handle, struct ucred *cred, vm_ooffset_t size, vm_ooffset_t offset) { vm_object_t object; if (cred != NULL) { if (!swap_reserve_by_cred(size, cred)) return (NULL); crhold(cred); } /* * The un_pager.swp.swp_blks trie is initialized by * vm_object_allocate() to ensure the correct order of * visibility to other threads. */ object = vm_object_allocate(OBJT_SWAP, OFF_TO_IDX(offset + PAGE_MASK + size)); object->un_pager.swp.writemappings = 0; object->handle = handle; if (cred != NULL) { object->cred = cred; object->charge = size; } return (object); } /* * SWAP_PAGER_ALLOC() - allocate a new OBJT_SWAP VM object and instantiate * its metadata structures. * * This routine is called from the mmap and fork code to create a new * OBJT_SWAP object. * * This routine must ensure that no live duplicate is created for * the named object request, which is protected against by * holding the sw_alloc_sx lock in case handle != NULL. */ static vm_object_t swap_pager_alloc(void *handle, vm_ooffset_t size, vm_prot_t prot, vm_ooffset_t offset, struct ucred *cred) { vm_object_t object; if (handle != NULL) { /* * Reference existing named region or allocate new one. There * should not be a race here against swp_pager_meta_build() * as called from vm_page_remove() in regards to the lookup * of the handle. */ sx_xlock(&sw_alloc_sx); object = vm_pager_object_lookup(NOBJLIST(handle), handle); if (object == NULL) { object = swap_pager_alloc_init(handle, cred, size, offset); if (object != NULL) { TAILQ_INSERT_TAIL(NOBJLIST(object->handle), object, pager_object_list); } } sx_xunlock(&sw_alloc_sx); } else { object = swap_pager_alloc_init(handle, cred, size, offset); } return (object); } /* * SWAP_PAGER_DEALLOC() - remove swap metadata from object * * The swap backing for the object is destroyed. The code is * designed such that we can reinstantiate it later, but this * routine is typically called only when the entire object is * about to be destroyed. * * The object must be locked. */ static void swap_pager_dealloc(vm_object_t object) { VM_OBJECT_ASSERT_WLOCKED(object); KASSERT((object->flags & OBJ_DEAD) != 0, ("dealloc of reachable obj")); /* * Remove from list right away so lookups will fail if we block for * pageout completion. */ if ((object->flags & OBJ_ANON) == 0 && object->handle != NULL) { VM_OBJECT_WUNLOCK(object); sx_xlock(&sw_alloc_sx); TAILQ_REMOVE(NOBJLIST(object->handle), object, pager_object_list); sx_xunlock(&sw_alloc_sx); VM_OBJECT_WLOCK(object); } vm_object_pip_wait(object, "swpdea"); /* * Free all remaining metadata. We only bother to free it from * the swap meta data. We do not attempt to free swapblk's still * associated with vm_page_t's for this object. We do not care * if paging is still in progress on some objects. */ swp_pager_meta_free_all(object); object->handle = NULL; object->type = OBJT_DEAD; } /************************************************************************ * SWAP PAGER BITMAP ROUTINES * ************************************************************************/ /* * SWP_PAGER_GETSWAPSPACE() - allocate raw swap space * * Allocate swap for up to the requested number of pages. The * starting swap block number (a page index) is returned or * SWAPBLK_NONE if the allocation failed. * * Also has the side effect of advising that somebody made a mistake * when they configured swap and didn't configure enough. * * This routine may not sleep. * * We allocate in round-robin fashion from the configured devices. */ static daddr_t swp_pager_getswapspace(int *io_npages) { daddr_t blk; struct swdevt *sp; int mpages, npages; KASSERT(*io_npages >= 1, ("%s: npages not positive", __func__)); blk = SWAPBLK_NONE; mpages = *io_npages; npages = imin(BLIST_MAX_ALLOC, mpages); mtx_lock(&sw_dev_mtx); sp = swdevhd; while (!TAILQ_EMPTY(&swtailq)) { if (sp == NULL) sp = TAILQ_FIRST(&swtailq); if ((sp->sw_flags & SW_CLOSING) == 0) blk = blist_alloc(sp->sw_blist, &npages, mpages); if (blk != SWAPBLK_NONE) break; sp = TAILQ_NEXT(sp, sw_list); if (swdevhd == sp) { if (npages == 1) break; mpages = npages - 1; npages >>= 1; } } if (blk != SWAPBLK_NONE) { *io_npages = npages; blk += sp->sw_first; sp->sw_used += npages; swap_pager_avail -= npages; swp_sizecheck(); swdevhd = TAILQ_NEXT(sp, sw_list); } else { if (swap_pager_full != 2) { printf("swp_pager_getswapspace(%d): failed\n", *io_npages); swap_pager_full = 2; swap_pager_almost_full = 1; } swdevhd = NULL; } mtx_unlock(&sw_dev_mtx); return (blk); } static bool swp_pager_isondev(daddr_t blk, struct swdevt *sp) { return (blk >= sp->sw_first && blk < sp->sw_end); } static void swp_pager_strategy(struct buf *bp) { struct swdevt *sp; mtx_lock(&sw_dev_mtx); TAILQ_FOREACH(sp, &swtailq, sw_list) { if (swp_pager_isondev(bp->b_blkno, sp)) { mtx_unlock(&sw_dev_mtx); if ((sp->sw_flags & SW_UNMAPPED) != 0 && unmapped_buf_allowed) { bp->b_data = unmapped_buf; bp->b_offset = 0; } else { pmap_qenter((vm_offset_t)bp->b_data, &bp->b_pages[0], bp->b_bcount / PAGE_SIZE); } sp->sw_strategy(bp, sp); return; } } panic("Swapdev not found"); } /* * SWP_PAGER_FREESWAPSPACE() - free raw swap space * * This routine returns the specified swap blocks back to the bitmap. * * This routine may not sleep. */ static void swp_pager_freeswapspace(daddr_t blk, daddr_t npages) { struct swdevt *sp; if (npages == 0) return; mtx_lock(&sw_dev_mtx); TAILQ_FOREACH(sp, &swtailq, sw_list) { if (swp_pager_isondev(blk, sp)) { sp->sw_used -= npages; /* * If we are attempting to stop swapping on * this device, we don't want to mark any * blocks free lest they be reused. */ if ((sp->sw_flags & SW_CLOSING) == 0) { blist_free(sp->sw_blist, blk - sp->sw_first, npages); swap_pager_avail += npages; swp_sizecheck(); } mtx_unlock(&sw_dev_mtx); return; } } panic("Swapdev not found"); } /* * SYSCTL_SWAP_FRAGMENTATION() - produce raw swap space stats */ static int sysctl_swap_fragmentation(SYSCTL_HANDLER_ARGS) { struct sbuf sbuf; struct swdevt *sp; const char *devname; int error; error = sysctl_wire_old_buffer(req, 0); if (error != 0) return (error); sbuf_new_for_sysctl(&sbuf, NULL, 128, req); mtx_lock(&sw_dev_mtx); TAILQ_FOREACH(sp, &swtailq, sw_list) { if (vn_isdisk(sp->sw_vp, NULL)) devname = devtoname(sp->sw_vp->v_rdev); else devname = "[file]"; sbuf_printf(&sbuf, "\nFree space on device %s:\n", devname); blist_stats(sp->sw_blist, &sbuf); } mtx_unlock(&sw_dev_mtx); error = sbuf_finish(&sbuf); sbuf_delete(&sbuf); return (error); } /* * SWAP_PAGER_FREESPACE() - frees swap blocks associated with a page * range within an object. * * This is a globally accessible routine. * * This routine removes swapblk assignments from swap metadata. * * The external callers of this routine typically have already destroyed * or renamed vm_page_t's associated with this range in the object so * we should be ok. * * The object must be locked. */ void swap_pager_freespace(vm_object_t object, vm_pindex_t start, vm_size_t size) { swp_pager_meta_free(object, start, size); } /* * SWAP_PAGER_RESERVE() - reserve swap blocks in object * * Assigns swap blocks to the specified range within the object. The * swap blocks are not zeroed. Any previous swap assignment is destroyed. * * Returns 0 on success, -1 on failure. */ int swap_pager_reserve(vm_object_t object, vm_pindex_t start, vm_size_t size) { daddr_t addr, blk, n_free, s_free; int i, j, n; swp_pager_init_freerange(&s_free, &n_free); VM_OBJECT_WLOCK(object); for (i = 0; i < size; i += n) { n = size - i; blk = swp_pager_getswapspace(&n); if (blk == SWAPBLK_NONE) { swp_pager_meta_free(object, start, i); VM_OBJECT_WUNLOCK(object); return (-1); } for (j = 0; j < n; ++j) { addr = swp_pager_meta_build(object, start + i + j, blk + j); if (addr != SWAPBLK_NONE) swp_pager_update_freerange(&s_free, &n_free, addr); } } swp_pager_freeswapspace(s_free, n_free); VM_OBJECT_WUNLOCK(object); return (0); } static bool swp_pager_xfer_source(vm_object_t srcobject, vm_object_t dstobject, vm_pindex_t pindex, daddr_t addr) { daddr_t dstaddr; KASSERT(srcobject->type == OBJT_SWAP, ("%s: Srcobject not swappable", __func__)); if (dstobject->type == OBJT_SWAP && swp_pager_meta_lookup(dstobject, pindex) != SWAPBLK_NONE) { /* Caller should destroy the source block. */ return (false); } /* * Destination has no swapblk and is not resident, transfer source. * swp_pager_meta_build() can sleep. */ VM_OBJECT_WUNLOCK(srcobject); dstaddr = swp_pager_meta_build(dstobject, pindex, addr); KASSERT(dstaddr == SWAPBLK_NONE, ("Unexpected destination swapblk")); VM_OBJECT_WLOCK(srcobject); return (true); } /* * SWAP_PAGER_COPY() - copy blocks from source pager to destination pager * and destroy the source. * * Copy any valid swapblks from the source to the destination. In * cases where both the source and destination have a valid swapblk, * we keep the destination's. * * This routine is allowed to sleep. It may sleep allocating metadata * indirectly through swp_pager_meta_build(). * * The source object contains no vm_page_t's (which is just as well) * * The source object is of type OBJT_SWAP. * * The source and destination objects must be locked. * Both object locks may temporarily be released. */ void swap_pager_copy(vm_object_t srcobject, vm_object_t dstobject, vm_pindex_t offset, int destroysource) { VM_OBJECT_ASSERT_WLOCKED(srcobject); VM_OBJECT_ASSERT_WLOCKED(dstobject); /* * If destroysource is set, we remove the source object from the * swap_pager internal queue now. */ if (destroysource && (srcobject->flags & OBJ_ANON) == 0 && srcobject->handle != NULL) { VM_OBJECT_WUNLOCK(srcobject); VM_OBJECT_WUNLOCK(dstobject); sx_xlock(&sw_alloc_sx); TAILQ_REMOVE(NOBJLIST(srcobject->handle), srcobject, pager_object_list); sx_xunlock(&sw_alloc_sx); VM_OBJECT_WLOCK(dstobject); VM_OBJECT_WLOCK(srcobject); } /* * Transfer source to destination. */ swp_pager_meta_transfer(srcobject, dstobject, offset, dstobject->size); /* * Free left over swap blocks in source. * * We have to revert the type to OBJT_DEFAULT so we do not accidentally * double-remove the object from the swap queues. */ if (destroysource) { swp_pager_meta_free_all(srcobject); /* * Reverting the type is not necessary, the caller is going * to destroy srcobject directly, but I'm doing it here * for consistency since we've removed the object from its * queues. */ srcobject->type = OBJT_DEFAULT; } } /* * SWAP_PAGER_HASPAGE() - determine if we have good backing store for * the requested page. * * We determine whether good backing store exists for the requested * page and return TRUE if it does, FALSE if it doesn't. * * If TRUE, we also try to determine how much valid, contiguous backing * store exists before and after the requested page. */ static boolean_t swap_pager_haspage(vm_object_t object, vm_pindex_t pindex, int *before, int *after) { daddr_t blk, blk0; int i; VM_OBJECT_ASSERT_LOCKED(object); KASSERT(object->type == OBJT_SWAP, ("%s: object not swappable", __func__)); /* * do we have good backing store at the requested index ? */ blk0 = swp_pager_meta_lookup(object, pindex); if (blk0 == SWAPBLK_NONE) { if (before) *before = 0; if (after) *after = 0; return (FALSE); } /* * find backwards-looking contiguous good backing store */ if (before != NULL) { for (i = 1; i < SWB_NPAGES; i++) { if (i > pindex) break; blk = swp_pager_meta_lookup(object, pindex - i); if (blk != blk0 - i) break; } *before = i - 1; } /* * find forward-looking contiguous good backing store */ if (after != NULL) { for (i = 1; i < SWB_NPAGES; i++) { blk = swp_pager_meta_lookup(object, pindex + i); if (blk != blk0 + i) break; } *after = i - 1; } return (TRUE); } /* * SWAP_PAGER_PAGE_UNSWAPPED() - remove swap backing store related to page * * This removes any associated swap backing store, whether valid or * not, from the page. * * This routine is typically called when a page is made dirty, at * which point any associated swap can be freed. MADV_FREE also * calls us in a special-case situation * * NOTE!!! If the page is clean and the swap was valid, the caller * should make the page dirty before calling this routine. This routine * does NOT change the m->dirty status of the page. Also: MADV_FREE * depends on it. * * This routine may not sleep. * * The object containing the page may be locked. */ static void swap_pager_unswapped(vm_page_t m) { struct swblk *sb; vm_object_t obj; /* * Handle enqueing deferred frees first. If we do not have the * object lock we wait for the page daemon to clear the space. */ obj = m->object; if (!VM_OBJECT_WOWNED(obj)) { VM_PAGE_OBJECT_BUSY_ASSERT(m); /* * The caller is responsible for synchronization but we * will harmlessly handle races. This is typically provided * by only calling unswapped() when a page transitions from * clean to dirty. */ if ((m->a.flags & (PGA_SWAP_SPACE | PGA_SWAP_FREE)) == PGA_SWAP_SPACE) { vm_page_aflag_set(m, PGA_SWAP_FREE); counter_u64_add(swap_free_deferred, 1); } return; } if ((m->a.flags & PGA_SWAP_FREE) != 0) counter_u64_add(swap_free_completed, 1); vm_page_aflag_clear(m, PGA_SWAP_FREE | PGA_SWAP_SPACE); /* * The meta data only exists if the object is OBJT_SWAP * and even then might not be allocated yet. */ KASSERT(m->object->type == OBJT_SWAP, ("Free object not swappable")); sb = SWAP_PCTRIE_LOOKUP(&m->object->un_pager.swp.swp_blks, rounddown(m->pindex, SWAP_META_PAGES)); if (sb == NULL) return; if (sb->d[m->pindex % SWAP_META_PAGES] == SWAPBLK_NONE) return; swp_pager_freeswapspace(sb->d[m->pindex % SWAP_META_PAGES], 1); sb->d[m->pindex % SWAP_META_PAGES] = SWAPBLK_NONE; swp_pager_free_empty_swblk(m->object, sb); } /* * swap_pager_getpages() - bring pages in from swap * * Attempt to page in the pages in array "ma" of length "count". The * caller may optionally specify that additional pages preceding and * succeeding the specified range be paged in. The number of such pages * is returned in the "rbehind" and "rahead" parameters, and they will * be in the inactive queue upon return. * * The pages in "ma" must be busied and will remain busied upon return. */ static int swap_pager_getpages_locked(vm_object_t object, vm_page_t *ma, int count, int *rbehind, int *rahead) { struct buf *bp; vm_page_t bm, mpred, msucc, p; vm_pindex_t pindex; daddr_t blk; int i, maxahead, maxbehind, reqcount; VM_OBJECT_ASSERT_WLOCKED(object); reqcount = count; KASSERT(object->type == OBJT_SWAP, ("%s: object not swappable", __func__)); if (!swap_pager_haspage(object, ma[0]->pindex, &maxbehind, &maxahead)) { VM_OBJECT_WUNLOCK(object); return (VM_PAGER_FAIL); } KASSERT(reqcount - 1 <= maxahead, ("page count %d extends beyond swap block", reqcount)); /* * Do not transfer any pages other than those that are xbusied * when running during a split or collapse operation. This * prevents clustering from re-creating pages which are being * moved into another object. */ if ((object->flags & (OBJ_SPLIT | OBJ_DEAD)) != 0) { maxahead = reqcount - 1; maxbehind = 0; } /* * Clip the readahead and readbehind ranges to exclude resident pages. */ if (rahead != NULL) { *rahead = imin(*rahead, maxahead - (reqcount - 1)); pindex = ma[reqcount - 1]->pindex; msucc = TAILQ_NEXT(ma[reqcount - 1], listq); if (msucc != NULL && msucc->pindex - pindex - 1 < *rahead) *rahead = msucc->pindex - pindex - 1; } if (rbehind != NULL) { *rbehind = imin(*rbehind, maxbehind); pindex = ma[0]->pindex; mpred = TAILQ_PREV(ma[0], pglist, listq); if (mpred != NULL && pindex - mpred->pindex - 1 < *rbehind) *rbehind = pindex - mpred->pindex - 1; } bm = ma[0]; for (i = 0; i < count; i++) ma[i]->oflags |= VPO_SWAPINPROG; /* * Allocate readahead and readbehind pages. */ if (rbehind != NULL) { for (i = 1; i <= *rbehind; i++) { p = vm_page_alloc(object, ma[0]->pindex - i, VM_ALLOC_NORMAL); if (p == NULL) break; p->oflags |= VPO_SWAPINPROG; bm = p; } *rbehind = i - 1; } if (rahead != NULL) { for (i = 0; i < *rahead; i++) { p = vm_page_alloc(object, ma[reqcount - 1]->pindex + i + 1, VM_ALLOC_NORMAL); if (p == NULL) break; p->oflags |= VPO_SWAPINPROG; } *rahead = i; } if (rbehind != NULL) count += *rbehind; if (rahead != NULL) count += *rahead; vm_object_pip_add(object, count); pindex = bm->pindex; blk = swp_pager_meta_lookup(object, pindex); KASSERT(blk != SWAPBLK_NONE, ("no swap blocking containing %p(%jx)", object, (uintmax_t)pindex)); VM_OBJECT_WUNLOCK(object); bp = uma_zalloc(swrbuf_zone, M_WAITOK); /* Pages cannot leave the object while busy. */ for (i = 0, p = bm; i < count; i++, p = TAILQ_NEXT(p, listq)) { MPASS(p->pindex == bm->pindex + i); bp->b_pages[i] = p; } bp->b_flags |= B_PAGING; bp->b_iocmd = BIO_READ; bp->b_iodone = swp_pager_async_iodone; bp->b_rcred = crhold(thread0.td_ucred); bp->b_wcred = crhold(thread0.td_ucred); bp->b_blkno = blk; bp->b_bcount = PAGE_SIZE * count; bp->b_bufsize = PAGE_SIZE * count; bp->b_npages = count; bp->b_pgbefore = rbehind != NULL ? *rbehind : 0; bp->b_pgafter = rahead != NULL ? *rahead : 0; VM_CNT_INC(v_swapin); VM_CNT_ADD(v_swappgsin, count); /* * perform the I/O. NOTE!!! bp cannot be considered valid after * this point because we automatically release it on completion. * Instead, we look at the one page we are interested in which we * still hold a lock on even through the I/O completion. * * The other pages in our ma[] array are also released on completion, * so we cannot assume they are valid anymore either. * * NOTE: b_blkno is destroyed by the call to swapdev_strategy */ BUF_KERNPROC(bp); swp_pager_strategy(bp); /* * Wait for the pages we want to complete. VPO_SWAPINPROG is always * cleared on completion. If an I/O error occurs, SWAPBLK_NONE * is set in the metadata for each page in the request. */ VM_OBJECT_WLOCK(object); /* This could be implemented more efficiently with aflags */ while ((ma[0]->oflags & VPO_SWAPINPROG) != 0) { ma[0]->oflags |= VPO_SWAPSLEEP; VM_CNT_INC(v_intrans); if (VM_OBJECT_SLEEP(object, &object->handle, PSWP, "swread", hz * 20)) { printf( "swap_pager: indefinite wait buffer: bufobj: %p, blkno: %jd, size: %ld\n", bp->b_bufobj, (intmax_t)bp->b_blkno, bp->b_bcount); } } VM_OBJECT_WUNLOCK(object); /* * If we had an unrecoverable read error pages will not be valid. */ for (i = 0; i < reqcount; i++) if (ma[i]->valid != VM_PAGE_BITS_ALL) return (VM_PAGER_ERROR); return (VM_PAGER_OK); /* * A final note: in a low swap situation, we cannot deallocate swap * and mark a page dirty here because the caller is likely to mark * the page clean when we return, causing the page to possibly revert * to all-zero's later. */ } static int swap_pager_getpages(vm_object_t object, vm_page_t *ma, int count, int *rbehind, int *rahead) { VM_OBJECT_WLOCK(object); return (swap_pager_getpages_locked(object, ma, count, rbehind, rahead)); } /* * swap_pager_getpages_async(): * * Right now this is emulation of asynchronous operation on top of * swap_pager_getpages(). */ static int swap_pager_getpages_async(vm_object_t object, vm_page_t *ma, int count, int *rbehind, int *rahead, pgo_getpages_iodone_t iodone, void *arg) { int r, error; r = swap_pager_getpages(object, ma, count, rbehind, rahead); switch (r) { case VM_PAGER_OK: error = 0; break; case VM_PAGER_ERROR: error = EIO; break; case VM_PAGER_FAIL: error = EINVAL; break; default: panic("unhandled swap_pager_getpages() error %d", r); } (iodone)(arg, ma, count, error); return (r); } /* * swap_pager_putpages: * * Assign swap (if necessary) and initiate I/O on the specified pages. * * We support both OBJT_DEFAULT and OBJT_SWAP objects. DEFAULT objects * are automatically converted to SWAP objects. * * In a low memory situation we may block in VOP_STRATEGY(), but the new * vm_page reservation system coupled with properly written VFS devices * should ensure that no low-memory deadlock occurs. This is an area * which needs work. * * The parent has N vm_object_pip_add() references prior to * calling us and will remove references for rtvals[] that are * not set to VM_PAGER_PEND. We need to remove the rest on I/O * completion. * * The parent has soft-busy'd the pages it passes us and will unbusy * those whose rtvals[] entry is not set to VM_PAGER_PEND on return. * We need to unbusy the rest on I/O completion. */ static void swap_pager_putpages(vm_object_t object, vm_page_t *ma, int count, int flags, int *rtvals) { struct buf *bp; daddr_t addr, blk, n_free, s_free; vm_page_t mreq; int i, j, n; bool async; KASSERT(count == 0 || ma[0]->object == object, ("%s: object mismatch %p/%p", __func__, object, ma[0]->object)); /* * Step 1 * * Turn object into OBJT_SWAP. Force sync if not a pageout process. */ if (object->type != OBJT_SWAP) { addr = swp_pager_meta_build(object, 0, SWAPBLK_NONE); KASSERT(addr == SWAPBLK_NONE, ("unexpected object swap block")); } VM_OBJECT_WUNLOCK(object); async = curproc == pageproc && (flags & VM_PAGER_PUT_SYNC) == 0; swp_pager_init_freerange(&s_free, &n_free); /* * Step 2 * * Assign swap blocks and issue I/O. We reallocate swap on the fly. * The page is left dirty until the pageout operation completes * successfully. */ for (i = 0; i < count; i += n) { /* Maximum I/O size is limited by maximum swap block size. */ n = min(count - i, nsw_cluster_max); if (async) { mtx_lock(&swbuf_mtx); while (nsw_wcount_async == 0) msleep(&nsw_wcount_async, &swbuf_mtx, PVM, "swbufa", 0); nsw_wcount_async--; mtx_unlock(&swbuf_mtx); } /* Get a block of swap of size up to size n. */ VM_OBJECT_WLOCK(object); blk = swp_pager_getswapspace(&n); if (blk == SWAPBLK_NONE) { VM_OBJECT_WUNLOCK(object); mtx_lock(&swbuf_mtx); if (++nsw_wcount_async == 1) wakeup(&nsw_wcount_async); mtx_unlock(&swbuf_mtx); for (j = 0; j < n; ++j) rtvals[i + j] = VM_PAGER_FAIL; continue; } for (j = 0; j < n; ++j) { mreq = ma[i + j]; vm_page_aflag_clear(mreq, PGA_SWAP_FREE); addr = swp_pager_meta_build(mreq->object, mreq->pindex, blk + j); if (addr != SWAPBLK_NONE) swp_pager_update_freerange(&s_free, &n_free, addr); MPASS(mreq->dirty == VM_PAGE_BITS_ALL); mreq->oflags |= VPO_SWAPINPROG; } VM_OBJECT_WUNLOCK(object); bp = uma_zalloc(swwbuf_zone, M_WAITOK); if (async) bp->b_flags = B_ASYNC; bp->b_flags |= B_PAGING; bp->b_iocmd = BIO_WRITE; bp->b_rcred = crhold(thread0.td_ucred); bp->b_wcred = crhold(thread0.td_ucred); bp->b_bcount = PAGE_SIZE * n; bp->b_bufsize = PAGE_SIZE * n; bp->b_blkno = blk; for (j = 0; j < n; j++) bp->b_pages[j] = ma[i + j]; bp->b_npages = n; /* * Must set dirty range for NFS to work. */ bp->b_dirtyoff = 0; bp->b_dirtyend = bp->b_bcount; VM_CNT_INC(v_swapout); VM_CNT_ADD(v_swappgsout, bp->b_npages); /* * We unconditionally set rtvals[] to VM_PAGER_PEND so that we * can call the async completion routine at the end of a * synchronous I/O operation. Otherwise, our caller would * perform duplicate unbusy and wakeup operations on the page * and object, respectively. */ for (j = 0; j < n; j++) rtvals[i + j] = VM_PAGER_PEND; /* * asynchronous * * NOTE: b_blkno is destroyed by the call to swapdev_strategy. */ if (async) { bp->b_iodone = swp_pager_async_iodone; BUF_KERNPROC(bp); swp_pager_strategy(bp); continue; } /* * synchronous * * NOTE: b_blkno is destroyed by the call to swapdev_strategy. */ bp->b_iodone = bdone; swp_pager_strategy(bp); /* * Wait for the sync I/O to complete. */ bwait(bp, PVM, "swwrt"); /* * Now that we are through with the bp, we can call the * normal async completion, which frees everything up. */ swp_pager_async_iodone(bp); } swp_pager_freeswapspace(s_free, n_free); VM_OBJECT_WLOCK(object); } /* * swp_pager_async_iodone: * * Completion routine for asynchronous reads and writes from/to swap. * Also called manually by synchronous code to finish up a bp. * * This routine may not sleep. */ static void swp_pager_async_iodone(struct buf *bp) { int i; vm_object_t object = NULL; /* * Report error - unless we ran out of memory, in which case * we've already logged it in swapgeom_strategy(). */ if (bp->b_ioflags & BIO_ERROR && bp->b_error != ENOMEM) { printf( "swap_pager: I/O error - %s failed; blkno %ld," "size %ld, error %d\n", ((bp->b_iocmd == BIO_READ) ? "pagein" : "pageout"), (long)bp->b_blkno, (long)bp->b_bcount, bp->b_error ); } /* * remove the mapping for kernel virtual */ if (buf_mapped(bp)) pmap_qremove((vm_offset_t)bp->b_data, bp->b_npages); else bp->b_data = bp->b_kvabase; if (bp->b_npages) { object = bp->b_pages[0]->object; VM_OBJECT_WLOCK(object); } /* * cleanup pages. If an error occurs writing to swap, we are in * very serious trouble. If it happens to be a disk error, though, * we may be able to recover by reassigning the swap later on. So * in this case we remove the m->swapblk assignment for the page * but do not free it in the rlist. The errornous block(s) are thus * never reallocated as swap. Redirty the page and continue. */ for (i = 0; i < bp->b_npages; ++i) { vm_page_t m = bp->b_pages[i]; m->oflags &= ~VPO_SWAPINPROG; if (m->oflags & VPO_SWAPSLEEP) { m->oflags &= ~VPO_SWAPSLEEP; wakeup(&object->handle); } /* We always have space after I/O, successful or not. */ vm_page_aflag_set(m, PGA_SWAP_SPACE); if (bp->b_ioflags & BIO_ERROR) { /* * If an error occurs I'd love to throw the swapblk * away without freeing it back to swapspace, so it * can never be used again. But I can't from an * interrupt. */ if (bp->b_iocmd == BIO_READ) { /* * NOTE: for reads, m->dirty will probably * be overridden by the original caller of * getpages so don't play cute tricks here. */ vm_page_invalid(m); } else { /* * If a write error occurs, reactivate page * so it doesn't clog the inactive list, * then finish the I/O. */ MPASS(m->dirty == VM_PAGE_BITS_ALL); /* PQ_UNSWAPPABLE? */ vm_page_activate(m); vm_page_sunbusy(m); } } else if (bp->b_iocmd == BIO_READ) { /* * NOTE: for reads, m->dirty will probably be * overridden by the original caller of getpages so * we cannot set them in order to free the underlying * swap in a low-swap situation. I don't think we'd * want to do that anyway, but it was an optimization * that existed in the old swapper for a time before * it got ripped out due to precisely this problem. */ KASSERT(!pmap_page_is_mapped(m), ("swp_pager_async_iodone: page %p is mapped", m)); KASSERT(m->dirty == 0, ("swp_pager_async_iodone: page %p is dirty", m)); vm_page_valid(m); if (i < bp->b_pgbefore || i >= bp->b_npages - bp->b_pgafter) vm_page_readahead_finish(m); } else { /* * For write success, clear the dirty * status, then finish the I/O ( which decrements the * busy count and possibly wakes waiter's up ). * A page is only written to swap after a period of * inactivity. Therefore, we do not expect it to be * reused. */ KASSERT(!pmap_page_is_write_mapped(m), ("swp_pager_async_iodone: page %p is not write" " protected", m)); vm_page_undirty(m); vm_page_deactivate_noreuse(m); vm_page_sunbusy(m); } } /* * adjust pip. NOTE: the original parent may still have its own * pip refs on the object. */ if (object != NULL) { vm_object_pip_wakeupn(object, bp->b_npages); VM_OBJECT_WUNLOCK(object); } /* * swapdev_strategy() manually sets b_vp and b_bufobj before calling * bstrategy(). Set them back to NULL now we're done with it, or we'll * trigger a KASSERT in relpbuf(). */ if (bp->b_vp) { bp->b_vp = NULL; bp->b_bufobj = NULL; } /* * release the physical I/O buffer */ if (bp->b_flags & B_ASYNC) { mtx_lock(&swbuf_mtx); if (++nsw_wcount_async == 1) wakeup(&nsw_wcount_async); mtx_unlock(&swbuf_mtx); } uma_zfree((bp->b_iocmd == BIO_READ) ? swrbuf_zone : swwbuf_zone, bp); } int swap_pager_nswapdev(void) { return (nswapdev); } static void swp_pager_force_dirty(vm_page_t m) { vm_page_dirty(m); swap_pager_unswapped(m); vm_page_launder(m); } /* * swap_pager_swapoff_object: * * Page in all of the pages that have been paged out for an object * to a swap device. */ static void swap_pager_swapoff_object(struct swdevt *sp, vm_object_t object) { struct swblk *sb; vm_page_t m; vm_pindex_t pi; daddr_t blk; int i, nv, rahead, rv; KASSERT(object->type == OBJT_SWAP, ("%s: Object not swappable", __func__)); for (pi = 0; (sb = SWAP_PCTRIE_LOOKUP_GE( &object->un_pager.swp.swp_blks, pi)) != NULL; ) { if ((object->flags & OBJ_DEAD) != 0) { /* * Make sure that pending writes finish before * returning. */ vm_object_pip_wait(object, "swpoff"); swp_pager_meta_free_all(object); break; } for (i = 0; i < SWAP_META_PAGES; i++) { /* * Count the number of contiguous valid blocks. */ for (nv = 0; nv < SWAP_META_PAGES - i; nv++) { blk = sb->d[i + nv]; if (!swp_pager_isondev(blk, sp) || blk == SWAPBLK_NONE) break; } if (nv == 0) continue; /* * Look for a page corresponding to the first * valid block and ensure that any pending paging * operations on it are complete. If the page is valid, * mark it dirty and free the swap block. Try to batch * this operation since it may cause sp to be freed, * meaning that we must restart the scan. Avoid busying * valid pages since we may block forever on kernel * stack pages. */ m = vm_page_lookup(object, sb->p + i); if (m == NULL) { m = vm_page_alloc(object, sb->p + i, VM_ALLOC_NORMAL | VM_ALLOC_WAITFAIL); if (m == NULL) break; } else { if ((m->oflags & VPO_SWAPINPROG) != 0) { m->oflags |= VPO_SWAPSLEEP; VM_OBJECT_SLEEP(object, &object->handle, PSWP, "swpoff", 0); break; } if (vm_page_all_valid(m)) { do { swp_pager_force_dirty(m); } while (--nv > 0 && (m = vm_page_next(m)) != NULL && vm_page_all_valid(m) && (m->oflags & VPO_SWAPINPROG) == 0); break; } if (!vm_page_busy_acquire(m, VM_ALLOC_WAITFAIL)) break; } vm_object_pip_add(object, 1); rahead = SWAP_META_PAGES; rv = swap_pager_getpages_locked(object, &m, 1, NULL, &rahead); if (rv != VM_PAGER_OK) panic("%s: read from swap failed: %d", __func__, rv); vm_object_pip_wakeupn(object, 1); VM_OBJECT_WLOCK(object); vm_page_xunbusy(m); /* * The object lock was dropped so we must restart the * scan of this swap block. Pages paged in during this * iteration will be marked dirty in a future iteration. */ break; } if (i == SWAP_META_PAGES) pi = sb->p + SWAP_META_PAGES; } } /* * swap_pager_swapoff: * * Page in all of the pages that have been paged out to the * given device. The corresponding blocks in the bitmap must be * marked as allocated and the device must be flagged SW_CLOSING. * There may be no processes swapped out to the device. * * This routine may block. */ static void swap_pager_swapoff(struct swdevt *sp) { vm_object_t object; int retries; sx_assert(&swdev_syscall_lock, SA_XLOCKED); retries = 0; full_rescan: mtx_lock(&vm_object_list_mtx); TAILQ_FOREACH(object, &vm_object_list, object_list) { if (object->type != OBJT_SWAP) continue; mtx_unlock(&vm_object_list_mtx); /* Depends on type-stability. */ VM_OBJECT_WLOCK(object); /* * Dead objects are eventually terminated on their own. */ if ((object->flags & OBJ_DEAD) != 0) goto next_obj; /* * Sync with fences placed after pctrie * initialization. We must not access pctrie below * unless we checked that our object is swap and not * dead. */ atomic_thread_fence_acq(); if (object->type != OBJT_SWAP) goto next_obj; swap_pager_swapoff_object(sp, object); next_obj: VM_OBJECT_WUNLOCK(object); mtx_lock(&vm_object_list_mtx); } mtx_unlock(&vm_object_list_mtx); if (sp->sw_used) { /* * Objects may be locked or paging to the device being * removed, so we will miss their pages and need to * make another pass. We have marked this device as * SW_CLOSING, so the activity should finish soon. */ retries++; if (retries > 100) { panic("swapoff: failed to locate %d swap blocks", sp->sw_used); } pause("swpoff", hz / 20); goto full_rescan; } EVENTHANDLER_INVOKE(swapoff, sp); } /************************************************************************ * SWAP META DATA * ************************************************************************ * * These routines manipulate the swap metadata stored in the * OBJT_SWAP object. * * Swap metadata is implemented with a global hash and not directly * linked into the object. Instead the object simply contains * appropriate tracking counters. */ /* * SWP_PAGER_SWBLK_EMPTY() - is a range of blocks free? */ static bool swp_pager_swblk_empty(struct swblk *sb, int start, int limit) { int i; MPASS(0 <= start && start <= limit && limit <= SWAP_META_PAGES); for (i = start; i < limit; i++) { if (sb->d[i] != SWAPBLK_NONE) return (false); } return (true); } /* * SWP_PAGER_FREE_EMPTY_SWBLK() - frees if a block is free * * Nothing is done if the block is still in use. */ static void swp_pager_free_empty_swblk(vm_object_t object, struct swblk *sb) { if (swp_pager_swblk_empty(sb, 0, SWAP_META_PAGES)) { SWAP_PCTRIE_REMOVE(&object->un_pager.swp.swp_blks, sb->p); uma_zfree(swblk_zone, sb); } } /* * SWP_PAGER_META_BUILD() - add swap block to swap meta data for object * * We first convert the object to a swap object if it is a default * object. * * The specified swapblk is added to the object's swap metadata. If * the swapblk is not valid, it is freed instead. Any previously * assigned swapblk is returned. */ static daddr_t swp_pager_meta_build(vm_object_t object, vm_pindex_t pindex, daddr_t swapblk) { static volatile int swblk_zone_exhausted, swpctrie_zone_exhausted; struct swblk *sb, *sb1; vm_pindex_t modpi, rdpi; daddr_t prev_swapblk; int error, i; VM_OBJECT_ASSERT_WLOCKED(object); /* * Convert default object to swap object if necessary */ if (object->type != OBJT_SWAP) { pctrie_init(&object->un_pager.swp.swp_blks); /* * Ensure that swap_pager_swapoff()'s iteration over * object_list does not see a garbage pctrie. */ atomic_thread_fence_rel(); object->type = OBJT_SWAP; object->un_pager.swp.writemappings = 0; KASSERT((object->flags & OBJ_ANON) != 0 || object->handle == NULL, ("default pager %p with handle %p", object, object->handle)); } rdpi = rounddown(pindex, SWAP_META_PAGES); sb = SWAP_PCTRIE_LOOKUP(&object->un_pager.swp.swp_blks, rdpi); if (sb == NULL) { if (swapblk == SWAPBLK_NONE) return (SWAPBLK_NONE); for (;;) { sb = uma_zalloc(swblk_zone, M_NOWAIT | (curproc == pageproc ? M_USE_RESERVE : 0)); if (sb != NULL) { sb->p = rdpi; for (i = 0; i < SWAP_META_PAGES; i++) sb->d[i] = SWAPBLK_NONE; if (atomic_cmpset_int(&swblk_zone_exhausted, 1, 0)) printf("swblk zone ok\n"); break; } VM_OBJECT_WUNLOCK(object); if (uma_zone_exhausted(swblk_zone)) { if (atomic_cmpset_int(&swblk_zone_exhausted, 0, 1)) printf("swap blk zone exhausted, " "increase kern.maxswzone\n"); vm_pageout_oom(VM_OOM_SWAPZ); pause("swzonxb", 10); } else uma_zwait(swblk_zone); VM_OBJECT_WLOCK(object); sb = SWAP_PCTRIE_LOOKUP(&object->un_pager.swp.swp_blks, rdpi); if (sb != NULL) /* * Somebody swapped out a nearby page, * allocating swblk at the rdpi index, * while we dropped the object lock. */ goto allocated; } for (;;) { error = SWAP_PCTRIE_INSERT( &object->un_pager.swp.swp_blks, sb); if (error == 0) { if (atomic_cmpset_int(&swpctrie_zone_exhausted, 1, 0)) printf("swpctrie zone ok\n"); break; } VM_OBJECT_WUNLOCK(object); if (uma_zone_exhausted(swpctrie_zone)) { if (atomic_cmpset_int(&swpctrie_zone_exhausted, 0, 1)) printf("swap pctrie zone exhausted, " "increase kern.maxswzone\n"); vm_pageout_oom(VM_OOM_SWAPZ); pause("swzonxp", 10); } else uma_zwait(swpctrie_zone); VM_OBJECT_WLOCK(object); sb1 = SWAP_PCTRIE_LOOKUP(&object->un_pager.swp.swp_blks, rdpi); if (sb1 != NULL) { uma_zfree(swblk_zone, sb); sb = sb1; goto allocated; } } } allocated: MPASS(sb->p == rdpi); modpi = pindex % SWAP_META_PAGES; /* Return prior contents of metadata. */ prev_swapblk = sb->d[modpi]; /* Enter block into metadata. */ sb->d[modpi] = swapblk; /* * Free the swblk if we end up with the empty page run. */ if (swapblk == SWAPBLK_NONE) swp_pager_free_empty_swblk(object, sb); return (prev_swapblk); } /* * SWP_PAGER_META_TRANSFER() - free a range of blocks in the srcobject's swap * metadata, or transfer it into dstobject. * * This routine will free swap metadata structures as they are cleaned * out. */ static void swp_pager_meta_transfer(vm_object_t srcobject, vm_object_t dstobject, vm_pindex_t pindex, vm_pindex_t count) { struct swblk *sb; daddr_t n_free, s_free; vm_pindex_t offset, last; int i, limit, start; VM_OBJECT_ASSERT_WLOCKED(srcobject); if (srcobject->type != OBJT_SWAP || count == 0) return; swp_pager_init_freerange(&s_free, &n_free); offset = pindex; last = pindex + count; for (;;) { sb = SWAP_PCTRIE_LOOKUP_GE(&srcobject->un_pager.swp.swp_blks, rounddown(pindex, SWAP_META_PAGES)); if (sb == NULL || sb->p >= last) break; start = pindex > sb->p ? pindex - sb->p : 0; limit = last - sb->p < SWAP_META_PAGES ? last - sb->p : SWAP_META_PAGES; for (i = start; i < limit; i++) { if (sb->d[i] == SWAPBLK_NONE) continue; if (dstobject == NULL || !swp_pager_xfer_source(srcobject, dstobject, sb->p + i - offset, sb->d[i])) { swp_pager_update_freerange(&s_free, &n_free, sb->d[i]); } sb->d[i] = SWAPBLK_NONE; } pindex = sb->p + SWAP_META_PAGES; if (swp_pager_swblk_empty(sb, 0, start) && swp_pager_swblk_empty(sb, limit, SWAP_META_PAGES)) { SWAP_PCTRIE_REMOVE(&srcobject->un_pager.swp.swp_blks, sb->p); uma_zfree(swblk_zone, sb); } } swp_pager_freeswapspace(s_free, n_free); } /* * SWP_PAGER_META_FREE() - free a range of blocks in the object's swap metadata * * The requested range of blocks is freed, with any associated swap * returned to the swap bitmap. * * This routine will free swap metadata structures as they are cleaned * out. This routine does *NOT* operate on swap metadata associated * with resident pages. */ static void swp_pager_meta_free(vm_object_t object, vm_pindex_t pindex, vm_pindex_t count) { swp_pager_meta_transfer(object, NULL, pindex, count); } /* * SWP_PAGER_META_FREE_ALL() - destroy all swap metadata associated with object * * This routine locates and destroys all swap metadata associated with * an object. */ static void swp_pager_meta_free_all(vm_object_t object) { struct swblk *sb; daddr_t n_free, s_free; vm_pindex_t pindex; int i; VM_OBJECT_ASSERT_WLOCKED(object); if (object->type != OBJT_SWAP) return; swp_pager_init_freerange(&s_free, &n_free); for (pindex = 0; (sb = SWAP_PCTRIE_LOOKUP_GE( &object->un_pager.swp.swp_blks, pindex)) != NULL;) { pindex = sb->p + SWAP_META_PAGES; for (i = 0; i < SWAP_META_PAGES; i++) { if (sb->d[i] == SWAPBLK_NONE) continue; swp_pager_update_freerange(&s_free, &n_free, sb->d[i]); } SWAP_PCTRIE_REMOVE(&object->un_pager.swp.swp_blks, sb->p); uma_zfree(swblk_zone, sb); } swp_pager_freeswapspace(s_free, n_free); } /* * SWP_PAGER_METACTL() - misc control of swap meta data. * * This routine is capable of looking up, or removing swapblk * assignments in the swap meta data. It returns the swapblk being * looked-up, popped, or SWAPBLK_NONE if the block was invalid. * * When acting on a busy resident page and paging is in progress, we * have to wait until paging is complete but otherwise can act on the * busy page. */ static daddr_t swp_pager_meta_lookup(vm_object_t object, vm_pindex_t pindex) { struct swblk *sb; VM_OBJECT_ASSERT_LOCKED(object); /* * The meta data only exists if the object is OBJT_SWAP * and even then might not be allocated yet. */ KASSERT(object->type == OBJT_SWAP, ("Lookup object not swappable")); sb = SWAP_PCTRIE_LOOKUP(&object->un_pager.swp.swp_blks, rounddown(pindex, SWAP_META_PAGES)); if (sb == NULL) return (SWAPBLK_NONE); return (sb->d[pindex % SWAP_META_PAGES]); } /* * Returns the least page index which is greater than or equal to the * parameter pindex and for which there is a swap block allocated. * Returns object's size if the object's type is not swap or if there * are no allocated swap blocks for the object after the requested * pindex. */ vm_pindex_t swap_pager_find_least(vm_object_t object, vm_pindex_t pindex) { struct swblk *sb; int i; VM_OBJECT_ASSERT_LOCKED(object); if (object->type != OBJT_SWAP) return (object->size); sb = SWAP_PCTRIE_LOOKUP_GE(&object->un_pager.swp.swp_blks, rounddown(pindex, SWAP_META_PAGES)); if (sb == NULL) return (object->size); if (sb->p < pindex) { for (i = pindex % SWAP_META_PAGES; i < SWAP_META_PAGES; i++) { if (sb->d[i] != SWAPBLK_NONE) return (sb->p + i); } sb = SWAP_PCTRIE_LOOKUP_GE(&object->un_pager.swp.swp_blks, roundup(pindex, SWAP_META_PAGES)); if (sb == NULL) return (object->size); } for (i = 0; i < SWAP_META_PAGES; i++) { if (sb->d[i] != SWAPBLK_NONE) return (sb->p + i); } /* * We get here if a swblk is present in the trie but it * doesn't map any blocks. */ MPASS(0); return (object->size); } /* * System call swapon(name) enables swapping on device name, * which must be in the swdevsw. Return EBUSY * if already swapping on this device. */ #ifndef _SYS_SYSPROTO_H_ struct swapon_args { char *name; }; #endif /* * MPSAFE */ /* ARGSUSED */ int sys_swapon(struct thread *td, struct swapon_args *uap) { struct vattr attr; struct vnode *vp; struct nameidata nd; int error; error = priv_check(td, PRIV_SWAPON); if (error) return (error); sx_xlock(&swdev_syscall_lock); /* * Swap metadata may not fit in the KVM if we have physical * memory of >1GB. */ if (swblk_zone == NULL) { error = ENOMEM; goto done; } NDINIT(&nd, LOOKUP, ISOPEN | FOLLOW | AUDITVNODE1, UIO_USERSPACE, uap->name, td); error = namei(&nd); if (error) goto done; NDFREE(&nd, NDF_ONLY_PNBUF); vp = nd.ni_vp; if (vn_isdisk(vp, &error)) { error = swapongeom(vp); } else if (vp->v_type == VREG && (vp->v_mount->mnt_vfc->vfc_flags & VFCF_NETWORK) != 0 && (error = VOP_GETATTR(vp, &attr, td->td_ucred)) == 0) { /* * Allow direct swapping to NFS regular files in the same * way that nfs_mountroot() sets up diskless swapping. */ error = swaponvp(td, vp, attr.va_size / DEV_BSIZE); } if (error) vrele(vp); done: sx_xunlock(&swdev_syscall_lock); return (error); } /* * Check that the total amount of swap currently configured does not * exceed half the theoretical maximum. If it does, print a warning * message. */ static void swapon_check_swzone(void) { /* recommend using no more than half that amount */ if (swap_total > swap_maxpages / 2) { printf("warning: total configured swap (%lu pages) " "exceeds maximum recommended amount (%lu pages).\n", swap_total, swap_maxpages / 2); printf("warning: increase kern.maxswzone " "or reduce amount of swap.\n"); } } static void swaponsomething(struct vnode *vp, void *id, u_long nblks, sw_strategy_t *strategy, sw_close_t *close, dev_t dev, int flags) { struct swdevt *sp, *tsp; daddr_t dvbase; u_long mblocks; /* * nblks is in DEV_BSIZE'd chunks, convert to PAGE_SIZE'd chunks. * First chop nblks off to page-align it, then convert. * * sw->sw_nblks is in page-sized chunks now too. */ nblks &= ~(ctodb(1) - 1); nblks = dbtoc(nblks); /* * If we go beyond this, we get overflows in the radix * tree bitmap code. */ mblocks = 0x40000000 / BLIST_META_RADIX; if (nblks > mblocks) { printf( "WARNING: reducing swap size to maximum of %luMB per unit\n", mblocks / 1024 / 1024 * PAGE_SIZE); nblks = mblocks; } sp = malloc(sizeof *sp, M_VMPGDATA, M_WAITOK | M_ZERO); sp->sw_vp = vp; sp->sw_id = id; sp->sw_dev = dev; sp->sw_nblks = nblks; sp->sw_used = 0; sp->sw_strategy = strategy; sp->sw_close = close; sp->sw_flags = flags; sp->sw_blist = blist_create(nblks, M_WAITOK); /* * Do not free the first blocks in order to avoid overwriting * any bsd label at the front of the partition */ blist_free(sp->sw_blist, howmany(BBSIZE, PAGE_SIZE), nblks - howmany(BBSIZE, PAGE_SIZE)); dvbase = 0; mtx_lock(&sw_dev_mtx); TAILQ_FOREACH(tsp, &swtailq, sw_list) { if (tsp->sw_end >= dvbase) { /* * We put one uncovered page between the devices * in order to definitively prevent any cross-device * I/O requests */ dvbase = tsp->sw_end + 1; } } sp->sw_first = dvbase; sp->sw_end = dvbase + nblks; TAILQ_INSERT_TAIL(&swtailq, sp, sw_list); nswapdev++; swap_pager_avail += nblks - howmany(BBSIZE, PAGE_SIZE); swap_total += nblks; swapon_check_swzone(); swp_sizecheck(); mtx_unlock(&sw_dev_mtx); EVENTHANDLER_INVOKE(swapon, sp); } /* * SYSCALL: swapoff(devname) * * Disable swapping on the given device. * * XXX: Badly designed system call: it should use a device index * rather than filename as specification. We keep sw_vp around * only to make this work. */ #ifndef _SYS_SYSPROTO_H_ struct swapoff_args { char *name; }; #endif /* * MPSAFE */ /* ARGSUSED */ int sys_swapoff(struct thread *td, struct swapoff_args *uap) { struct vnode *vp; struct nameidata nd; struct swdevt *sp; int error; error = priv_check(td, PRIV_SWAPOFF); if (error) return (error); sx_xlock(&swdev_syscall_lock); NDINIT(&nd, LOOKUP, FOLLOW | AUDITVNODE1, UIO_USERSPACE, uap->name, td); error = namei(&nd); if (error) goto done; NDFREE(&nd, NDF_ONLY_PNBUF); vp = nd.ni_vp; mtx_lock(&sw_dev_mtx); TAILQ_FOREACH(sp, &swtailq, sw_list) { if (sp->sw_vp == vp) break; } mtx_unlock(&sw_dev_mtx); if (sp == NULL) { error = EINVAL; goto done; } error = swapoff_one(sp, td->td_ucred); done: sx_xunlock(&swdev_syscall_lock); return (error); } static int swapoff_one(struct swdevt *sp, struct ucred *cred) { u_long nblks; #ifdef MAC int error; #endif sx_assert(&swdev_syscall_lock, SA_XLOCKED); #ifdef MAC (void) vn_lock(sp->sw_vp, LK_EXCLUSIVE | LK_RETRY); error = mac_system_check_swapoff(cred, sp->sw_vp); (void) VOP_UNLOCK(sp->sw_vp); if (error != 0) return (error); #endif nblks = sp->sw_nblks; /* * We can turn off this swap device safely only if the * available virtual memory in the system will fit the amount * of data we will have to page back in, plus an epsilon so * the system doesn't become critically low on swap space. */ if (vm_free_count() + swap_pager_avail < nblks + nswap_lowat) return (ENOMEM); /* * Prevent further allocations on this device. */ mtx_lock(&sw_dev_mtx); sp->sw_flags |= SW_CLOSING; swap_pager_avail -= blist_fill(sp->sw_blist, 0, nblks); swap_total -= nblks; mtx_unlock(&sw_dev_mtx); /* * Page in the contents of the device and close it. */ swap_pager_swapoff(sp); sp->sw_close(curthread, sp); mtx_lock(&sw_dev_mtx); sp->sw_id = NULL; TAILQ_REMOVE(&swtailq, sp, sw_list); nswapdev--; if (nswapdev == 0) { swap_pager_full = 2; swap_pager_almost_full = 1; } if (swdevhd == sp) swdevhd = NULL; mtx_unlock(&sw_dev_mtx); blist_destroy(sp->sw_blist); free(sp, M_VMPGDATA); return (0); } void swapoff_all(void) { struct swdevt *sp, *spt; const char *devname; int error; sx_xlock(&swdev_syscall_lock); mtx_lock(&sw_dev_mtx); TAILQ_FOREACH_SAFE(sp, &swtailq, sw_list, spt) { mtx_unlock(&sw_dev_mtx); if (vn_isdisk(sp->sw_vp, NULL)) devname = devtoname(sp->sw_vp->v_rdev); else devname = "[file]"; error = swapoff_one(sp, thread0.td_ucred); if (error != 0) { printf("Cannot remove swap device %s (error=%d), " "skipping.\n", devname, error); } else if (bootverbose) { printf("Swap device %s removed.\n", devname); } mtx_lock(&sw_dev_mtx); } mtx_unlock(&sw_dev_mtx); sx_xunlock(&swdev_syscall_lock); } void swap_pager_status(int *total, int *used) { struct swdevt *sp; *total = 0; *used = 0; mtx_lock(&sw_dev_mtx); TAILQ_FOREACH(sp, &swtailq, sw_list) { *total += sp->sw_nblks; *used += sp->sw_used; } mtx_unlock(&sw_dev_mtx); } int swap_dev_info(int name, struct xswdev *xs, char *devname, size_t len) { struct swdevt *sp; const char *tmp_devname; int error, n; n = 0; error = ENOENT; mtx_lock(&sw_dev_mtx); TAILQ_FOREACH(sp, &swtailq, sw_list) { if (n != name) { n++; continue; } xs->xsw_version = XSWDEV_VERSION; xs->xsw_dev = sp->sw_dev; xs->xsw_flags = sp->sw_flags; xs->xsw_nblks = sp->sw_nblks; xs->xsw_used = sp->sw_used; if (devname != NULL) { if (vn_isdisk(sp->sw_vp, NULL)) tmp_devname = devtoname(sp->sw_vp->v_rdev); else tmp_devname = "[file]"; strncpy(devname, tmp_devname, len); } error = 0; break; } mtx_unlock(&sw_dev_mtx); return (error); } #if defined(COMPAT_FREEBSD11) #define XSWDEV_VERSION_11 1 struct xswdev11 { u_int xsw_version; uint32_t xsw_dev; int xsw_flags; int xsw_nblks; int xsw_used; }; #endif #if defined(__amd64__) && defined(COMPAT_FREEBSD32) struct xswdev32 { u_int xsw_version; u_int xsw_dev1, xsw_dev2; int xsw_flags; int xsw_nblks; int xsw_used; }; #endif static int sysctl_vm_swap_info(SYSCTL_HANDLER_ARGS) { struct xswdev xs; #if defined(__amd64__) && defined(COMPAT_FREEBSD32) struct xswdev32 xs32; #endif #if defined(COMPAT_FREEBSD11) struct xswdev11 xs11; #endif int error; if (arg2 != 1) /* name length */ return (EINVAL); error = swap_dev_info(*(int *)arg1, &xs, NULL, 0); if (error != 0) return (error); #if defined(__amd64__) && defined(COMPAT_FREEBSD32) if (req->oldlen == sizeof(xs32)) { xs32.xsw_version = XSWDEV_VERSION; xs32.xsw_dev1 = xs.xsw_dev; xs32.xsw_dev2 = xs.xsw_dev >> 32; xs32.xsw_flags = xs.xsw_flags; xs32.xsw_nblks = xs.xsw_nblks; xs32.xsw_used = xs.xsw_used; error = SYSCTL_OUT(req, &xs32, sizeof(xs32)); return (error); } #endif #if defined(COMPAT_FREEBSD11) if (req->oldlen == sizeof(xs11)) { xs11.xsw_version = XSWDEV_VERSION_11; xs11.xsw_dev = xs.xsw_dev; /* truncation */ xs11.xsw_flags = xs.xsw_flags; xs11.xsw_nblks = xs.xsw_nblks; xs11.xsw_used = xs.xsw_used; error = SYSCTL_OUT(req, &xs11, sizeof(xs11)); return (error); } #endif error = SYSCTL_OUT(req, &xs, sizeof(xs)); return (error); } SYSCTL_INT(_vm, OID_AUTO, nswapdev, CTLFLAG_RD, &nswapdev, 0, "Number of swap devices"); SYSCTL_NODE(_vm, OID_AUTO, swap_info, CTLFLAG_RD | CTLFLAG_MPSAFE, sysctl_vm_swap_info, "Swap statistics by device"); /* * Count the approximate swap usage in pages for a vmspace. The * shadowed or not yet copied on write swap blocks are not accounted. * The map must be locked. */ long vmspace_swap_count(struct vmspace *vmspace) { vm_map_t map; vm_map_entry_t cur; vm_object_t object; struct swblk *sb; vm_pindex_t e, pi; long count; int i; map = &vmspace->vm_map; count = 0; VM_MAP_ENTRY_FOREACH(cur, map) { if ((cur->eflags & MAP_ENTRY_IS_SUB_MAP) != 0) continue; object = cur->object.vm_object; if (object == NULL || object->type != OBJT_SWAP) continue; VM_OBJECT_RLOCK(object); if (object->type != OBJT_SWAP) goto unlock; pi = OFF_TO_IDX(cur->offset); e = pi + OFF_TO_IDX(cur->end - cur->start); for (;; pi = sb->p + SWAP_META_PAGES) { sb = SWAP_PCTRIE_LOOKUP_GE( &object->un_pager.swp.swp_blks, pi); if (sb == NULL || sb->p >= e) break; for (i = 0; i < SWAP_META_PAGES; i++) { if (sb->p + i < e && sb->d[i] != SWAPBLK_NONE) count++; } } unlock: VM_OBJECT_RUNLOCK(object); } return (count); } /* * GEOM backend * * Swapping onto disk devices. * */ static g_orphan_t swapgeom_orphan; static struct g_class g_swap_class = { .name = "SWAP", .version = G_VERSION, .orphan = swapgeom_orphan, }; DECLARE_GEOM_CLASS(g_swap_class, g_class); static void swapgeom_close_ev(void *arg, int flags) { struct g_consumer *cp; cp = arg; g_access(cp, -1, -1, 0); g_detach(cp); g_destroy_consumer(cp); } /* * Add a reference to the g_consumer for an inflight transaction. */ static void swapgeom_acquire(struct g_consumer *cp) { mtx_assert(&sw_dev_mtx, MA_OWNED); cp->index++; } /* * Remove a reference from the g_consumer. Post a close event if all * references go away, since the function might be called from the * biodone context. */ static void swapgeom_release(struct g_consumer *cp, struct swdevt *sp) { mtx_assert(&sw_dev_mtx, MA_OWNED); cp->index--; if (cp->index == 0) { if (g_post_event(swapgeom_close_ev, cp, M_NOWAIT, NULL) == 0) sp->sw_id = NULL; } } static void swapgeom_done(struct bio *bp2) { struct swdevt *sp; struct buf *bp; struct g_consumer *cp; bp = bp2->bio_caller2; cp = bp2->bio_from; bp->b_ioflags = bp2->bio_flags; if (bp2->bio_error) bp->b_ioflags |= BIO_ERROR; bp->b_resid = bp->b_bcount - bp2->bio_completed; bp->b_error = bp2->bio_error; bp->b_caller1 = NULL; bufdone(bp); sp = bp2->bio_caller1; mtx_lock(&sw_dev_mtx); swapgeom_release(cp, sp); mtx_unlock(&sw_dev_mtx); g_destroy_bio(bp2); } static void swapgeom_strategy(struct buf *bp, struct swdevt *sp) { struct bio *bio; struct g_consumer *cp; mtx_lock(&sw_dev_mtx); cp = sp->sw_id; if (cp == NULL) { mtx_unlock(&sw_dev_mtx); bp->b_error = ENXIO; bp->b_ioflags |= BIO_ERROR; bufdone(bp); return; } swapgeom_acquire(cp); mtx_unlock(&sw_dev_mtx); if (bp->b_iocmd == BIO_WRITE) bio = g_new_bio(); else bio = g_alloc_bio(); if (bio == NULL) { mtx_lock(&sw_dev_mtx); swapgeom_release(cp, sp); mtx_unlock(&sw_dev_mtx); bp->b_error = ENOMEM; bp->b_ioflags |= BIO_ERROR; printf("swap_pager: cannot allocate bio\n"); bufdone(bp); return; } bp->b_caller1 = bio; bio->bio_caller1 = sp; bio->bio_caller2 = bp; bio->bio_cmd = bp->b_iocmd; bio->bio_offset = (bp->b_blkno - sp->sw_first) * PAGE_SIZE; bio->bio_length = bp->b_bcount; bio->bio_done = swapgeom_done; if (!buf_mapped(bp)) { bio->bio_ma = bp->b_pages; bio->bio_data = unmapped_buf; bio->bio_ma_offset = (vm_offset_t)bp->b_offset & PAGE_MASK; bio->bio_ma_n = bp->b_npages; bio->bio_flags |= BIO_UNMAPPED; } else { bio->bio_data = bp->b_data; bio->bio_ma = NULL; } g_io_request(bio, cp); return; } static void swapgeom_orphan(struct g_consumer *cp) { struct swdevt *sp; int destroy; mtx_lock(&sw_dev_mtx); TAILQ_FOREACH(sp, &swtailq, sw_list) { if (sp->sw_id == cp) { sp->sw_flags |= SW_CLOSING; break; } } /* * Drop reference we were created with. Do directly since we're in a * special context where we don't have to queue the call to * swapgeom_close_ev(). */ cp->index--; destroy = ((sp != NULL) && (cp->index == 0)); if (destroy) sp->sw_id = NULL; mtx_unlock(&sw_dev_mtx); if (destroy) swapgeom_close_ev(cp, 0); } static void swapgeom_close(struct thread *td, struct swdevt *sw) { struct g_consumer *cp; mtx_lock(&sw_dev_mtx); cp = sw->sw_id; sw->sw_id = NULL; mtx_unlock(&sw_dev_mtx); /* * swapgeom_close() may be called from the biodone context, * where we cannot perform topology changes. Delegate the * work to the events thread. */ if (cp != NULL) g_waitfor_event(swapgeom_close_ev, cp, M_WAITOK, NULL); } static int swapongeom_locked(struct cdev *dev, struct vnode *vp) { struct g_provider *pp; struct g_consumer *cp; static struct g_geom *gp; struct swdevt *sp; u_long nblks; int error; pp = g_dev_getprovider(dev); if (pp == NULL) return (ENODEV); mtx_lock(&sw_dev_mtx); TAILQ_FOREACH(sp, &swtailq, sw_list) { cp = sp->sw_id; if (cp != NULL && cp->provider == pp) { mtx_unlock(&sw_dev_mtx); return (EBUSY); } } mtx_unlock(&sw_dev_mtx); if (gp == NULL) gp = g_new_geomf(&g_swap_class, "swap"); cp = g_new_consumer(gp); cp->index = 1; /* Number of active I/Os, plus one for being active. */ cp->flags |= G_CF_DIRECT_SEND | G_CF_DIRECT_RECEIVE; g_attach(cp, pp); /* * XXX: Every time you think you can improve the margin for * footshooting, somebody depends on the ability to do so: * savecore(8) wants to write to our swapdev so we cannot * set an exclusive count :-( */ error = g_access(cp, 1, 1, 0); if (error != 0) { g_detach(cp); g_destroy_consumer(cp); return (error); } nblks = pp->mediasize / DEV_BSIZE; swaponsomething(vp, cp, nblks, swapgeom_strategy, swapgeom_close, dev2udev(dev), (pp->flags & G_PF_ACCEPT_UNMAPPED) != 0 ? SW_UNMAPPED : 0); return (0); } static int swapongeom(struct vnode *vp) { int error; vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); if (vp->v_type != VCHR || VN_IS_DOOMED(vp)) { error = ENOENT; } else { g_topology_lock(); error = swapongeom_locked(vp->v_rdev, vp); g_topology_unlock(); } VOP_UNLOCK(vp); return (error); } /* * VNODE backend * * This is used mainly for network filesystem (read: probably only tested * with NFS) swapfiles. * */ static void swapdev_strategy(struct buf *bp, struct swdevt *sp) { struct vnode *vp2; bp->b_blkno = ctodb(bp->b_blkno - sp->sw_first); vp2 = sp->sw_id; vhold(vp2); if (bp->b_iocmd == BIO_WRITE) { if (bp->b_bufobj) bufobj_wdrop(bp->b_bufobj); bufobj_wref(&vp2->v_bufobj); } if (bp->b_bufobj != &vp2->v_bufobj) bp->b_bufobj = &vp2->v_bufobj; bp->b_vp = vp2; bp->b_iooffset = dbtob(bp->b_blkno); bstrategy(bp); return; } static void swapdev_close(struct thread *td, struct swdevt *sp) { VOP_CLOSE(sp->sw_vp, FREAD | FWRITE, td->td_ucred, td); vrele(sp->sw_vp); } static int swaponvp(struct thread *td, struct vnode *vp, u_long nblks) { struct swdevt *sp; int error; if (nblks == 0) return (ENXIO); mtx_lock(&sw_dev_mtx); TAILQ_FOREACH(sp, &swtailq, sw_list) { if (sp->sw_id == vp) { mtx_unlock(&sw_dev_mtx); return (EBUSY); } } mtx_unlock(&sw_dev_mtx); (void) vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); #ifdef MAC error = mac_system_check_swapon(td->td_ucred, vp); if (error == 0) #endif error = VOP_OPEN(vp, FREAD | FWRITE, td->td_ucred, td, NULL); (void) VOP_UNLOCK(vp); if (error) return (error); swaponsomething(vp, vp, nblks, swapdev_strategy, swapdev_close, NODEV, 0); return (0); } static int sysctl_swap_async_max(SYSCTL_HANDLER_ARGS) { int error, new, n; new = nsw_wcount_async_max; error = sysctl_handle_int(oidp, &new, 0, req); if (error != 0 || req->newptr == NULL) return (error); if (new > nswbuf / 2 || new < 1) return (EINVAL); mtx_lock(&swbuf_mtx); while (nsw_wcount_async_max != new) { /* * Adjust difference. If the current async count is too low, * we will need to sqeeze our update slowly in. Sleep with a * higher priority than getpbuf() to finish faster. */ n = new - nsw_wcount_async_max; if (nsw_wcount_async + n >= 0) { nsw_wcount_async += n; nsw_wcount_async_max += n; wakeup(&nsw_wcount_async); } else { nsw_wcount_async_max -= nsw_wcount_async; nsw_wcount_async = 0; msleep(&nsw_wcount_async, &swbuf_mtx, PSWP, "swpsysctl", 0); } } mtx_unlock(&swbuf_mtx); return (0); } static void swap_pager_update_writecount(vm_object_t object, vm_offset_t start, vm_offset_t end) { VM_OBJECT_WLOCK(object); KASSERT((object->flags & OBJ_ANON) == 0, ("Splittable object with writecount")); object->un_pager.swp.writemappings += (vm_ooffset_t)end - start; VM_OBJECT_WUNLOCK(object); } static void swap_pager_release_writecount(vm_object_t object, vm_offset_t start, vm_offset_t end) { VM_OBJECT_WLOCK(object); KASSERT((object->flags & OBJ_ANON) == 0, ("Splittable object with writecount")); object->un_pager.swp.writemappings -= (vm_ooffset_t)end - start; VM_OBJECT_WUNLOCK(object); } diff --git a/sys/vm/uma_core.c b/sys/vm/uma_core.c index 8116c7672616..c93383567b8b 100644 --- a/sys/vm/uma_core.c +++ b/sys/vm/uma_core.c @@ -1,5448 +1,5445 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2002-2019 Jeffrey Roberson * Copyright (c) 2004, 2005 Bosko Milekic * Copyright (c) 2004-2006 Robert N. M. Watson * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice unmodified, this list of conditions, and the following * disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */ /* * uma_core.c Implementation of the Universal Memory allocator * * This allocator is intended to replace the multitude of similar object caches * in the standard FreeBSD kernel. The intent is to be flexible as well as * efficient. A primary design goal is to return unused memory to the rest of * the system. This will make the system as a whole more flexible due to the * ability to move memory to subsystems which most need it instead of leaving * pools of reserved memory unused. * * The basic ideas stem from similar slab/zone based allocators whose algorithms * are well known. * */ /* * TODO: * - Improve memory usage for large allocations * - Investigate cache size adjustments */ #include __FBSDID("$FreeBSD$"); #include "opt_ddb.h" #include "opt_param.h" #include "opt_vm.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifdef DEBUG_MEMGUARD #include #endif #include #ifdef INVARIANTS #define UMA_ALWAYS_CTORDTOR 1 #else #define UMA_ALWAYS_CTORDTOR 0 #endif /* * This is the zone and keg from which all zones are spawned. */ static uma_zone_t kegs; static uma_zone_t zones; /* * These are the two zones from which all offpage uma_slab_ts are allocated. * * One zone is for slab headers that can represent a larger number of items, * making the slabs themselves more efficient, and the other zone is for * headers that are smaller and represent fewer items, making the headers more * efficient. */ #define SLABZONE_SIZE(setsize) \ (sizeof(struct uma_hash_slab) + BITSET_SIZE(setsize) * SLAB_BITSETS) #define SLABZONE0_SETSIZE (PAGE_SIZE / 16) #define SLABZONE1_SETSIZE SLAB_MAX_SETSIZE #define SLABZONE0_SIZE SLABZONE_SIZE(SLABZONE0_SETSIZE) #define SLABZONE1_SIZE SLABZONE_SIZE(SLABZONE1_SETSIZE) static uma_zone_t slabzones[2]; /* * The initial hash tables come out of this zone so they can be allocated * prior to malloc coming up. */ static uma_zone_t hashzone; /* The boot-time adjusted value for cache line alignment. */ int uma_align_cache = 64 - 1; static MALLOC_DEFINE(M_UMAHASH, "UMAHash", "UMA Hash Buckets"); static MALLOC_DEFINE(M_UMA, "UMA", "UMA Misc"); /* * Are we allowed to allocate buckets? */ static int bucketdisable = 1; /* Linked list of all kegs in the system */ static LIST_HEAD(,uma_keg) uma_kegs = LIST_HEAD_INITIALIZER(uma_kegs); /* Linked list of all cache-only zones in the system */ static LIST_HEAD(,uma_zone) uma_cachezones = LIST_HEAD_INITIALIZER(uma_cachezones); /* This RW lock protects the keg list */ static struct rwlock_padalign __exclusive_cache_line uma_rwlock; /* * First available virual address for boot time allocations. */ static vm_offset_t bootstart; static vm_offset_t bootmem; static struct sx uma_reclaim_lock; /* * kmem soft limit, initialized by uma_set_limit(). Ensure that early * allocations don't trigger a wakeup of the reclaim thread. */ unsigned long uma_kmem_limit = LONG_MAX; SYSCTL_ULONG(_vm, OID_AUTO, uma_kmem_limit, CTLFLAG_RD, &uma_kmem_limit, 0, "UMA kernel memory soft limit"); unsigned long uma_kmem_total; SYSCTL_ULONG(_vm, OID_AUTO, uma_kmem_total, CTLFLAG_RD, &uma_kmem_total, 0, "UMA kernel memory usage"); /* Is the VM done starting up? */ static enum { BOOT_COLD, BOOT_KVA, BOOT_RUNNING, BOOT_SHUTDOWN, } booted = BOOT_COLD; /* * This is the handle used to schedule events that need to happen * outside of the allocation fast path. */ static struct callout uma_callout; #define UMA_TIMEOUT 20 /* Seconds for callout interval. */ /* * This structure is passed as the zone ctor arg so that I don't have to create * a special allocation function just for zones. */ struct uma_zctor_args { const char *name; size_t size; uma_ctor ctor; uma_dtor dtor; uma_init uminit; uma_fini fini; uma_import import; uma_release release; void *arg; uma_keg_t keg; int align; uint32_t flags; }; struct uma_kctor_args { uma_zone_t zone; size_t size; uma_init uminit; uma_fini fini; int align; uint32_t flags; }; struct uma_bucket_zone { uma_zone_t ubz_zone; const char *ubz_name; int ubz_entries; /* Number of items it can hold. */ int ubz_maxsize; /* Maximum allocation size per-item. */ }; /* * Compute the actual number of bucket entries to pack them in power * of two sizes for more efficient space utilization. */ #define BUCKET_SIZE(n) \ (((sizeof(void *) * (n)) - sizeof(struct uma_bucket)) / sizeof(void *)) #define BUCKET_MAX BUCKET_SIZE(256) #define BUCKET_MIN 2 struct uma_bucket_zone bucket_zones[] = { /* Literal bucket sizes. */ { NULL, "2 Bucket", 2, 4096 }, { NULL, "4 Bucket", 4, 3072 }, { NULL, "8 Bucket", 8, 2048 }, { NULL, "16 Bucket", 16, 1024 }, /* Rounded down power of 2 sizes for efficiency. */ { NULL, "32 Bucket", BUCKET_SIZE(32), 512 }, { NULL, "64 Bucket", BUCKET_SIZE(64), 256 }, { NULL, "128 Bucket", BUCKET_SIZE(128), 128 }, { NULL, "256 Bucket", BUCKET_SIZE(256), 64 }, { NULL, NULL, 0} }; /* * Flags and enumerations to be passed to internal functions. */ enum zfreeskip { SKIP_NONE = 0, SKIP_CNT = 0x00000001, SKIP_DTOR = 0x00010000, SKIP_FINI = 0x00020000, }; /* Prototypes.. */ void uma_startup1(vm_offset_t); void uma_startup2(void); static void *noobj_alloc(uma_zone_t, vm_size_t, int, uint8_t *, int); static void *page_alloc(uma_zone_t, vm_size_t, int, uint8_t *, int); static void *pcpu_page_alloc(uma_zone_t, vm_size_t, int, uint8_t *, int); static void *startup_alloc(uma_zone_t, vm_size_t, int, uint8_t *, int); static void *contig_alloc(uma_zone_t, vm_size_t, int, uint8_t *, int); static void page_free(void *, vm_size_t, uint8_t); static void pcpu_page_free(void *, vm_size_t, uint8_t); static uma_slab_t keg_alloc_slab(uma_keg_t, uma_zone_t, int, int, int); static void cache_drain(uma_zone_t); static void bucket_drain(uma_zone_t, uma_bucket_t); static void bucket_cache_reclaim(uma_zone_t zone, bool); static int keg_ctor(void *, int, void *, int); static void keg_dtor(void *, int, void *); static int zone_ctor(void *, int, void *, int); static void zone_dtor(void *, int, void *); static inline void item_dtor(uma_zone_t zone, void *item, int size, void *udata, enum zfreeskip skip); static int zero_init(void *, int, int); static void zone_free_bucket(uma_zone_t zone, uma_bucket_t bucket, void *udata, int itemdomain, bool ws); static void zone_foreach(void (*zfunc)(uma_zone_t, void *), void *); static void zone_foreach_unlocked(void (*zfunc)(uma_zone_t, void *), void *); static void zone_timeout(uma_zone_t zone, void *); static int hash_alloc(struct uma_hash *, u_int); static int hash_expand(struct uma_hash *, struct uma_hash *); static void hash_free(struct uma_hash *hash); static void uma_timeout(void *); static void uma_startup3(void); static void uma_shutdown(void); static void *zone_alloc_item(uma_zone_t, void *, int, int); static void zone_free_item(uma_zone_t, void *, void *, enum zfreeskip); static int zone_alloc_limit(uma_zone_t zone, int count, int flags); static void zone_free_limit(uma_zone_t zone, int count); static void bucket_enable(void); static void bucket_init(void); static uma_bucket_t bucket_alloc(uma_zone_t zone, void *, int); static void bucket_free(uma_zone_t zone, uma_bucket_t, void *); static void bucket_zone_drain(void); static uma_bucket_t zone_alloc_bucket(uma_zone_t, void *, int, int); static void *slab_alloc_item(uma_keg_t keg, uma_slab_t slab); static void slab_free_item(uma_zone_t zone, uma_slab_t slab, void *item); static uma_keg_t uma_kcreate(uma_zone_t zone, size_t size, uma_init uminit, uma_fini fini, int align, uint32_t flags); static int zone_import(void *, void **, int, int, int); static void zone_release(void *, void **, int); static bool cache_alloc(uma_zone_t, uma_cache_t, void *, int); static bool cache_free(uma_zone_t, uma_cache_t, void *, void *, int); static int sysctl_vm_zone_count(SYSCTL_HANDLER_ARGS); static int sysctl_vm_zone_stats(SYSCTL_HANDLER_ARGS); static int sysctl_handle_uma_zone_allocs(SYSCTL_HANDLER_ARGS); static int sysctl_handle_uma_zone_frees(SYSCTL_HANDLER_ARGS); static int sysctl_handle_uma_zone_flags(SYSCTL_HANDLER_ARGS); static int sysctl_handle_uma_slab_efficiency(SYSCTL_HANDLER_ARGS); static int sysctl_handle_uma_zone_items(SYSCTL_HANDLER_ARGS); static uint64_t uma_zone_get_allocs(uma_zone_t zone); static SYSCTL_NODE(_vm, OID_AUTO, debug, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "Memory allocation debugging"); #ifdef INVARIANTS static uint64_t uma_keg_get_allocs(uma_keg_t zone); static inline struct noslabbits *slab_dbg_bits(uma_slab_t slab, uma_keg_t keg); static bool uma_dbg_kskip(uma_keg_t keg, void *mem); static bool uma_dbg_zskip(uma_zone_t zone, void *mem); static void uma_dbg_free(uma_zone_t zone, uma_slab_t slab, void *item); static void uma_dbg_alloc(uma_zone_t zone, uma_slab_t slab, void *item); static u_int dbg_divisor = 1; SYSCTL_UINT(_vm_debug, OID_AUTO, divisor, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, &dbg_divisor, 0, "Debug & thrash every this item in memory allocator"); static counter_u64_t uma_dbg_cnt = EARLY_COUNTER; static counter_u64_t uma_skip_cnt = EARLY_COUNTER; SYSCTL_COUNTER_U64(_vm_debug, OID_AUTO, trashed, CTLFLAG_RD, &uma_dbg_cnt, "memory items debugged"); SYSCTL_COUNTER_U64(_vm_debug, OID_AUTO, skipped, CTLFLAG_RD, &uma_skip_cnt, "memory items skipped, not debugged"); #endif SYSINIT(uma_startup3, SI_SUB_VM_CONF, SI_ORDER_SECOND, uma_startup3, NULL); SYSCTL_NODE(_vm, OID_AUTO, uma, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "Universal Memory Allocator"); SYSCTL_PROC(_vm, OID_AUTO, zone_count, CTLFLAG_RD|CTLFLAG_MPSAFE|CTLTYPE_INT, 0, 0, sysctl_vm_zone_count, "I", "Number of UMA zones"); SYSCTL_PROC(_vm, OID_AUTO, zone_stats, CTLFLAG_RD|CTLFLAG_MPSAFE|CTLTYPE_STRUCT, 0, 0, sysctl_vm_zone_stats, "s,struct uma_type_header", "Zone Stats"); static int zone_warnings = 1; SYSCTL_INT(_vm, OID_AUTO, zone_warnings, CTLFLAG_RWTUN, &zone_warnings, 0, "Warn when UMA zones becomes full"); static int multipage_slabs = 1; TUNABLE_INT("vm.debug.uma_multipage_slabs", &multipage_slabs); SYSCTL_INT(_vm_debug, OID_AUTO, uma_multipage_slabs, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, &multipage_slabs, 0, "UMA may choose larger slab sizes for better efficiency"); /* * Select the slab zone for an offpage slab with the given maximum item count. */ static inline uma_zone_t slabzone(int ipers) { return (slabzones[ipers > SLABZONE0_SETSIZE]); } /* * This routine checks to see whether or not it's safe to enable buckets. */ static void bucket_enable(void) { KASSERT(booted >= BOOT_KVA, ("Bucket enable before init")); bucketdisable = vm_page_count_min(); } /* * Initialize bucket_zones, the array of zones of buckets of various sizes. * * For each zone, calculate the memory required for each bucket, consisting * of the header and an array of pointers. */ static void bucket_init(void) { struct uma_bucket_zone *ubz; int size; for (ubz = &bucket_zones[0]; ubz->ubz_entries != 0; ubz++) { size = roundup(sizeof(struct uma_bucket), sizeof(void *)); size += sizeof(void *) * ubz->ubz_entries; ubz->ubz_zone = uma_zcreate(ubz->ubz_name, size, NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_MTXCLASS | UMA_ZFLAG_BUCKET | UMA_ZONE_FIRSTTOUCH); } } /* * Given a desired number of entries for a bucket, return the zone from which * to allocate the bucket. */ static struct uma_bucket_zone * bucket_zone_lookup(int entries) { struct uma_bucket_zone *ubz; for (ubz = &bucket_zones[0]; ubz->ubz_entries != 0; ubz++) if (ubz->ubz_entries >= entries) return (ubz); ubz--; return (ubz); } static struct uma_bucket_zone * bucket_zone_max(uma_zone_t zone, int nitems) { struct uma_bucket_zone *ubz; int bpcpu; bpcpu = 2; if ((zone->uz_flags & UMA_ZONE_FIRSTTOUCH) != 0) /* Count the cross-domain bucket. */ bpcpu++; for (ubz = &bucket_zones[0]; ubz->ubz_entries != 0; ubz++) if (ubz->ubz_entries * bpcpu * mp_ncpus > nitems) break; if (ubz == &bucket_zones[0]) ubz = NULL; else ubz--; return (ubz); } static int bucket_select(int size) { struct uma_bucket_zone *ubz; ubz = &bucket_zones[0]; if (size > ubz->ubz_maxsize) return MAX((ubz->ubz_maxsize * ubz->ubz_entries) / size, 1); for (; ubz->ubz_entries != 0; ubz++) if (ubz->ubz_maxsize < size) break; ubz--; return (ubz->ubz_entries); } static uma_bucket_t bucket_alloc(uma_zone_t zone, void *udata, int flags) { struct uma_bucket_zone *ubz; uma_bucket_t bucket; /* * Don't allocate buckets early in boot. */ if (__predict_false(booted < BOOT_KVA)) return (NULL); /* * To limit bucket recursion we store the original zone flags * in a cookie passed via zalloc_arg/zfree_arg. This allows the * NOVM flag to persist even through deep recursions. We also * store ZFLAG_BUCKET once we have recursed attempting to allocate * a bucket for a bucket zone so we do not allow infinite bucket * recursion. This cookie will even persist to frees of unused * buckets via the allocation path or bucket allocations in the * free path. */ if ((zone->uz_flags & UMA_ZFLAG_BUCKET) == 0) udata = (void *)(uintptr_t)zone->uz_flags; else { if ((uintptr_t)udata & UMA_ZFLAG_BUCKET) return (NULL); udata = (void *)((uintptr_t)udata | UMA_ZFLAG_BUCKET); } if (((uintptr_t)udata & UMA_ZONE_VM) != 0) flags |= M_NOVM; ubz = bucket_zone_lookup(zone->uz_bucket_size); if (ubz->ubz_zone == zone && (ubz + 1)->ubz_entries != 0) ubz++; bucket = uma_zalloc_arg(ubz->ubz_zone, udata, flags); if (bucket) { #ifdef INVARIANTS bzero(bucket->ub_bucket, sizeof(void *) * ubz->ubz_entries); #endif bucket->ub_cnt = 0; bucket->ub_entries = ubz->ubz_entries; bucket->ub_seq = SMR_SEQ_INVALID; CTR3(KTR_UMA, "bucket_alloc: zone %s(%p) allocated bucket %p", zone->uz_name, zone, bucket); } return (bucket); } static void bucket_free(uma_zone_t zone, uma_bucket_t bucket, void *udata) { struct uma_bucket_zone *ubz; if (bucket->ub_cnt != 0) bucket_drain(zone, bucket); KASSERT(bucket->ub_cnt == 0, ("bucket_free: Freeing a non free bucket.")); KASSERT(bucket->ub_seq == SMR_SEQ_INVALID, ("bucket_free: Freeing an SMR bucket.")); if ((zone->uz_flags & UMA_ZFLAG_BUCKET) == 0) udata = (void *)(uintptr_t)zone->uz_flags; ubz = bucket_zone_lookup(bucket->ub_entries); uma_zfree_arg(ubz->ubz_zone, bucket, udata); } static void bucket_zone_drain(void) { struct uma_bucket_zone *ubz; for (ubz = &bucket_zones[0]; ubz->ubz_entries != 0; ubz++) uma_zone_reclaim(ubz->ubz_zone, UMA_RECLAIM_DRAIN); } /* * Acquire the domain lock and record contention. */ static uma_zone_domain_t zone_domain_lock(uma_zone_t zone, int domain) { uma_zone_domain_t zdom; bool lockfail; zdom = ZDOM_GET(zone, domain); lockfail = false; if (ZDOM_OWNED(zdom)) lockfail = true; ZDOM_LOCK(zdom); /* This is unsynchronized. The counter does not need to be precise. */ if (lockfail && zone->uz_bucket_size < zone->uz_bucket_size_max) zone->uz_bucket_size++; return (zdom); } /* * Search for the domain with the least cached items and return it if it * is out of balance with the preferred domain. */ static __noinline int zone_domain_lowest(uma_zone_t zone, int pref) { long least, nitems, prefitems; int domain; int i; prefitems = least = LONG_MAX; domain = 0; for (i = 0; i < vm_ndomains; i++) { nitems = ZDOM_GET(zone, i)->uzd_nitems; if (nitems < least) { domain = i; least = nitems; } if (domain == pref) prefitems = nitems; } if (prefitems < least * 2) return (pref); return (domain); } /* * Search for the domain with the most cached items and return it or the * preferred domain if it has enough to proceed. */ static __noinline int zone_domain_highest(uma_zone_t zone, int pref) { long most, nitems; int domain; int i; if (ZDOM_GET(zone, pref)->uzd_nitems > BUCKET_MAX) return (pref); most = 0; domain = 0; for (i = 0; i < vm_ndomains; i++) { nitems = ZDOM_GET(zone, i)->uzd_nitems; if (nitems > most) { domain = i; most = nitems; } } return (domain); } /* * Safely subtract cnt from imax. */ static void zone_domain_imax_sub(uma_zone_domain_t zdom, int cnt) { long new; long old; old = zdom->uzd_imax; do { if (old <= cnt) new = 0; else new = old - cnt; } while (atomic_fcmpset_long(&zdom->uzd_imax, &old, new) == 0); } /* * Set the maximum imax value. */ static void zone_domain_imax_set(uma_zone_domain_t zdom, int nitems) { long old; old = zdom->uzd_imax; do { if (old >= nitems) break; } while (atomic_fcmpset_long(&zdom->uzd_imax, &old, nitems) == 0); } /* * Attempt to satisfy an allocation by retrieving a full bucket from one of the * zone's caches. If a bucket is found the zone is not locked on return. */ static uma_bucket_t zone_fetch_bucket(uma_zone_t zone, uma_zone_domain_t zdom, bool reclaim) { uma_bucket_t bucket; int i; bool dtor = false; ZDOM_LOCK_ASSERT(zdom); if ((bucket = STAILQ_FIRST(&zdom->uzd_buckets)) == NULL) return (NULL); /* SMR Buckets can not be re-used until readers expire. */ if ((zone->uz_flags & UMA_ZONE_SMR) != 0 && bucket->ub_seq != SMR_SEQ_INVALID) { if (!smr_poll(zone->uz_smr, bucket->ub_seq, false)) return (NULL); bucket->ub_seq = SMR_SEQ_INVALID; dtor = (zone->uz_dtor != NULL) || UMA_ALWAYS_CTORDTOR; if (STAILQ_NEXT(bucket, ub_link) != NULL) zdom->uzd_seq = STAILQ_NEXT(bucket, ub_link)->ub_seq; } MPASS(zdom->uzd_nitems >= bucket->ub_cnt); STAILQ_REMOVE_HEAD(&zdom->uzd_buckets, ub_link); zdom->uzd_nitems -= bucket->ub_cnt; /* * Shift the bounds of the current WSS interval to avoid * perturbing the estimate. */ if (reclaim) { zdom->uzd_imin -= lmin(zdom->uzd_imin, bucket->ub_cnt); zone_domain_imax_sub(zdom, bucket->ub_cnt); } else if (zdom->uzd_imin > zdom->uzd_nitems) zdom->uzd_imin = zdom->uzd_nitems; ZDOM_UNLOCK(zdom); if (dtor) for (i = 0; i < bucket->ub_cnt; i++) item_dtor(zone, bucket->ub_bucket[i], zone->uz_size, NULL, SKIP_NONE); return (bucket); } /* * Insert a full bucket into the specified cache. The "ws" parameter indicates * whether the bucket's contents should be counted as part of the zone's working * set. The bucket may be freed if it exceeds the bucket limit. */ static void zone_put_bucket(uma_zone_t zone, int domain, uma_bucket_t bucket, void *udata, const bool ws) { uma_zone_domain_t zdom; /* We don't cache empty buckets. This can happen after a reclaim. */ if (bucket->ub_cnt == 0) goto out; zdom = zone_domain_lock(zone, domain); - KASSERT(!ws || zdom->uzd_nitems < zone->uz_bucket_max, - ("%s: zone %p overflow", __func__, zone)); - /* * Conditionally set the maximum number of items. */ zdom->uzd_nitems += bucket->ub_cnt; if (__predict_true(zdom->uzd_nitems < zone->uz_bucket_max)) { if (ws) zone_domain_imax_set(zdom, zdom->uzd_nitems); if (STAILQ_EMPTY(&zdom->uzd_buckets)) zdom->uzd_seq = bucket->ub_seq; STAILQ_INSERT_TAIL(&zdom->uzd_buckets, bucket, ub_link); ZDOM_UNLOCK(zdom); return; } zdom->uzd_nitems -= bucket->ub_cnt; ZDOM_UNLOCK(zdom); out: bucket_free(zone, bucket, udata); } /* Pops an item out of a per-cpu cache bucket. */ static inline void * cache_bucket_pop(uma_cache_t cache, uma_cache_bucket_t bucket) { void *item; CRITICAL_ASSERT(curthread); bucket->ucb_cnt--; item = bucket->ucb_bucket->ub_bucket[bucket->ucb_cnt]; #ifdef INVARIANTS bucket->ucb_bucket->ub_bucket[bucket->ucb_cnt] = NULL; KASSERT(item != NULL, ("uma_zalloc: Bucket pointer mangled.")); #endif cache->uc_allocs++; return (item); } /* Pushes an item into a per-cpu cache bucket. */ static inline void cache_bucket_push(uma_cache_t cache, uma_cache_bucket_t bucket, void *item) { CRITICAL_ASSERT(curthread); KASSERT(bucket->ucb_bucket->ub_bucket[bucket->ucb_cnt] == NULL, ("uma_zfree: Freeing to non free bucket index.")); bucket->ucb_bucket->ub_bucket[bucket->ucb_cnt] = item; bucket->ucb_cnt++; cache->uc_frees++; } /* * Unload a UMA bucket from a per-cpu cache. */ static inline uma_bucket_t cache_bucket_unload(uma_cache_bucket_t bucket) { uma_bucket_t b; b = bucket->ucb_bucket; if (b != NULL) { MPASS(b->ub_entries == bucket->ucb_entries); b->ub_cnt = bucket->ucb_cnt; bucket->ucb_bucket = NULL; bucket->ucb_entries = bucket->ucb_cnt = 0; } return (b); } static inline uma_bucket_t cache_bucket_unload_alloc(uma_cache_t cache) { return (cache_bucket_unload(&cache->uc_allocbucket)); } static inline uma_bucket_t cache_bucket_unload_free(uma_cache_t cache) { return (cache_bucket_unload(&cache->uc_freebucket)); } static inline uma_bucket_t cache_bucket_unload_cross(uma_cache_t cache) { return (cache_bucket_unload(&cache->uc_crossbucket)); } /* * Load a bucket into a per-cpu cache bucket. */ static inline void cache_bucket_load(uma_cache_bucket_t bucket, uma_bucket_t b) { CRITICAL_ASSERT(curthread); MPASS(bucket->ucb_bucket == NULL); MPASS(b->ub_seq == SMR_SEQ_INVALID); bucket->ucb_bucket = b; bucket->ucb_cnt = b->ub_cnt; bucket->ucb_entries = b->ub_entries; } static inline void cache_bucket_load_alloc(uma_cache_t cache, uma_bucket_t b) { cache_bucket_load(&cache->uc_allocbucket, b); } static inline void cache_bucket_load_free(uma_cache_t cache, uma_bucket_t b) { cache_bucket_load(&cache->uc_freebucket, b); } #ifdef NUMA static inline void cache_bucket_load_cross(uma_cache_t cache, uma_bucket_t b) { cache_bucket_load(&cache->uc_crossbucket, b); } #endif /* * Copy and preserve ucb_spare. */ static inline void cache_bucket_copy(uma_cache_bucket_t b1, uma_cache_bucket_t b2) { b1->ucb_bucket = b2->ucb_bucket; b1->ucb_entries = b2->ucb_entries; b1->ucb_cnt = b2->ucb_cnt; } /* * Swap two cache buckets. */ static inline void cache_bucket_swap(uma_cache_bucket_t b1, uma_cache_bucket_t b2) { struct uma_cache_bucket b3; CRITICAL_ASSERT(curthread); cache_bucket_copy(&b3, b1); cache_bucket_copy(b1, b2); cache_bucket_copy(b2, &b3); } /* * Attempt to fetch a bucket from a zone on behalf of the current cpu cache. */ static uma_bucket_t cache_fetch_bucket(uma_zone_t zone, uma_cache_t cache, int domain) { uma_zone_domain_t zdom; uma_bucket_t bucket; /* * Avoid the lock if possible. */ zdom = ZDOM_GET(zone, domain); if (zdom->uzd_nitems == 0) return (NULL); if ((cache_uz_flags(cache) & UMA_ZONE_SMR) != 0 && !smr_poll(zone->uz_smr, zdom->uzd_seq, false)) return (NULL); /* * Check the zone's cache of buckets. */ zdom = zone_domain_lock(zone, domain); if ((bucket = zone_fetch_bucket(zone, zdom, false)) != NULL) { KASSERT(bucket->ub_cnt != 0, ("cache_fetch_bucket: Returning an empty bucket.")); return (bucket); } ZDOM_UNLOCK(zdom); return (NULL); } static void zone_log_warning(uma_zone_t zone) { static const struct timeval warninterval = { 300, 0 }; if (!zone_warnings || zone->uz_warning == NULL) return; if (ratecheck(&zone->uz_ratecheck, &warninterval)) printf("[zone: %s] %s\n", zone->uz_name, zone->uz_warning); } static inline void zone_maxaction(uma_zone_t zone) { if (zone->uz_maxaction.ta_func != NULL) taskqueue_enqueue(taskqueue_thread, &zone->uz_maxaction); } /* * Routine called by timeout which is used to fire off some time interval * based calculations. (stats, hash size, etc.) * * Arguments: * arg Unused * * Returns: * Nothing */ static void uma_timeout(void *unused) { bucket_enable(); zone_foreach(zone_timeout, NULL); /* Reschedule this event */ callout_reset(&uma_callout, UMA_TIMEOUT * hz, uma_timeout, NULL); } /* * Update the working set size estimate for the zone's bucket cache. * The constants chosen here are somewhat arbitrary. With an update period of * 20s (UMA_TIMEOUT), this estimate is dominated by zone activity over the * last 100s. */ static void zone_domain_update_wss(uma_zone_domain_t zdom) { long wss; ZDOM_LOCK(zdom); MPASS(zdom->uzd_imax >= zdom->uzd_imin); wss = zdom->uzd_imax - zdom->uzd_imin; zdom->uzd_imax = zdom->uzd_imin = zdom->uzd_nitems; zdom->uzd_wss = (4 * wss + zdom->uzd_wss) / 5; ZDOM_UNLOCK(zdom); } /* * Routine to perform timeout driven calculations. This expands the * hashes and does per cpu statistics aggregation. * * Returns nothing. */ static void zone_timeout(uma_zone_t zone, void *unused) { uma_keg_t keg; u_int slabs, pages; if ((zone->uz_flags & UMA_ZFLAG_HASH) == 0) goto update_wss; keg = zone->uz_keg; /* * Hash zones are non-numa by definition so the first domain * is the only one present. */ KEG_LOCK(keg, 0); pages = keg->uk_domain[0].ud_pages; /* * Expand the keg hash table. * * This is done if the number of slabs is larger than the hash size. * What I'm trying to do here is completely reduce collisions. This * may be a little aggressive. Should I allow for two collisions max? */ if ((slabs = pages / keg->uk_ppera) > keg->uk_hash.uh_hashsize) { struct uma_hash newhash; struct uma_hash oldhash; int ret; /* * This is so involved because allocating and freeing * while the keg lock is held will lead to deadlock. * I have to do everything in stages and check for * races. */ KEG_UNLOCK(keg, 0); ret = hash_alloc(&newhash, 1 << fls(slabs)); KEG_LOCK(keg, 0); if (ret) { if (hash_expand(&keg->uk_hash, &newhash)) { oldhash = keg->uk_hash; keg->uk_hash = newhash; } else oldhash = newhash; KEG_UNLOCK(keg, 0); hash_free(&oldhash); goto update_wss; } } KEG_UNLOCK(keg, 0); update_wss: for (int i = 0; i < vm_ndomains; i++) zone_domain_update_wss(ZDOM_GET(zone, i)); } /* * Allocate and zero fill the next sized hash table from the appropriate * backing store. * * Arguments: * hash A new hash structure with the old hash size in uh_hashsize * * Returns: * 1 on success and 0 on failure. */ static int hash_alloc(struct uma_hash *hash, u_int size) { size_t alloc; KASSERT(powerof2(size), ("hash size must be power of 2")); if (size > UMA_HASH_SIZE_INIT) { hash->uh_hashsize = size; alloc = sizeof(hash->uh_slab_hash[0]) * hash->uh_hashsize; hash->uh_slab_hash = malloc(alloc, M_UMAHASH, M_NOWAIT); } else { alloc = sizeof(hash->uh_slab_hash[0]) * UMA_HASH_SIZE_INIT; hash->uh_slab_hash = zone_alloc_item(hashzone, NULL, UMA_ANYDOMAIN, M_WAITOK); hash->uh_hashsize = UMA_HASH_SIZE_INIT; } if (hash->uh_slab_hash) { bzero(hash->uh_slab_hash, alloc); hash->uh_hashmask = hash->uh_hashsize - 1; return (1); } return (0); } /* * Expands the hash table for HASH zones. This is done from zone_timeout * to reduce collisions. This must not be done in the regular allocation * path, otherwise, we can recurse on the vm while allocating pages. * * Arguments: * oldhash The hash you want to expand * newhash The hash structure for the new table * * Returns: * Nothing * * Discussion: */ static int hash_expand(struct uma_hash *oldhash, struct uma_hash *newhash) { uma_hash_slab_t slab; u_int hval; u_int idx; if (!newhash->uh_slab_hash) return (0); if (oldhash->uh_hashsize >= newhash->uh_hashsize) return (0); /* * I need to investigate hash algorithms for resizing without a * full rehash. */ for (idx = 0; idx < oldhash->uh_hashsize; idx++) while (!LIST_EMPTY(&oldhash->uh_slab_hash[idx])) { slab = LIST_FIRST(&oldhash->uh_slab_hash[idx]); LIST_REMOVE(slab, uhs_hlink); hval = UMA_HASH(newhash, slab->uhs_data); LIST_INSERT_HEAD(&newhash->uh_slab_hash[hval], slab, uhs_hlink); } return (1); } /* * Free the hash bucket to the appropriate backing store. * * Arguments: * slab_hash The hash bucket we're freeing * hashsize The number of entries in that hash bucket * * Returns: * Nothing */ static void hash_free(struct uma_hash *hash) { if (hash->uh_slab_hash == NULL) return; if (hash->uh_hashsize == UMA_HASH_SIZE_INIT) zone_free_item(hashzone, hash->uh_slab_hash, NULL, SKIP_NONE); else free(hash->uh_slab_hash, M_UMAHASH); } /* * Frees all outstanding items in a bucket * * Arguments: * zone The zone to free to, must be unlocked. * bucket The free/alloc bucket with items. * * Returns: * Nothing */ static void bucket_drain(uma_zone_t zone, uma_bucket_t bucket) { int i; if (bucket->ub_cnt == 0) return; if ((zone->uz_flags & UMA_ZONE_SMR) != 0 && bucket->ub_seq != SMR_SEQ_INVALID) { smr_wait(zone->uz_smr, bucket->ub_seq); bucket->ub_seq = SMR_SEQ_INVALID; for (i = 0; i < bucket->ub_cnt; i++) item_dtor(zone, bucket->ub_bucket[i], zone->uz_size, NULL, SKIP_NONE); } if (zone->uz_fini) for (i = 0; i < bucket->ub_cnt; i++) zone->uz_fini(bucket->ub_bucket[i], zone->uz_size); zone->uz_release(zone->uz_arg, bucket->ub_bucket, bucket->ub_cnt); if (zone->uz_max_items > 0) zone_free_limit(zone, bucket->ub_cnt); #ifdef INVARIANTS bzero(bucket->ub_bucket, sizeof(void *) * bucket->ub_cnt); #endif bucket->ub_cnt = 0; } /* * Drains the per cpu caches for a zone. * * NOTE: This may only be called while the zone is being torn down, and not * during normal operation. This is necessary in order that we do not have * to migrate CPUs to drain the per-CPU caches. * * Arguments: * zone The zone to drain, must be unlocked. * * Returns: * Nothing */ static void cache_drain(uma_zone_t zone) { uma_cache_t cache; uma_bucket_t bucket; smr_seq_t seq; int cpu; /* * XXX: It is safe to not lock the per-CPU caches, because we're * tearing down the zone anyway. I.e., there will be no further use * of the caches at this point. * * XXX: It would good to be able to assert that the zone is being * torn down to prevent improper use of cache_drain(). */ seq = SMR_SEQ_INVALID; if ((zone->uz_flags & UMA_ZONE_SMR) != 0) seq = smr_advance(zone->uz_smr); CPU_FOREACH(cpu) { cache = &zone->uz_cpu[cpu]; bucket = cache_bucket_unload_alloc(cache); if (bucket != NULL) bucket_free(zone, bucket, NULL); bucket = cache_bucket_unload_free(cache); if (bucket != NULL) { bucket->ub_seq = seq; bucket_free(zone, bucket, NULL); } bucket = cache_bucket_unload_cross(cache); if (bucket != NULL) { bucket->ub_seq = seq; bucket_free(zone, bucket, NULL); } } bucket_cache_reclaim(zone, true); } static void cache_shrink(uma_zone_t zone, void *unused) { if (zone->uz_flags & UMA_ZFLAG_INTERNAL) return; zone->uz_bucket_size = (zone->uz_bucket_size_min + zone->uz_bucket_size) / 2; } static void cache_drain_safe_cpu(uma_zone_t zone, void *unused) { uma_cache_t cache; uma_bucket_t b1, b2, b3; int domain; if (zone->uz_flags & UMA_ZFLAG_INTERNAL) return; b1 = b2 = b3 = NULL; critical_enter(); cache = &zone->uz_cpu[curcpu]; domain = PCPU_GET(domain); b1 = cache_bucket_unload_alloc(cache); /* * Don't flush SMR zone buckets. This leaves the zone without a * bucket and forces every free to synchronize(). */ if ((zone->uz_flags & UMA_ZONE_SMR) == 0) { b2 = cache_bucket_unload_free(cache); b3 = cache_bucket_unload_cross(cache); } critical_exit(); if (b1 != NULL) zone_free_bucket(zone, b1, NULL, domain, false); if (b2 != NULL) zone_free_bucket(zone, b2, NULL, domain, false); if (b3 != NULL) { /* Adjust the domain so it goes to zone_free_cross. */ domain = (domain + 1) % vm_ndomains; zone_free_bucket(zone, b3, NULL, domain, false); } } /* * Safely drain per-CPU caches of a zone(s) to alloc bucket. * This is an expensive call because it needs to bind to all CPUs * one by one and enter a critical section on each of them in order * to safely access their cache buckets. * Zone lock must not be held on call this function. */ static void pcpu_cache_drain_safe(uma_zone_t zone) { int cpu; /* * Polite bucket sizes shrinking was not enough, shrink aggressively. */ if (zone) cache_shrink(zone, NULL); else zone_foreach(cache_shrink, NULL); CPU_FOREACH(cpu) { thread_lock(curthread); sched_bind(curthread, cpu); thread_unlock(curthread); if (zone) cache_drain_safe_cpu(zone, NULL); else zone_foreach(cache_drain_safe_cpu, NULL); } thread_lock(curthread); sched_unbind(curthread); thread_unlock(curthread); } /* * Reclaim cached buckets from a zone. All buckets are reclaimed if the caller * requested a drain, otherwise the per-domain caches are trimmed to either * estimated working set size. */ static void bucket_cache_reclaim(uma_zone_t zone, bool drain) { uma_zone_domain_t zdom; uma_bucket_t bucket; long target; int i; /* * Shrink the zone bucket size to ensure that the per-CPU caches * don't grow too large. */ if (zone->uz_bucket_size > zone->uz_bucket_size_min) zone->uz_bucket_size--; for (i = 0; i < vm_ndomains; i++) { /* * The cross bucket is partially filled and not part of * the item count. Reclaim it individually here. */ zdom = ZDOM_GET(zone, i); if ((zone->uz_flags & UMA_ZONE_SMR) == 0 || drain) { ZONE_CROSS_LOCK(zone); bucket = zdom->uzd_cross; zdom->uzd_cross = NULL; ZONE_CROSS_UNLOCK(zone); if (bucket != NULL) bucket_free(zone, bucket, NULL); } /* * If we were asked to drain the zone, we are done only once * this bucket cache is empty. Otherwise, we reclaim items in * excess of the zone's estimated working set size. If the * difference nitems - imin is larger than the WSS estimate, * then the estimate will grow at the end of this interval and * we ignore the historical average. */ ZDOM_LOCK(zdom); target = drain ? 0 : lmax(zdom->uzd_wss, zdom->uzd_nitems - zdom->uzd_imin); while (zdom->uzd_nitems > target) { bucket = zone_fetch_bucket(zone, zdom, true); if (bucket == NULL) break; bucket_free(zone, bucket, NULL); ZDOM_LOCK(zdom); } ZDOM_UNLOCK(zdom); } } static void keg_free_slab(uma_keg_t keg, uma_slab_t slab, int start) { uint8_t *mem; int i; uint8_t flags; CTR4(KTR_UMA, "keg_free_slab keg %s(%p) slab %p, returning %d bytes", keg->uk_name, keg, slab, PAGE_SIZE * keg->uk_ppera); mem = slab_data(slab, keg); flags = slab->us_flags; i = start; if (keg->uk_fini != NULL) { for (i--; i > -1; i--) #ifdef INVARIANTS /* * trash_fini implies that dtor was trash_dtor. trash_fini * would check that memory hasn't been modified since free, * which executed trash_dtor. * That's why we need to run uma_dbg_kskip() check here, * albeit we don't make skip check for other init/fini * invocations. */ if (!uma_dbg_kskip(keg, slab_item(slab, keg, i)) || keg->uk_fini != trash_fini) #endif keg->uk_fini(slab_item(slab, keg, i), keg->uk_size); } if (keg->uk_flags & UMA_ZFLAG_OFFPAGE) zone_free_item(slabzone(keg->uk_ipers), slab_tohashslab(slab), NULL, SKIP_NONE); keg->uk_freef(mem, PAGE_SIZE * keg->uk_ppera, flags); uma_total_dec(PAGE_SIZE * keg->uk_ppera); } /* * Frees pages from a keg back to the system. This is done on demand from * the pageout daemon. * * Returns nothing. */ static void keg_drain(uma_keg_t keg) { struct slabhead freeslabs; uma_domain_t dom; uma_slab_t slab, tmp; int i, n; if (keg->uk_flags & UMA_ZONE_NOFREE || keg->uk_freef == NULL) return; for (i = 0; i < vm_ndomains; i++) { CTR4(KTR_UMA, "keg_drain %s(%p) domain %d free items: %u", keg->uk_name, keg, i, dom->ud_free_items); dom = &keg->uk_domain[i]; LIST_INIT(&freeslabs); KEG_LOCK(keg, i); if ((keg->uk_flags & UMA_ZFLAG_HASH) != 0) { LIST_FOREACH(slab, &dom->ud_free_slab, us_link) UMA_HASH_REMOVE(&keg->uk_hash, slab); } n = dom->ud_free_slabs; LIST_SWAP(&freeslabs, &dom->ud_free_slab, uma_slab, us_link); dom->ud_free_slabs = 0; dom->ud_free_items -= n * keg->uk_ipers; dom->ud_pages -= n * keg->uk_ppera; KEG_UNLOCK(keg, i); LIST_FOREACH_SAFE(slab, &freeslabs, us_link, tmp) keg_free_slab(keg, slab, keg->uk_ipers); } } static void zone_reclaim(uma_zone_t zone, int waitok, bool drain) { /* * Set draining to interlock with zone_dtor() so we can release our * locks as we go. Only dtor() should do a WAITOK call since it * is the only call that knows the structure will still be available * when it wakes up. */ ZONE_LOCK(zone); while (zone->uz_flags & UMA_ZFLAG_RECLAIMING) { if (waitok == M_NOWAIT) goto out; msleep(zone, &ZDOM_GET(zone, 0)->uzd_lock, PVM, "zonedrain", 1); } zone->uz_flags |= UMA_ZFLAG_RECLAIMING; ZONE_UNLOCK(zone); bucket_cache_reclaim(zone, drain); /* * The DRAINING flag protects us from being freed while * we're running. Normally the uma_rwlock would protect us but we * must be able to release and acquire the right lock for each keg. */ if ((zone->uz_flags & UMA_ZFLAG_CACHE) == 0) keg_drain(zone->uz_keg); ZONE_LOCK(zone); zone->uz_flags &= ~UMA_ZFLAG_RECLAIMING; wakeup(zone); out: ZONE_UNLOCK(zone); } static void zone_drain(uma_zone_t zone, void *unused) { zone_reclaim(zone, M_NOWAIT, true); } static void zone_trim(uma_zone_t zone, void *unused) { zone_reclaim(zone, M_NOWAIT, false); } /* * Allocate a new slab for a keg and inserts it into the partial slab list. * The keg should be unlocked on entry. If the allocation succeeds it will * be locked on return. * * Arguments: * flags Wait flags for the item initialization routine * aflags Wait flags for the slab allocation * * Returns: * The slab that was allocated or NULL if there is no memory and the * caller specified M_NOWAIT. */ static uma_slab_t keg_alloc_slab(uma_keg_t keg, uma_zone_t zone, int domain, int flags, int aflags) { uma_domain_t dom; uma_alloc allocf; uma_slab_t slab; unsigned long size; uint8_t *mem; uint8_t sflags; int i; KASSERT(domain >= 0 && domain < vm_ndomains, ("keg_alloc_slab: domain %d out of range", domain)); allocf = keg->uk_allocf; slab = NULL; mem = NULL; if (keg->uk_flags & UMA_ZFLAG_OFFPAGE) { uma_hash_slab_t hslab; hslab = zone_alloc_item(slabzone(keg->uk_ipers), NULL, domain, aflags); if (hslab == NULL) goto fail; slab = &hslab->uhs_slab; } /* * This reproduces the old vm_zone behavior of zero filling pages the * first time they are added to a zone. * * Malloced items are zeroed in uma_zalloc. */ if ((keg->uk_flags & UMA_ZONE_MALLOC) == 0) aflags |= M_ZERO; else aflags &= ~M_ZERO; if (keg->uk_flags & UMA_ZONE_NODUMP) aflags |= M_NODUMP; /* zone is passed for legacy reasons. */ size = keg->uk_ppera * PAGE_SIZE; mem = allocf(zone, size, domain, &sflags, aflags); if (mem == NULL) { if (keg->uk_flags & UMA_ZFLAG_OFFPAGE) zone_free_item(slabzone(keg->uk_ipers), slab_tohashslab(slab), NULL, SKIP_NONE); goto fail; } uma_total_inc(size); /* For HASH zones all pages go to the same uma_domain. */ if ((keg->uk_flags & UMA_ZFLAG_HASH) != 0) domain = 0; /* Point the slab into the allocated memory */ if (!(keg->uk_flags & UMA_ZFLAG_OFFPAGE)) slab = (uma_slab_t )(mem + keg->uk_pgoff); else slab_tohashslab(slab)->uhs_data = mem; if (keg->uk_flags & UMA_ZFLAG_VTOSLAB) for (i = 0; i < keg->uk_ppera; i++) vsetzoneslab((vm_offset_t)mem + (i * PAGE_SIZE), zone, slab); slab->us_freecount = keg->uk_ipers; slab->us_flags = sflags; slab->us_domain = domain; BIT_FILL(keg->uk_ipers, &slab->us_free); #ifdef INVARIANTS BIT_ZERO(keg->uk_ipers, slab_dbg_bits(slab, keg)); #endif if (keg->uk_init != NULL) { for (i = 0; i < keg->uk_ipers; i++) if (keg->uk_init(slab_item(slab, keg, i), keg->uk_size, flags) != 0) break; if (i != keg->uk_ipers) { keg_free_slab(keg, slab, i); goto fail; } } KEG_LOCK(keg, domain); CTR3(KTR_UMA, "keg_alloc_slab: allocated slab %p for %s(%p)", slab, keg->uk_name, keg); if (keg->uk_flags & UMA_ZFLAG_HASH) UMA_HASH_INSERT(&keg->uk_hash, slab, mem); /* * If we got a slab here it's safe to mark it partially used * and return. We assume that the caller is going to remove * at least one item. */ dom = &keg->uk_domain[domain]; LIST_INSERT_HEAD(&dom->ud_part_slab, slab, us_link); dom->ud_pages += keg->uk_ppera; dom->ud_free_items += keg->uk_ipers; return (slab); fail: return (NULL); } /* * This function is intended to be used early on in place of page_alloc() so * that we may use the boot time page cache to satisfy allocations before * the VM is ready. */ static void * startup_alloc(uma_zone_t zone, vm_size_t bytes, int domain, uint8_t *pflag, int wait) { vm_paddr_t pa; vm_page_t m; void *mem; int pages; int i; pages = howmany(bytes, PAGE_SIZE); KASSERT(pages > 0, ("%s can't reserve 0 pages", __func__)); *pflag = UMA_SLAB_BOOT; m = vm_page_alloc_contig_domain(NULL, 0, domain, malloc2vm_flags(wait) | VM_ALLOC_NOOBJ | VM_ALLOC_WIRED, pages, (vm_paddr_t)0, ~(vm_paddr_t)0, 1, 0, VM_MEMATTR_DEFAULT); if (m == NULL) return (NULL); pa = VM_PAGE_TO_PHYS(m); for (i = 0; i < pages; i++, pa += PAGE_SIZE) { #if defined(__aarch64__) || defined(__amd64__) || defined(__mips__) || \ defined(__riscv) || defined(__powerpc64__) if ((wait & M_NODUMP) == 0) dump_add_page(pa); #endif } /* Allocate KVA and indirectly advance bootmem. */ mem = (void *)pmap_map(&bootmem, m->phys_addr, m->phys_addr + (pages * PAGE_SIZE), VM_PROT_READ | VM_PROT_WRITE); if ((wait & M_ZERO) != 0) bzero(mem, pages * PAGE_SIZE); return (mem); } static void startup_free(void *mem, vm_size_t bytes) { vm_offset_t va; vm_page_t m; va = (vm_offset_t)mem; m = PHYS_TO_VM_PAGE(pmap_kextract(va)); pmap_remove(kernel_pmap, va, va + bytes); for (; bytes != 0; bytes -= PAGE_SIZE, m++) { #if defined(__aarch64__) || defined(__amd64__) || defined(__mips__) || \ defined(__riscv) || defined(__powerpc64__) dump_drop_page(VM_PAGE_TO_PHYS(m)); #endif vm_page_unwire_noq(m); vm_page_free(m); } } /* * Allocates a number of pages from the system * * Arguments: * bytes The number of bytes requested * wait Shall we wait? * * Returns: * A pointer to the alloced memory or possibly * NULL if M_NOWAIT is set. */ static void * page_alloc(uma_zone_t zone, vm_size_t bytes, int domain, uint8_t *pflag, int wait) { void *p; /* Returned page */ *pflag = UMA_SLAB_KERNEL; p = (void *)kmem_malloc_domainset(DOMAINSET_FIXED(domain), bytes, wait); return (p); } static void * pcpu_page_alloc(uma_zone_t zone, vm_size_t bytes, int domain, uint8_t *pflag, int wait) { struct pglist alloctail; vm_offset_t addr, zkva; int cpu, flags; vm_page_t p, p_next; #ifdef NUMA struct pcpu *pc; #endif MPASS(bytes == (mp_maxid + 1) * PAGE_SIZE); TAILQ_INIT(&alloctail); flags = VM_ALLOC_SYSTEM | VM_ALLOC_WIRED | VM_ALLOC_NOOBJ | malloc2vm_flags(wait); *pflag = UMA_SLAB_KERNEL; for (cpu = 0; cpu <= mp_maxid; cpu++) { if (CPU_ABSENT(cpu)) { p = vm_page_alloc(NULL, 0, flags); } else { #ifndef NUMA p = vm_page_alloc(NULL, 0, flags); #else pc = pcpu_find(cpu); if (__predict_false(VM_DOMAIN_EMPTY(pc->pc_domain))) p = NULL; else p = vm_page_alloc_domain(NULL, 0, pc->pc_domain, flags); if (__predict_false(p == NULL)) p = vm_page_alloc(NULL, 0, flags); #endif } if (__predict_false(p == NULL)) goto fail; TAILQ_INSERT_TAIL(&alloctail, p, listq); } if ((addr = kva_alloc(bytes)) == 0) goto fail; zkva = addr; TAILQ_FOREACH(p, &alloctail, listq) { pmap_qenter(zkva, &p, 1); zkva += PAGE_SIZE; } return ((void*)addr); fail: TAILQ_FOREACH_SAFE(p, &alloctail, listq, p_next) { vm_page_unwire_noq(p); vm_page_free(p); } return (NULL); } /* * Allocates a number of pages from within an object * * Arguments: * bytes The number of bytes requested * wait Shall we wait? * * Returns: * A pointer to the alloced memory or possibly * NULL if M_NOWAIT is set. */ static void * noobj_alloc(uma_zone_t zone, vm_size_t bytes, int domain, uint8_t *flags, int wait) { TAILQ_HEAD(, vm_page) alloctail; u_long npages; vm_offset_t retkva, zkva; vm_page_t p, p_next; uma_keg_t keg; TAILQ_INIT(&alloctail); keg = zone->uz_keg; npages = howmany(bytes, PAGE_SIZE); while (npages > 0) { p = vm_page_alloc_domain(NULL, 0, domain, VM_ALLOC_INTERRUPT | VM_ALLOC_WIRED | VM_ALLOC_NOOBJ | ((wait & M_WAITOK) != 0 ? VM_ALLOC_WAITOK : VM_ALLOC_NOWAIT)); if (p != NULL) { /* * Since the page does not belong to an object, its * listq is unused. */ TAILQ_INSERT_TAIL(&alloctail, p, listq); npages--; continue; } /* * Page allocation failed, free intermediate pages and * exit. */ TAILQ_FOREACH_SAFE(p, &alloctail, listq, p_next) { vm_page_unwire_noq(p); vm_page_free(p); } return (NULL); } *flags = UMA_SLAB_PRIV; zkva = keg->uk_kva + atomic_fetchadd_long(&keg->uk_offset, round_page(bytes)); retkva = zkva; TAILQ_FOREACH(p, &alloctail, listq) { pmap_qenter(zkva, &p, 1); zkva += PAGE_SIZE; } return ((void *)retkva); } /* * Allocate physically contiguous pages. */ static void * contig_alloc(uma_zone_t zone, vm_size_t bytes, int domain, uint8_t *pflag, int wait) { *pflag = UMA_SLAB_KERNEL; return ((void *)kmem_alloc_contig_domainset(DOMAINSET_FIXED(domain), bytes, wait, 0, ~(vm_paddr_t)0, 1, 0, VM_MEMATTR_DEFAULT)); } /* * Frees a number of pages to the system * * Arguments: * mem A pointer to the memory to be freed * size The size of the memory being freed * flags The original p->us_flags field * * Returns: * Nothing */ static void page_free(void *mem, vm_size_t size, uint8_t flags) { if ((flags & UMA_SLAB_BOOT) != 0) { startup_free(mem, size); return; } KASSERT((flags & UMA_SLAB_KERNEL) != 0, ("UMA: page_free used with invalid flags %x", flags)); kmem_free((vm_offset_t)mem, size); } /* * Frees pcpu zone allocations * * Arguments: * mem A pointer to the memory to be freed * size The size of the memory being freed * flags The original p->us_flags field * * Returns: * Nothing */ static void pcpu_page_free(void *mem, vm_size_t size, uint8_t flags) { vm_offset_t sva, curva; vm_paddr_t paddr; vm_page_t m; MPASS(size == (mp_maxid+1)*PAGE_SIZE); if ((flags & UMA_SLAB_BOOT) != 0) { startup_free(mem, size); return; } sva = (vm_offset_t)mem; for (curva = sva; curva < sva + size; curva += PAGE_SIZE) { paddr = pmap_kextract(curva); m = PHYS_TO_VM_PAGE(paddr); vm_page_unwire_noq(m); vm_page_free(m); } pmap_qremove(sva, size >> PAGE_SHIFT); kva_free(sva, size); } /* * Zero fill initializer * * Arguments/Returns follow uma_init specifications */ static int zero_init(void *mem, int size, int flags) { bzero(mem, size); return (0); } #ifdef INVARIANTS struct noslabbits * slab_dbg_bits(uma_slab_t slab, uma_keg_t keg) { return ((void *)((char *)&slab->us_free + BITSET_SIZE(keg->uk_ipers))); } #endif /* * Actual size of embedded struct slab (!OFFPAGE). */ size_t slab_sizeof(int nitems) { size_t s; s = sizeof(struct uma_slab) + BITSET_SIZE(nitems) * SLAB_BITSETS; return (roundup(s, UMA_ALIGN_PTR + 1)); } /* * Size of memory for embedded slabs (!OFFPAGE). */ size_t slab_space(int nitems) { return (UMA_SLAB_SIZE - slab_sizeof(nitems)); } #define UMA_FIXPT_SHIFT 31 #define UMA_FRAC_FIXPT(n, d) \ ((uint32_t)(((uint64_t)(n) << UMA_FIXPT_SHIFT) / (d))) #define UMA_FIXPT_PCT(f) \ ((u_int)(((uint64_t)100 * (f)) >> UMA_FIXPT_SHIFT)) #define UMA_PCT_FIXPT(pct) UMA_FRAC_FIXPT((pct), 100) #define UMA_MIN_EFF UMA_PCT_FIXPT(100 - UMA_MAX_WASTE) /* * Compute the number of items that will fit in a slab. If hdr is true, the * item count may be limited to provide space in the slab for an inline slab * header. Otherwise, all slab space will be provided for item storage. */ static u_int slab_ipers_hdr(u_int size, u_int rsize, u_int slabsize, bool hdr) { u_int ipers; u_int padpi; /* The padding between items is not needed after the last item. */ padpi = rsize - size; if (hdr) { /* * Start with the maximum item count and remove items until * the slab header first alongside the allocatable memory. */ for (ipers = MIN(SLAB_MAX_SETSIZE, (slabsize + padpi - slab_sizeof(1)) / rsize); ipers > 0 && ipers * rsize - padpi + slab_sizeof(ipers) > slabsize; ipers--) continue; } else { ipers = MIN((slabsize + padpi) / rsize, SLAB_MAX_SETSIZE); } return (ipers); } /* * Compute the number of items that will fit in a slab for a startup zone. */ int slab_ipers(size_t size, int align) { int rsize; rsize = roundup(size, align + 1); /* Assume no CACHESPREAD */ return (slab_ipers_hdr(size, rsize, UMA_SLAB_SIZE, true)); } struct keg_layout_result { u_int format; u_int slabsize; u_int ipers; u_int eff; }; static void keg_layout_one(uma_keg_t keg, u_int rsize, u_int slabsize, u_int fmt, struct keg_layout_result *kl) { u_int total; kl->format = fmt; kl->slabsize = slabsize; /* Handle INTERNAL as inline with an extra page. */ if ((fmt & UMA_ZFLAG_INTERNAL) != 0) { kl->format &= ~UMA_ZFLAG_INTERNAL; kl->slabsize += PAGE_SIZE; } kl->ipers = slab_ipers_hdr(keg->uk_size, rsize, kl->slabsize, (fmt & UMA_ZFLAG_OFFPAGE) == 0); /* Account for memory used by an offpage slab header. */ total = kl->slabsize; if ((fmt & UMA_ZFLAG_OFFPAGE) != 0) total += slabzone(kl->ipers)->uz_keg->uk_rsize; kl->eff = UMA_FRAC_FIXPT(kl->ipers * rsize, total); } /* * Determine the format of a uma keg. This determines where the slab header * will be placed (inline or offpage) and calculates ipers, rsize, and ppera. * * Arguments * keg The zone we should initialize * * Returns * Nothing */ static void keg_layout(uma_keg_t keg) { struct keg_layout_result kl = {}, kl_tmp; u_int fmts[2]; u_int alignsize; u_int nfmt; u_int pages; u_int rsize; u_int slabsize; u_int i, j; KASSERT((keg->uk_flags & UMA_ZONE_PCPU) == 0 || (keg->uk_size <= UMA_PCPU_ALLOC_SIZE && (keg->uk_flags & UMA_ZONE_CACHESPREAD) == 0), ("%s: cannot configure for PCPU: keg=%s, size=%u, flags=0x%b", __func__, keg->uk_name, keg->uk_size, keg->uk_flags, PRINT_UMA_ZFLAGS)); KASSERT((keg->uk_flags & (UMA_ZFLAG_INTERNAL | UMA_ZONE_VM)) == 0 || (keg->uk_flags & (UMA_ZONE_NOTOUCH | UMA_ZONE_PCPU)) == 0, ("%s: incompatible flags 0x%b", __func__, keg->uk_flags, PRINT_UMA_ZFLAGS)); alignsize = keg->uk_align + 1; /* * Calculate the size of each allocation (rsize) according to * alignment. If the requested size is smaller than we have * allocation bits for we round it up. */ rsize = MAX(keg->uk_size, UMA_SMALLEST_UNIT); rsize = roundup2(rsize, alignsize); if ((keg->uk_flags & UMA_ZONE_CACHESPREAD) != 0) { /* * We want one item to start on every align boundary in a page. * To do this we will span pages. We will also extend the item * by the size of align if it is an even multiple of align. * Otherwise, it would fall on the same boundary every time. */ if ((rsize & alignsize) == 0) rsize += alignsize; slabsize = rsize * (PAGE_SIZE / alignsize); slabsize = MIN(slabsize, rsize * SLAB_MAX_SETSIZE); slabsize = MIN(slabsize, UMA_CACHESPREAD_MAX_SIZE); slabsize = round_page(slabsize); } else { /* * Start with a slab size of as many pages as it takes to * represent a single item. We will try to fit as many * additional items into the slab as possible. */ slabsize = round_page(keg->uk_size); } /* Build a list of all of the available formats for this keg. */ nfmt = 0; /* Evaluate an inline slab layout. */ if ((keg->uk_flags & (UMA_ZONE_NOTOUCH | UMA_ZONE_PCPU)) == 0) fmts[nfmt++] = 0; /* TODO: vm_page-embedded slab. */ /* * We can't do OFFPAGE if we're internal or if we've been * asked to not go to the VM for buckets. If we do this we * may end up going to the VM for slabs which we do not want * to do if we're UMA_ZONE_VM, which clearly forbids it. * In those cases, evaluate a pseudo-format called INTERNAL * which has an inline slab header and one extra page to * guarantee that it fits. * * Otherwise, see if using an OFFPAGE slab will improve our * efficiency. */ if ((keg->uk_flags & (UMA_ZFLAG_INTERNAL | UMA_ZONE_VM)) != 0) fmts[nfmt++] = UMA_ZFLAG_INTERNAL; else fmts[nfmt++] = UMA_ZFLAG_OFFPAGE; /* * Choose a slab size and format which satisfy the minimum efficiency. * Prefer the smallest slab size that meets the constraints. * * Start with a minimum slab size, to accommodate CACHESPREAD. Then, * for small items (up to PAGE_SIZE), the iteration increment is one * page; and for large items, the increment is one item. */ i = (slabsize + rsize - keg->uk_size) / MAX(PAGE_SIZE, rsize); KASSERT(i >= 1, ("keg %s(%p) flags=0x%b slabsize=%u, rsize=%u, i=%u", keg->uk_name, keg, keg->uk_flags, PRINT_UMA_ZFLAGS, slabsize, rsize, i)); for ( ; ; i++) { slabsize = (rsize <= PAGE_SIZE) ? ptoa(i) : round_page(rsize * (i - 1) + keg->uk_size); for (j = 0; j < nfmt; j++) { /* Only if we have no viable format yet. */ if ((fmts[j] & UMA_ZFLAG_INTERNAL) != 0 && kl.ipers > 0) continue; keg_layout_one(keg, rsize, slabsize, fmts[j], &kl_tmp); if (kl_tmp.eff <= kl.eff) continue; kl = kl_tmp; CTR6(KTR_UMA, "keg %s layout: format %#x " "(ipers %u * rsize %u) / slabsize %#x = %u%% eff", keg->uk_name, kl.format, kl.ipers, rsize, kl.slabsize, UMA_FIXPT_PCT(kl.eff)); /* Stop when we reach the minimum efficiency. */ if (kl.eff >= UMA_MIN_EFF) break; } if (kl.eff >= UMA_MIN_EFF || !multipage_slabs || slabsize >= SLAB_MAX_SETSIZE * rsize || (keg->uk_flags & (UMA_ZONE_PCPU | UMA_ZONE_CONTIG)) != 0) break; } pages = atop(kl.slabsize); if ((keg->uk_flags & UMA_ZONE_PCPU) != 0) pages *= mp_maxid + 1; keg->uk_rsize = rsize; keg->uk_ipers = kl.ipers; keg->uk_ppera = pages; keg->uk_flags |= kl.format; /* * How do we find the slab header if it is offpage or if not all item * start addresses are in the same page? We could solve the latter * case with vaddr alignment, but we don't. */ if ((keg->uk_flags & UMA_ZFLAG_OFFPAGE) != 0 || (keg->uk_ipers - 1) * rsize >= PAGE_SIZE) { if ((keg->uk_flags & UMA_ZONE_NOTPAGE) != 0) keg->uk_flags |= UMA_ZFLAG_HASH; else keg->uk_flags |= UMA_ZFLAG_VTOSLAB; } CTR6(KTR_UMA, "%s: keg=%s, flags=%#x, rsize=%u, ipers=%u, ppera=%u", __func__, keg->uk_name, keg->uk_flags, rsize, keg->uk_ipers, pages); KASSERT(keg->uk_ipers > 0 && keg->uk_ipers <= SLAB_MAX_SETSIZE, ("%s: keg=%s, flags=0x%b, rsize=%u, ipers=%u, ppera=%u", __func__, keg->uk_name, keg->uk_flags, PRINT_UMA_ZFLAGS, rsize, keg->uk_ipers, pages)); } /* * Keg header ctor. This initializes all fields, locks, etc. And inserts * the keg onto the global keg list. * * Arguments/Returns follow uma_ctor specifications * udata Actually uma_kctor_args */ static int keg_ctor(void *mem, int size, void *udata, int flags) { struct uma_kctor_args *arg = udata; uma_keg_t keg = mem; uma_zone_t zone; int i; bzero(keg, size); keg->uk_size = arg->size; keg->uk_init = arg->uminit; keg->uk_fini = arg->fini; keg->uk_align = arg->align; keg->uk_reserve = 0; keg->uk_flags = arg->flags; /* * We use a global round-robin policy by default. Zones with * UMA_ZONE_FIRSTTOUCH set will use first-touch instead, in which * case the iterator is never run. */ keg->uk_dr.dr_policy = DOMAINSET_RR(); keg->uk_dr.dr_iter = 0; /* * The master zone is passed to us at keg-creation time. */ zone = arg->zone; keg->uk_name = zone->uz_name; if (arg->flags & UMA_ZONE_ZINIT) keg->uk_init = zero_init; if (arg->flags & UMA_ZONE_MALLOC) keg->uk_flags |= UMA_ZFLAG_VTOSLAB; #ifndef SMP keg->uk_flags &= ~UMA_ZONE_PCPU; #endif keg_layout(keg); /* * Use a first-touch NUMA policy for kegs that pmap_extract() will * work on. Use round-robin for everything else. * * Zones may override the default by specifying either. */ #ifdef NUMA if ((keg->uk_flags & (UMA_ZONE_ROUNDROBIN | UMA_ZFLAG_CACHE | UMA_ZONE_NOTPAGE)) == 0) keg->uk_flags |= UMA_ZONE_FIRSTTOUCH; else if ((keg->uk_flags & UMA_ZONE_FIRSTTOUCH) == 0) keg->uk_flags |= UMA_ZONE_ROUNDROBIN; #endif /* * If we haven't booted yet we need allocations to go through the * startup cache until the vm is ready. */ #ifdef UMA_MD_SMALL_ALLOC if (keg->uk_ppera == 1) keg->uk_allocf = uma_small_alloc; else #endif if (booted < BOOT_KVA) keg->uk_allocf = startup_alloc; else if (keg->uk_flags & UMA_ZONE_PCPU) keg->uk_allocf = pcpu_page_alloc; else if ((keg->uk_flags & UMA_ZONE_CONTIG) != 0 && keg->uk_ppera > 1) keg->uk_allocf = contig_alloc; else keg->uk_allocf = page_alloc; #ifdef UMA_MD_SMALL_ALLOC if (keg->uk_ppera == 1) keg->uk_freef = uma_small_free; else #endif if (keg->uk_flags & UMA_ZONE_PCPU) keg->uk_freef = pcpu_page_free; else keg->uk_freef = page_free; /* * Initialize keg's locks. */ for (i = 0; i < vm_ndomains; i++) KEG_LOCK_INIT(keg, i, (arg->flags & UMA_ZONE_MTXCLASS)); /* * If we're putting the slab header in the actual page we need to * figure out where in each page it goes. See slab_sizeof * definition. */ if (!(keg->uk_flags & UMA_ZFLAG_OFFPAGE)) { size_t shsize; shsize = slab_sizeof(keg->uk_ipers); keg->uk_pgoff = (PAGE_SIZE * keg->uk_ppera) - shsize; /* * The only way the following is possible is if with our * UMA_ALIGN_PTR adjustments we are now bigger than * UMA_SLAB_SIZE. I haven't checked whether this is * mathematically possible for all cases, so we make * sure here anyway. */ KASSERT(keg->uk_pgoff + shsize <= PAGE_SIZE * keg->uk_ppera, ("zone %s ipers %d rsize %d size %d slab won't fit", zone->uz_name, keg->uk_ipers, keg->uk_rsize, keg->uk_size)); } if (keg->uk_flags & UMA_ZFLAG_HASH) hash_alloc(&keg->uk_hash, 0); CTR3(KTR_UMA, "keg_ctor %p zone %s(%p)", keg, zone->uz_name, zone); LIST_INSERT_HEAD(&keg->uk_zones, zone, uz_link); rw_wlock(&uma_rwlock); LIST_INSERT_HEAD(&uma_kegs, keg, uk_link); rw_wunlock(&uma_rwlock); return (0); } static void zone_kva_available(uma_zone_t zone, void *unused) { uma_keg_t keg; if ((zone->uz_flags & UMA_ZFLAG_CACHE) != 0) return; KEG_GET(zone, keg); if (keg->uk_allocf == startup_alloc) { /* Switch to the real allocator. */ if (keg->uk_flags & UMA_ZONE_PCPU) keg->uk_allocf = pcpu_page_alloc; else if ((keg->uk_flags & UMA_ZONE_CONTIG) != 0 && keg->uk_ppera > 1) keg->uk_allocf = contig_alloc; else keg->uk_allocf = page_alloc; } } static void zone_alloc_counters(uma_zone_t zone, void *unused) { zone->uz_allocs = counter_u64_alloc(M_WAITOK); zone->uz_frees = counter_u64_alloc(M_WAITOK); zone->uz_fails = counter_u64_alloc(M_WAITOK); zone->uz_xdomain = counter_u64_alloc(M_WAITOK); } static void zone_alloc_sysctl(uma_zone_t zone, void *unused) { uma_zone_domain_t zdom; uma_domain_t dom; uma_keg_t keg; struct sysctl_oid *oid, *domainoid; int domains, i, cnt; static const char *nokeg = "cache zone"; char *c; /* * Make a sysctl safe copy of the zone name by removing * any special characters and handling dups by appending * an index. */ if (zone->uz_namecnt != 0) { /* Count the number of decimal digits and '_' separator. */ for (i = 1, cnt = zone->uz_namecnt; cnt != 0; i++) cnt /= 10; zone->uz_ctlname = malloc(strlen(zone->uz_name) + i + 1, M_UMA, M_WAITOK); sprintf(zone->uz_ctlname, "%s_%d", zone->uz_name, zone->uz_namecnt); } else zone->uz_ctlname = strdup(zone->uz_name, M_UMA); for (c = zone->uz_ctlname; *c != '\0'; c++) if (strchr("./\\ -", *c) != NULL) *c = '_'; /* * Basic parameters at the root. */ zone->uz_oid = SYSCTL_ADD_NODE(NULL, SYSCTL_STATIC_CHILDREN(_vm_uma), OID_AUTO, zone->uz_ctlname, CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, ""); oid = zone->uz_oid; SYSCTL_ADD_U32(NULL, SYSCTL_CHILDREN(oid), OID_AUTO, "size", CTLFLAG_RD, &zone->uz_size, 0, "Allocation size"); SYSCTL_ADD_PROC(NULL, SYSCTL_CHILDREN(oid), OID_AUTO, "flags", CTLFLAG_RD | CTLTYPE_STRING | CTLFLAG_MPSAFE, zone, 0, sysctl_handle_uma_zone_flags, "A", "Allocator configuration flags"); SYSCTL_ADD_U16(NULL, SYSCTL_CHILDREN(oid), OID_AUTO, "bucket_size", CTLFLAG_RD, &zone->uz_bucket_size, 0, "Desired per-cpu cache size"); SYSCTL_ADD_U16(NULL, SYSCTL_CHILDREN(oid), OID_AUTO, "bucket_size_max", CTLFLAG_RD, &zone->uz_bucket_size_max, 0, "Maximum allowed per-cpu cache size"); /* * keg if present. */ if ((zone->uz_flags & UMA_ZFLAG_HASH) == 0) domains = vm_ndomains; else domains = 1; oid = SYSCTL_ADD_NODE(NULL, SYSCTL_CHILDREN(zone->uz_oid), OID_AUTO, "keg", CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, ""); keg = zone->uz_keg; if ((zone->uz_flags & UMA_ZFLAG_CACHE) == 0) { SYSCTL_ADD_CONST_STRING(NULL, SYSCTL_CHILDREN(oid), OID_AUTO, "name", CTLFLAG_RD, keg->uk_name, "Keg name"); SYSCTL_ADD_U32(NULL, SYSCTL_CHILDREN(oid), OID_AUTO, "rsize", CTLFLAG_RD, &keg->uk_rsize, 0, "Real object size with alignment"); SYSCTL_ADD_U16(NULL, SYSCTL_CHILDREN(oid), OID_AUTO, "ppera", CTLFLAG_RD, &keg->uk_ppera, 0, "pages per-slab allocation"); SYSCTL_ADD_U16(NULL, SYSCTL_CHILDREN(oid), OID_AUTO, "ipers", CTLFLAG_RD, &keg->uk_ipers, 0, "items available per-slab"); SYSCTL_ADD_U32(NULL, SYSCTL_CHILDREN(oid), OID_AUTO, "align", CTLFLAG_RD, &keg->uk_align, 0, "item alignment mask"); SYSCTL_ADD_PROC(NULL, SYSCTL_CHILDREN(oid), OID_AUTO, "efficiency", CTLFLAG_RD | CTLTYPE_INT | CTLFLAG_MPSAFE, keg, 0, sysctl_handle_uma_slab_efficiency, "I", "Slab utilization (100 - internal fragmentation %)"); domainoid = SYSCTL_ADD_NODE(NULL, SYSCTL_CHILDREN(oid), OID_AUTO, "domain", CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, ""); for (i = 0; i < domains; i++) { dom = &keg->uk_domain[i]; oid = SYSCTL_ADD_NODE(NULL, SYSCTL_CHILDREN(domainoid), OID_AUTO, VM_DOMAIN(i)->vmd_name, CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, ""); SYSCTL_ADD_U32(NULL, SYSCTL_CHILDREN(oid), OID_AUTO, "pages", CTLFLAG_RD, &dom->ud_pages, 0, "Total pages currently allocated from VM"); SYSCTL_ADD_U32(NULL, SYSCTL_CHILDREN(oid), OID_AUTO, "free_items", CTLFLAG_RD, &dom->ud_free_items, 0, "items free in the slab layer"); } } else SYSCTL_ADD_CONST_STRING(NULL, SYSCTL_CHILDREN(oid), OID_AUTO, "name", CTLFLAG_RD, nokeg, "Keg name"); /* * Information about zone limits. */ oid = SYSCTL_ADD_NODE(NULL, SYSCTL_CHILDREN(zone->uz_oid), OID_AUTO, "limit", CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, ""); SYSCTL_ADD_PROC(NULL, SYSCTL_CHILDREN(oid), OID_AUTO, "items", CTLFLAG_RD | CTLTYPE_U64 | CTLFLAG_MPSAFE, zone, 0, sysctl_handle_uma_zone_items, "QU", "current number of allocated items if limit is set"); SYSCTL_ADD_U64(NULL, SYSCTL_CHILDREN(oid), OID_AUTO, "max_items", CTLFLAG_RD, &zone->uz_max_items, 0, "Maximum number of cached items"); SYSCTL_ADD_U32(NULL, SYSCTL_CHILDREN(oid), OID_AUTO, "sleepers", CTLFLAG_RD, &zone->uz_sleepers, 0, "Number of threads sleeping at limit"); SYSCTL_ADD_U64(NULL, SYSCTL_CHILDREN(oid), OID_AUTO, "sleeps", CTLFLAG_RD, &zone->uz_sleeps, 0, "Total zone limit sleeps"); SYSCTL_ADD_U64(NULL, SYSCTL_CHILDREN(oid), OID_AUTO, "bucket_max", CTLFLAG_RD, &zone->uz_bucket_max, 0, "Maximum number of items in each domain's bucket cache"); /* * Per-domain zone information. */ domainoid = SYSCTL_ADD_NODE(NULL, SYSCTL_CHILDREN(zone->uz_oid), OID_AUTO, "domain", CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, ""); for (i = 0; i < domains; i++) { zdom = ZDOM_GET(zone, i); oid = SYSCTL_ADD_NODE(NULL, SYSCTL_CHILDREN(domainoid), OID_AUTO, VM_DOMAIN(i)->vmd_name, CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, ""); SYSCTL_ADD_LONG(NULL, SYSCTL_CHILDREN(oid), OID_AUTO, "nitems", CTLFLAG_RD, &zdom->uzd_nitems, "number of items in this domain"); SYSCTL_ADD_LONG(NULL, SYSCTL_CHILDREN(oid), OID_AUTO, "imax", CTLFLAG_RD, &zdom->uzd_imax, "maximum item count in this period"); SYSCTL_ADD_LONG(NULL, SYSCTL_CHILDREN(oid), OID_AUTO, "imin", CTLFLAG_RD, &zdom->uzd_imin, "minimum item count in this period"); SYSCTL_ADD_LONG(NULL, SYSCTL_CHILDREN(oid), OID_AUTO, "wss", CTLFLAG_RD, &zdom->uzd_wss, "Working set size"); } /* * General statistics. */ oid = SYSCTL_ADD_NODE(NULL, SYSCTL_CHILDREN(zone->uz_oid), OID_AUTO, "stats", CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, ""); SYSCTL_ADD_PROC(NULL, SYSCTL_CHILDREN(oid), OID_AUTO, "current", CTLFLAG_RD | CTLTYPE_INT | CTLFLAG_MPSAFE, zone, 1, sysctl_handle_uma_zone_cur, "I", "Current number of allocated items"); SYSCTL_ADD_PROC(NULL, SYSCTL_CHILDREN(oid), OID_AUTO, "allocs", CTLFLAG_RD | CTLTYPE_U64 | CTLFLAG_MPSAFE, zone, 0, sysctl_handle_uma_zone_allocs, "QU", "Total allocation calls"); SYSCTL_ADD_PROC(NULL, SYSCTL_CHILDREN(oid), OID_AUTO, "frees", CTLFLAG_RD | CTLTYPE_U64 | CTLFLAG_MPSAFE, zone, 0, sysctl_handle_uma_zone_frees, "QU", "Total free calls"); SYSCTL_ADD_COUNTER_U64(NULL, SYSCTL_CHILDREN(oid), OID_AUTO, "fails", CTLFLAG_RD, &zone->uz_fails, "Number of allocation failures"); SYSCTL_ADD_COUNTER_U64(NULL, SYSCTL_CHILDREN(oid), OID_AUTO, "xdomain", CTLFLAG_RD, &zone->uz_xdomain, "Free calls from the wrong domain"); } struct uma_zone_count { const char *name; int count; }; static void zone_count(uma_zone_t zone, void *arg) { struct uma_zone_count *cnt; cnt = arg; /* * Some zones are rapidly created with identical names and * destroyed out of order. This can lead to gaps in the count. * Use one greater than the maximum observed for this name. */ if (strcmp(zone->uz_name, cnt->name) == 0) cnt->count = MAX(cnt->count, zone->uz_namecnt + 1); } static void zone_update_caches(uma_zone_t zone) { int i; for (i = 0; i <= mp_maxid; i++) { cache_set_uz_size(&zone->uz_cpu[i], zone->uz_size); cache_set_uz_flags(&zone->uz_cpu[i], zone->uz_flags); } } /* * Zone header ctor. This initializes all fields, locks, etc. * * Arguments/Returns follow uma_ctor specifications * udata Actually uma_zctor_args */ static int zone_ctor(void *mem, int size, void *udata, int flags) { struct uma_zone_count cnt; struct uma_zctor_args *arg = udata; uma_zone_domain_t zdom; uma_zone_t zone = mem; uma_zone_t z; uma_keg_t keg; int i; bzero(zone, size); zone->uz_name = arg->name; zone->uz_ctor = arg->ctor; zone->uz_dtor = arg->dtor; zone->uz_init = NULL; zone->uz_fini = NULL; zone->uz_sleeps = 0; zone->uz_bucket_size = 0; zone->uz_bucket_size_min = 0; zone->uz_bucket_size_max = BUCKET_MAX; zone->uz_flags = (arg->flags & UMA_ZONE_SMR); zone->uz_warning = NULL; /* The domain structures follow the cpu structures. */ zone->uz_bucket_max = ULONG_MAX; timevalclear(&zone->uz_ratecheck); /* Count the number of duplicate names. */ cnt.name = arg->name; cnt.count = 0; zone_foreach(zone_count, &cnt); zone->uz_namecnt = cnt.count; ZONE_CROSS_LOCK_INIT(zone); for (i = 0; i < vm_ndomains; i++) { zdom = ZDOM_GET(zone, i); ZDOM_LOCK_INIT(zone, zdom, (arg->flags & UMA_ZONE_MTXCLASS)); STAILQ_INIT(&zdom->uzd_buckets); } #ifdef INVARIANTS if (arg->uminit == trash_init && arg->fini == trash_fini) zone->uz_flags |= UMA_ZFLAG_TRASH | UMA_ZFLAG_CTORDTOR; #endif /* * This is a pure cache zone, no kegs. */ if (arg->import) { KASSERT((arg->flags & UMA_ZFLAG_CACHE) != 0, ("zone_ctor: Import specified for non-cache zone.")); zone->uz_flags = arg->flags; zone->uz_size = arg->size; zone->uz_import = arg->import; zone->uz_release = arg->release; zone->uz_arg = arg->arg; #ifdef NUMA /* * Cache zones are round-robin unless a policy is * specified because they may have incompatible * constraints. */ if ((zone->uz_flags & UMA_ZONE_FIRSTTOUCH) == 0) zone->uz_flags |= UMA_ZONE_ROUNDROBIN; #endif rw_wlock(&uma_rwlock); LIST_INSERT_HEAD(&uma_cachezones, zone, uz_link); rw_wunlock(&uma_rwlock); goto out; } /* * Use the regular zone/keg/slab allocator. */ zone->uz_import = zone_import; zone->uz_release = zone_release; zone->uz_arg = zone; keg = arg->keg; if (arg->flags & UMA_ZONE_SECONDARY) { KASSERT((zone->uz_flags & UMA_ZONE_SECONDARY) == 0, ("Secondary zone requested UMA_ZFLAG_INTERNAL")); KASSERT(arg->keg != NULL, ("Secondary zone on zero'd keg")); zone->uz_init = arg->uminit; zone->uz_fini = arg->fini; zone->uz_flags |= UMA_ZONE_SECONDARY; rw_wlock(&uma_rwlock); ZONE_LOCK(zone); LIST_FOREACH(z, &keg->uk_zones, uz_link) { if (LIST_NEXT(z, uz_link) == NULL) { LIST_INSERT_AFTER(z, zone, uz_link); break; } } ZONE_UNLOCK(zone); rw_wunlock(&uma_rwlock); } else if (keg == NULL) { if ((keg = uma_kcreate(zone, arg->size, arg->uminit, arg->fini, arg->align, arg->flags)) == NULL) return (ENOMEM); } else { struct uma_kctor_args karg; int error; /* We should only be here from uma_startup() */ karg.size = arg->size; karg.uminit = arg->uminit; karg.fini = arg->fini; karg.align = arg->align; karg.flags = (arg->flags & ~UMA_ZONE_SMR); karg.zone = zone; error = keg_ctor(arg->keg, sizeof(struct uma_keg), &karg, flags); if (error) return (error); } /* Inherit properties from the keg. */ zone->uz_keg = keg; zone->uz_size = keg->uk_size; zone->uz_flags |= (keg->uk_flags & (UMA_ZONE_INHERIT | UMA_ZFLAG_INHERIT)); out: if (__predict_true(booted >= BOOT_RUNNING)) { zone_alloc_counters(zone, NULL); zone_alloc_sysctl(zone, NULL); } else { zone->uz_allocs = EARLY_COUNTER; zone->uz_frees = EARLY_COUNTER; zone->uz_fails = EARLY_COUNTER; } /* Caller requests a private SMR context. */ if ((zone->uz_flags & UMA_ZONE_SMR) != 0) zone->uz_smr = smr_create(zone->uz_name, 0, 0); KASSERT((arg->flags & (UMA_ZONE_MAXBUCKET | UMA_ZONE_NOBUCKET)) != (UMA_ZONE_MAXBUCKET | UMA_ZONE_NOBUCKET), ("Invalid zone flag combination")); if (arg->flags & UMA_ZFLAG_INTERNAL) zone->uz_bucket_size_max = zone->uz_bucket_size = 0; if ((arg->flags & UMA_ZONE_MAXBUCKET) != 0) zone->uz_bucket_size = BUCKET_MAX; else if ((arg->flags & UMA_ZONE_MINBUCKET) != 0) zone->uz_bucket_size_max = zone->uz_bucket_size = BUCKET_MIN; else if ((arg->flags & UMA_ZONE_NOBUCKET) != 0) zone->uz_bucket_size = 0; else zone->uz_bucket_size = bucket_select(zone->uz_size); zone->uz_bucket_size_min = zone->uz_bucket_size; if (zone->uz_dtor != NULL || zone->uz_ctor != NULL) zone->uz_flags |= UMA_ZFLAG_CTORDTOR; zone_update_caches(zone); return (0); } /* * Keg header dtor. This frees all data, destroys locks, frees the hash * table and removes the keg from the global list. * * Arguments/Returns follow uma_dtor specifications * udata unused */ static void keg_dtor(void *arg, int size, void *udata) { uma_keg_t keg; uint32_t free, pages; int i; keg = (uma_keg_t)arg; free = pages = 0; for (i = 0; i < vm_ndomains; i++) { free += keg->uk_domain[i].ud_free_items; pages += keg->uk_domain[i].ud_pages; KEG_LOCK_FINI(keg, i); } if (pages != 0) printf("Freed UMA keg (%s) was not empty (%u items). " " Lost %u pages of memory.\n", keg->uk_name ? keg->uk_name : "", pages / keg->uk_ppera * keg->uk_ipers - free, pages); hash_free(&keg->uk_hash); } /* * Zone header dtor. * * Arguments/Returns follow uma_dtor specifications * udata unused */ static void zone_dtor(void *arg, int size, void *udata) { uma_zone_t zone; uma_keg_t keg; int i; zone = (uma_zone_t)arg; sysctl_remove_oid(zone->uz_oid, 1, 1); if (!(zone->uz_flags & UMA_ZFLAG_INTERNAL)) cache_drain(zone); rw_wlock(&uma_rwlock); LIST_REMOVE(zone, uz_link); rw_wunlock(&uma_rwlock); zone_reclaim(zone, M_WAITOK, true); /* * We only destroy kegs from non secondary/non cache zones. */ if ((zone->uz_flags & (UMA_ZONE_SECONDARY | UMA_ZFLAG_CACHE)) == 0) { keg = zone->uz_keg; rw_wlock(&uma_rwlock); LIST_REMOVE(keg, uk_link); rw_wunlock(&uma_rwlock); zone_free_item(kegs, keg, NULL, SKIP_NONE); } counter_u64_free(zone->uz_allocs); counter_u64_free(zone->uz_frees); counter_u64_free(zone->uz_fails); counter_u64_free(zone->uz_xdomain); free(zone->uz_ctlname, M_UMA); for (i = 0; i < vm_ndomains; i++) ZDOM_LOCK_FINI(ZDOM_GET(zone, i)); ZONE_CROSS_LOCK_FINI(zone); } static void zone_foreach_unlocked(void (*zfunc)(uma_zone_t, void *arg), void *arg) { uma_keg_t keg; uma_zone_t zone; LIST_FOREACH(keg, &uma_kegs, uk_link) { LIST_FOREACH(zone, &keg->uk_zones, uz_link) zfunc(zone, arg); } LIST_FOREACH(zone, &uma_cachezones, uz_link) zfunc(zone, arg); } /* * Traverses every zone in the system and calls a callback * * Arguments: * zfunc A pointer to a function which accepts a zone * as an argument. * * Returns: * Nothing */ static void zone_foreach(void (*zfunc)(uma_zone_t, void *arg), void *arg) { rw_rlock(&uma_rwlock); zone_foreach_unlocked(zfunc, arg); rw_runlock(&uma_rwlock); } /* * Initialize the kernel memory allocator. This is done after pages can be * allocated but before general KVA is available. */ void uma_startup1(vm_offset_t virtual_avail) { struct uma_zctor_args args; size_t ksize, zsize, size; uma_keg_t masterkeg; uintptr_t m; uint8_t pflag; bootstart = bootmem = virtual_avail; rw_init(&uma_rwlock, "UMA lock"); sx_init(&uma_reclaim_lock, "umareclaim"); ksize = sizeof(struct uma_keg) + (sizeof(struct uma_domain) * vm_ndomains); ksize = roundup(ksize, UMA_SUPER_ALIGN); zsize = sizeof(struct uma_zone) + (sizeof(struct uma_cache) * (mp_maxid + 1)) + (sizeof(struct uma_zone_domain) * vm_ndomains); zsize = roundup(zsize, UMA_SUPER_ALIGN); /* Allocate the zone of zones, zone of kegs, and zone of zones keg. */ size = (zsize * 2) + ksize; m = (uintptr_t)startup_alloc(NULL, size, 0, &pflag, M_NOWAIT | M_ZERO); zones = (uma_zone_t)m; m += zsize; kegs = (uma_zone_t)m; m += zsize; masterkeg = (uma_keg_t)m; /* "manually" create the initial zone */ memset(&args, 0, sizeof(args)); args.name = "UMA Kegs"; args.size = ksize; args.ctor = keg_ctor; args.dtor = keg_dtor; args.uminit = zero_init; args.fini = NULL; args.keg = masterkeg; args.align = UMA_SUPER_ALIGN - 1; args.flags = UMA_ZFLAG_INTERNAL; zone_ctor(kegs, zsize, &args, M_WAITOK); args.name = "UMA Zones"; args.size = zsize; args.ctor = zone_ctor; args.dtor = zone_dtor; args.uminit = zero_init; args.fini = NULL; args.keg = NULL; args.align = UMA_SUPER_ALIGN - 1; args.flags = UMA_ZFLAG_INTERNAL; zone_ctor(zones, zsize, &args, M_WAITOK); /* Now make zones for slab headers */ slabzones[0] = uma_zcreate("UMA Slabs 0", SLABZONE0_SIZE, NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZFLAG_INTERNAL); slabzones[1] = uma_zcreate("UMA Slabs 1", SLABZONE1_SIZE, NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZFLAG_INTERNAL); hashzone = uma_zcreate("UMA Hash", sizeof(struct slabhead *) * UMA_HASH_SIZE_INIT, NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZFLAG_INTERNAL); bucket_init(); smr_init(); } #ifndef UMA_MD_SMALL_ALLOC extern void vm_radix_reserve_kva(void); #endif /* * Advertise the availability of normal kva allocations and switch to * the default back-end allocator. Marks the KVA we consumed on startup * as used in the map. */ void uma_startup2(void) { if (bootstart != bootmem) { vm_map_lock(kernel_map); (void)vm_map_insert(kernel_map, NULL, 0, bootstart, bootmem, VM_PROT_RW, VM_PROT_RW, MAP_NOFAULT); vm_map_unlock(kernel_map); } #ifndef UMA_MD_SMALL_ALLOC /* Set up radix zone to use noobj_alloc. */ vm_radix_reserve_kva(); #endif booted = BOOT_KVA; zone_foreach_unlocked(zone_kva_available, NULL); bucket_enable(); } /* * Finish our initialization steps. */ static void uma_startup3(void) { #ifdef INVARIANTS TUNABLE_INT_FETCH("vm.debug.divisor", &dbg_divisor); uma_dbg_cnt = counter_u64_alloc(M_WAITOK); uma_skip_cnt = counter_u64_alloc(M_WAITOK); #endif zone_foreach_unlocked(zone_alloc_counters, NULL); zone_foreach_unlocked(zone_alloc_sysctl, NULL); callout_init(&uma_callout, 1); callout_reset(&uma_callout, UMA_TIMEOUT * hz, uma_timeout, NULL); booted = BOOT_RUNNING; EVENTHANDLER_REGISTER(shutdown_post_sync, uma_shutdown, NULL, EVENTHANDLER_PRI_FIRST); } static void uma_shutdown(void) { booted = BOOT_SHUTDOWN; } static uma_keg_t uma_kcreate(uma_zone_t zone, size_t size, uma_init uminit, uma_fini fini, int align, uint32_t flags) { struct uma_kctor_args args; args.size = size; args.uminit = uminit; args.fini = fini; args.align = (align == UMA_ALIGN_CACHE) ? uma_align_cache : align; args.flags = flags; args.zone = zone; return (zone_alloc_item(kegs, &args, UMA_ANYDOMAIN, M_WAITOK)); } /* Public functions */ /* See uma.h */ void uma_set_align(int align) { if (align != UMA_ALIGN_CACHE) uma_align_cache = align; } /* See uma.h */ uma_zone_t uma_zcreate(const char *name, size_t size, uma_ctor ctor, uma_dtor dtor, uma_init uminit, uma_fini fini, int align, uint32_t flags) { struct uma_zctor_args args; uma_zone_t res; KASSERT(powerof2(align + 1), ("invalid zone alignment %d for \"%s\"", align, name)); /* This stuff is essential for the zone ctor */ memset(&args, 0, sizeof(args)); args.name = name; args.size = size; args.ctor = ctor; args.dtor = dtor; args.uminit = uminit; args.fini = fini; #ifdef INVARIANTS /* * Inject procedures which check for memory use after free if we are * allowed to scramble the memory while it is not allocated. This * requires that: UMA is actually able to access the memory, no init * or fini procedures, no dependency on the initial value of the * memory, and no (legitimate) use of the memory after free. Note, * the ctor and dtor do not need to be empty. */ if ((!(flags & (UMA_ZONE_ZINIT | UMA_ZONE_NOTOUCH | UMA_ZONE_NOFREE))) && uminit == NULL && fini == NULL) { args.uminit = trash_init; args.fini = trash_fini; } #endif args.align = align; args.flags = flags; args.keg = NULL; sx_slock(&uma_reclaim_lock); res = zone_alloc_item(zones, &args, UMA_ANYDOMAIN, M_WAITOK); sx_sunlock(&uma_reclaim_lock); return (res); } /* See uma.h */ uma_zone_t uma_zsecond_create(const char *name, uma_ctor ctor, uma_dtor dtor, uma_init zinit, uma_fini zfini, uma_zone_t master) { struct uma_zctor_args args; uma_keg_t keg; uma_zone_t res; keg = master->uz_keg; memset(&args, 0, sizeof(args)); args.name = name; args.size = keg->uk_size; args.ctor = ctor; args.dtor = dtor; args.uminit = zinit; args.fini = zfini; args.align = keg->uk_align; args.flags = keg->uk_flags | UMA_ZONE_SECONDARY; args.keg = keg; sx_slock(&uma_reclaim_lock); res = zone_alloc_item(zones, &args, UMA_ANYDOMAIN, M_WAITOK); sx_sunlock(&uma_reclaim_lock); return (res); } /* See uma.h */ uma_zone_t uma_zcache_create(const char *name, int size, uma_ctor ctor, uma_dtor dtor, uma_init zinit, uma_fini zfini, uma_import zimport, uma_release zrelease, void *arg, int flags) { struct uma_zctor_args args; memset(&args, 0, sizeof(args)); args.name = name; args.size = size; args.ctor = ctor; args.dtor = dtor; args.uminit = zinit; args.fini = zfini; args.import = zimport; args.release = zrelease; args.arg = arg; args.align = 0; args.flags = flags | UMA_ZFLAG_CACHE; return (zone_alloc_item(zones, &args, UMA_ANYDOMAIN, M_WAITOK)); } /* See uma.h */ void uma_zdestroy(uma_zone_t zone) { /* * Large slabs are expensive to reclaim, so don't bother doing * unnecessary work if we're shutting down. */ if (booted == BOOT_SHUTDOWN && zone->uz_fini == NULL && zone->uz_release == zone_release) return; sx_slock(&uma_reclaim_lock); zone_free_item(zones, zone, NULL, SKIP_NONE); sx_sunlock(&uma_reclaim_lock); } void uma_zwait(uma_zone_t zone) { if ((zone->uz_flags & UMA_ZONE_SMR) != 0) uma_zfree_smr(zone, uma_zalloc_smr(zone, M_WAITOK)); else if ((zone->uz_flags & UMA_ZONE_PCPU) != 0) uma_zfree_pcpu(zone, uma_zalloc_pcpu(zone, M_WAITOK)); else uma_zfree(zone, uma_zalloc(zone, M_WAITOK)); } void * uma_zalloc_pcpu_arg(uma_zone_t zone, void *udata, int flags) { void *item, *pcpu_item; #ifdef SMP int i; MPASS(zone->uz_flags & UMA_ZONE_PCPU); #endif item = uma_zalloc_arg(zone, udata, flags & ~M_ZERO); if (item == NULL) return (NULL); pcpu_item = zpcpu_base_to_offset(item); if (flags & M_ZERO) { #ifdef SMP for (i = 0; i <= mp_maxid; i++) bzero(zpcpu_get_cpu(pcpu_item, i), zone->uz_size); #else bzero(item, zone->uz_size); #endif } return (pcpu_item); } /* * A stub while both regular and pcpu cases are identical. */ void uma_zfree_pcpu_arg(uma_zone_t zone, void *pcpu_item, void *udata) { void *item; #ifdef SMP MPASS(zone->uz_flags & UMA_ZONE_PCPU); #endif item = zpcpu_offset_to_base(pcpu_item); uma_zfree_arg(zone, item, udata); } static inline void * item_ctor(uma_zone_t zone, int uz_flags, int size, void *udata, int flags, void *item) { #ifdef INVARIANTS bool skipdbg; skipdbg = uma_dbg_zskip(zone, item); if (!skipdbg && (zone->uz_flags & UMA_ZFLAG_TRASH) != 0 && zone->uz_ctor != trash_ctor) trash_ctor(item, size, udata, flags); #endif /* Check flags before loading ctor pointer. */ if (__predict_false((uz_flags & UMA_ZFLAG_CTORDTOR) != 0) && __predict_false(zone->uz_ctor != NULL) && zone->uz_ctor(item, size, udata, flags) != 0) { counter_u64_add(zone->uz_fails, 1); zone_free_item(zone, item, udata, SKIP_DTOR | SKIP_CNT); return (NULL); } #ifdef INVARIANTS if (!skipdbg) uma_dbg_alloc(zone, NULL, item); #endif if (__predict_false(flags & M_ZERO)) return (memset(item, 0, size)); return (item); } static inline void item_dtor(uma_zone_t zone, void *item, int size, void *udata, enum zfreeskip skip) { #ifdef INVARIANTS bool skipdbg; skipdbg = uma_dbg_zskip(zone, item); if (skip == SKIP_NONE && !skipdbg) { if ((zone->uz_flags & UMA_ZONE_MALLOC) != 0) uma_dbg_free(zone, udata, item); else uma_dbg_free(zone, NULL, item); } #endif if (__predict_true(skip < SKIP_DTOR)) { if (zone->uz_dtor != NULL) zone->uz_dtor(item, size, udata); #ifdef INVARIANTS if (!skipdbg && (zone->uz_flags & UMA_ZFLAG_TRASH) != 0 && zone->uz_dtor != trash_dtor) trash_dtor(item, size, udata); #endif } } #if defined(INVARIANTS) || defined(DEBUG_MEMGUARD) || defined(WITNESS) #define UMA_ZALLOC_DEBUG static int uma_zalloc_debug(uma_zone_t zone, void **itemp, void *udata, int flags) { int error; error = 0; #ifdef WITNESS if (flags & M_WAITOK) { WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "uma_zalloc_debug: zone \"%s\"", zone->uz_name); } #endif #ifdef INVARIANTS KASSERT((flags & M_EXEC) == 0, ("uma_zalloc_debug: called with M_EXEC")); KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(), ("uma_zalloc_debug: called within spinlock or critical section")); KASSERT((zone->uz_flags & UMA_ZONE_PCPU) == 0 || (flags & M_ZERO) == 0, ("uma_zalloc_debug: allocating from a pcpu zone with M_ZERO")); #endif #ifdef DEBUG_MEMGUARD if ((zone->uz_flags & UMA_ZONE_SMR) == 0 && memguard_cmp_zone(zone)) { void *item; item = memguard_alloc(zone->uz_size, flags); if (item != NULL) { error = EJUSTRETURN; if (zone->uz_init != NULL && zone->uz_init(item, zone->uz_size, flags) != 0) { *itemp = NULL; return (error); } if (zone->uz_ctor != NULL && zone->uz_ctor(item, zone->uz_size, udata, flags) != 0) { counter_u64_add(zone->uz_fails, 1); zone->uz_fini(item, zone->uz_size); *itemp = NULL; return (error); } *itemp = item; return (error); } /* This is unfortunate but should not be fatal. */ } #endif return (error); } static int uma_zfree_debug(uma_zone_t zone, void *item, void *udata) { KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(), ("uma_zfree_debug: called with spinlock or critical section held")); #ifdef DEBUG_MEMGUARD if ((zone->uz_flags & UMA_ZONE_SMR) == 0 && is_memguard_addr(item)) { if (zone->uz_dtor != NULL) zone->uz_dtor(item, zone->uz_size, udata); if (zone->uz_fini != NULL) zone->uz_fini(item, zone->uz_size); memguard_free(item); return (EJUSTRETURN); } #endif return (0); } #endif static inline void * cache_alloc_item(uma_zone_t zone, uma_cache_t cache, uma_cache_bucket_t bucket, void *udata, int flags) { void *item; int size, uz_flags; item = cache_bucket_pop(cache, bucket); size = cache_uz_size(cache); uz_flags = cache_uz_flags(cache); critical_exit(); return (item_ctor(zone, uz_flags, size, udata, flags, item)); } static __noinline void * cache_alloc_retry(uma_zone_t zone, uma_cache_t cache, void *udata, int flags) { uma_cache_bucket_t bucket; int domain; while (cache_alloc(zone, cache, udata, flags)) { cache = &zone->uz_cpu[curcpu]; bucket = &cache->uc_allocbucket; if (__predict_false(bucket->ucb_cnt == 0)) continue; return (cache_alloc_item(zone, cache, bucket, udata, flags)); } critical_exit(); /* * We can not get a bucket so try to return a single item. */ if (zone->uz_flags & UMA_ZONE_FIRSTTOUCH) domain = PCPU_GET(domain); else domain = UMA_ANYDOMAIN; return (zone_alloc_item(zone, udata, domain, flags)); } /* See uma.h */ void * uma_zalloc_smr(uma_zone_t zone, int flags) { uma_cache_bucket_t bucket; uma_cache_t cache; #ifdef UMA_ZALLOC_DEBUG void *item; KASSERT((zone->uz_flags & UMA_ZONE_SMR) != 0, ("uma_zalloc_arg: called with non-SMR zone.\n")); if (uma_zalloc_debug(zone, &item, NULL, flags) == EJUSTRETURN) return (item); #endif critical_enter(); cache = &zone->uz_cpu[curcpu]; bucket = &cache->uc_allocbucket; if (__predict_false(bucket->ucb_cnt == 0)) return (cache_alloc_retry(zone, cache, NULL, flags)); return (cache_alloc_item(zone, cache, bucket, NULL, flags)); } /* See uma.h */ void * uma_zalloc_arg(uma_zone_t zone, void *udata, int flags) { uma_cache_bucket_t bucket; uma_cache_t cache; /* Enable entropy collection for RANDOM_ENABLE_UMA kernel option */ random_harvest_fast_uma(&zone, sizeof(zone), RANDOM_UMA); /* This is the fast path allocation */ CTR3(KTR_UMA, "uma_zalloc_arg zone %s(%p) flags %d", zone->uz_name, zone, flags); #ifdef UMA_ZALLOC_DEBUG void *item; KASSERT((zone->uz_flags & UMA_ZONE_SMR) == 0, ("uma_zalloc_arg: called with SMR zone.\n")); if (uma_zalloc_debug(zone, &item, udata, flags) == EJUSTRETURN) return (item); #endif /* * If possible, allocate from the per-CPU cache. There are two * requirements for safe access to the per-CPU cache: (1) the thread * accessing the cache must not be preempted or yield during access, * and (2) the thread must not migrate CPUs without switching which * cache it accesses. We rely on a critical section to prevent * preemption and migration. We release the critical section in * order to acquire the zone mutex if we are unable to allocate from * the current cache; when we re-acquire the critical section, we * must detect and handle migration if it has occurred. */ critical_enter(); cache = &zone->uz_cpu[curcpu]; bucket = &cache->uc_allocbucket; if (__predict_false(bucket->ucb_cnt == 0)) return (cache_alloc_retry(zone, cache, udata, flags)); return (cache_alloc_item(zone, cache, bucket, udata, flags)); } /* * Replenish an alloc bucket and possibly restore an old one. Called in * a critical section. Returns in a critical section. * * A false return value indicates an allocation failure. * A true return value indicates success and the caller should retry. */ static __noinline bool cache_alloc(uma_zone_t zone, uma_cache_t cache, void *udata, int flags) { uma_bucket_t bucket; int domain; bool new; CRITICAL_ASSERT(curthread); /* * If we have run out of items in our alloc bucket see * if we can switch with the free bucket. * * SMR Zones can't re-use the free bucket until the sequence has * expired. */ if ((cache_uz_flags(cache) & UMA_ZONE_SMR) == 0 && cache->uc_freebucket.ucb_cnt != 0) { cache_bucket_swap(&cache->uc_freebucket, &cache->uc_allocbucket); return (true); } /* * Discard any empty allocation bucket while we hold no locks. */ bucket = cache_bucket_unload_alloc(cache); critical_exit(); if (bucket != NULL) { KASSERT(bucket->ub_cnt == 0, ("cache_alloc: Entered with non-empty alloc bucket.")); bucket_free(zone, bucket, udata); } /* Short-circuit for zones without buckets and low memory. */ if (zone->uz_bucket_size == 0 || bucketdisable) { critical_enter(); return (false); } /* * Attempt to retrieve the item from the per-CPU cache has failed, so * we must go back to the zone. This requires the zdom lock, so we * must drop the critical section, then re-acquire it when we go back * to the cache. Since the critical section is released, we may be * preempted or migrate. As such, make sure not to maintain any * thread-local state specific to the cache from prior to releasing * the critical section. */ domain = PCPU_GET(domain); if ((cache_uz_flags(cache) & UMA_ZONE_ROUNDROBIN) != 0) domain = zone_domain_highest(zone, domain); bucket = cache_fetch_bucket(zone, cache, domain); if (bucket == NULL) { bucket = zone_alloc_bucket(zone, udata, domain, flags); new = true; } else new = false; CTR3(KTR_UMA, "uma_zalloc: zone %s(%p) bucket zone returned %p", zone->uz_name, zone, bucket); if (bucket == NULL) { critical_enter(); return (false); } /* * See if we lost the race or were migrated. Cache the * initialized bucket to make this less likely or claim * the memory directly. */ critical_enter(); cache = &zone->uz_cpu[curcpu]; if (cache->uc_allocbucket.ucb_bucket == NULL && ((cache_uz_flags(cache) & UMA_ZONE_FIRSTTOUCH) == 0 || domain == PCPU_GET(domain))) { if (new) atomic_add_long(&ZDOM_GET(zone, domain)->uzd_imax, bucket->ub_cnt); cache_bucket_load_alloc(cache, bucket); return (true); } /* * We lost the race, release this bucket and start over. */ critical_exit(); zone_put_bucket(zone, domain, bucket, udata, false); critical_enter(); return (true); } void * uma_zalloc_domain(uma_zone_t zone, void *udata, int domain, int flags) { /* Enable entropy collection for RANDOM_ENABLE_UMA kernel option */ random_harvest_fast_uma(&zone, sizeof(zone), RANDOM_UMA); /* This is the fast path allocation */ CTR4(KTR_UMA, "uma_zalloc_domain zone %s(%p) domain %d flags %d", zone->uz_name, zone, domain, flags); if (flags & M_WAITOK) { WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL, "uma_zalloc_domain: zone \"%s\"", zone->uz_name); } KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(), ("uma_zalloc_domain: called with spinlock or critical section held")); return (zone_alloc_item(zone, udata, domain, flags)); } /* * Find a slab with some space. Prefer slabs that are partially used over those * that are totally full. This helps to reduce fragmentation. * * If 'rr' is 1, search all domains starting from 'domain'. Otherwise check * only 'domain'. */ static uma_slab_t keg_first_slab(uma_keg_t keg, int domain, bool rr) { uma_domain_t dom; uma_slab_t slab; int start; KASSERT(domain >= 0 && domain < vm_ndomains, ("keg_first_slab: domain %d out of range", domain)); KEG_LOCK_ASSERT(keg, domain); slab = NULL; start = domain; do { dom = &keg->uk_domain[domain]; if ((slab = LIST_FIRST(&dom->ud_part_slab)) != NULL) return (slab); if ((slab = LIST_FIRST(&dom->ud_free_slab)) != NULL) { LIST_REMOVE(slab, us_link); dom->ud_free_slabs--; LIST_INSERT_HEAD(&dom->ud_part_slab, slab, us_link); return (slab); } if (rr) domain = (domain + 1) % vm_ndomains; } while (domain != start); return (NULL); } /* * Fetch an existing slab from a free or partial list. Returns with the * keg domain lock held if a slab was found or unlocked if not. */ static uma_slab_t keg_fetch_free_slab(uma_keg_t keg, int domain, bool rr, int flags) { uma_slab_t slab; uint32_t reserve; /* HASH has a single free list. */ if ((keg->uk_flags & UMA_ZFLAG_HASH) != 0) domain = 0; KEG_LOCK(keg, domain); reserve = (flags & M_USE_RESERVE) != 0 ? 0 : keg->uk_reserve; if (keg->uk_domain[domain].ud_free_items <= reserve || (slab = keg_first_slab(keg, domain, rr)) == NULL) { KEG_UNLOCK(keg, domain); return (NULL); } return (slab); } static uma_slab_t keg_fetch_slab(uma_keg_t keg, uma_zone_t zone, int rdomain, const int flags) { struct vm_domainset_iter di; uma_slab_t slab; int aflags, domain; bool rr; restart: /* * Use the keg's policy if upper layers haven't already specified a * domain (as happens with first-touch zones). * * To avoid races we run the iterator with the keg lock held, but that * means that we cannot allow the vm_domainset layer to sleep. Thus, * clear M_WAITOK and handle low memory conditions locally. */ rr = rdomain == UMA_ANYDOMAIN; if (rr) { aflags = (flags & ~M_WAITOK) | M_NOWAIT; vm_domainset_iter_policy_ref_init(&di, &keg->uk_dr, &domain, &aflags); } else { aflags = flags; domain = rdomain; } for (;;) { slab = keg_fetch_free_slab(keg, domain, rr, flags); if (slab != NULL) return (slab); /* * M_NOVM means don't ask at all! */ if (flags & M_NOVM) break; slab = keg_alloc_slab(keg, zone, domain, flags, aflags); if (slab != NULL) return (slab); if (!rr && (flags & M_WAITOK) == 0) break; if (rr && vm_domainset_iter_policy(&di, &domain) != 0) { if ((flags & M_WAITOK) != 0) { vm_wait_doms(&keg->uk_dr.dr_policy->ds_mask); goto restart; } break; } } /* * We might not have been able to get a slab but another cpu * could have while we were unlocked. Check again before we * fail. */ if ((slab = keg_fetch_free_slab(keg, domain, rr, flags)) != NULL) return (slab); return (NULL); } static void * slab_alloc_item(uma_keg_t keg, uma_slab_t slab) { uma_domain_t dom; void *item; int freei; KEG_LOCK_ASSERT(keg, slab->us_domain); dom = &keg->uk_domain[slab->us_domain]; freei = BIT_FFS(keg->uk_ipers, &slab->us_free) - 1; BIT_CLR(keg->uk_ipers, freei, &slab->us_free); item = slab_item(slab, keg, freei); slab->us_freecount--; dom->ud_free_items--; /* * Move this slab to the full list. It must be on the partial list, so * we do not need to update the free slab count. In particular, * keg_fetch_slab() always returns slabs on the partial list. */ if (slab->us_freecount == 0) { LIST_REMOVE(slab, us_link); LIST_INSERT_HEAD(&dom->ud_full_slab, slab, us_link); } return (item); } static int zone_import(void *arg, void **bucket, int max, int domain, int flags) { uma_domain_t dom; uma_zone_t zone; uma_slab_t slab; uma_keg_t keg; #ifdef NUMA int stripe; #endif int i; zone = arg; slab = NULL; keg = zone->uz_keg; /* Try to keep the buckets totally full */ for (i = 0; i < max; ) { if ((slab = keg_fetch_slab(keg, zone, domain, flags)) == NULL) break; #ifdef NUMA stripe = howmany(max, vm_ndomains); #endif dom = &keg->uk_domain[slab->us_domain]; while (slab->us_freecount && i < max) { bucket[i++] = slab_alloc_item(keg, slab); if (dom->ud_free_items <= keg->uk_reserve) break; #ifdef NUMA /* * If the zone is striped we pick a new slab for every * N allocations. Eliminating this conditional will * instead pick a new domain for each bucket rather * than stripe within each bucket. The current option * produces more fragmentation and requires more cpu * time but yields better distribution. */ if ((zone->uz_flags & UMA_ZONE_ROUNDROBIN) != 0 && vm_ndomains > 1 && --stripe == 0) break; #endif } KEG_UNLOCK(keg, slab->us_domain); /* Don't block if we allocated any successfully. */ flags &= ~M_WAITOK; flags |= M_NOWAIT; } return i; } static int zone_alloc_limit_hard(uma_zone_t zone, int count, int flags) { uint64_t old, new, total, max; /* * The hard case. We're going to sleep because there were existing * sleepers or because we ran out of items. This routine enforces * fairness by keeping fifo order. * * First release our ill gotten gains and make some noise. */ for (;;) { zone_free_limit(zone, count); zone_log_warning(zone); zone_maxaction(zone); if (flags & M_NOWAIT) return (0); /* * We need to allocate an item or set ourself as a sleeper * while the sleepq lock is held to avoid wakeup races. This * is essentially a home rolled semaphore. */ sleepq_lock(&zone->uz_max_items); old = zone->uz_items; do { MPASS(UZ_ITEMS_SLEEPERS(old) < UZ_ITEMS_SLEEPERS_MAX); /* Cache the max since we will evaluate twice. */ max = zone->uz_max_items; if (UZ_ITEMS_SLEEPERS(old) != 0 || UZ_ITEMS_COUNT(old) >= max) new = old + UZ_ITEMS_SLEEPER; else new = old + MIN(count, max - old); } while (atomic_fcmpset_64(&zone->uz_items, &old, new) == 0); /* We may have successfully allocated under the sleepq lock. */ if (UZ_ITEMS_SLEEPERS(new) == 0) { sleepq_release(&zone->uz_max_items); return (new - old); } /* * This is in a different cacheline from uz_items so that we * don't constantly invalidate the fastpath cacheline when we * adjust item counts. This could be limited to toggling on * transitions. */ atomic_add_32(&zone->uz_sleepers, 1); atomic_add_64(&zone->uz_sleeps, 1); /* * We have added ourselves as a sleeper. The sleepq lock * protects us from wakeup races. Sleep now and then retry. */ sleepq_add(&zone->uz_max_items, NULL, "zonelimit", 0, 0); sleepq_wait(&zone->uz_max_items, PVM); /* * After wakeup, remove ourselves as a sleeper and try * again. We no longer have the sleepq lock for protection. * * Subract ourselves as a sleeper while attempting to add * our count. */ atomic_subtract_32(&zone->uz_sleepers, 1); old = atomic_fetchadd_64(&zone->uz_items, -(UZ_ITEMS_SLEEPER - count)); /* We're no longer a sleeper. */ old -= UZ_ITEMS_SLEEPER; /* * If we're still at the limit, restart. Notably do not * block on other sleepers. Cache the max value to protect * against changes via sysctl. */ total = UZ_ITEMS_COUNT(old); max = zone->uz_max_items; if (total >= max) continue; /* Truncate if necessary, otherwise wake other sleepers. */ if (total + count > max) { zone_free_limit(zone, total + count - max); count = max - total; } else if (total + count < max && UZ_ITEMS_SLEEPERS(old) != 0) wakeup_one(&zone->uz_max_items); return (count); } } /* * Allocate 'count' items from our max_items limit. Returns the number * available. If M_NOWAIT is not specified it will sleep until at least * one item can be allocated. */ static int zone_alloc_limit(uma_zone_t zone, int count, int flags) { uint64_t old; uint64_t max; max = zone->uz_max_items; MPASS(max > 0); /* * We expect normal allocations to succeed with a simple * fetchadd. */ old = atomic_fetchadd_64(&zone->uz_items, count); if (__predict_true(old + count <= max)) return (count); /* * If we had some items and no sleepers just return the * truncated value. We have to release the excess space * though because that may wake sleepers who weren't woken * because we were temporarily over the limit. */ if (old < max) { zone_free_limit(zone, (old + count) - max); return (max - old); } return (zone_alloc_limit_hard(zone, count, flags)); } /* * Free a number of items back to the limit. */ static void zone_free_limit(uma_zone_t zone, int count) { uint64_t old; MPASS(count > 0); /* * In the common case we either have no sleepers or * are still over the limit and can just return. */ old = atomic_fetchadd_64(&zone->uz_items, -count); if (__predict_true(UZ_ITEMS_SLEEPERS(old) == 0 || UZ_ITEMS_COUNT(old) - count >= zone->uz_max_items)) return; /* * Moderate the rate of wakeups. Sleepers will continue * to generate wakeups if necessary. */ wakeup_one(&zone->uz_max_items); } static uma_bucket_t zone_alloc_bucket(uma_zone_t zone, void *udata, int domain, int flags) { uma_bucket_t bucket; int maxbucket, cnt; CTR3(KTR_UMA, "zone_alloc_bucket zone %s(%p) domain %d", zone->uz_name, zone, domain); /* Avoid allocs targeting empty domains. */ if (domain != UMA_ANYDOMAIN && VM_DOMAIN_EMPTY(domain)) domain = UMA_ANYDOMAIN; if ((zone->uz_flags & UMA_ZONE_ROUNDROBIN) != 0) domain = UMA_ANYDOMAIN; if (zone->uz_max_items > 0) maxbucket = zone_alloc_limit(zone, zone->uz_bucket_size, M_NOWAIT); else maxbucket = zone->uz_bucket_size; if (maxbucket == 0) return (false); /* Don't wait for buckets, preserve caller's NOVM setting. */ bucket = bucket_alloc(zone, udata, M_NOWAIT | (flags & M_NOVM)); if (bucket == NULL) { cnt = 0; goto out; } bucket->ub_cnt = zone->uz_import(zone->uz_arg, bucket->ub_bucket, MIN(maxbucket, bucket->ub_entries), domain, flags); /* * Initialize the memory if necessary. */ if (bucket->ub_cnt != 0 && zone->uz_init != NULL) { int i; for (i = 0; i < bucket->ub_cnt; i++) if (zone->uz_init(bucket->ub_bucket[i], zone->uz_size, flags) != 0) break; /* * If we couldn't initialize the whole bucket, put the * rest back onto the freelist. */ if (i != bucket->ub_cnt) { zone->uz_release(zone->uz_arg, &bucket->ub_bucket[i], bucket->ub_cnt - i); #ifdef INVARIANTS bzero(&bucket->ub_bucket[i], sizeof(void *) * (bucket->ub_cnt - i)); #endif bucket->ub_cnt = i; } } cnt = bucket->ub_cnt; if (bucket->ub_cnt == 0) { bucket_free(zone, bucket, udata); counter_u64_add(zone->uz_fails, 1); bucket = NULL; } out: if (zone->uz_max_items > 0 && cnt < maxbucket) zone_free_limit(zone, maxbucket - cnt); return (bucket); } /* * Allocates a single item from a zone. * * Arguments * zone The zone to alloc for. * udata The data to be passed to the constructor. * domain The domain to allocate from or UMA_ANYDOMAIN. * flags M_WAITOK, M_NOWAIT, M_ZERO. * * Returns * NULL if there is no memory and M_NOWAIT is set * An item if successful */ static void * zone_alloc_item(uma_zone_t zone, void *udata, int domain, int flags) { void *item; if (zone->uz_max_items > 0 && zone_alloc_limit(zone, 1, flags) == 0) return (NULL); /* Avoid allocs targeting empty domains. */ if (domain != UMA_ANYDOMAIN && VM_DOMAIN_EMPTY(domain)) domain = UMA_ANYDOMAIN; if (zone->uz_import(zone->uz_arg, &item, 1, domain, flags) != 1) goto fail_cnt; /* * We have to call both the zone's init (not the keg's init) * and the zone's ctor. This is because the item is going from * a keg slab directly to the user, and the user is expecting it * to be both zone-init'd as well as zone-ctor'd. */ if (zone->uz_init != NULL) { if (zone->uz_init(item, zone->uz_size, flags) != 0) { zone_free_item(zone, item, udata, SKIP_FINI | SKIP_CNT); goto fail_cnt; } } item = item_ctor(zone, zone->uz_flags, zone->uz_size, udata, flags, item); if (item == NULL) goto fail; counter_u64_add(zone->uz_allocs, 1); CTR3(KTR_UMA, "zone_alloc_item item %p from %s(%p)", item, zone->uz_name, zone); return (item); fail_cnt: counter_u64_add(zone->uz_fails, 1); fail: if (zone->uz_max_items > 0) zone_free_limit(zone, 1); CTR2(KTR_UMA, "zone_alloc_item failed from %s(%p)", zone->uz_name, zone); return (NULL); } /* See uma.h */ void uma_zfree_smr(uma_zone_t zone, void *item) { uma_cache_t cache; uma_cache_bucket_t bucket; int itemdomain, uz_flags; #ifdef UMA_ZALLOC_DEBUG KASSERT((zone->uz_flags & UMA_ZONE_SMR) != 0, ("uma_zfree_smr: called with non-SMR zone.\n")); KASSERT(item != NULL, ("uma_zfree_smr: Called with NULL pointer.")); SMR_ASSERT_NOT_ENTERED(zone->uz_smr); if (uma_zfree_debug(zone, item, NULL) == EJUSTRETURN) return; #endif cache = &zone->uz_cpu[curcpu]; uz_flags = cache_uz_flags(cache); itemdomain = 0; #ifdef NUMA if ((uz_flags & UMA_ZONE_FIRSTTOUCH) != 0) itemdomain = _vm_phys_domain(pmap_kextract((vm_offset_t)item)); #endif critical_enter(); do { cache = &zone->uz_cpu[curcpu]; /* SMR Zones must free to the free bucket. */ bucket = &cache->uc_freebucket; #ifdef NUMA if ((uz_flags & UMA_ZONE_FIRSTTOUCH) != 0 && PCPU_GET(domain) != itemdomain) { bucket = &cache->uc_crossbucket; } #endif if (__predict_true(bucket->ucb_cnt < bucket->ucb_entries)) { cache_bucket_push(cache, bucket, item); critical_exit(); return; } } while (cache_free(zone, cache, NULL, item, itemdomain)); critical_exit(); /* * If nothing else caught this, we'll just do an internal free. */ zone_free_item(zone, item, NULL, SKIP_NONE); } /* See uma.h */ void uma_zfree_arg(uma_zone_t zone, void *item, void *udata) { uma_cache_t cache; uma_cache_bucket_t bucket; int itemdomain, uz_flags; /* Enable entropy collection for RANDOM_ENABLE_UMA kernel option */ random_harvest_fast_uma(&zone, sizeof(zone), RANDOM_UMA); CTR2(KTR_UMA, "uma_zfree_arg zone %s(%p)", zone->uz_name, zone); #ifdef UMA_ZALLOC_DEBUG KASSERT((zone->uz_flags & UMA_ZONE_SMR) == 0, ("uma_zfree_arg: called with SMR zone.\n")); if (uma_zfree_debug(zone, item, udata) == EJUSTRETURN) return; #endif /* uma_zfree(..., NULL) does nothing, to match free(9). */ if (item == NULL) return; /* * We are accessing the per-cpu cache without a critical section to * fetch size and flags. This is acceptable, if we are preempted we * will simply read another cpu's line. */ cache = &zone->uz_cpu[curcpu]; uz_flags = cache_uz_flags(cache); if (UMA_ALWAYS_CTORDTOR || __predict_false((uz_flags & UMA_ZFLAG_CTORDTOR) != 0)) item_dtor(zone, item, cache_uz_size(cache), udata, SKIP_NONE); /* * The race here is acceptable. If we miss it we'll just have to wait * a little longer for the limits to be reset. */ if (__predict_false(uz_flags & UMA_ZFLAG_LIMIT)) { if (zone->uz_sleepers > 0) goto zfree_item; } /* * If possible, free to the per-CPU cache. There are two * requirements for safe access to the per-CPU cache: (1) the thread * accessing the cache must not be preempted or yield during access, * and (2) the thread must not migrate CPUs without switching which * cache it accesses. We rely on a critical section to prevent * preemption and migration. We release the critical section in * order to acquire the zone mutex if we are unable to free to the * current cache; when we re-acquire the critical section, we must * detect and handle migration if it has occurred. */ itemdomain = 0; #ifdef NUMA if ((uz_flags & UMA_ZONE_FIRSTTOUCH) != 0) itemdomain = _vm_phys_domain(pmap_kextract((vm_offset_t)item)); #endif critical_enter(); do { cache = &zone->uz_cpu[curcpu]; /* * Try to free into the allocbucket first to give LIFO * ordering for cache-hot datastructures. Spill over * into the freebucket if necessary. Alloc will swap * them if one runs dry. */ bucket = &cache->uc_allocbucket; #ifdef NUMA if ((uz_flags & UMA_ZONE_FIRSTTOUCH) != 0 && PCPU_GET(domain) != itemdomain) { bucket = &cache->uc_crossbucket; } else #endif if (bucket->ucb_cnt == bucket->ucb_entries && cache->uc_freebucket.ucb_cnt < cache->uc_freebucket.ucb_entries) cache_bucket_swap(&cache->uc_freebucket, &cache->uc_allocbucket); if (__predict_true(bucket->ucb_cnt < bucket->ucb_entries)) { cache_bucket_push(cache, bucket, item); critical_exit(); return; } } while (cache_free(zone, cache, udata, item, itemdomain)); critical_exit(); /* * If nothing else caught this, we'll just do an internal free. */ zfree_item: zone_free_item(zone, item, udata, SKIP_DTOR); } #ifdef NUMA /* * sort crossdomain free buckets to domain correct buckets and cache * them. */ static void zone_free_cross(uma_zone_t zone, uma_bucket_t bucket, void *udata) { struct uma_bucketlist fullbuckets; uma_zone_domain_t zdom; uma_bucket_t b; smr_seq_t seq; void *item; int domain; CTR3(KTR_UMA, "uma_zfree: zone %s(%p) draining cross bucket %p", zone->uz_name, zone, bucket); /* * It is possible for buckets to arrive here out of order so we fetch * the current smr seq rather than accepting the bucket's. */ seq = SMR_SEQ_INVALID; if ((zone->uz_flags & UMA_ZONE_SMR) != 0) seq = smr_advance(zone->uz_smr); /* * To avoid having ndomain * ndomain buckets for sorting we have a * lock on the current crossfree bucket. A full matrix with * per-domain locking could be used if necessary. */ STAILQ_INIT(&fullbuckets); ZONE_CROSS_LOCK(zone); while (bucket->ub_cnt > 0) { item = bucket->ub_bucket[bucket->ub_cnt - 1]; domain = _vm_phys_domain(pmap_kextract((vm_offset_t)item)); zdom = ZDOM_GET(zone, domain); if (zdom->uzd_cross == NULL) { zdom->uzd_cross = bucket_alloc(zone, udata, M_NOWAIT); if (zdom->uzd_cross == NULL) break; } b = zdom->uzd_cross; b->ub_bucket[b->ub_cnt++] = item; b->ub_seq = seq; if (b->ub_cnt == b->ub_entries) { STAILQ_INSERT_HEAD(&fullbuckets, b, ub_link); zdom->uzd_cross = NULL; } bucket->ub_cnt--; } ZONE_CROSS_UNLOCK(zone); if (bucket->ub_cnt == 0) bucket->ub_seq = SMR_SEQ_INVALID; bucket_free(zone, bucket, udata); while ((b = STAILQ_FIRST(&fullbuckets)) != NULL) { STAILQ_REMOVE_HEAD(&fullbuckets, ub_link); domain = _vm_phys_domain(pmap_kextract( (vm_offset_t)b->ub_bucket[0])); zone_put_bucket(zone, domain, b, udata, true); } } #endif static void zone_free_bucket(uma_zone_t zone, uma_bucket_t bucket, void *udata, int itemdomain, bool ws) { #ifdef NUMA /* * Buckets coming from the wrong domain will be entirely for the * only other domain on two domain systems. In this case we can * simply cache them. Otherwise we need to sort them back to * correct domains. */ if ((zone->uz_flags & UMA_ZONE_FIRSTTOUCH) != 0 && vm_ndomains > 2 && PCPU_GET(domain) != itemdomain) { zone_free_cross(zone, bucket, udata); return; } #endif /* * Attempt to save the bucket in the zone's domain bucket cache. */ CTR3(KTR_UMA, "uma_zfree: zone %s(%p) putting bucket %p on free list", zone->uz_name, zone, bucket); /* ub_cnt is pointing to the last free item */ if ((zone->uz_flags & UMA_ZONE_ROUNDROBIN) != 0) itemdomain = zone_domain_lowest(zone, itemdomain); zone_put_bucket(zone, itemdomain, bucket, udata, ws); } /* * Populate a free or cross bucket for the current cpu cache. Free any * existing full bucket either to the zone cache or back to the slab layer. * * Enters and returns in a critical section. false return indicates that * we can not satisfy this free in the cache layer. true indicates that * the caller should retry. */ static __noinline bool cache_free(uma_zone_t zone, uma_cache_t cache, void *udata, void *item, int itemdomain) { uma_cache_bucket_t cbucket; uma_bucket_t newbucket, bucket; CRITICAL_ASSERT(curthread); if (zone->uz_bucket_size == 0) return false; cache = &zone->uz_cpu[curcpu]; newbucket = NULL; /* * FIRSTTOUCH domains need to free to the correct zdom. When * enabled this is the zdom of the item. The bucket is the * cross bucket if the current domain and itemdomain do not match. */ cbucket = &cache->uc_freebucket; #ifdef NUMA if ((cache_uz_flags(cache) & UMA_ZONE_FIRSTTOUCH) != 0) { if (PCPU_GET(domain) != itemdomain) { cbucket = &cache->uc_crossbucket; if (cbucket->ucb_cnt != 0) counter_u64_add(zone->uz_xdomain, cbucket->ucb_cnt); } } #endif bucket = cache_bucket_unload(cbucket); KASSERT(bucket == NULL || bucket->ub_cnt == bucket->ub_entries, ("cache_free: Entered with non-full free bucket.")); /* We are no longer associated with this CPU. */ critical_exit(); /* * Don't let SMR zones operate without a free bucket. Force * a synchronize and re-use this one. We will only degrade * to a synchronize every bucket_size items rather than every * item if we fail to allocate a bucket. */ if ((zone->uz_flags & UMA_ZONE_SMR) != 0) { if (bucket != NULL) bucket->ub_seq = smr_advance(zone->uz_smr); newbucket = bucket_alloc(zone, udata, M_NOWAIT); if (newbucket == NULL && bucket != NULL) { bucket_drain(zone, bucket); newbucket = bucket; bucket = NULL; } } else if (!bucketdisable) newbucket = bucket_alloc(zone, udata, M_NOWAIT); if (bucket != NULL) zone_free_bucket(zone, bucket, udata, itemdomain, true); critical_enter(); if ((bucket = newbucket) == NULL) return (false); cache = &zone->uz_cpu[curcpu]; #ifdef NUMA /* * Check to see if we should be populating the cross bucket. If it * is already populated we will fall through and attempt to populate * the free bucket. */ if ((cache_uz_flags(cache) & UMA_ZONE_FIRSTTOUCH) != 0) { if (PCPU_GET(domain) != itemdomain && cache->uc_crossbucket.ucb_bucket == NULL) { cache_bucket_load_cross(cache, bucket); return (true); } } #endif /* * We may have lost the race to fill the bucket or switched CPUs. */ if (cache->uc_freebucket.ucb_bucket != NULL) { critical_exit(); bucket_free(zone, bucket, udata); critical_enter(); } else cache_bucket_load_free(cache, bucket); return (true); } void uma_zfree_domain(uma_zone_t zone, void *item, void *udata) { /* Enable entropy collection for RANDOM_ENABLE_UMA kernel option */ random_harvest_fast_uma(&zone, sizeof(zone), RANDOM_UMA); CTR2(KTR_UMA, "uma_zfree_domain zone %s(%p)", zone->uz_name, zone); KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(), ("uma_zfree_domain: called with spinlock or critical section held")); /* uma_zfree(..., NULL) does nothing, to match free(9). */ if (item == NULL) return; zone_free_item(zone, item, udata, SKIP_NONE); } static void slab_free_item(uma_zone_t zone, uma_slab_t slab, void *item) { uma_keg_t keg; uma_domain_t dom; int freei; keg = zone->uz_keg; KEG_LOCK_ASSERT(keg, slab->us_domain); /* Do we need to remove from any lists? */ dom = &keg->uk_domain[slab->us_domain]; if (slab->us_freecount + 1 == keg->uk_ipers) { LIST_REMOVE(slab, us_link); LIST_INSERT_HEAD(&dom->ud_free_slab, slab, us_link); dom->ud_free_slabs++; } else if (slab->us_freecount == 0) { LIST_REMOVE(slab, us_link); LIST_INSERT_HEAD(&dom->ud_part_slab, slab, us_link); } /* Slab management. */ freei = slab_item_index(slab, keg, item); BIT_SET(keg->uk_ipers, freei, &slab->us_free); slab->us_freecount++; /* Keg statistics. */ dom->ud_free_items++; } static void zone_release(void *arg, void **bucket, int cnt) { struct mtx *lock; uma_zone_t zone; uma_slab_t slab; uma_keg_t keg; uint8_t *mem; void *item; int i; zone = arg; keg = zone->uz_keg; lock = NULL; if (__predict_false((zone->uz_flags & UMA_ZFLAG_HASH) != 0)) lock = KEG_LOCK(keg, 0); for (i = 0; i < cnt; i++) { item = bucket[i]; if (__predict_true((zone->uz_flags & UMA_ZFLAG_VTOSLAB) != 0)) { slab = vtoslab((vm_offset_t)item); } else { mem = (uint8_t *)((uintptr_t)item & (~UMA_SLAB_MASK)); if ((zone->uz_flags & UMA_ZFLAG_HASH) != 0) slab = hash_sfind(&keg->uk_hash, mem); else slab = (uma_slab_t)(mem + keg->uk_pgoff); } if (lock != KEG_LOCKPTR(keg, slab->us_domain)) { if (lock != NULL) mtx_unlock(lock); lock = KEG_LOCK(keg, slab->us_domain); } slab_free_item(zone, slab, item); } if (lock != NULL) mtx_unlock(lock); } /* * Frees a single item to any zone. * * Arguments: * zone The zone to free to * item The item we're freeing * udata User supplied data for the dtor * skip Skip dtors and finis */ static __noinline void zone_free_item(uma_zone_t zone, void *item, void *udata, enum zfreeskip skip) { /* * If a free is sent directly to an SMR zone we have to * synchronize immediately because the item can instantly * be reallocated. This should only happen in degenerate * cases when no memory is available for per-cpu caches. */ if ((zone->uz_flags & UMA_ZONE_SMR) != 0 && skip == SKIP_NONE) smr_synchronize(zone->uz_smr); item_dtor(zone, item, zone->uz_size, udata, skip); if (skip < SKIP_FINI && zone->uz_fini) zone->uz_fini(item, zone->uz_size); zone->uz_release(zone->uz_arg, &item, 1); if (skip & SKIP_CNT) return; counter_u64_add(zone->uz_frees, 1); if (zone->uz_max_items > 0) zone_free_limit(zone, 1); } /* See uma.h */ int uma_zone_set_max(uma_zone_t zone, int nitems) { struct uma_bucket_zone *ubz; int count; /* * XXX This can misbehave if the zone has any allocations with * no limit and a limit is imposed. There is currently no * way to clear a limit. */ ZONE_LOCK(zone); ubz = bucket_zone_max(zone, nitems); count = ubz != NULL ? ubz->ubz_entries : 0; zone->uz_bucket_size_max = zone->uz_bucket_size = count; if (zone->uz_bucket_size_min > zone->uz_bucket_size_max) zone->uz_bucket_size_min = zone->uz_bucket_size_max; zone->uz_max_items = nitems; zone->uz_flags |= UMA_ZFLAG_LIMIT; zone_update_caches(zone); /* We may need to wake waiters. */ wakeup(&zone->uz_max_items); ZONE_UNLOCK(zone); return (nitems); } /* See uma.h */ void uma_zone_set_maxcache(uma_zone_t zone, int nitems) { struct uma_bucket_zone *ubz; int bpcpu; ZONE_LOCK(zone); ubz = bucket_zone_max(zone, nitems); if (ubz != NULL) { bpcpu = 2; if ((zone->uz_flags & UMA_ZONE_FIRSTTOUCH) != 0) /* Count the cross-domain bucket. */ bpcpu++; nitems -= ubz->ubz_entries * bpcpu * mp_ncpus; zone->uz_bucket_size_max = ubz->ubz_entries; } else { zone->uz_bucket_size_max = zone->uz_bucket_size = 0; } if (zone->uz_bucket_size_min > zone->uz_bucket_size_max) zone->uz_bucket_size_min = zone->uz_bucket_size_max; zone->uz_bucket_max = nitems / vm_ndomains; ZONE_UNLOCK(zone); } /* See uma.h */ int uma_zone_get_max(uma_zone_t zone) { int nitems; nitems = atomic_load_64(&zone->uz_max_items); return (nitems); } /* See uma.h */ void uma_zone_set_warning(uma_zone_t zone, const char *warning) { ZONE_ASSERT_COLD(zone); zone->uz_warning = warning; } /* See uma.h */ void uma_zone_set_maxaction(uma_zone_t zone, uma_maxaction_t maxaction) { ZONE_ASSERT_COLD(zone); TASK_INIT(&zone->uz_maxaction, 0, (task_fn_t *)maxaction, zone); } /* See uma.h */ int uma_zone_get_cur(uma_zone_t zone) { int64_t nitems; u_int i; nitems = 0; if (zone->uz_allocs != EARLY_COUNTER && zone->uz_frees != EARLY_COUNTER) nitems = counter_u64_fetch(zone->uz_allocs) - counter_u64_fetch(zone->uz_frees); CPU_FOREACH(i) nitems += atomic_load_64(&zone->uz_cpu[i].uc_allocs) - atomic_load_64(&zone->uz_cpu[i].uc_frees); return (nitems < 0 ? 0 : nitems); } static uint64_t uma_zone_get_allocs(uma_zone_t zone) { uint64_t nitems; u_int i; nitems = 0; if (zone->uz_allocs != EARLY_COUNTER) nitems = counter_u64_fetch(zone->uz_allocs); CPU_FOREACH(i) nitems += atomic_load_64(&zone->uz_cpu[i].uc_allocs); return (nitems); } static uint64_t uma_zone_get_frees(uma_zone_t zone) { uint64_t nitems; u_int i; nitems = 0; if (zone->uz_frees != EARLY_COUNTER) nitems = counter_u64_fetch(zone->uz_frees); CPU_FOREACH(i) nitems += atomic_load_64(&zone->uz_cpu[i].uc_frees); return (nitems); } #ifdef INVARIANTS /* Used only for KEG_ASSERT_COLD(). */ static uint64_t uma_keg_get_allocs(uma_keg_t keg) { uma_zone_t z; uint64_t nitems; nitems = 0; LIST_FOREACH(z, &keg->uk_zones, uz_link) nitems += uma_zone_get_allocs(z); return (nitems); } #endif /* See uma.h */ void uma_zone_set_init(uma_zone_t zone, uma_init uminit) { uma_keg_t keg; KEG_GET(zone, keg); KEG_ASSERT_COLD(keg); keg->uk_init = uminit; } /* See uma.h */ void uma_zone_set_fini(uma_zone_t zone, uma_fini fini) { uma_keg_t keg; KEG_GET(zone, keg); KEG_ASSERT_COLD(keg); keg->uk_fini = fini; } /* See uma.h */ void uma_zone_set_zinit(uma_zone_t zone, uma_init zinit) { ZONE_ASSERT_COLD(zone); zone->uz_init = zinit; } /* See uma.h */ void uma_zone_set_zfini(uma_zone_t zone, uma_fini zfini) { ZONE_ASSERT_COLD(zone); zone->uz_fini = zfini; } /* See uma.h */ void uma_zone_set_freef(uma_zone_t zone, uma_free freef) { uma_keg_t keg; KEG_GET(zone, keg); KEG_ASSERT_COLD(keg); keg->uk_freef = freef; } /* See uma.h */ void uma_zone_set_allocf(uma_zone_t zone, uma_alloc allocf) { uma_keg_t keg; KEG_GET(zone, keg); KEG_ASSERT_COLD(keg); keg->uk_allocf = allocf; } /* See uma.h */ void uma_zone_set_smr(uma_zone_t zone, smr_t smr) { ZONE_ASSERT_COLD(zone); KASSERT(smr != NULL, ("Got NULL smr")); KASSERT((zone->uz_flags & UMA_ZONE_SMR) == 0, ("zone %p (%s) already uses SMR", zone, zone->uz_name)); zone->uz_flags |= UMA_ZONE_SMR; zone->uz_smr = smr; zone_update_caches(zone); } smr_t uma_zone_get_smr(uma_zone_t zone) { return (zone->uz_smr); } /* See uma.h */ void uma_zone_reserve(uma_zone_t zone, int items) { uma_keg_t keg; KEG_GET(zone, keg); KEG_ASSERT_COLD(keg); keg->uk_reserve = items; } /* See uma.h */ int uma_zone_reserve_kva(uma_zone_t zone, int count) { uma_keg_t keg; vm_offset_t kva; u_int pages; KEG_GET(zone, keg); KEG_ASSERT_COLD(keg); ZONE_ASSERT_COLD(zone); pages = howmany(count, keg->uk_ipers) * keg->uk_ppera; #ifdef UMA_MD_SMALL_ALLOC if (keg->uk_ppera > 1) { #else if (1) { #endif kva = kva_alloc((vm_size_t)pages * PAGE_SIZE); if (kva == 0) return (0); } else kva = 0; MPASS(keg->uk_kva == 0); keg->uk_kva = kva; keg->uk_offset = 0; zone->uz_max_items = pages * keg->uk_ipers; #ifdef UMA_MD_SMALL_ALLOC keg->uk_allocf = (keg->uk_ppera > 1) ? noobj_alloc : uma_small_alloc; #else keg->uk_allocf = noobj_alloc; #endif keg->uk_flags |= UMA_ZFLAG_LIMIT | UMA_ZONE_NOFREE; zone->uz_flags |= UMA_ZFLAG_LIMIT | UMA_ZONE_NOFREE; zone_update_caches(zone); return (1); } /* See uma.h */ void uma_prealloc(uma_zone_t zone, int items) { struct vm_domainset_iter di; uma_domain_t dom; uma_slab_t slab; uma_keg_t keg; int aflags, domain, slabs; KEG_GET(zone, keg); slabs = howmany(items, keg->uk_ipers); while (slabs-- > 0) { aflags = M_NOWAIT; vm_domainset_iter_policy_ref_init(&di, &keg->uk_dr, &domain, &aflags); for (;;) { slab = keg_alloc_slab(keg, zone, domain, M_WAITOK, aflags); if (slab != NULL) { dom = &keg->uk_domain[slab->us_domain]; /* * keg_alloc_slab() always returns a slab on the * partial list. */ LIST_REMOVE(slab, us_link); LIST_INSERT_HEAD(&dom->ud_free_slab, slab, us_link); dom->ud_free_slabs++; KEG_UNLOCK(keg, slab->us_domain); break; } if (vm_domainset_iter_policy(&di, &domain) != 0) vm_wait_doms(&keg->uk_dr.dr_policy->ds_mask); } } } /* * Returns a snapshot of memory consumption in bytes. */ size_t uma_zone_memory(uma_zone_t zone) { size_t sz; int i; sz = 0; if (zone->uz_flags & UMA_ZFLAG_CACHE) { for (i = 0; i < vm_ndomains; i++) sz += ZDOM_GET(zone, i)->uzd_nitems; return (sz * zone->uz_size); } for (i = 0; i < vm_ndomains; i++) sz += zone->uz_keg->uk_domain[i].ud_pages; return (sz * PAGE_SIZE); } /* See uma.h */ void uma_reclaim(int req) { CTR0(KTR_UMA, "UMA: vm asked us to release pages!"); sx_xlock(&uma_reclaim_lock); bucket_enable(); switch (req) { case UMA_RECLAIM_TRIM: zone_foreach(zone_trim, NULL); break; case UMA_RECLAIM_DRAIN: case UMA_RECLAIM_DRAIN_CPU: zone_foreach(zone_drain, NULL); if (req == UMA_RECLAIM_DRAIN_CPU) { pcpu_cache_drain_safe(NULL); zone_foreach(zone_drain, NULL); } break; default: panic("unhandled reclamation request %d", req); } /* * Some slabs may have been freed but this zone will be visited early * we visit again so that we can free pages that are empty once other * zones are drained. We have to do the same for buckets. */ zone_drain(slabzones[0], NULL); zone_drain(slabzones[1], NULL); bucket_zone_drain(); sx_xunlock(&uma_reclaim_lock); } static volatile int uma_reclaim_needed; void uma_reclaim_wakeup(void) { if (atomic_fetchadd_int(&uma_reclaim_needed, 1) == 0) wakeup(uma_reclaim); } void uma_reclaim_worker(void *arg __unused) { for (;;) { sx_xlock(&uma_reclaim_lock); while (atomic_load_int(&uma_reclaim_needed) == 0) sx_sleep(uma_reclaim, &uma_reclaim_lock, PVM, "umarcl", hz); sx_xunlock(&uma_reclaim_lock); EVENTHANDLER_INVOKE(vm_lowmem, VM_LOW_KMEM); uma_reclaim(UMA_RECLAIM_DRAIN_CPU); atomic_store_int(&uma_reclaim_needed, 0); /* Don't fire more than once per-second. */ pause("umarclslp", hz); } } /* See uma.h */ void uma_zone_reclaim(uma_zone_t zone, int req) { switch (req) { case UMA_RECLAIM_TRIM: zone_trim(zone, NULL); break; case UMA_RECLAIM_DRAIN: zone_drain(zone, NULL); break; case UMA_RECLAIM_DRAIN_CPU: pcpu_cache_drain_safe(zone); zone_drain(zone, NULL); break; default: panic("unhandled reclamation request %d", req); } } /* See uma.h */ int uma_zone_exhausted(uma_zone_t zone) { return (atomic_load_32(&zone->uz_sleepers) > 0); } unsigned long uma_limit(void) { return (uma_kmem_limit); } void uma_set_limit(unsigned long limit) { uma_kmem_limit = limit; } unsigned long uma_size(void) { return (atomic_load_long(&uma_kmem_total)); } long uma_avail(void) { return (uma_kmem_limit - uma_size()); } #ifdef DDB /* * Generate statistics across both the zone and its per-cpu cache's. Return * desired statistics if the pointer is non-NULL for that statistic. * * Note: does not update the zone statistics, as it can't safely clear the * per-CPU cache statistic. * */ static void uma_zone_sumstat(uma_zone_t z, long *cachefreep, uint64_t *allocsp, uint64_t *freesp, uint64_t *sleepsp, uint64_t *xdomainp) { uma_cache_t cache; uint64_t allocs, frees, sleeps, xdomain; int cachefree, cpu; allocs = frees = sleeps = xdomain = 0; cachefree = 0; CPU_FOREACH(cpu) { cache = &z->uz_cpu[cpu]; cachefree += cache->uc_allocbucket.ucb_cnt; cachefree += cache->uc_freebucket.ucb_cnt; xdomain += cache->uc_crossbucket.ucb_cnt; cachefree += cache->uc_crossbucket.ucb_cnt; allocs += cache->uc_allocs; frees += cache->uc_frees; } allocs += counter_u64_fetch(z->uz_allocs); frees += counter_u64_fetch(z->uz_frees); xdomain += counter_u64_fetch(z->uz_xdomain); sleeps += z->uz_sleeps; if (cachefreep != NULL) *cachefreep = cachefree; if (allocsp != NULL) *allocsp = allocs; if (freesp != NULL) *freesp = frees; if (sleepsp != NULL) *sleepsp = sleeps; if (xdomainp != NULL) *xdomainp = xdomain; } #endif /* DDB */ static int sysctl_vm_zone_count(SYSCTL_HANDLER_ARGS) { uma_keg_t kz; uma_zone_t z; int count; count = 0; rw_rlock(&uma_rwlock); LIST_FOREACH(kz, &uma_kegs, uk_link) { LIST_FOREACH(z, &kz->uk_zones, uz_link) count++; } LIST_FOREACH(z, &uma_cachezones, uz_link) count++; rw_runlock(&uma_rwlock); return (sysctl_handle_int(oidp, &count, 0, req)); } static void uma_vm_zone_stats(struct uma_type_header *uth, uma_zone_t z, struct sbuf *sbuf, struct uma_percpu_stat *ups, bool internal) { uma_zone_domain_t zdom; uma_cache_t cache; int i; for (i = 0; i < vm_ndomains; i++) { zdom = ZDOM_GET(z, i); uth->uth_zone_free += zdom->uzd_nitems; } uth->uth_allocs = counter_u64_fetch(z->uz_allocs); uth->uth_frees = counter_u64_fetch(z->uz_frees); uth->uth_fails = counter_u64_fetch(z->uz_fails); uth->uth_xdomain = counter_u64_fetch(z->uz_xdomain); uth->uth_sleeps = z->uz_sleeps; for (i = 0; i < mp_maxid + 1; i++) { bzero(&ups[i], sizeof(*ups)); if (internal || CPU_ABSENT(i)) continue; cache = &z->uz_cpu[i]; ups[i].ups_cache_free += cache->uc_allocbucket.ucb_cnt; ups[i].ups_cache_free += cache->uc_freebucket.ucb_cnt; ups[i].ups_cache_free += cache->uc_crossbucket.ucb_cnt; ups[i].ups_allocs = cache->uc_allocs; ups[i].ups_frees = cache->uc_frees; } } static int sysctl_vm_zone_stats(SYSCTL_HANDLER_ARGS) { struct uma_stream_header ush; struct uma_type_header uth; struct uma_percpu_stat *ups; struct sbuf sbuf; uma_keg_t kz; uma_zone_t z; uint64_t items; uint32_t kfree, pages; int count, error, i; error = sysctl_wire_old_buffer(req, 0); if (error != 0) return (error); sbuf_new_for_sysctl(&sbuf, NULL, 128, req); sbuf_clear_flags(&sbuf, SBUF_INCLUDENUL); ups = malloc((mp_maxid + 1) * sizeof(*ups), M_TEMP, M_WAITOK); count = 0; rw_rlock(&uma_rwlock); LIST_FOREACH(kz, &uma_kegs, uk_link) { LIST_FOREACH(z, &kz->uk_zones, uz_link) count++; } LIST_FOREACH(z, &uma_cachezones, uz_link) count++; /* * Insert stream header. */ bzero(&ush, sizeof(ush)); ush.ush_version = UMA_STREAM_VERSION; ush.ush_maxcpus = (mp_maxid + 1); ush.ush_count = count; (void)sbuf_bcat(&sbuf, &ush, sizeof(ush)); LIST_FOREACH(kz, &uma_kegs, uk_link) { kfree = pages = 0; for (i = 0; i < vm_ndomains; i++) { kfree += kz->uk_domain[i].ud_free_items; pages += kz->uk_domain[i].ud_pages; } LIST_FOREACH(z, &kz->uk_zones, uz_link) { bzero(&uth, sizeof(uth)); strlcpy(uth.uth_name, z->uz_name, UTH_MAX_NAME); uth.uth_align = kz->uk_align; uth.uth_size = kz->uk_size; uth.uth_rsize = kz->uk_rsize; if (z->uz_max_items > 0) { items = UZ_ITEMS_COUNT(z->uz_items); uth.uth_pages = (items / kz->uk_ipers) * kz->uk_ppera; } else uth.uth_pages = pages; uth.uth_maxpages = (z->uz_max_items / kz->uk_ipers) * kz->uk_ppera; uth.uth_limit = z->uz_max_items; uth.uth_keg_free = kfree; /* * A zone is secondary is it is not the first entry * on the keg's zone list. */ if ((z->uz_flags & UMA_ZONE_SECONDARY) && (LIST_FIRST(&kz->uk_zones) != z)) uth.uth_zone_flags = UTH_ZONE_SECONDARY; uma_vm_zone_stats(&uth, z, &sbuf, ups, kz->uk_flags & UMA_ZFLAG_INTERNAL); (void)sbuf_bcat(&sbuf, &uth, sizeof(uth)); for (i = 0; i < mp_maxid + 1; i++) (void)sbuf_bcat(&sbuf, &ups[i], sizeof(ups[i])); } } LIST_FOREACH(z, &uma_cachezones, uz_link) { bzero(&uth, sizeof(uth)); strlcpy(uth.uth_name, z->uz_name, UTH_MAX_NAME); uth.uth_size = z->uz_size; uma_vm_zone_stats(&uth, z, &sbuf, ups, false); (void)sbuf_bcat(&sbuf, &uth, sizeof(uth)); for (i = 0; i < mp_maxid + 1; i++) (void)sbuf_bcat(&sbuf, &ups[i], sizeof(ups[i])); } rw_runlock(&uma_rwlock); error = sbuf_finish(&sbuf); sbuf_delete(&sbuf); free(ups, M_TEMP); return (error); } int sysctl_handle_uma_zone_max(SYSCTL_HANDLER_ARGS) { uma_zone_t zone = *(uma_zone_t *)arg1; int error, max; max = uma_zone_get_max(zone); error = sysctl_handle_int(oidp, &max, 0, req); if (error || !req->newptr) return (error); uma_zone_set_max(zone, max); return (0); } int sysctl_handle_uma_zone_cur(SYSCTL_HANDLER_ARGS) { uma_zone_t zone; int cur; /* * Some callers want to add sysctls for global zones that * may not yet exist so they pass a pointer to a pointer. */ if (arg2 == 0) zone = *(uma_zone_t *)arg1; else zone = arg1; cur = uma_zone_get_cur(zone); return (sysctl_handle_int(oidp, &cur, 0, req)); } static int sysctl_handle_uma_zone_allocs(SYSCTL_HANDLER_ARGS) { uma_zone_t zone = arg1; uint64_t cur; cur = uma_zone_get_allocs(zone); return (sysctl_handle_64(oidp, &cur, 0, req)); } static int sysctl_handle_uma_zone_frees(SYSCTL_HANDLER_ARGS) { uma_zone_t zone = arg1; uint64_t cur; cur = uma_zone_get_frees(zone); return (sysctl_handle_64(oidp, &cur, 0, req)); } static int sysctl_handle_uma_zone_flags(SYSCTL_HANDLER_ARGS) { struct sbuf sbuf; uma_zone_t zone = arg1; int error; sbuf_new_for_sysctl(&sbuf, NULL, 0, req); if (zone->uz_flags != 0) sbuf_printf(&sbuf, "0x%b", zone->uz_flags, PRINT_UMA_ZFLAGS); else sbuf_printf(&sbuf, "0"); error = sbuf_finish(&sbuf); sbuf_delete(&sbuf); return (error); } static int sysctl_handle_uma_slab_efficiency(SYSCTL_HANDLER_ARGS) { uma_keg_t keg = arg1; int avail, effpct, total; total = keg->uk_ppera * PAGE_SIZE; if ((keg->uk_flags & UMA_ZFLAG_OFFPAGE) != 0) total += slabzone(keg->uk_ipers)->uz_keg->uk_rsize; /* * We consider the client's requested size and alignment here, not the * real size determination uk_rsize, because we also adjust the real * size for internal implementation reasons (max bitset size). */ avail = keg->uk_ipers * roundup2(keg->uk_size, keg->uk_align + 1); if ((keg->uk_flags & UMA_ZONE_PCPU) != 0) avail *= mp_maxid + 1; effpct = 100 * avail / total; return (sysctl_handle_int(oidp, &effpct, 0, req)); } static int sysctl_handle_uma_zone_items(SYSCTL_HANDLER_ARGS) { uma_zone_t zone = arg1; uint64_t cur; cur = UZ_ITEMS_COUNT(atomic_load_64(&zone->uz_items)); return (sysctl_handle_64(oidp, &cur, 0, req)); } #ifdef INVARIANTS static uma_slab_t uma_dbg_getslab(uma_zone_t zone, void *item) { uma_slab_t slab; uma_keg_t keg; uint8_t *mem; /* * It is safe to return the slab here even though the * zone is unlocked because the item's allocation state * essentially holds a reference. */ mem = (uint8_t *)((uintptr_t)item & (~UMA_SLAB_MASK)); if ((zone->uz_flags & UMA_ZFLAG_CACHE) != 0) return (NULL); if (zone->uz_flags & UMA_ZFLAG_VTOSLAB) return (vtoslab((vm_offset_t)mem)); keg = zone->uz_keg; if ((keg->uk_flags & UMA_ZFLAG_HASH) == 0) return ((uma_slab_t)(mem + keg->uk_pgoff)); KEG_LOCK(keg, 0); slab = hash_sfind(&keg->uk_hash, mem); KEG_UNLOCK(keg, 0); return (slab); } static bool uma_dbg_zskip(uma_zone_t zone, void *mem) { if ((zone->uz_flags & UMA_ZFLAG_CACHE) != 0) return (true); return (uma_dbg_kskip(zone->uz_keg, mem)); } static bool uma_dbg_kskip(uma_keg_t keg, void *mem) { uintptr_t idx; if (dbg_divisor == 0) return (true); if (dbg_divisor == 1) return (false); idx = (uintptr_t)mem >> PAGE_SHIFT; if (keg->uk_ipers > 1) { idx *= keg->uk_ipers; idx += ((uintptr_t)mem & PAGE_MASK) / keg->uk_rsize; } if ((idx / dbg_divisor) * dbg_divisor != idx) { counter_u64_add(uma_skip_cnt, 1); return (true); } counter_u64_add(uma_dbg_cnt, 1); return (false); } /* * Set up the slab's freei data such that uma_dbg_free can function. * */ static void uma_dbg_alloc(uma_zone_t zone, uma_slab_t slab, void *item) { uma_keg_t keg; int freei; if (slab == NULL) { slab = uma_dbg_getslab(zone, item); if (slab == NULL) panic("uma: item %p did not belong to zone %s\n", item, zone->uz_name); } keg = zone->uz_keg; freei = slab_item_index(slab, keg, item); if (BIT_ISSET(keg->uk_ipers, freei, slab_dbg_bits(slab, keg))) panic("Duplicate alloc of %p from zone %p(%s) slab %p(%d)\n", item, zone, zone->uz_name, slab, freei); BIT_SET_ATOMIC(keg->uk_ipers, freei, slab_dbg_bits(slab, keg)); } /* * Verifies freed addresses. Checks for alignment, valid slab membership * and duplicate frees. * */ static void uma_dbg_free(uma_zone_t zone, uma_slab_t slab, void *item) { uma_keg_t keg; int freei; if (slab == NULL) { slab = uma_dbg_getslab(zone, item); if (slab == NULL) panic("uma: Freed item %p did not belong to zone %s\n", item, zone->uz_name); } keg = zone->uz_keg; freei = slab_item_index(slab, keg, item); if (freei >= keg->uk_ipers) panic("Invalid free of %p from zone %p(%s) slab %p(%d)\n", item, zone, zone->uz_name, slab, freei); if (slab_item(slab, keg, freei) != item) panic("Unaligned free of %p from zone %p(%s) slab %p(%d)\n", item, zone, zone->uz_name, slab, freei); if (!BIT_ISSET(keg->uk_ipers, freei, slab_dbg_bits(slab, keg))) panic("Duplicate free of %p from zone %p(%s) slab %p(%d)\n", item, zone, zone->uz_name, slab, freei); BIT_CLR_ATOMIC(keg->uk_ipers, freei, slab_dbg_bits(slab, keg)); } #endif /* INVARIANTS */ #ifdef DDB static int64_t get_uma_stats(uma_keg_t kz, uma_zone_t z, uint64_t *allocs, uint64_t *used, uint64_t *sleeps, long *cachefree, uint64_t *xdomain) { uint64_t frees; int i; if (kz->uk_flags & UMA_ZFLAG_INTERNAL) { *allocs = counter_u64_fetch(z->uz_allocs); frees = counter_u64_fetch(z->uz_frees); *sleeps = z->uz_sleeps; *cachefree = 0; *xdomain = 0; } else uma_zone_sumstat(z, cachefree, allocs, &frees, sleeps, xdomain); for (i = 0; i < vm_ndomains; i++) { *cachefree += ZDOM_GET(z, i)->uzd_nitems; if (!((z->uz_flags & UMA_ZONE_SECONDARY) && (LIST_FIRST(&kz->uk_zones) != z))) *cachefree += kz->uk_domain[i].ud_free_items; } *used = *allocs - frees; return (((int64_t)*used + *cachefree) * kz->uk_size); } DB_SHOW_COMMAND(uma, db_show_uma) { const char *fmt_hdr, *fmt_entry; uma_keg_t kz; uma_zone_t z; uint64_t allocs, used, sleeps, xdomain; long cachefree; /* variables for sorting */ uma_keg_t cur_keg; uma_zone_t cur_zone, last_zone; int64_t cur_size, last_size, size; int ties; /* /i option produces machine-parseable CSV output */ if (modif[0] == 'i') { fmt_hdr = "%s,%s,%s,%s,%s,%s,%s,%s,%s\n"; fmt_entry = "\"%s\",%ju,%jd,%ld,%ju,%ju,%u,%jd,%ju\n"; } else { fmt_hdr = "%18s %6s %7s %7s %11s %7s %7s %10s %8s\n"; fmt_entry = "%18s %6ju %7jd %7ld %11ju %7ju %7u %10jd %8ju\n"; } db_printf(fmt_hdr, "Zone", "Size", "Used", "Free", "Requests", "Sleeps", "Bucket", "Total Mem", "XFree"); /* Sort the zones with largest size first. */ last_zone = NULL; last_size = INT64_MAX; for (;;) { cur_zone = NULL; cur_size = -1; ties = 0; LIST_FOREACH(kz, &uma_kegs, uk_link) { LIST_FOREACH(z, &kz->uk_zones, uz_link) { /* * In the case of size ties, print out zones * in the order they are encountered. That is, * when we encounter the most recently output * zone, we have already printed all preceding * ties, and we must print all following ties. */ if (z == last_zone) { ties = 1; continue; } size = get_uma_stats(kz, z, &allocs, &used, &sleeps, &cachefree, &xdomain); if (size > cur_size && size < last_size + ties) { cur_size = size; cur_zone = z; cur_keg = kz; } } } if (cur_zone == NULL) break; size = get_uma_stats(cur_keg, cur_zone, &allocs, &used, &sleeps, &cachefree, &xdomain); db_printf(fmt_entry, cur_zone->uz_name, (uintmax_t)cur_keg->uk_size, (intmax_t)used, cachefree, (uintmax_t)allocs, (uintmax_t)sleeps, (unsigned)cur_zone->uz_bucket_size, (intmax_t)size, xdomain); if (db_pager_quit) return; last_zone = cur_zone; last_size = cur_size; } } DB_SHOW_COMMAND(umacache, db_show_umacache) { uma_zone_t z; uint64_t allocs, frees; long cachefree; int i; db_printf("%18s %8s %8s %8s %12s %8s\n", "Zone", "Size", "Used", "Free", "Requests", "Bucket"); LIST_FOREACH(z, &uma_cachezones, uz_link) { uma_zone_sumstat(z, &cachefree, &allocs, &frees, NULL, NULL); for (i = 0; i < vm_ndomains; i++) cachefree += ZDOM_GET(z, i)->uzd_nitems; db_printf("%18s %8ju %8jd %8ld %12ju %8u\n", z->uz_name, (uintmax_t)z->uz_size, (intmax_t)(allocs - frees), cachefree, (uintmax_t)allocs, z->uz_bucket_size); if (db_pager_quit) return; } } #endif /* DDB */ diff --git a/sys/vm/vm_object.c b/sys/vm/vm_object.c index b974a1f9e88a..a823c55d72ef 100644 --- a/sys/vm/vm_object.c +++ b/sys/vm/vm_object.c @@ -1,2849 +1,2839 @@ /*- * SPDX-License-Identifier: (BSD-3-Clause AND MIT-CMU) * * Copyright (c) 1991, 1993 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed to Berkeley by * The Mach Operating System project at Carnegie-Mellon University. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * from: @(#)vm_object.c 8.5 (Berkeley) 3/22/94 * * * Copyright (c) 1987, 1990 Carnegie-Mellon University. * All rights reserved. * * Authors: Avadis Tevanian, Jr., Michael Wayne Young * * Permission to use, copy, modify and distribute this software and * its documentation is hereby granted, provided that both the copyright * notice and this permission notice appear in all copies of the * software, derivative works or modified versions, and any portions * thereof, and that both notices appear in supporting documentation. * * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. * * Carnegie Mellon requests users of this software to return to * * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU * School of Computer Science * Carnegie Mellon University * Pittsburgh PA 15213-3890 * * any improvements or extensions that they make and grant Carnegie the * rights to redistribute these changes. */ /* * Virtual memory object module. */ #include __FBSDID("$FreeBSD$"); #include "opt_vm.h" #include #include #include #include #include #include #include #include #include #include #include #include /* for curproc, pageproc */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include static int old_msync; SYSCTL_INT(_vm, OID_AUTO, old_msync, CTLFLAG_RW, &old_msync, 0, "Use old (insecure) msync behavior"); static int vm_object_page_collect_flush(vm_object_t object, vm_page_t p, int pagerflags, int flags, boolean_t *allclean, boolean_t *eio); static boolean_t vm_object_page_remove_write(vm_page_t p, int flags, boolean_t *allclean); static void vm_object_backing_remove(vm_object_t object); /* * Virtual memory objects maintain the actual data * associated with allocated virtual memory. A given * page of memory exists within exactly one object. * * An object is only deallocated when all "references" * are given up. Only one "reference" to a given * region of an object should be writeable. * * Associated with each object is a list of all resident * memory pages belonging to that object; this list is * maintained by the "vm_page" module, and locked by the object's * lock. * * Each object also records a "pager" routine which is * used to retrieve (and store) pages to the proper backing * storage. In addition, objects may be backed by other * objects from which they were virtual-copied. * * The only items within the object structure which are * modified after time of creation are: * reference count locked by object's lock * pager routine locked by object's lock * */ struct object_q vm_object_list; struct mtx vm_object_list_mtx; /* lock for object list and count */ struct vm_object kernel_object_store; static SYSCTL_NODE(_vm_stats, OID_AUTO, object, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "VM object stats"); -static counter_u64_t object_collapses = EARLY_COUNTER; +static COUNTER_U64_DEFINE_EARLY(object_collapses); SYSCTL_COUNTER_U64(_vm_stats_object, OID_AUTO, collapses, CTLFLAG_RD, &object_collapses, "VM object collapses"); -static counter_u64_t object_bypasses = EARLY_COUNTER; +static COUNTER_U64_DEFINE_EARLY(object_bypasses); SYSCTL_COUNTER_U64(_vm_stats_object, OID_AUTO, bypasses, CTLFLAG_RD, &object_bypasses, "VM object bypasses"); -static counter_u64_t object_collapse_waits = EARLY_COUNTER; +static COUNTER_U64_DEFINE_EARLY(object_collapse_waits); SYSCTL_COUNTER_U64(_vm_stats_object, OID_AUTO, collapse_waits, CTLFLAG_RD, &object_collapse_waits, "Number of sleeps for collapse"); -static void -counter_startup(void) -{ - - object_collapses = counter_u64_alloc(M_WAITOK); - object_bypasses = counter_u64_alloc(M_WAITOK); - object_collapse_waits = counter_u64_alloc(M_WAITOK); -} -SYSINIT(object_counters, SI_SUB_CPU, SI_ORDER_ANY, counter_startup, NULL); - static uma_zone_t obj_zone; static int vm_object_zinit(void *mem, int size, int flags); #ifdef INVARIANTS static void vm_object_zdtor(void *mem, int size, void *arg); static void vm_object_zdtor(void *mem, int size, void *arg) { vm_object_t object; object = (vm_object_t)mem; KASSERT(object->ref_count == 0, ("object %p ref_count = %d", object, object->ref_count)); KASSERT(TAILQ_EMPTY(&object->memq), ("object %p has resident pages in its memq", object)); KASSERT(vm_radix_is_empty(&object->rtree), ("object %p has resident pages in its trie", object)); #if VM_NRESERVLEVEL > 0 KASSERT(LIST_EMPTY(&object->rvq), ("object %p has reservations", object)); #endif KASSERT(blockcount_read(&object->paging_in_progress) == 0, ("object %p paging_in_progress = %d", object, blockcount_read(&object->paging_in_progress))); KASSERT(!vm_object_busied(object), ("object %p busy = %d", object, blockcount_read(&object->busy))); KASSERT(object->resident_page_count == 0, ("object %p resident_page_count = %d", object, object->resident_page_count)); KASSERT(object->shadow_count == 0, ("object %p shadow_count = %d", object, object->shadow_count)); KASSERT(object->type == OBJT_DEAD, ("object %p has non-dead type %d", object, object->type)); } #endif static int vm_object_zinit(void *mem, int size, int flags) { vm_object_t object; object = (vm_object_t)mem; rw_init_flags(&object->lock, "vm object", RW_DUPOK | RW_NEW); /* These are true for any object that has been freed */ object->type = OBJT_DEAD; vm_radix_init(&object->rtree); refcount_init(&object->ref_count, 0); blockcount_init(&object->paging_in_progress); blockcount_init(&object->busy); object->resident_page_count = 0; object->shadow_count = 0; object->flags = OBJ_DEAD; mtx_lock(&vm_object_list_mtx); TAILQ_INSERT_TAIL(&vm_object_list, object, object_list); mtx_unlock(&vm_object_list_mtx); return (0); } static void _vm_object_allocate(objtype_t type, vm_pindex_t size, u_short flags, vm_object_t object, void *handle) { TAILQ_INIT(&object->memq); LIST_INIT(&object->shadow_head); object->type = type; if (type == OBJT_SWAP) pctrie_init(&object->un_pager.swp.swp_blks); /* * Ensure that swap_pager_swapoff() iteration over object_list * sees up to date type and pctrie head if it observed * non-dead object. */ atomic_thread_fence_rel(); object->pg_color = 0; object->flags = flags; object->size = size; object->domain.dr_policy = NULL; object->generation = 1; object->cleangeneration = 1; refcount_init(&object->ref_count, 1); object->memattr = VM_MEMATTR_DEFAULT; object->cred = NULL; object->charge = 0; object->handle = handle; object->backing_object = NULL; object->backing_object_offset = (vm_ooffset_t) 0; #if VM_NRESERVLEVEL > 0 LIST_INIT(&object->rvq); #endif umtx_shm_object_init(object); } /* * vm_object_init: * * Initialize the VM objects module. */ void vm_object_init(void) { TAILQ_INIT(&vm_object_list); mtx_init(&vm_object_list_mtx, "vm object_list", NULL, MTX_DEF); rw_init(&kernel_object->lock, "kernel vm object"); _vm_object_allocate(OBJT_PHYS, atop(VM_MAX_KERNEL_ADDRESS - VM_MIN_KERNEL_ADDRESS), OBJ_UNMANAGED, kernel_object, NULL); #if VM_NRESERVLEVEL > 0 kernel_object->flags |= OBJ_COLORED; kernel_object->pg_color = (u_short)atop(VM_MIN_KERNEL_ADDRESS); #endif /* * The lock portion of struct vm_object must be type stable due * to vm_pageout_fallback_object_lock locking a vm object * without holding any references to it. */ obj_zone = uma_zcreate("VM OBJECT", sizeof (struct vm_object), NULL, #ifdef INVARIANTS vm_object_zdtor, #else NULL, #endif vm_object_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); vm_radix_zinit(); } void vm_object_clear_flag(vm_object_t object, u_short bits) { VM_OBJECT_ASSERT_WLOCKED(object); object->flags &= ~bits; } /* * Sets the default memory attribute for the specified object. Pages * that are allocated to this object are by default assigned this memory * attribute. * * Presently, this function must be called before any pages are allocated * to the object. In the future, this requirement may be relaxed for * "default" and "swap" objects. */ int vm_object_set_memattr(vm_object_t object, vm_memattr_t memattr) { VM_OBJECT_ASSERT_WLOCKED(object); switch (object->type) { case OBJT_DEFAULT: case OBJT_DEVICE: case OBJT_MGTDEVICE: case OBJT_PHYS: case OBJT_SG: case OBJT_SWAP: case OBJT_VNODE: if (!TAILQ_EMPTY(&object->memq)) return (KERN_FAILURE); break; case OBJT_DEAD: return (KERN_INVALID_ARGUMENT); default: panic("vm_object_set_memattr: object %p is of undefined type", object); } object->memattr = memattr; return (KERN_SUCCESS); } void vm_object_pip_add(vm_object_t object, short i) { if (i > 0) blockcount_acquire(&object->paging_in_progress, i); } void vm_object_pip_wakeup(vm_object_t object) { vm_object_pip_wakeupn(object, 1); } void vm_object_pip_wakeupn(vm_object_t object, short i) { if (i > 0) blockcount_release(&object->paging_in_progress, i); } /* * Atomically drop the object lock and wait for pip to drain. This protects * from sleep/wakeup races due to identity changes. The lock is not re-acquired * on return. */ static void vm_object_pip_sleep(vm_object_t object, const char *waitid) { (void)blockcount_sleep(&object->paging_in_progress, &object->lock, waitid, PVM | PDROP); } void vm_object_pip_wait(vm_object_t object, const char *waitid) { VM_OBJECT_ASSERT_WLOCKED(object); blockcount_wait(&object->paging_in_progress, &object->lock, waitid, PVM); } void vm_object_pip_wait_unlocked(vm_object_t object, const char *waitid) { VM_OBJECT_ASSERT_UNLOCKED(object); blockcount_wait(&object->paging_in_progress, NULL, waitid, PVM); } /* * vm_object_allocate: * * Returns a new object with the given size. */ vm_object_t vm_object_allocate(objtype_t type, vm_pindex_t size) { vm_object_t object; u_short flags; switch (type) { case OBJT_DEAD: panic("vm_object_allocate: can't create OBJT_DEAD"); case OBJT_DEFAULT: case OBJT_SWAP: flags = OBJ_COLORED; break; case OBJT_DEVICE: case OBJT_SG: flags = OBJ_FICTITIOUS | OBJ_UNMANAGED; break; case OBJT_MGTDEVICE: flags = OBJ_FICTITIOUS; break; case OBJT_PHYS: flags = OBJ_UNMANAGED; break; case OBJT_VNODE: flags = 0; break; default: panic("vm_object_allocate: type %d is undefined", type); } object = (vm_object_t)uma_zalloc(obj_zone, M_WAITOK); _vm_object_allocate(type, size, flags, object, NULL); return (object); } /* * vm_object_allocate_anon: * * Returns a new default object of the given size and marked as * anonymous memory for special split/collapse handling. Color * to be initialized by the caller. */ vm_object_t vm_object_allocate_anon(vm_pindex_t size, vm_object_t backing_object, struct ucred *cred, vm_size_t charge) { vm_object_t handle, object; if (backing_object == NULL) handle = NULL; else if ((backing_object->flags & OBJ_ANON) != 0) handle = backing_object->handle; else handle = backing_object; object = uma_zalloc(obj_zone, M_WAITOK); _vm_object_allocate(OBJT_DEFAULT, size, OBJ_ANON | OBJ_ONEMAPPING, object, handle); object->cred = cred; object->charge = cred != NULL ? charge : 0; return (object); } static void vm_object_reference_vnode(vm_object_t object) { u_int old; /* * vnode objects need the lock for the first reference * to serialize with vnode_object_deallocate(). */ if (!refcount_acquire_if_gt(&object->ref_count, 0)) { VM_OBJECT_RLOCK(object); old = refcount_acquire(&object->ref_count); if (object->type == OBJT_VNODE && old == 0) vref(object->handle); VM_OBJECT_RUNLOCK(object); } } /* * vm_object_reference: * * Acquires a reference to the given object. */ void vm_object_reference(vm_object_t object) { if (object == NULL) return; if (object->type == OBJT_VNODE) vm_object_reference_vnode(object); else refcount_acquire(&object->ref_count); KASSERT((object->flags & OBJ_DEAD) == 0, ("vm_object_reference: Referenced dead object.")); } /* * vm_object_reference_locked: * * Gets another reference to the given object. * * The object must be locked. */ void vm_object_reference_locked(vm_object_t object) { u_int old; VM_OBJECT_ASSERT_LOCKED(object); old = refcount_acquire(&object->ref_count); if (object->type == OBJT_VNODE && old == 0) vref(object->handle); KASSERT((object->flags & OBJ_DEAD) == 0, ("vm_object_reference: Referenced dead object.")); } /* * Handle deallocating an object of type OBJT_VNODE. */ static void vm_object_deallocate_vnode(vm_object_t object) { struct vnode *vp = (struct vnode *) object->handle; bool last; KASSERT(object->type == OBJT_VNODE, ("vm_object_deallocate_vnode: not a vnode object")); KASSERT(vp != NULL, ("vm_object_deallocate_vnode: missing vp")); /* Object lock to protect handle lookup. */ last = refcount_release(&object->ref_count); VM_OBJECT_RUNLOCK(object); if (!last) return; if (!umtx_shm_vnobj_persistent) umtx_shm_object_terminated(object); /* vrele may need the vnode lock. */ vrele(vp); } /* * We dropped a reference on an object and discovered that it had a * single remaining shadow. This is a sibling of the reference we * dropped. Attempt to collapse the sibling and backing object. */ static vm_object_t vm_object_deallocate_anon(vm_object_t backing_object) { vm_object_t object; /* Fetch the final shadow. */ object = LIST_FIRST(&backing_object->shadow_head); KASSERT(object != NULL && backing_object->shadow_count == 1, ("vm_object_anon_deallocate: ref_count: %d, shadow_count: %d", backing_object->ref_count, backing_object->shadow_count)); KASSERT((object->flags & (OBJ_TMPFS_NODE | OBJ_ANON)) == OBJ_ANON, ("invalid shadow object %p", object)); if (!VM_OBJECT_TRYWLOCK(object)) { /* * Prevent object from disappearing since we do not have a * reference. */ vm_object_pip_add(object, 1); VM_OBJECT_WUNLOCK(backing_object); VM_OBJECT_WLOCK(object); vm_object_pip_wakeup(object); } else VM_OBJECT_WUNLOCK(backing_object); /* * Check for a collapse/terminate race with the last reference holder. */ if ((object->flags & (OBJ_DEAD | OBJ_COLLAPSING)) != 0 || !refcount_acquire_if_not_zero(&object->ref_count)) { VM_OBJECT_WUNLOCK(object); return (NULL); } backing_object = object->backing_object; if (backing_object != NULL && (backing_object->flags & OBJ_ANON) != 0) vm_object_collapse(object); VM_OBJECT_WUNLOCK(object); return (object); } /* * vm_object_deallocate: * * Release a reference to the specified object, * gained either through a vm_object_allocate * or a vm_object_reference call. When all references * are gone, storage associated with this object * may be relinquished. * * No object may be locked. */ void vm_object_deallocate(vm_object_t object) { vm_object_t temp; bool released; while (object != NULL) { /* * If the reference count goes to 0 we start calling * vm_object_terminate() on the object chain. A ref count * of 1 may be a special case depending on the shadow count * being 0 or 1. These cases require a write lock on the * object. */ if ((object->flags & OBJ_ANON) == 0) released = refcount_release_if_gt(&object->ref_count, 1); else released = refcount_release_if_gt(&object->ref_count, 2); if (released) return; if (object->type == OBJT_VNODE) { VM_OBJECT_RLOCK(object); if (object->type == OBJT_VNODE) { vm_object_deallocate_vnode(object); return; } VM_OBJECT_RUNLOCK(object); } VM_OBJECT_WLOCK(object); KASSERT(object->ref_count > 0, ("vm_object_deallocate: object deallocated too many times: %d", object->type)); /* * If this is not the final reference to an anonymous * object we may need to collapse the shadow chain. */ if (!refcount_release(&object->ref_count)) { if (object->ref_count > 1 || object->shadow_count == 0) { if ((object->flags & OBJ_ANON) != 0 && object->ref_count == 1) vm_object_set_flag(object, OBJ_ONEMAPPING); VM_OBJECT_WUNLOCK(object); return; } /* Handle collapsing last ref on anonymous objects. */ object = vm_object_deallocate_anon(object); continue; } /* * Handle the final reference to an object. We restart * the loop with the backing object to avoid recursion. */ umtx_shm_object_terminated(object); temp = object->backing_object; if (temp != NULL) { KASSERT((object->flags & OBJ_TMPFS_NODE) == 0, ("shadowed tmpfs v_object 2 %p", object)); vm_object_backing_remove(object); } KASSERT((object->flags & OBJ_DEAD) == 0, ("vm_object_deallocate: Terminating dead object.")); vm_object_set_flag(object, OBJ_DEAD); vm_object_terminate(object); object = temp; } } /* * vm_object_destroy removes the object from the global object list * and frees the space for the object. */ void vm_object_destroy(vm_object_t object) { /* * Release the allocation charge. */ if (object->cred != NULL) { swap_release_by_cred(object->charge, object->cred); object->charge = 0; crfree(object->cred); object->cred = NULL; } /* * Free the space for the object. */ uma_zfree(obj_zone, object); } static void vm_object_backing_remove_locked(vm_object_t object) { vm_object_t backing_object; backing_object = object->backing_object; VM_OBJECT_ASSERT_WLOCKED(object); VM_OBJECT_ASSERT_WLOCKED(backing_object); KASSERT((object->flags & OBJ_COLLAPSING) == 0, ("vm_object_backing_remove: Removing collapsing object.")); if ((object->flags & OBJ_SHADOWLIST) != 0) { LIST_REMOVE(object, shadow_list); backing_object->shadow_count--; object->flags &= ~OBJ_SHADOWLIST; } object->backing_object = NULL; } static void vm_object_backing_remove(vm_object_t object) { vm_object_t backing_object; VM_OBJECT_ASSERT_WLOCKED(object); if ((object->flags & OBJ_SHADOWLIST) != 0) { backing_object = object->backing_object; VM_OBJECT_WLOCK(backing_object); vm_object_backing_remove_locked(object); VM_OBJECT_WUNLOCK(backing_object); } else object->backing_object = NULL; } static void vm_object_backing_insert_locked(vm_object_t object, vm_object_t backing_object) { VM_OBJECT_ASSERT_WLOCKED(object); if ((backing_object->flags & OBJ_ANON) != 0) { VM_OBJECT_ASSERT_WLOCKED(backing_object); LIST_INSERT_HEAD(&backing_object->shadow_head, object, shadow_list); backing_object->shadow_count++; object->flags |= OBJ_SHADOWLIST; } object->backing_object = backing_object; } static void vm_object_backing_insert(vm_object_t object, vm_object_t backing_object) { VM_OBJECT_ASSERT_WLOCKED(object); if ((backing_object->flags & OBJ_ANON) != 0) { VM_OBJECT_WLOCK(backing_object); vm_object_backing_insert_locked(object, backing_object); VM_OBJECT_WUNLOCK(backing_object); } else object->backing_object = backing_object; } /* * Insert an object into a backing_object's shadow list with an additional * reference to the backing_object added. */ static void vm_object_backing_insert_ref(vm_object_t object, vm_object_t backing_object) { VM_OBJECT_ASSERT_WLOCKED(object); if ((backing_object->flags & OBJ_ANON) != 0) { VM_OBJECT_WLOCK(backing_object); KASSERT((backing_object->flags & OBJ_DEAD) == 0, ("shadowing dead anonymous object")); vm_object_reference_locked(backing_object); vm_object_backing_insert_locked(object, backing_object); vm_object_clear_flag(backing_object, OBJ_ONEMAPPING); VM_OBJECT_WUNLOCK(backing_object); } else { vm_object_reference(backing_object); object->backing_object = backing_object; } } /* * Transfer a backing reference from backing_object to object. */ static void vm_object_backing_transfer(vm_object_t object, vm_object_t backing_object) { vm_object_t new_backing_object; /* * Note that the reference to backing_object->backing_object * moves from within backing_object to within object. */ vm_object_backing_remove_locked(object); new_backing_object = backing_object->backing_object; if (new_backing_object == NULL) return; if ((new_backing_object->flags & OBJ_ANON) != 0) { VM_OBJECT_WLOCK(new_backing_object); vm_object_backing_remove_locked(backing_object); vm_object_backing_insert_locked(object, new_backing_object); VM_OBJECT_WUNLOCK(new_backing_object); } else { object->backing_object = new_backing_object; backing_object->backing_object = NULL; } } /* * Wait for a concurrent collapse to settle. */ static void vm_object_collapse_wait(vm_object_t object) { VM_OBJECT_ASSERT_WLOCKED(object); while ((object->flags & OBJ_COLLAPSING) != 0) { vm_object_pip_wait(object, "vmcolwait"); counter_u64_add(object_collapse_waits, 1); } } /* * Waits for a backing object to clear a pending collapse and returns * it locked if it is an ANON object. */ static vm_object_t vm_object_backing_collapse_wait(vm_object_t object) { vm_object_t backing_object; VM_OBJECT_ASSERT_WLOCKED(object); for (;;) { backing_object = object->backing_object; if (backing_object == NULL || (backing_object->flags & OBJ_ANON) == 0) return (NULL); VM_OBJECT_WLOCK(backing_object); if ((backing_object->flags & (OBJ_DEAD | OBJ_COLLAPSING)) == 0) break; VM_OBJECT_WUNLOCK(object); vm_object_pip_sleep(backing_object, "vmbckwait"); counter_u64_add(object_collapse_waits, 1); VM_OBJECT_WLOCK(object); } return (backing_object); } /* * vm_object_terminate_pages removes any remaining pageable pages * from the object and resets the object to an empty state. */ static void vm_object_terminate_pages(vm_object_t object) { vm_page_t p, p_next; VM_OBJECT_ASSERT_WLOCKED(object); /* * Free any remaining pageable pages. This also removes them from the * paging queues. However, don't free wired pages, just remove them * from the object. Rather than incrementally removing each page from * the object, the page and object are reset to any empty state. */ TAILQ_FOREACH_SAFE(p, &object->memq, listq, p_next) { vm_page_assert_unbusied(p); KASSERT(p->object == object && (p->ref_count & VPRC_OBJREF) != 0, ("vm_object_terminate_pages: page %p is inconsistent", p)); p->object = NULL; if (vm_page_drop(p, VPRC_OBJREF) == VPRC_OBJREF) { VM_CNT_INC(v_pfree); vm_page_free(p); } } /* * If the object contained any pages, then reset it to an empty state. * None of the object's fields, including "resident_page_count", were * modified by the preceding loop. */ if (object->resident_page_count != 0) { vm_radix_reclaim_allnodes(&object->rtree); TAILQ_INIT(&object->memq); object->resident_page_count = 0; if (object->type == OBJT_VNODE) vdrop(object->handle); } } /* * vm_object_terminate actually destroys the specified object, freeing * up all previously used resources. * * The object must be locked. * This routine may block. */ void vm_object_terminate(vm_object_t object) { VM_OBJECT_ASSERT_WLOCKED(object); KASSERT((object->flags & OBJ_DEAD) != 0, ("terminating non-dead obj %p", object)); KASSERT((object->flags & OBJ_COLLAPSING) == 0, ("terminating collapsing obj %p", object)); KASSERT(object->backing_object == NULL, ("terminating shadow obj %p", object)); /* * wait for the pageout daemon to be done with the object */ vm_object_pip_wait(object, "objtrm"); KASSERT(!blockcount_read(&object->paging_in_progress), ("vm_object_terminate: pageout in progress")); KASSERT(object->ref_count == 0, ("vm_object_terminate: object with references, ref_count=%d", object->ref_count)); if ((object->flags & OBJ_PG_DTOR) == 0) vm_object_terminate_pages(object); #if VM_NRESERVLEVEL > 0 if (__predict_false(!LIST_EMPTY(&object->rvq))) vm_reserv_break_all(object); #endif KASSERT(object->cred == NULL || object->type == OBJT_DEFAULT || object->type == OBJT_SWAP, ("%s: non-swap obj %p has cred", __func__, object)); /* * Let the pager know object is dead. */ vm_pager_deallocate(object); VM_OBJECT_WUNLOCK(object); vm_object_destroy(object); } /* * Make the page read-only so that we can clear the object flags. However, if * this is a nosync mmap then the object is likely to stay dirty so do not * mess with the page and do not clear the object flags. Returns TRUE if the * page should be flushed, and FALSE otherwise. */ static boolean_t vm_object_page_remove_write(vm_page_t p, int flags, boolean_t *allclean) { vm_page_assert_busied(p); /* * If we have been asked to skip nosync pages and this is a * nosync page, skip it. Note that the object flags were not * cleared in this case so we do not have to set them. */ if ((flags & OBJPC_NOSYNC) != 0 && (p->a.flags & PGA_NOSYNC) != 0) { *allclean = FALSE; return (FALSE); } else { pmap_remove_write(p); return (p->dirty != 0); } } /* * vm_object_page_clean * * Clean all dirty pages in the specified range of object. Leaves page * on whatever queue it is currently on. If NOSYNC is set then do not * write out pages with PGA_NOSYNC set (originally comes from MAP_NOSYNC), * leaving the object dirty. * * For swap objects backing tmpfs regular files, do not flush anything, * but remove write protection on the mapped pages to update mtime through * mmaped writes. * * When stuffing pages asynchronously, allow clustering. XXX we need a * synchronous clustering mode implementation. * * Odd semantics: if start == end, we clean everything. * * The object must be locked. * * Returns FALSE if some page from the range was not written, as * reported by the pager, and TRUE otherwise. */ boolean_t vm_object_page_clean(vm_object_t object, vm_ooffset_t start, vm_ooffset_t end, int flags) { vm_page_t np, p; vm_pindex_t pi, tend, tstart; int curgeneration, n, pagerflags; boolean_t eio, res, allclean; VM_OBJECT_ASSERT_WLOCKED(object); if (!vm_object_mightbedirty(object) || object->resident_page_count == 0) return (TRUE); pagerflags = (flags & (OBJPC_SYNC | OBJPC_INVAL)) != 0 ? VM_PAGER_PUT_SYNC : VM_PAGER_CLUSTER_OK; pagerflags |= (flags & OBJPC_INVAL) != 0 ? VM_PAGER_PUT_INVAL : 0; tstart = OFF_TO_IDX(start); tend = (end == 0) ? object->size : OFF_TO_IDX(end + PAGE_MASK); allclean = tstart == 0 && tend >= object->size; res = TRUE; rescan: curgeneration = object->generation; for (p = vm_page_find_least(object, tstart); p != NULL; p = np) { pi = p->pindex; if (pi >= tend) break; np = TAILQ_NEXT(p, listq); if (vm_page_none_valid(p)) continue; if (vm_page_busy_acquire(p, VM_ALLOC_WAITFAIL) == 0) { if (object->generation != curgeneration && (flags & OBJPC_SYNC) != 0) goto rescan; np = vm_page_find_least(object, pi); continue; } if (!vm_object_page_remove_write(p, flags, &allclean)) { vm_page_xunbusy(p); continue; } if (object->type == OBJT_VNODE) { n = vm_object_page_collect_flush(object, p, pagerflags, flags, &allclean, &eio); if (eio) { res = FALSE; allclean = FALSE; } if (object->generation != curgeneration && (flags & OBJPC_SYNC) != 0) goto rescan; /* * If the VOP_PUTPAGES() did a truncated write, so * that even the first page of the run is not fully * written, vm_pageout_flush() returns 0 as the run * length. Since the condition that caused truncated * write may be permanent, e.g. exhausted free space, * accepting n == 0 would cause an infinite loop. * * Forwarding the iterator leaves the unwritten page * behind, but there is not much we can do there if * filesystem refuses to write it. */ if (n == 0) { n = 1; allclean = FALSE; } } else { n = 1; vm_page_xunbusy(p); } np = vm_page_find_least(object, pi + n); } #if 0 VOP_FSYNC(vp, (pagerflags & VM_PAGER_PUT_SYNC) ? MNT_WAIT : 0); #endif /* * Leave updating cleangeneration for tmpfs objects to tmpfs * scan. It needs to update mtime, which happens for other * filesystems during page writeouts. */ if (allclean && object->type == OBJT_VNODE) object->cleangeneration = curgeneration; return (res); } static int vm_object_page_collect_flush(vm_object_t object, vm_page_t p, int pagerflags, int flags, boolean_t *allclean, boolean_t *eio) { vm_page_t ma[vm_pageout_page_count], p_first, tp; int count, i, mreq, runlen; vm_page_lock_assert(p, MA_NOTOWNED); vm_page_assert_xbusied(p); VM_OBJECT_ASSERT_WLOCKED(object); count = 1; mreq = 0; for (tp = p; count < vm_pageout_page_count; count++) { tp = vm_page_next(tp); if (tp == NULL || vm_page_tryxbusy(tp) == 0) break; if (!vm_object_page_remove_write(tp, flags, allclean)) { vm_page_xunbusy(tp); break; } } for (p_first = p; count < vm_pageout_page_count; count++) { tp = vm_page_prev(p_first); if (tp == NULL || vm_page_tryxbusy(tp) == 0) break; if (!vm_object_page_remove_write(tp, flags, allclean)) { vm_page_xunbusy(tp); break; } p_first = tp; mreq++; } for (tp = p_first, i = 0; i < count; tp = TAILQ_NEXT(tp, listq), i++) ma[i] = tp; vm_pageout_flush(ma, count, pagerflags, mreq, &runlen, eio); return (runlen); } /* * Note that there is absolutely no sense in writing out * anonymous objects, so we track down the vnode object * to write out. * We invalidate (remove) all pages from the address space * for semantic correctness. * * If the backing object is a device object with unmanaged pages, then any * mappings to the specified range of pages must be removed before this * function is called. * * Note: certain anonymous maps, such as MAP_NOSYNC maps, * may start out with a NULL object. */ boolean_t vm_object_sync(vm_object_t object, vm_ooffset_t offset, vm_size_t size, boolean_t syncio, boolean_t invalidate) { vm_object_t backing_object; struct vnode *vp; struct mount *mp; int error, flags, fsync_after; boolean_t res; if (object == NULL) return (TRUE); res = TRUE; error = 0; VM_OBJECT_WLOCK(object); while ((backing_object = object->backing_object) != NULL) { VM_OBJECT_WLOCK(backing_object); offset += object->backing_object_offset; VM_OBJECT_WUNLOCK(object); object = backing_object; if (object->size < OFF_TO_IDX(offset + size)) size = IDX_TO_OFF(object->size) - offset; } /* * Flush pages if writing is allowed, invalidate them * if invalidation requested. Pages undergoing I/O * will be ignored by vm_object_page_remove(). * * We cannot lock the vnode and then wait for paging * to complete without deadlocking against vm_fault. * Instead we simply call vm_object_page_remove() and * allow it to block internally on a page-by-page * basis when it encounters pages undergoing async * I/O. */ if (object->type == OBJT_VNODE && vm_object_mightbedirty(object) != 0 && ((vp = object->handle)->v_vflag & VV_NOSYNC) == 0) { VM_OBJECT_WUNLOCK(object); (void) vn_start_write(vp, &mp, V_WAIT); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); if (syncio && !invalidate && offset == 0 && atop(size) == object->size) { /* * If syncing the whole mapping of the file, * it is faster to schedule all the writes in * async mode, also allowing the clustering, * and then wait for i/o to complete. */ flags = 0; fsync_after = TRUE; } else { flags = (syncio || invalidate) ? OBJPC_SYNC : 0; flags |= invalidate ? (OBJPC_SYNC | OBJPC_INVAL) : 0; fsync_after = FALSE; } VM_OBJECT_WLOCK(object); res = vm_object_page_clean(object, offset, offset + size, flags); VM_OBJECT_WUNLOCK(object); if (fsync_after) error = VOP_FSYNC(vp, MNT_WAIT, curthread); VOP_UNLOCK(vp); vn_finished_write(mp); if (error != 0) res = FALSE; VM_OBJECT_WLOCK(object); } if ((object->type == OBJT_VNODE || object->type == OBJT_DEVICE) && invalidate) { if (object->type == OBJT_DEVICE) /* * The option OBJPR_NOTMAPPED must be passed here * because vm_object_page_remove() cannot remove * unmanaged mappings. */ flags = OBJPR_NOTMAPPED; else if (old_msync) flags = 0; else flags = OBJPR_CLEANONLY; vm_object_page_remove(object, OFF_TO_IDX(offset), OFF_TO_IDX(offset + size + PAGE_MASK), flags); } VM_OBJECT_WUNLOCK(object); return (res); } /* * Determine whether the given advice can be applied to the object. Advice is * not applied to unmanaged pages since they never belong to page queues, and * since MADV_FREE is destructive, it can apply only to anonymous pages that * have been mapped at most once. */ static bool vm_object_advice_applies(vm_object_t object, int advice) { if ((object->flags & OBJ_UNMANAGED) != 0) return (false); if (advice != MADV_FREE) return (true); return ((object->flags & (OBJ_ONEMAPPING | OBJ_ANON)) == (OBJ_ONEMAPPING | OBJ_ANON)); } static void vm_object_madvise_freespace(vm_object_t object, int advice, vm_pindex_t pindex, vm_size_t size) { if (advice == MADV_FREE && object->type == OBJT_SWAP) swap_pager_freespace(object, pindex, size); } /* * vm_object_madvise: * * Implements the madvise function at the object/page level. * * MADV_WILLNEED (any object) * * Activate the specified pages if they are resident. * * MADV_DONTNEED (any object) * * Deactivate the specified pages if they are resident. * * MADV_FREE (OBJT_DEFAULT/OBJT_SWAP objects, * OBJ_ONEMAPPING only) * * Deactivate and clean the specified pages if they are * resident. This permits the process to reuse the pages * without faulting or the kernel to reclaim the pages * without I/O. */ void vm_object_madvise(vm_object_t object, vm_pindex_t pindex, vm_pindex_t end, int advice) { vm_pindex_t tpindex; vm_object_t backing_object, tobject; vm_page_t m, tm; if (object == NULL) return; relookup: VM_OBJECT_WLOCK(object); if (!vm_object_advice_applies(object, advice)) { VM_OBJECT_WUNLOCK(object); return; } for (m = vm_page_find_least(object, pindex); pindex < end; pindex++) { tobject = object; /* * If the next page isn't resident in the top-level object, we * need to search the shadow chain. When applying MADV_FREE, we * take care to release any swap space used to store * non-resident pages. */ if (m == NULL || pindex < m->pindex) { /* * Optimize a common case: if the top-level object has * no backing object, we can skip over the non-resident * range in constant time. */ if (object->backing_object == NULL) { tpindex = (m != NULL && m->pindex < end) ? m->pindex : end; vm_object_madvise_freespace(object, advice, pindex, tpindex - pindex); if ((pindex = tpindex) == end) break; goto next_page; } tpindex = pindex; do { vm_object_madvise_freespace(tobject, advice, tpindex, 1); /* * Prepare to search the next object in the * chain. */ backing_object = tobject->backing_object; if (backing_object == NULL) goto next_pindex; VM_OBJECT_WLOCK(backing_object); tpindex += OFF_TO_IDX(tobject->backing_object_offset); if (tobject != object) VM_OBJECT_WUNLOCK(tobject); tobject = backing_object; if (!vm_object_advice_applies(tobject, advice)) goto next_pindex; } while ((tm = vm_page_lookup(tobject, tpindex)) == NULL); } else { next_page: tm = m; m = TAILQ_NEXT(m, listq); } /* * If the page is not in a normal state, skip it. The page * can not be invalidated while the object lock is held. */ if (!vm_page_all_valid(tm) || vm_page_wired(tm)) goto next_pindex; KASSERT((tm->flags & PG_FICTITIOUS) == 0, ("vm_object_madvise: page %p is fictitious", tm)); KASSERT((tm->oflags & VPO_UNMANAGED) == 0, ("vm_object_madvise: page %p is not managed", tm)); if (vm_page_tryxbusy(tm) == 0) { if (object != tobject) VM_OBJECT_WUNLOCK(object); if (advice == MADV_WILLNEED) { /* * Reference the page before unlocking and * sleeping so that the page daemon is less * likely to reclaim it. */ vm_page_aflag_set(tm, PGA_REFERENCED); } vm_page_busy_sleep(tm, "madvpo", false); goto relookup; } vm_page_advise(tm, advice); vm_page_xunbusy(tm); vm_object_madvise_freespace(tobject, advice, tm->pindex, 1); next_pindex: if (tobject != object) VM_OBJECT_WUNLOCK(tobject); } VM_OBJECT_WUNLOCK(object); } /* * vm_object_shadow: * * Create a new object which is backed by the * specified existing object range. The source * object reference is deallocated. * * The new object and offset into that object * are returned in the source parameters. */ void vm_object_shadow(vm_object_t *object, vm_ooffset_t *offset, vm_size_t length, struct ucred *cred, bool shared) { vm_object_t source; vm_object_t result; source = *object; /* * Don't create the new object if the old object isn't shared. * * If we hold the only reference we can guarantee that it won't * increase while we have the map locked. Otherwise the race is * harmless and we will end up with an extra shadow object that * will be collapsed later. */ if (source != NULL && source->ref_count == 1 && (source->flags & OBJ_ANON) != 0) return; /* * Allocate a new object with the given length. */ result = vm_object_allocate_anon(atop(length), source, cred, length); /* * Store the offset into the source object, and fix up the offset into * the new object. */ result->backing_object_offset = *offset; if (shared || source != NULL) { VM_OBJECT_WLOCK(result); /* * The new object shadows the source object, adding a * reference to it. Our caller changes his reference * to point to the new object, removing a reference to * the source object. Net result: no change of * reference count, unless the caller needs to add one * more reference due to forking a shared map entry. */ if (shared) { vm_object_reference_locked(result); vm_object_clear_flag(result, OBJ_ONEMAPPING); } /* * Try to optimize the result object's page color when * shadowing in order to maintain page coloring * consistency in the combined shadowed object. */ if (source != NULL) { vm_object_backing_insert(result, source); result->domain = source->domain; #if VM_NRESERVLEVEL > 0 result->flags |= source->flags & OBJ_COLORED; result->pg_color = (source->pg_color + OFF_TO_IDX(*offset)) & ((1 << (VM_NFREEORDER - 1)) - 1); #endif } VM_OBJECT_WUNLOCK(result); } /* * Return the new things */ *offset = 0; *object = result; } /* * vm_object_split: * * Split the pages in a map entry into a new object. This affords * easier removal of unused pages, and keeps object inheritance from * being a negative impact on memory usage. */ void vm_object_split(vm_map_entry_t entry) { vm_page_t m, m_next; vm_object_t orig_object, new_object, backing_object; vm_pindex_t idx, offidxstart; vm_size_t size; orig_object = entry->object.vm_object; KASSERT((orig_object->flags & OBJ_ONEMAPPING) != 0, ("vm_object_split: Splitting object with multiple mappings.")); if ((orig_object->flags & OBJ_ANON) == 0) return; if (orig_object->ref_count <= 1) return; VM_OBJECT_WUNLOCK(orig_object); offidxstart = OFF_TO_IDX(entry->offset); size = atop(entry->end - entry->start); /* * If swap_pager_copy() is later called, it will convert new_object * into a swap object. */ new_object = vm_object_allocate_anon(size, orig_object, orig_object->cred, ptoa(size)); /* * We must wait for the orig_object to complete any in-progress * collapse so that the swap blocks are stable below. The * additional reference on backing_object by new object will * prevent further collapse operations until split completes. */ VM_OBJECT_WLOCK(orig_object); vm_object_collapse_wait(orig_object); /* * At this point, the new object is still private, so the order in * which the original and new objects are locked does not matter. */ VM_OBJECT_WLOCK(new_object); new_object->domain = orig_object->domain; backing_object = orig_object->backing_object; if (backing_object != NULL) { vm_object_backing_insert_ref(new_object, backing_object); new_object->backing_object_offset = orig_object->backing_object_offset + entry->offset; } if (orig_object->cred != NULL) { crhold(orig_object->cred); KASSERT(orig_object->charge >= ptoa(size), ("orig_object->charge < 0")); orig_object->charge -= ptoa(size); } /* * Mark the split operation so that swap_pager_getpages() knows * that the object is in transition. */ vm_object_set_flag(orig_object, OBJ_SPLIT); retry: m = vm_page_find_least(orig_object, offidxstart); for (; m != NULL && (idx = m->pindex - offidxstart) < size; m = m_next) { m_next = TAILQ_NEXT(m, listq); /* * We must wait for pending I/O to complete before we can * rename the page. * * We do not have to VM_PROT_NONE the page as mappings should * not be changed by this operation. */ if (vm_page_tryxbusy(m) == 0) { VM_OBJECT_WUNLOCK(new_object); vm_page_sleep_if_busy(m, "spltwt"); VM_OBJECT_WLOCK(new_object); goto retry; } /* * The page was left invalid. Likely placed there by * an incomplete fault. Just remove and ignore. */ if (vm_page_none_valid(m)) { if (vm_page_remove(m)) vm_page_free(m); continue; } /* vm_page_rename() will dirty the page. */ if (vm_page_rename(m, new_object, idx)) { vm_page_xunbusy(m); VM_OBJECT_WUNLOCK(new_object); VM_OBJECT_WUNLOCK(orig_object); vm_radix_wait(); VM_OBJECT_WLOCK(orig_object); VM_OBJECT_WLOCK(new_object); goto retry; } #if VM_NRESERVLEVEL > 0 /* * If some of the reservation's allocated pages remain with * the original object, then transferring the reservation to * the new object is neither particularly beneficial nor * particularly harmful as compared to leaving the reservation * with the original object. If, however, all of the * reservation's allocated pages are transferred to the new * object, then transferring the reservation is typically * beneficial. Determining which of these two cases applies * would be more costly than unconditionally renaming the * reservation. */ vm_reserv_rename(m, new_object, orig_object, offidxstart); #endif if (orig_object->type != OBJT_SWAP) vm_page_xunbusy(m); } if (orig_object->type == OBJT_SWAP) { /* * swap_pager_copy() can sleep, in which case the orig_object's * and new_object's locks are released and reacquired. */ swap_pager_copy(orig_object, new_object, offidxstart, 0); TAILQ_FOREACH(m, &new_object->memq, listq) vm_page_xunbusy(m); } vm_object_clear_flag(orig_object, OBJ_SPLIT); VM_OBJECT_WUNLOCK(orig_object); VM_OBJECT_WUNLOCK(new_object); entry->object.vm_object = new_object; entry->offset = 0LL; vm_object_deallocate(orig_object); VM_OBJECT_WLOCK(new_object); } static vm_page_t vm_object_collapse_scan_wait(vm_object_t object, vm_page_t p) { vm_object_t backing_object; VM_OBJECT_ASSERT_WLOCKED(object); backing_object = object->backing_object; VM_OBJECT_ASSERT_WLOCKED(backing_object); KASSERT(p == NULL || p->object == object || p->object == backing_object, ("invalid ownership %p %p %p", p, object, backing_object)); /* The page is only NULL when rename fails. */ if (p == NULL) { VM_OBJECT_WUNLOCK(object); VM_OBJECT_WUNLOCK(backing_object); vm_radix_wait(); } else { if (p->object == object) VM_OBJECT_WUNLOCK(backing_object); else VM_OBJECT_WUNLOCK(object); vm_page_busy_sleep(p, "vmocol", false); } VM_OBJECT_WLOCK(object); VM_OBJECT_WLOCK(backing_object); return (TAILQ_FIRST(&backing_object->memq)); } static bool vm_object_scan_all_shadowed(vm_object_t object) { vm_object_t backing_object; vm_page_t p, pp; vm_pindex_t backing_offset_index, new_pindex, pi, ps; VM_OBJECT_ASSERT_WLOCKED(object); VM_OBJECT_ASSERT_WLOCKED(object->backing_object); backing_object = object->backing_object; if ((backing_object->flags & OBJ_ANON) == 0) return (false); pi = backing_offset_index = OFF_TO_IDX(object->backing_object_offset); p = vm_page_find_least(backing_object, pi); ps = swap_pager_find_least(backing_object, pi); /* * Only check pages inside the parent object's range and * inside the parent object's mapping of the backing object. */ for (;; pi++) { if (p != NULL && p->pindex < pi) p = TAILQ_NEXT(p, listq); if (ps < pi) ps = swap_pager_find_least(backing_object, pi); if (p == NULL && ps >= backing_object->size) break; else if (p == NULL) pi = ps; else pi = MIN(p->pindex, ps); new_pindex = pi - backing_offset_index; if (new_pindex >= object->size) break; if (p != NULL) { /* * If the backing object page is busy a * grandparent or older page may still be * undergoing CoW. It is not safe to collapse * the backing object until it is quiesced. */ if (vm_page_tryxbusy(p) == 0) return (false); /* * We raced with the fault handler that left * newly allocated invalid page on the object * queue and retried. */ if (!vm_page_all_valid(p)) goto unbusy_ret; } /* * See if the parent has the page or if the parent's object * pager has the page. If the parent has the page but the page * is not valid, the parent's object pager must have the page. * * If this fails, the parent does not completely shadow the * object and we might as well give up now. */ pp = vm_page_lookup(object, new_pindex); /* * The valid check here is stable due to object lock * being required to clear valid and initiate paging. * Busy of p disallows fault handler to validate pp. */ if ((pp == NULL || vm_page_none_valid(pp)) && !vm_pager_has_page(object, new_pindex, NULL, NULL)) goto unbusy_ret; if (p != NULL) vm_page_xunbusy(p); } return (true); unbusy_ret: if (p != NULL) vm_page_xunbusy(p); return (false); } static void vm_object_collapse_scan(vm_object_t object) { vm_object_t backing_object; vm_page_t next, p, pp; vm_pindex_t backing_offset_index, new_pindex; VM_OBJECT_ASSERT_WLOCKED(object); VM_OBJECT_ASSERT_WLOCKED(object->backing_object); backing_object = object->backing_object; backing_offset_index = OFF_TO_IDX(object->backing_object_offset); /* * Our scan */ for (p = TAILQ_FIRST(&backing_object->memq); p != NULL; p = next) { next = TAILQ_NEXT(p, listq); new_pindex = p->pindex - backing_offset_index; /* * Check for busy page */ if (vm_page_tryxbusy(p) == 0) { next = vm_object_collapse_scan_wait(object, p); continue; } KASSERT(object->backing_object == backing_object, ("vm_object_collapse_scan: backing object mismatch %p != %p", object->backing_object, backing_object)); KASSERT(p->object == backing_object, ("vm_object_collapse_scan: object mismatch %p != %p", p->object, backing_object)); if (p->pindex < backing_offset_index || new_pindex >= object->size) { if (backing_object->type == OBJT_SWAP) swap_pager_freespace(backing_object, p->pindex, 1); KASSERT(!pmap_page_is_mapped(p), ("freeing mapped page %p", p)); if (vm_page_remove(p)) vm_page_free(p); continue; } if (!vm_page_all_valid(p)) { KASSERT(!pmap_page_is_mapped(p), ("freeing mapped page %p", p)); if (vm_page_remove(p)) vm_page_free(p); continue; } pp = vm_page_lookup(object, new_pindex); if (pp != NULL && vm_page_tryxbusy(pp) == 0) { vm_page_xunbusy(p); /* * The page in the parent is busy and possibly not * (yet) valid. Until its state is finalized by the * busy bit owner, we can't tell whether it shadows the * original page. */ next = vm_object_collapse_scan_wait(object, pp); continue; } if (pp != NULL && vm_page_none_valid(pp)) { /* * The page was invalid in the parent. Likely placed * there by an incomplete fault. Just remove and * ignore. p can replace it. */ if (vm_page_remove(pp)) vm_page_free(pp); pp = NULL; } if (pp != NULL || vm_pager_has_page(object, new_pindex, NULL, NULL)) { /* * The page already exists in the parent OR swap exists * for this location in the parent. Leave the parent's * page alone. Destroy the original page from the * backing object. */ if (backing_object->type == OBJT_SWAP) swap_pager_freespace(backing_object, p->pindex, 1); KASSERT(!pmap_page_is_mapped(p), ("freeing mapped page %p", p)); if (vm_page_remove(p)) vm_page_free(p); if (pp != NULL) vm_page_xunbusy(pp); continue; } /* * Page does not exist in parent, rename the page from the * backing object to the main object. * * If the page was mapped to a process, it can remain mapped * through the rename. vm_page_rename() will dirty the page. */ if (vm_page_rename(p, object, new_pindex)) { vm_page_xunbusy(p); next = vm_object_collapse_scan_wait(object, NULL); continue; } /* Use the old pindex to free the right page. */ if (backing_object->type == OBJT_SWAP) swap_pager_freespace(backing_object, new_pindex + backing_offset_index, 1); #if VM_NRESERVLEVEL > 0 /* * Rename the reservation. */ vm_reserv_rename(p, object, backing_object, backing_offset_index); #endif vm_page_xunbusy(p); } return; } /* * vm_object_collapse: * * Collapse an object with the object backing it. * Pages in the backing object are moved into the * parent, and the backing object is deallocated. */ void vm_object_collapse(vm_object_t object) { vm_object_t backing_object, new_backing_object; VM_OBJECT_ASSERT_WLOCKED(object); while (TRUE) { KASSERT((object->flags & (OBJ_DEAD | OBJ_ANON)) == OBJ_ANON, ("collapsing invalid object")); /* * Wait for the backing_object to finish any pending * collapse so that the caller sees the shortest possible * shadow chain. */ backing_object = vm_object_backing_collapse_wait(object); if (backing_object == NULL) return; KASSERT(object->ref_count > 0 && object->ref_count > object->shadow_count, ("collapse with invalid ref %d or shadow %d count.", object->ref_count, object->shadow_count)); KASSERT((backing_object->flags & (OBJ_COLLAPSING | OBJ_DEAD)) == 0, ("vm_object_collapse: Backing object already collapsing.")); KASSERT((object->flags & (OBJ_COLLAPSING | OBJ_DEAD)) == 0, ("vm_object_collapse: object is already collapsing.")); /* * We know that we can either collapse the backing object if * the parent is the only reference to it, or (perhaps) have * the parent bypass the object if the parent happens to shadow * all the resident pages in the entire backing object. */ if (backing_object->ref_count == 1) { KASSERT(backing_object->shadow_count == 1, ("vm_object_collapse: shadow_count: %d", backing_object->shadow_count)); vm_object_pip_add(object, 1); vm_object_set_flag(object, OBJ_COLLAPSING); vm_object_pip_add(backing_object, 1); vm_object_set_flag(backing_object, OBJ_DEAD); /* * If there is exactly one reference to the backing * object, we can collapse it into the parent. */ vm_object_collapse_scan(object); #if VM_NRESERVLEVEL > 0 /* * Break any reservations from backing_object. */ if (__predict_false(!LIST_EMPTY(&backing_object->rvq))) vm_reserv_break_all(backing_object); #endif /* * Move the pager from backing_object to object. */ if (backing_object->type == OBJT_SWAP) { /* * swap_pager_copy() can sleep, in which case * the backing_object's and object's locks are * released and reacquired. * Since swap_pager_copy() is being asked to * destroy backing_object, it will change the * type to OBJT_DEFAULT. */ swap_pager_copy( backing_object, object, OFF_TO_IDX(object->backing_object_offset), TRUE); } /* * Object now shadows whatever backing_object did. */ vm_object_clear_flag(object, OBJ_COLLAPSING); vm_object_backing_transfer(object, backing_object); object->backing_object_offset += backing_object->backing_object_offset; VM_OBJECT_WUNLOCK(object); vm_object_pip_wakeup(object); /* * Discard backing_object. * * Since the backing object has no pages, no pager left, * and no object references within it, all that is * necessary is to dispose of it. */ KASSERT(backing_object->ref_count == 1, ( "backing_object %p was somehow re-referenced during collapse!", backing_object)); vm_object_pip_wakeup(backing_object); (void)refcount_release(&backing_object->ref_count); vm_object_terminate(backing_object); counter_u64_add(object_collapses, 1); VM_OBJECT_WLOCK(object); } else { /* * If we do not entirely shadow the backing object, * there is nothing we can do so we give up. * * The object lock and backing_object lock must not * be dropped during this sequence. */ if (!vm_object_scan_all_shadowed(object)) { VM_OBJECT_WUNLOCK(backing_object); break; } /* * Make the parent shadow the next object in the * chain. Deallocating backing_object will not remove * it, since its reference count is at least 2. */ vm_object_backing_remove_locked(object); new_backing_object = backing_object->backing_object; if (new_backing_object != NULL) { vm_object_backing_insert_ref(object, new_backing_object); object->backing_object_offset += backing_object->backing_object_offset; } /* * Drop the reference count on backing_object. Since * its ref_count was at least 2, it will not vanish. */ (void)refcount_release(&backing_object->ref_count); KASSERT(backing_object->ref_count >= 1, ( "backing_object %p was somehow dereferenced during collapse!", backing_object)); VM_OBJECT_WUNLOCK(backing_object); counter_u64_add(object_bypasses, 1); } /* * Try again with this object's new backing object. */ } } /* * vm_object_page_remove: * * For the given object, either frees or invalidates each of the * specified pages. In general, a page is freed. However, if a page is * wired for any reason other than the existence of a managed, wired * mapping, then it may be invalidated but not removed from the object. * Pages are specified by the given range ["start", "end") and the option * OBJPR_CLEANONLY. As a special case, if "end" is zero, then the range * extends from "start" to the end of the object. If the option * OBJPR_CLEANONLY is specified, then only the non-dirty pages within the * specified range are affected. If the option OBJPR_NOTMAPPED is * specified, then the pages within the specified range must have no * mappings. Otherwise, if this option is not specified, any mappings to * the specified pages are removed before the pages are freed or * invalidated. * * In general, this operation should only be performed on objects that * contain managed pages. There are, however, two exceptions. First, it * is performed on the kernel and kmem objects by vm_map_entry_delete(). * Second, it is used by msync(..., MS_INVALIDATE) to invalidate device- * backed pages. In both of these cases, the option OBJPR_CLEANONLY must * not be specified and the option OBJPR_NOTMAPPED must be specified. * * The object must be locked. */ void vm_object_page_remove(vm_object_t object, vm_pindex_t start, vm_pindex_t end, int options) { vm_page_t p, next; VM_OBJECT_ASSERT_WLOCKED(object); KASSERT((object->flags & OBJ_UNMANAGED) == 0 || (options & (OBJPR_CLEANONLY | OBJPR_NOTMAPPED)) == OBJPR_NOTMAPPED, ("vm_object_page_remove: illegal options for object %p", object)); if (object->resident_page_count == 0) return; vm_object_pip_add(object, 1); again: p = vm_page_find_least(object, start); /* * Here, the variable "p" is either (1) the page with the least pindex * greater than or equal to the parameter "start" or (2) NULL. */ for (; p != NULL && (p->pindex < end || end == 0); p = next) { next = TAILQ_NEXT(p, listq); /* * If the page is wired for any reason besides the existence * of managed, wired mappings, then it cannot be freed. For * example, fictitious pages, which represent device memory, * are inherently wired and cannot be freed. They can, * however, be invalidated if the option OBJPR_CLEANONLY is * not specified. */ if (vm_page_tryxbusy(p) == 0) { vm_page_sleep_if_busy(p, "vmopar"); goto again; } if (vm_page_wired(p)) { wired: if ((options & OBJPR_NOTMAPPED) == 0 && object->ref_count != 0) pmap_remove_all(p); if ((options & OBJPR_CLEANONLY) == 0) { vm_page_invalid(p); vm_page_undirty(p); } vm_page_xunbusy(p); continue; } KASSERT((p->flags & PG_FICTITIOUS) == 0, ("vm_object_page_remove: page %p is fictitious", p)); if ((options & OBJPR_CLEANONLY) != 0 && !vm_page_none_valid(p)) { if ((options & OBJPR_NOTMAPPED) == 0 && object->ref_count != 0 && !vm_page_try_remove_write(p)) goto wired; if (p->dirty != 0) { vm_page_xunbusy(p); continue; } } if ((options & OBJPR_NOTMAPPED) == 0 && object->ref_count != 0 && !vm_page_try_remove_all(p)) goto wired; vm_page_free(p); } vm_object_pip_wakeup(object); } /* * vm_object_page_noreuse: * * For the given object, attempt to move the specified pages to * the head of the inactive queue. This bypasses regular LRU * operation and allows the pages to be reused quickly under memory * pressure. If a page is wired for any reason, then it will not * be queued. Pages are specified by the range ["start", "end"). * As a special case, if "end" is zero, then the range extends from * "start" to the end of the object. * * This operation should only be performed on objects that * contain non-fictitious, managed pages. * * The object must be locked. */ void vm_object_page_noreuse(vm_object_t object, vm_pindex_t start, vm_pindex_t end) { vm_page_t p, next; VM_OBJECT_ASSERT_LOCKED(object); KASSERT((object->flags & (OBJ_FICTITIOUS | OBJ_UNMANAGED)) == 0, ("vm_object_page_noreuse: illegal object %p", object)); if (object->resident_page_count == 0) return; p = vm_page_find_least(object, start); /* * Here, the variable "p" is either (1) the page with the least pindex * greater than or equal to the parameter "start" or (2) NULL. */ for (; p != NULL && (p->pindex < end || end == 0); p = next) { next = TAILQ_NEXT(p, listq); vm_page_deactivate_noreuse(p); } } /* * Populate the specified range of the object with valid pages. Returns * TRUE if the range is successfully populated and FALSE otherwise. * * Note: This function should be optimized to pass a larger array of * pages to vm_pager_get_pages() before it is applied to a non- * OBJT_DEVICE object. * * The object must be locked. */ boolean_t vm_object_populate(vm_object_t object, vm_pindex_t start, vm_pindex_t end) { vm_page_t m; vm_pindex_t pindex; int rv; VM_OBJECT_ASSERT_WLOCKED(object); for (pindex = start; pindex < end; pindex++) { rv = vm_page_grab_valid(&m, object, pindex, VM_ALLOC_NORMAL); if (rv != VM_PAGER_OK) break; /* * Keep "m" busy because a subsequent iteration may unlock * the object. */ } if (pindex > start) { m = vm_page_lookup(object, start); while (m != NULL && m->pindex < pindex) { vm_page_xunbusy(m); m = TAILQ_NEXT(m, listq); } } return (pindex == end); } /* * Routine: vm_object_coalesce * Function: Coalesces two objects backing up adjoining * regions of memory into a single object. * * returns TRUE if objects were combined. * * NOTE: Only works at the moment if the second object is NULL - * if it's not, which object do we lock first? * * Parameters: * prev_object First object to coalesce * prev_offset Offset into prev_object * prev_size Size of reference to prev_object * next_size Size of reference to the second object * reserved Indicator that extension region has * swap accounted for * * Conditions: * The object must *not* be locked. */ boolean_t vm_object_coalesce(vm_object_t prev_object, vm_ooffset_t prev_offset, vm_size_t prev_size, vm_size_t next_size, boolean_t reserved) { vm_pindex_t next_pindex; if (prev_object == NULL) return (TRUE); if ((prev_object->flags & OBJ_ANON) == 0) return (FALSE); VM_OBJECT_WLOCK(prev_object); /* * Try to collapse the object first. */ vm_object_collapse(prev_object); /* * Can't coalesce if: . more than one reference . paged out . shadows * another object . has a copy elsewhere (any of which mean that the * pages not mapped to prev_entry may be in use anyway) */ if (prev_object->backing_object != NULL) { VM_OBJECT_WUNLOCK(prev_object); return (FALSE); } prev_size >>= PAGE_SHIFT; next_size >>= PAGE_SHIFT; next_pindex = OFF_TO_IDX(prev_offset) + prev_size; if (prev_object->ref_count > 1 && prev_object->size != next_pindex && (prev_object->flags & OBJ_ONEMAPPING) == 0) { VM_OBJECT_WUNLOCK(prev_object); return (FALSE); } /* * Account for the charge. */ if (prev_object->cred != NULL) { /* * If prev_object was charged, then this mapping, * although not charged now, may become writable * later. Non-NULL cred in the object would prevent * swap reservation during enabling of the write * access, so reserve swap now. Failed reservation * cause allocation of the separate object for the map * entry, and swap reservation for this entry is * managed in appropriate time. */ if (!reserved && !swap_reserve_by_cred(ptoa(next_size), prev_object->cred)) { VM_OBJECT_WUNLOCK(prev_object); return (FALSE); } prev_object->charge += ptoa(next_size); } /* * Remove any pages that may still be in the object from a previous * deallocation. */ if (next_pindex < prev_object->size) { vm_object_page_remove(prev_object, next_pindex, next_pindex + next_size, 0); if (prev_object->type == OBJT_SWAP) swap_pager_freespace(prev_object, next_pindex, next_size); #if 0 if (prev_object->cred != NULL) { KASSERT(prev_object->charge >= ptoa(prev_object->size - next_pindex), ("object %p overcharged 1 %jx %jx", prev_object, (uintmax_t)next_pindex, (uintmax_t)next_size)); prev_object->charge -= ptoa(prev_object->size - next_pindex); } #endif } /* * Extend the object if necessary. */ if (next_pindex + next_size > prev_object->size) prev_object->size = next_pindex + next_size; VM_OBJECT_WUNLOCK(prev_object); return (TRUE); } void vm_object_set_writeable_dirty(vm_object_t object) { /* Only set for vnodes & tmpfs */ if (object->type != OBJT_VNODE && (object->flags & OBJ_TMPFS_NODE) == 0) return; atomic_add_int(&object->generation, 1); } /* * vm_object_unwire: * * For each page offset within the specified range of the given object, * find the highest-level page in the shadow chain and unwire it. A page * must exist at every page offset, and the highest-level page must be * wired. */ void vm_object_unwire(vm_object_t object, vm_ooffset_t offset, vm_size_t length, uint8_t queue) { vm_object_t tobject, t1object; vm_page_t m, tm; vm_pindex_t end_pindex, pindex, tpindex; int depth, locked_depth; KASSERT((offset & PAGE_MASK) == 0, ("vm_object_unwire: offset is not page aligned")); KASSERT((length & PAGE_MASK) == 0, ("vm_object_unwire: length is not a multiple of PAGE_SIZE")); /* The wired count of a fictitious page never changes. */ if ((object->flags & OBJ_FICTITIOUS) != 0) return; pindex = OFF_TO_IDX(offset); end_pindex = pindex + atop(length); again: locked_depth = 1; VM_OBJECT_RLOCK(object); m = vm_page_find_least(object, pindex); while (pindex < end_pindex) { if (m == NULL || pindex < m->pindex) { /* * The first object in the shadow chain doesn't * contain a page at the current index. Therefore, * the page must exist in a backing object. */ tobject = object; tpindex = pindex; depth = 0; do { tpindex += OFF_TO_IDX(tobject->backing_object_offset); tobject = tobject->backing_object; KASSERT(tobject != NULL, ("vm_object_unwire: missing page")); if ((tobject->flags & OBJ_FICTITIOUS) != 0) goto next_page; depth++; if (depth == locked_depth) { locked_depth++; VM_OBJECT_RLOCK(tobject); } } while ((tm = vm_page_lookup(tobject, tpindex)) == NULL); } else { tm = m; m = TAILQ_NEXT(m, listq); } if (vm_page_trysbusy(tm) == 0) { for (tobject = object; locked_depth >= 1; locked_depth--) { t1object = tobject->backing_object; if (tm->object != tobject) VM_OBJECT_RUNLOCK(tobject); tobject = t1object; } vm_page_busy_sleep(tm, "unwbo", true); goto again; } vm_page_unwire(tm, queue); vm_page_sunbusy(tm); next_page: pindex++; } /* Release the accumulated object locks. */ for (tobject = object; locked_depth >= 1; locked_depth--) { t1object = tobject->backing_object; VM_OBJECT_RUNLOCK(tobject); tobject = t1object; } } /* * Return the vnode for the given object, or NULL if none exists. * For tmpfs objects, the function may return NULL if there is * no vnode allocated at the time of the call. */ struct vnode * vm_object_vnode(vm_object_t object) { struct vnode *vp; VM_OBJECT_ASSERT_LOCKED(object); if (object->type == OBJT_VNODE) { vp = object->handle; KASSERT(vp != NULL, ("%s: OBJT_VNODE has no vnode", __func__)); } else if (object->type == OBJT_SWAP && (object->flags & OBJ_TMPFS) != 0) { vp = object->un_pager.swp.swp_tmpfs; KASSERT(vp != NULL, ("%s: OBJT_TMPFS has no vnode", __func__)); } else { vp = NULL; } return (vp); } /* * Busy the vm object. This prevents new pages belonging to the object from * becoming busy. Existing pages persist as busy. Callers are responsible * for checking page state before proceeding. */ void vm_object_busy(vm_object_t obj) { VM_OBJECT_ASSERT_LOCKED(obj); blockcount_acquire(&obj->busy, 1); /* The fence is required to order loads of page busy. */ atomic_thread_fence_acq_rel(); } void vm_object_unbusy(vm_object_t obj) { blockcount_release(&obj->busy, 1); } void vm_object_busy_wait(vm_object_t obj, const char *wmesg) { VM_OBJECT_ASSERT_UNLOCKED(obj); (void)blockcount_sleep(&obj->busy, NULL, wmesg, PVM); } /* * Return the kvme type of the given object. * If vpp is not NULL, set it to the object's vm_object_vnode() or NULL. */ int vm_object_kvme_type(vm_object_t object, struct vnode **vpp) { VM_OBJECT_ASSERT_LOCKED(object); if (vpp != NULL) *vpp = vm_object_vnode(object); switch (object->type) { case OBJT_DEFAULT: return (KVME_TYPE_DEFAULT); case OBJT_VNODE: return (KVME_TYPE_VNODE); case OBJT_SWAP: if ((object->flags & OBJ_TMPFS_NODE) != 0) return (KVME_TYPE_VNODE); return (KVME_TYPE_SWAP); case OBJT_DEVICE: return (KVME_TYPE_DEVICE); case OBJT_PHYS: return (KVME_TYPE_PHYS); case OBJT_DEAD: return (KVME_TYPE_DEAD); case OBJT_SG: return (KVME_TYPE_SG); case OBJT_MGTDEVICE: return (KVME_TYPE_MGTDEVICE); default: return (KVME_TYPE_UNKNOWN); } } static int sysctl_vm_object_list(SYSCTL_HANDLER_ARGS) { struct kinfo_vmobject *kvo; char *fullpath, *freepath; struct vnode *vp; struct vattr va; vm_object_t obj; vm_page_t m; int count, error; if (req->oldptr == NULL) { /* * If an old buffer has not been provided, generate an * estimate of the space needed for a subsequent call. */ mtx_lock(&vm_object_list_mtx); count = 0; TAILQ_FOREACH(obj, &vm_object_list, object_list) { if (obj->type == OBJT_DEAD) continue; count++; } mtx_unlock(&vm_object_list_mtx); return (SYSCTL_OUT(req, NULL, sizeof(struct kinfo_vmobject) * count * 11 / 10)); } kvo = malloc(sizeof(*kvo), M_TEMP, M_WAITOK); error = 0; /* * VM objects are type stable and are never removed from the * list once added. This allows us to safely read obj->object_list * after reacquiring the VM object lock. */ mtx_lock(&vm_object_list_mtx); TAILQ_FOREACH(obj, &vm_object_list, object_list) { if (obj->type == OBJT_DEAD) continue; VM_OBJECT_RLOCK(obj); if (obj->type == OBJT_DEAD) { VM_OBJECT_RUNLOCK(obj); continue; } mtx_unlock(&vm_object_list_mtx); kvo->kvo_size = ptoa(obj->size); kvo->kvo_resident = obj->resident_page_count; kvo->kvo_ref_count = obj->ref_count; kvo->kvo_shadow_count = obj->shadow_count; kvo->kvo_memattr = obj->memattr; kvo->kvo_active = 0; kvo->kvo_inactive = 0; TAILQ_FOREACH(m, &obj->memq, listq) { /* * A page may belong to the object but be * dequeued and set to PQ_NONE while the * object lock is not held. This makes the * reads of m->queue below racy, and we do not * count pages set to PQ_NONE. However, this * sysctl is only meant to give an * approximation of the system anyway. */ if (m->a.queue == PQ_ACTIVE) kvo->kvo_active++; else if (m->a.queue == PQ_INACTIVE) kvo->kvo_inactive++; } kvo->kvo_vn_fileid = 0; kvo->kvo_vn_fsid = 0; kvo->kvo_vn_fsid_freebsd11 = 0; freepath = NULL; fullpath = ""; kvo->kvo_type = vm_object_kvme_type(obj, &vp); if (vp != NULL) vref(vp); VM_OBJECT_RUNLOCK(obj); if (vp != NULL) { vn_fullpath(curthread, vp, &fullpath, &freepath); vn_lock(vp, LK_SHARED | LK_RETRY); if (VOP_GETATTR(vp, &va, curthread->td_ucred) == 0) { kvo->kvo_vn_fileid = va.va_fileid; kvo->kvo_vn_fsid = va.va_fsid; kvo->kvo_vn_fsid_freebsd11 = va.va_fsid; /* truncate */ } vput(vp); } strlcpy(kvo->kvo_path, fullpath, sizeof(kvo->kvo_path)); if (freepath != NULL) free(freepath, M_TEMP); /* Pack record size down */ kvo->kvo_structsize = offsetof(struct kinfo_vmobject, kvo_path) + strlen(kvo->kvo_path) + 1; kvo->kvo_structsize = roundup(kvo->kvo_structsize, sizeof(uint64_t)); error = SYSCTL_OUT(req, kvo, kvo->kvo_structsize); mtx_lock(&vm_object_list_mtx); if (error) break; } mtx_unlock(&vm_object_list_mtx); free(kvo, M_TEMP); return (error); } SYSCTL_PROC(_vm, OID_AUTO, objects, CTLTYPE_STRUCT | CTLFLAG_RW | CTLFLAG_SKIP | CTLFLAG_MPSAFE, NULL, 0, sysctl_vm_object_list, "S,kinfo_vmobject", "List of VM objects"); #include "opt_ddb.h" #ifdef DDB #include #include #include static int _vm_object_in_map(vm_map_t map, vm_object_t object, vm_map_entry_t entry) { vm_map_t tmpm; vm_map_entry_t tmpe; vm_object_t obj; if (map == 0) return 0; if (entry == 0) { VM_MAP_ENTRY_FOREACH(tmpe, map) { if (_vm_object_in_map(map, object, tmpe)) { return 1; } } } else if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) { tmpm = entry->object.sub_map; VM_MAP_ENTRY_FOREACH(tmpe, tmpm) { if (_vm_object_in_map(tmpm, object, tmpe)) { return 1; } } } else if ((obj = entry->object.vm_object) != NULL) { for (; obj; obj = obj->backing_object) if (obj == object) { return 1; } } return 0; } static int vm_object_in_map(vm_object_t object) { struct proc *p; /* sx_slock(&allproc_lock); */ FOREACH_PROC_IN_SYSTEM(p) { if (!p->p_vmspace /* || (p->p_flag & (P_SYSTEM|P_WEXIT)) */) continue; if (_vm_object_in_map(&p->p_vmspace->vm_map, object, 0)) { /* sx_sunlock(&allproc_lock); */ return 1; } } /* sx_sunlock(&allproc_lock); */ if (_vm_object_in_map(kernel_map, object, 0)) return 1; return 0; } DB_SHOW_COMMAND(vmochk, vm_object_check) { vm_object_t object; /* * make sure that internal objs are in a map somewhere * and none have zero ref counts. */ TAILQ_FOREACH(object, &vm_object_list, object_list) { if ((object->flags & OBJ_ANON) != 0) { if (object->ref_count == 0) { db_printf("vmochk: internal obj has zero ref count: %ld\n", (long)object->size); } if (!vm_object_in_map(object)) { db_printf( "vmochk: internal obj is not in a map: " "ref: %d, size: %lu: 0x%lx, backing_object: %p\n", object->ref_count, (u_long)object->size, (u_long)object->size, (void *)object->backing_object); } } } } /* * vm_object_print: [ debug ] */ DB_SHOW_COMMAND(object, vm_object_print_static) { /* XXX convert args. */ vm_object_t object = (vm_object_t)addr; boolean_t full = have_addr; vm_page_t p; /* XXX count is an (unused) arg. Avoid shadowing it. */ #define count was_count int count; if (object == NULL) return; db_iprintf( "Object %p: type=%d, size=0x%jx, res=%d, ref=%d, flags=0x%x ruid %d charge %jx\n", object, (int)object->type, (uintmax_t)object->size, object->resident_page_count, object->ref_count, object->flags, object->cred ? object->cred->cr_ruid : -1, (uintmax_t)object->charge); db_iprintf(" sref=%d, backing_object(%d)=(%p)+0x%jx\n", object->shadow_count, object->backing_object ? object->backing_object->ref_count : 0, object->backing_object, (uintmax_t)object->backing_object_offset); if (!full) return; db_indent += 2; count = 0; TAILQ_FOREACH(p, &object->memq, listq) { if (count == 0) db_iprintf("memory:="); else if (count == 6) { db_printf("\n"); db_iprintf(" ..."); count = 0; } else db_printf(","); count++; db_printf("(off=0x%jx,page=0x%jx)", (uintmax_t)p->pindex, (uintmax_t)VM_PAGE_TO_PHYS(p)); } if (count != 0) db_printf("\n"); db_indent -= 2; } /* XXX. */ #undef count /* XXX need this non-static entry for calling from vm_map_print. */ void vm_object_print( /* db_expr_t */ long addr, boolean_t have_addr, /* db_expr_t */ long count, char *modif) { vm_object_print_static(addr, have_addr, count, modif); } DB_SHOW_COMMAND(vmopag, vm_object_print_pages) { vm_object_t object; vm_pindex_t fidx; vm_paddr_t pa; vm_page_t m, prev_m; int rcount, nl, c; nl = 0; TAILQ_FOREACH(object, &vm_object_list, object_list) { db_printf("new object: %p\n", (void *)object); if (nl > 18) { c = cngetc(); if (c != ' ') return; nl = 0; } nl++; rcount = 0; fidx = 0; pa = -1; TAILQ_FOREACH(m, &object->memq, listq) { if (m->pindex > 128) break; if ((prev_m = TAILQ_PREV(m, pglist, listq)) != NULL && prev_m->pindex + 1 != m->pindex) { if (rcount) { db_printf(" index(%ld)run(%d)pa(0x%lx)\n", (long)fidx, rcount, (long)pa); if (nl > 18) { c = cngetc(); if (c != ' ') return; nl = 0; } nl++; rcount = 0; } } if (rcount && (VM_PAGE_TO_PHYS(m) == pa + rcount * PAGE_SIZE)) { ++rcount; continue; } if (rcount) { db_printf(" index(%ld)run(%d)pa(0x%lx)\n", (long)fidx, rcount, (long)pa); if (nl > 18) { c = cngetc(); if (c != ' ') return; nl = 0; } nl++; } fidx = m->pindex; pa = VM_PAGE_TO_PHYS(m); rcount = 1; } if (rcount) { db_printf(" index(%ld)run(%d)pa(0x%lx)\n", (long)fidx, rcount, (long)pa); if (nl > 18) { c = cngetc(); if (c != ' ') return; nl = 0; } nl++; } } } #endif /* DDB */ diff --git a/sys/vm/vm_page.c b/sys/vm/vm_page.c index 41e7a5bb3099..5a45de6b2753 100644 --- a/sys/vm/vm_page.c +++ b/sys/vm/vm_page.c @@ -1,5472 +1,5462 @@ /*- * SPDX-License-Identifier: (BSD-3-Clause AND MIT-CMU) * * Copyright (c) 1991 Regents of the University of California. * All rights reserved. * Copyright (c) 1998 Matthew Dillon. All Rights Reserved. * * This code is derived from software contributed to Berkeley by * The Mach Operating System project at Carnegie-Mellon University. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * from: @(#)vm_page.c 7.4 (Berkeley) 5/7/91 */ /*- * Copyright (c) 1987, 1990 Carnegie-Mellon University. * All rights reserved. * * Authors: Avadis Tevanian, Jr., Michael Wayne Young * * Permission to use, copy, modify and distribute this software and * its documentation is hereby granted, provided that both the copyright * notice and this permission notice appear in all copies of the * software, derivative works or modified versions, and any portions * thereof, and that both notices appear in supporting documentation. * * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS" * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE. * * Carnegie Mellon requests users of this software to return to * * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU * School of Computer Science * Carnegie Mellon University * Pittsburgh PA 15213-3890 * * any improvements or extensions that they make and grant Carnegie the * rights to redistribute these changes. */ /* * Resident memory management module. */ #include __FBSDID("$FreeBSD$"); #include "opt_vm.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include struct vm_domain vm_dom[MAXMEMDOM]; DPCPU_DEFINE_STATIC(struct vm_batchqueue, pqbatch[MAXMEMDOM][PQ_COUNT]); struct mtx_padalign __exclusive_cache_line pa_lock[PA_LOCK_COUNT]; struct mtx_padalign __exclusive_cache_line vm_domainset_lock; /* The following fields are protected by the domainset lock. */ domainset_t __exclusive_cache_line vm_min_domains; domainset_t __exclusive_cache_line vm_severe_domains; static int vm_min_waiters; static int vm_severe_waiters; static int vm_pageproc_waiters; static SYSCTL_NODE(_vm_stats, OID_AUTO, page, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "VM page statistics"); -static counter_u64_t pqstate_commit_retries = EARLY_COUNTER; +static COUNTER_U64_DEFINE_EARLY(pqstate_commit_retries); SYSCTL_COUNTER_U64(_vm_stats_page, OID_AUTO, pqstate_commit_retries, CTLFLAG_RD, &pqstate_commit_retries, "Number of failed per-page atomic queue state updates"); -static counter_u64_t queue_ops = EARLY_COUNTER; +static COUNTER_U64_DEFINE_EARLY(queue_ops); SYSCTL_COUNTER_U64(_vm_stats_page, OID_AUTO, queue_ops, CTLFLAG_RD, &queue_ops, "Number of batched queue operations"); -static counter_u64_t queue_nops = EARLY_COUNTER; +static COUNTER_U64_DEFINE_EARLY(queue_nops); SYSCTL_COUNTER_U64(_vm_stats_page, OID_AUTO, queue_nops, CTLFLAG_RD, &queue_nops, "Number of batched queue operations with no effects"); -static void -counter_startup(void) -{ - - pqstate_commit_retries = counter_u64_alloc(M_WAITOK); - queue_ops = counter_u64_alloc(M_WAITOK); - queue_nops = counter_u64_alloc(M_WAITOK); -} -SYSINIT(page_counters, SI_SUB_CPU, SI_ORDER_ANY, counter_startup, NULL); - /* * bogus page -- for I/O to/from partially complete buffers, * or for paging into sparsely invalid regions. */ vm_page_t bogus_page; vm_page_t vm_page_array; long vm_page_array_size; long first_page; static TAILQ_HEAD(, vm_page) blacklist_head; static int sysctl_vm_page_blacklist(SYSCTL_HANDLER_ARGS); SYSCTL_PROC(_vm, OID_AUTO, page_blacklist, CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, 0, sysctl_vm_page_blacklist, "A", "Blacklist pages"); static uma_zone_t fakepg_zone; static void vm_page_alloc_check(vm_page_t m); static bool _vm_page_busy_sleep(vm_object_t obj, vm_page_t m, vm_pindex_t pindex, const char *wmesg, int allocflags, bool locked); static void vm_page_clear_dirty_mask(vm_page_t m, vm_page_bits_t pagebits); static void vm_page_enqueue(vm_page_t m, uint8_t queue); static bool vm_page_free_prep(vm_page_t m); static void vm_page_free_toq(vm_page_t m); static void vm_page_init(void *dummy); static int vm_page_insert_after(vm_page_t m, vm_object_t object, vm_pindex_t pindex, vm_page_t mpred); static void vm_page_insert_radixdone(vm_page_t m, vm_object_t object, vm_page_t mpred); static void vm_page_mvqueue(vm_page_t m, const uint8_t queue, const uint16_t nflag); static int vm_page_reclaim_run(int req_class, int domain, u_long npages, vm_page_t m_run, vm_paddr_t high); static void vm_page_release_toq(vm_page_t m, uint8_t nqueue, bool noreuse); static int vm_domain_alloc_fail(struct vm_domain *vmd, vm_object_t object, int req); static int vm_page_zone_import(void *arg, void **store, int cnt, int domain, int flags); static void vm_page_zone_release(void *arg, void **store, int cnt); SYSINIT(vm_page, SI_SUB_VM, SI_ORDER_SECOND, vm_page_init, NULL); static void vm_page_init(void *dummy) { fakepg_zone = uma_zcreate("fakepg", sizeof(struct vm_page), NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); bogus_page = vm_page_alloc(NULL, 0, VM_ALLOC_NOOBJ | VM_ALLOC_NORMAL | VM_ALLOC_WIRED); } /* * The cache page zone is initialized later since we need to be able to allocate * pages before UMA is fully initialized. */ static void vm_page_init_cache_zones(void *dummy __unused) { struct vm_domain *vmd; struct vm_pgcache *pgcache; int cache, domain, maxcache, pool; maxcache = 0; TUNABLE_INT_FETCH("vm.pgcache_zone_max_pcpu", &maxcache); maxcache *= mp_ncpus; for (domain = 0; domain < vm_ndomains; domain++) { vmd = VM_DOMAIN(domain); for (pool = 0; pool < VM_NFREEPOOL; pool++) { pgcache = &vmd->vmd_pgcache[pool]; pgcache->domain = domain; pgcache->pool = pool; pgcache->zone = uma_zcache_create("vm pgcache", PAGE_SIZE, NULL, NULL, NULL, NULL, vm_page_zone_import, vm_page_zone_release, pgcache, UMA_ZONE_VM); /* * Limit each pool's zone to 0.1% of the pages in the * domain. */ cache = maxcache != 0 ? maxcache : vmd->vmd_page_count / 1000; uma_zone_set_maxcache(pgcache->zone, cache); } } } SYSINIT(vm_page2, SI_SUB_VM_CONF, SI_ORDER_ANY, vm_page_init_cache_zones, NULL); /* Make sure that u_long is at least 64 bits when PAGE_SIZE is 32K. */ #if PAGE_SIZE == 32768 #ifdef CTASSERT CTASSERT(sizeof(u_long) >= 8); #endif #endif /* * vm_set_page_size: * * Sets the page size, perhaps based upon the memory * size. Must be called before any use of page-size * dependent functions. */ void vm_set_page_size(void) { if (vm_cnt.v_page_size == 0) vm_cnt.v_page_size = PAGE_SIZE; if (((vm_cnt.v_page_size - 1) & vm_cnt.v_page_size) != 0) panic("vm_set_page_size: page size not a power of two"); } /* * vm_page_blacklist_next: * * Find the next entry in the provided string of blacklist * addresses. Entries are separated by space, comma, or newline. * If an invalid integer is encountered then the rest of the * string is skipped. Updates the list pointer to the next * character, or NULL if the string is exhausted or invalid. */ static vm_paddr_t vm_page_blacklist_next(char **list, char *end) { vm_paddr_t bad; char *cp, *pos; if (list == NULL || *list == NULL) return (0); if (**list =='\0') { *list = NULL; return (0); } /* * If there's no end pointer then the buffer is coming from * the kenv and we know it's null-terminated. */ if (end == NULL) end = *list + strlen(*list); /* Ensure that strtoq() won't walk off the end */ if (*end != '\0') { if (*end == '\n' || *end == ' ' || *end == ',') *end = '\0'; else { printf("Blacklist not terminated, skipping\n"); *list = NULL; return (0); } } for (pos = *list; *pos != '\0'; pos = cp) { bad = strtoq(pos, &cp, 0); if (*cp == '\0' || *cp == ' ' || *cp == ',' || *cp == '\n') { if (bad == 0) { if (++cp < end) continue; else break; } } else break; if (*cp == '\0' || ++cp >= end) *list = NULL; else *list = cp; return (trunc_page(bad)); } printf("Garbage in RAM blacklist, skipping\n"); *list = NULL; return (0); } bool vm_page_blacklist_add(vm_paddr_t pa, bool verbose) { struct vm_domain *vmd; vm_page_t m; int ret; m = vm_phys_paddr_to_vm_page(pa); if (m == NULL) return (true); /* page does not exist, no failure */ vmd = vm_pagequeue_domain(m); vm_domain_free_lock(vmd); ret = vm_phys_unfree_page(m); vm_domain_free_unlock(vmd); if (ret != 0) { vm_domain_freecnt_inc(vmd, -1); TAILQ_INSERT_TAIL(&blacklist_head, m, listq); if (verbose) printf("Skipping page with pa 0x%jx\n", (uintmax_t)pa); } return (ret); } /* * vm_page_blacklist_check: * * Iterate through the provided string of blacklist addresses, pulling * each entry out of the physical allocator free list and putting it * onto a list for reporting via the vm.page_blacklist sysctl. */ static void vm_page_blacklist_check(char *list, char *end) { vm_paddr_t pa; char *next; next = list; while (next != NULL) { if ((pa = vm_page_blacklist_next(&next, end)) == 0) continue; vm_page_blacklist_add(pa, bootverbose); } } /* * vm_page_blacklist_load: * * Search for a special module named "ram_blacklist". It'll be a * plain text file provided by the user via the loader directive * of the same name. */ static void vm_page_blacklist_load(char **list, char **end) { void *mod; u_char *ptr; u_int len; mod = NULL; ptr = NULL; mod = preload_search_by_type("ram_blacklist"); if (mod != NULL) { ptr = preload_fetch_addr(mod); len = preload_fetch_size(mod); } *list = ptr; if (ptr != NULL) *end = ptr + len; else *end = NULL; return; } static int sysctl_vm_page_blacklist(SYSCTL_HANDLER_ARGS) { vm_page_t m; struct sbuf sbuf; int error, first; first = 1; error = sysctl_wire_old_buffer(req, 0); if (error != 0) return (error); sbuf_new_for_sysctl(&sbuf, NULL, 128, req); TAILQ_FOREACH(m, &blacklist_head, listq) { sbuf_printf(&sbuf, "%s%#jx", first ? "" : ",", (uintmax_t)m->phys_addr); first = 0; } error = sbuf_finish(&sbuf); sbuf_delete(&sbuf); return (error); } /* * Initialize a dummy page for use in scans of the specified paging queue. * In principle, this function only needs to set the flag PG_MARKER. * Nonetheless, it write busies the page as a safety precaution. */ static void vm_page_init_marker(vm_page_t marker, int queue, uint16_t aflags) { bzero(marker, sizeof(*marker)); marker->flags = PG_MARKER; marker->a.flags = aflags; marker->busy_lock = VPB_CURTHREAD_EXCLUSIVE; marker->a.queue = queue; } static void vm_page_domain_init(int domain) { struct vm_domain *vmd; struct vm_pagequeue *pq; int i; vmd = VM_DOMAIN(domain); bzero(vmd, sizeof(*vmd)); *__DECONST(const char **, &vmd->vmd_pagequeues[PQ_INACTIVE].pq_name) = "vm inactive pagequeue"; *__DECONST(const char **, &vmd->vmd_pagequeues[PQ_ACTIVE].pq_name) = "vm active pagequeue"; *__DECONST(const char **, &vmd->vmd_pagequeues[PQ_LAUNDRY].pq_name) = "vm laundry pagequeue"; *__DECONST(const char **, &vmd->vmd_pagequeues[PQ_UNSWAPPABLE].pq_name) = "vm unswappable pagequeue"; vmd->vmd_domain = domain; vmd->vmd_page_count = 0; vmd->vmd_free_count = 0; vmd->vmd_segs = 0; vmd->vmd_oom = FALSE; for (i = 0; i < PQ_COUNT; i++) { pq = &vmd->vmd_pagequeues[i]; TAILQ_INIT(&pq->pq_pl); mtx_init(&pq->pq_mutex, pq->pq_name, "vm pagequeue", MTX_DEF | MTX_DUPOK); pq->pq_pdpages = 0; vm_page_init_marker(&vmd->vmd_markers[i], i, 0); } mtx_init(&vmd->vmd_free_mtx, "vm page free queue", NULL, MTX_DEF); mtx_init(&vmd->vmd_pageout_mtx, "vm pageout lock", NULL, MTX_DEF); snprintf(vmd->vmd_name, sizeof(vmd->vmd_name), "%d", domain); /* * inacthead is used to provide FIFO ordering for LRU-bypassing * insertions. */ vm_page_init_marker(&vmd->vmd_inacthead, PQ_INACTIVE, PGA_ENQUEUED); TAILQ_INSERT_HEAD(&vmd->vmd_pagequeues[PQ_INACTIVE].pq_pl, &vmd->vmd_inacthead, plinks.q); /* * The clock pages are used to implement active queue scanning without * requeues. Scans start at clock[0], which is advanced after the scan * ends. When the two clock hands meet, they are reset and scanning * resumes from the head of the queue. */ vm_page_init_marker(&vmd->vmd_clock[0], PQ_ACTIVE, PGA_ENQUEUED); vm_page_init_marker(&vmd->vmd_clock[1], PQ_ACTIVE, PGA_ENQUEUED); TAILQ_INSERT_HEAD(&vmd->vmd_pagequeues[PQ_ACTIVE].pq_pl, &vmd->vmd_clock[0], plinks.q); TAILQ_INSERT_TAIL(&vmd->vmd_pagequeues[PQ_ACTIVE].pq_pl, &vmd->vmd_clock[1], plinks.q); } /* * Initialize a physical page in preparation for adding it to the free * lists. */ static void vm_page_init_page(vm_page_t m, vm_paddr_t pa, int segind) { m->object = NULL; m->ref_count = 0; m->busy_lock = VPB_FREED; m->flags = m->a.flags = 0; m->phys_addr = pa; m->a.queue = PQ_NONE; m->psind = 0; m->segind = segind; m->order = VM_NFREEORDER; m->pool = VM_FREEPOOL_DEFAULT; m->valid = m->dirty = 0; pmap_page_init(m); } #ifndef PMAP_HAS_PAGE_ARRAY static vm_paddr_t vm_page_array_alloc(vm_offset_t *vaddr, vm_paddr_t end, vm_paddr_t page_range) { vm_paddr_t new_end; /* * Reserve an unmapped guard page to trap access to vm_page_array[-1]. * However, because this page is allocated from KVM, out-of-bounds * accesses using the direct map will not be trapped. */ *vaddr += PAGE_SIZE; /* * Allocate physical memory for the page structures, and map it. */ new_end = trunc_page(end - page_range * sizeof(struct vm_page)); vm_page_array = (vm_page_t)pmap_map(vaddr, new_end, end, VM_PROT_READ | VM_PROT_WRITE); vm_page_array_size = page_range; return (new_end); } #endif /* * vm_page_startup: * * Initializes the resident memory module. Allocates physical memory for * bootstrapping UMA and some data structures that are used to manage * physical pages. Initializes these structures, and populates the free * page queues. */ vm_offset_t vm_page_startup(vm_offset_t vaddr) { struct vm_phys_seg *seg; vm_page_t m; char *list, *listend; vm_paddr_t end, high_avail, low_avail, new_end, size; vm_paddr_t page_range __unused; vm_paddr_t last_pa, pa; u_long pagecount; int biggestone, i, segind; #ifdef WITNESS vm_offset_t mapped; int witness_size; #endif #if defined(__i386__) && defined(VM_PHYSSEG_DENSE) long ii; #endif vaddr = round_page(vaddr); vm_phys_early_startup(); biggestone = vm_phys_avail_largest(); end = phys_avail[biggestone+1]; /* * Initialize the page and queue locks. */ mtx_init(&vm_domainset_lock, "vm domainset lock", NULL, MTX_DEF); for (i = 0; i < PA_LOCK_COUNT; i++) mtx_init(&pa_lock[i], "vm page", NULL, MTX_DEF); for (i = 0; i < vm_ndomains; i++) vm_page_domain_init(i); new_end = end; #ifdef WITNESS witness_size = round_page(witness_startup_count()); new_end -= witness_size; mapped = pmap_map(&vaddr, new_end, new_end + witness_size, VM_PROT_READ | VM_PROT_WRITE); bzero((void *)mapped, witness_size); witness_startup((void *)mapped); #endif #if defined(__aarch64__) || defined(__amd64__) || defined(__arm__) || \ defined(__i386__) || defined(__mips__) || defined(__riscv) || \ defined(__powerpc64__) /* * Allocate a bitmap to indicate that a random physical page * needs to be included in a minidump. * * The amd64 port needs this to indicate which direct map pages * need to be dumped, via calls to dump_add_page()/dump_drop_page(). * * However, i386 still needs this workspace internally within the * minidump code. In theory, they are not needed on i386, but are * included should the sf_buf code decide to use them. */ last_pa = 0; for (i = 0; dump_avail[i + 1] != 0; i += 2) if (dump_avail[i + 1] > last_pa) last_pa = dump_avail[i + 1]; page_range = last_pa / PAGE_SIZE; vm_page_dump_size = round_page(roundup2(page_range, NBBY) / NBBY); new_end -= vm_page_dump_size; vm_page_dump = (void *)(uintptr_t)pmap_map(&vaddr, new_end, new_end + vm_page_dump_size, VM_PROT_READ | VM_PROT_WRITE); bzero((void *)vm_page_dump, vm_page_dump_size); #else (void)last_pa; #endif #if defined(__aarch64__) || defined(__amd64__) || defined(__mips__) || \ defined(__riscv) || defined(__powerpc64__) /* * Include the UMA bootstrap pages, witness pages and vm_page_dump * in a crash dump. When pmap_map() uses the direct map, they are * not automatically included. */ for (pa = new_end; pa < end; pa += PAGE_SIZE) dump_add_page(pa); #endif phys_avail[biggestone + 1] = new_end; #ifdef __amd64__ /* * Request that the physical pages underlying the message buffer be * included in a crash dump. Since the message buffer is accessed * through the direct map, they are not automatically included. */ pa = DMAP_TO_PHYS((vm_offset_t)msgbufp->msg_ptr); last_pa = pa + round_page(msgbufsize); while (pa < last_pa) { dump_add_page(pa); pa += PAGE_SIZE; } #endif /* * Compute the number of pages of memory that will be available for * use, taking into account the overhead of a page structure per page. * In other words, solve * "available physical memory" - round_page(page_range * * sizeof(struct vm_page)) = page_range * PAGE_SIZE * for page_range. */ low_avail = phys_avail[0]; high_avail = phys_avail[1]; for (i = 0; i < vm_phys_nsegs; i++) { if (vm_phys_segs[i].start < low_avail) low_avail = vm_phys_segs[i].start; if (vm_phys_segs[i].end > high_avail) high_avail = vm_phys_segs[i].end; } /* Skip the first chunk. It is already accounted for. */ for (i = 2; phys_avail[i + 1] != 0; i += 2) { if (phys_avail[i] < low_avail) low_avail = phys_avail[i]; if (phys_avail[i + 1] > high_avail) high_avail = phys_avail[i + 1]; } first_page = low_avail / PAGE_SIZE; #ifdef VM_PHYSSEG_SPARSE size = 0; for (i = 0; i < vm_phys_nsegs; i++) size += vm_phys_segs[i].end - vm_phys_segs[i].start; for (i = 0; phys_avail[i + 1] != 0; i += 2) size += phys_avail[i + 1] - phys_avail[i]; #elif defined(VM_PHYSSEG_DENSE) size = high_avail - low_avail; #else #error "Either VM_PHYSSEG_DENSE or VM_PHYSSEG_SPARSE must be defined." #endif #ifdef PMAP_HAS_PAGE_ARRAY pmap_page_array_startup(size / PAGE_SIZE); biggestone = vm_phys_avail_largest(); end = new_end = phys_avail[biggestone + 1]; #else #ifdef VM_PHYSSEG_DENSE /* * In the VM_PHYSSEG_DENSE case, the number of pages can account for * the overhead of a page structure per page only if vm_page_array is * allocated from the last physical memory chunk. Otherwise, we must * allocate page structures representing the physical memory * underlying vm_page_array, even though they will not be used. */ if (new_end != high_avail) page_range = size / PAGE_SIZE; else #endif { page_range = size / (PAGE_SIZE + sizeof(struct vm_page)); /* * If the partial bytes remaining are large enough for * a page (PAGE_SIZE) without a corresponding * 'struct vm_page', then new_end will contain an * extra page after subtracting the length of the VM * page array. Compensate by subtracting an extra * page from new_end. */ if (size % (PAGE_SIZE + sizeof(struct vm_page)) >= PAGE_SIZE) { if (new_end == high_avail) high_avail -= PAGE_SIZE; new_end -= PAGE_SIZE; } } end = new_end; new_end = vm_page_array_alloc(&vaddr, end, page_range); #endif #if VM_NRESERVLEVEL > 0 /* * Allocate physical memory for the reservation management system's * data structures, and map it. */ new_end = vm_reserv_startup(&vaddr, new_end); #endif #if defined(__aarch64__) || defined(__amd64__) || defined(__mips__) || \ defined(__riscv) || defined(__powerpc64__) /* * Include vm_page_array and vm_reserv_array in a crash dump. */ for (pa = new_end; pa < end; pa += PAGE_SIZE) dump_add_page(pa); #endif phys_avail[biggestone + 1] = new_end; /* * Add physical memory segments corresponding to the available * physical pages. */ for (i = 0; phys_avail[i + 1] != 0; i += 2) if (vm_phys_avail_size(i) != 0) vm_phys_add_seg(phys_avail[i], phys_avail[i + 1]); /* * Initialize the physical memory allocator. */ vm_phys_init(); /* * Initialize the page structures and add every available page to the * physical memory allocator's free lists. */ #if defined(__i386__) && defined(VM_PHYSSEG_DENSE) for (ii = 0; ii < vm_page_array_size; ii++) { m = &vm_page_array[ii]; vm_page_init_page(m, (first_page + ii) << PAGE_SHIFT, 0); m->flags = PG_FICTITIOUS; } #endif vm_cnt.v_page_count = 0; for (segind = 0; segind < vm_phys_nsegs; segind++) { seg = &vm_phys_segs[segind]; for (m = seg->first_page, pa = seg->start; pa < seg->end; m++, pa += PAGE_SIZE) vm_page_init_page(m, pa, segind); /* * Add the segment to the free lists only if it is covered by * one of the ranges in phys_avail. Because we've added the * ranges to the vm_phys_segs array, we can assume that each * segment is either entirely contained in one of the ranges, * or doesn't overlap any of them. */ for (i = 0; phys_avail[i + 1] != 0; i += 2) { struct vm_domain *vmd; if (seg->start < phys_avail[i] || seg->end > phys_avail[i + 1]) continue; m = seg->first_page; pagecount = (u_long)atop(seg->end - seg->start); vmd = VM_DOMAIN(seg->domain); vm_domain_free_lock(vmd); vm_phys_enqueue_contig(m, pagecount); vm_domain_free_unlock(vmd); vm_domain_freecnt_inc(vmd, pagecount); vm_cnt.v_page_count += (u_int)pagecount; vmd = VM_DOMAIN(seg->domain); vmd->vmd_page_count += (u_int)pagecount; vmd->vmd_segs |= 1UL << m->segind; break; } } /* * Remove blacklisted pages from the physical memory allocator. */ TAILQ_INIT(&blacklist_head); vm_page_blacklist_load(&list, &listend); vm_page_blacklist_check(list, listend); list = kern_getenv("vm.blacklist"); vm_page_blacklist_check(list, NULL); freeenv(list); #if VM_NRESERVLEVEL > 0 /* * Initialize the reservation management system. */ vm_reserv_init(); #endif return (vaddr); } void vm_page_reference(vm_page_t m) { vm_page_aflag_set(m, PGA_REFERENCED); } /* * vm_page_trybusy * * Helper routine for grab functions to trylock busy. * * Returns true on success and false on failure. */ static bool vm_page_trybusy(vm_page_t m, int allocflags) { if ((allocflags & (VM_ALLOC_SBUSY | VM_ALLOC_IGN_SBUSY)) != 0) return (vm_page_trysbusy(m)); else return (vm_page_tryxbusy(m)); } /* * vm_page_tryacquire * * Helper routine for grab functions to trylock busy and wire. * * Returns true on success and false on failure. */ static inline bool vm_page_tryacquire(vm_page_t m, int allocflags) { bool locked; locked = vm_page_trybusy(m, allocflags); if (locked && (allocflags & VM_ALLOC_WIRED) != 0) vm_page_wire(m); return (locked); } /* * vm_page_busy_acquire: * * Acquire the busy lock as described by VM_ALLOC_* flags. Will loop * and drop the object lock if necessary. */ bool vm_page_busy_acquire(vm_page_t m, int allocflags) { vm_object_t obj; bool locked; /* * The page-specific object must be cached because page * identity can change during the sleep, causing the * re-lock of a different object. * It is assumed that a reference to the object is already * held by the callers. */ obj = m->object; for (;;) { if (vm_page_tryacquire(m, allocflags)) return (true); if ((allocflags & VM_ALLOC_NOWAIT) != 0) return (false); if (obj != NULL) locked = VM_OBJECT_WOWNED(obj); else locked = false; MPASS(locked || vm_page_wired(m)); if (_vm_page_busy_sleep(obj, m, m->pindex, "vmpba", allocflags, locked) && locked) VM_OBJECT_WLOCK(obj); if ((allocflags & VM_ALLOC_WAITFAIL) != 0) return (false); KASSERT(m->object == obj || m->object == NULL, ("vm_page_busy_acquire: page %p does not belong to %p", m, obj)); } } /* * vm_page_busy_downgrade: * * Downgrade an exclusive busy page into a single shared busy page. */ void vm_page_busy_downgrade(vm_page_t m) { u_int x; vm_page_assert_xbusied(m); x = m->busy_lock; for (;;) { if (atomic_fcmpset_rel_int(&m->busy_lock, &x, VPB_SHARERS_WORD(1))) break; } if ((x & VPB_BIT_WAITERS) != 0) wakeup(m); } /* * * vm_page_busy_tryupgrade: * * Attempt to upgrade a single shared busy into an exclusive busy. */ int vm_page_busy_tryupgrade(vm_page_t m) { u_int ce, x; vm_page_assert_sbusied(m); x = m->busy_lock; ce = VPB_CURTHREAD_EXCLUSIVE; for (;;) { if (VPB_SHARERS(x) > 1) return (0); KASSERT((x & ~VPB_BIT_WAITERS) == VPB_SHARERS_WORD(1), ("vm_page_busy_tryupgrade: invalid lock state")); if (!atomic_fcmpset_acq_int(&m->busy_lock, &x, ce | (x & VPB_BIT_WAITERS))) continue; return (1); } } /* * vm_page_sbusied: * * Return a positive value if the page is shared busied, 0 otherwise. */ int vm_page_sbusied(vm_page_t m) { u_int x; x = m->busy_lock; return ((x & VPB_BIT_SHARED) != 0 && x != VPB_UNBUSIED); } /* * vm_page_sunbusy: * * Shared unbusy a page. */ void vm_page_sunbusy(vm_page_t m) { u_int x; vm_page_assert_sbusied(m); x = m->busy_lock; for (;;) { KASSERT(x != VPB_FREED, ("vm_page_sunbusy: Unlocking freed page.")); if (VPB_SHARERS(x) > 1) { if (atomic_fcmpset_int(&m->busy_lock, &x, x - VPB_ONE_SHARER)) break; continue; } KASSERT((x & ~VPB_BIT_WAITERS) == VPB_SHARERS_WORD(1), ("vm_page_sunbusy: invalid lock state")); if (!atomic_fcmpset_rel_int(&m->busy_lock, &x, VPB_UNBUSIED)) continue; if ((x & VPB_BIT_WAITERS) == 0) break; wakeup(m); break; } } /* * vm_page_busy_sleep: * * Sleep if the page is busy, using the page pointer as wchan. * This is used to implement the hard-path of busying mechanism. * * If nonshared is true, sleep only if the page is xbusy. * * The object lock must be held on entry and will be released on exit. */ void vm_page_busy_sleep(vm_page_t m, const char *wmesg, bool nonshared) { vm_object_t obj; obj = m->object; VM_OBJECT_ASSERT_LOCKED(obj); vm_page_lock_assert(m, MA_NOTOWNED); if (!_vm_page_busy_sleep(obj, m, m->pindex, wmesg, nonshared ? VM_ALLOC_SBUSY : 0 , true)) VM_OBJECT_DROP(obj); } /* * vm_page_busy_sleep_unlocked: * * Sleep if the page is busy, using the page pointer as wchan. * This is used to implement the hard-path of busying mechanism. * * If nonshared is true, sleep only if the page is xbusy. * * The object lock must not be held on entry. The operation will * return if the page changes identity. */ void vm_page_busy_sleep_unlocked(vm_object_t obj, vm_page_t m, vm_pindex_t pindex, const char *wmesg, bool nonshared) { VM_OBJECT_ASSERT_UNLOCKED(obj); vm_page_lock_assert(m, MA_NOTOWNED); _vm_page_busy_sleep(obj, m, pindex, wmesg, nonshared ? VM_ALLOC_SBUSY : 0, false); } /* * _vm_page_busy_sleep: * * Internal busy sleep function. Verifies the page identity and * lockstate against parameters. Returns true if it sleeps and * false otherwise. * * If locked is true the lock will be dropped for any true returns * and held for any false returns. */ static bool _vm_page_busy_sleep(vm_object_t obj, vm_page_t m, vm_pindex_t pindex, const char *wmesg, int allocflags, bool locked) { bool xsleep; u_int x; /* * If the object is busy we must wait for that to drain to zero * before trying the page again. */ if (obj != NULL && vm_object_busied(obj)) { if (locked) VM_OBJECT_DROP(obj); vm_object_busy_wait(obj, wmesg); return (true); } if (!vm_page_busied(m)) return (false); xsleep = (allocflags & (VM_ALLOC_SBUSY | VM_ALLOC_IGN_SBUSY)) != 0; sleepq_lock(m); x = atomic_load_int(&m->busy_lock); do { /* * If the page changes objects or becomes unlocked we can * simply return. */ if (x == VPB_UNBUSIED || (xsleep && (x & VPB_BIT_SHARED) != 0) || m->object != obj || m->pindex != pindex) { sleepq_release(m); return (false); } if ((x & VPB_BIT_WAITERS) != 0) break; } while (!atomic_fcmpset_int(&m->busy_lock, &x, x | VPB_BIT_WAITERS)); if (locked) VM_OBJECT_DROP(obj); DROP_GIANT(); sleepq_add(m, NULL, wmesg, 0, 0); sleepq_wait(m, PVM); PICKUP_GIANT(); return (true); } /* * vm_page_trysbusy: * * Try to shared busy a page. * If the operation succeeds 1 is returned otherwise 0. * The operation never sleeps. */ int vm_page_trysbusy(vm_page_t m) { vm_object_t obj; u_int x; obj = m->object; x = m->busy_lock; for (;;) { if ((x & VPB_BIT_SHARED) == 0) return (0); /* * Reduce the window for transient busies that will trigger * false negatives in vm_page_ps_test(). */ if (obj != NULL && vm_object_busied(obj)) return (0); if (atomic_fcmpset_acq_int(&m->busy_lock, &x, x + VPB_ONE_SHARER)) break; } /* Refetch the object now that we're guaranteed that it is stable. */ obj = m->object; if (obj != NULL && vm_object_busied(obj)) { vm_page_sunbusy(m); return (0); } return (1); } /* * vm_page_tryxbusy: * * Try to exclusive busy a page. * If the operation succeeds 1 is returned otherwise 0. * The operation never sleeps. */ int vm_page_tryxbusy(vm_page_t m) { vm_object_t obj; if (atomic_cmpset_acq_int(&(m)->busy_lock, VPB_UNBUSIED, VPB_CURTHREAD_EXCLUSIVE) == 0) return (0); obj = m->object; if (obj != NULL && vm_object_busied(obj)) { vm_page_xunbusy(m); return (0); } return (1); } static void vm_page_xunbusy_hard_tail(vm_page_t m) { atomic_store_rel_int(&m->busy_lock, VPB_UNBUSIED); /* Wake the waiter. */ wakeup(m); } /* * vm_page_xunbusy_hard: * * Called when unbusy has failed because there is a waiter. */ void vm_page_xunbusy_hard(vm_page_t m) { vm_page_assert_xbusied(m); vm_page_xunbusy_hard_tail(m); } void vm_page_xunbusy_hard_unchecked(vm_page_t m) { vm_page_assert_xbusied_unchecked(m); vm_page_xunbusy_hard_tail(m); } static void vm_page_busy_free(vm_page_t m) { u_int x; atomic_thread_fence_rel(); x = atomic_swap_int(&m->busy_lock, VPB_FREED); if ((x & VPB_BIT_WAITERS) != 0) wakeup(m); } /* * vm_page_unhold_pages: * * Unhold each of the pages that is referenced by the given array. */ void vm_page_unhold_pages(vm_page_t *ma, int count) { for (; count != 0; count--) { vm_page_unwire(*ma, PQ_ACTIVE); ma++; } } vm_page_t PHYS_TO_VM_PAGE(vm_paddr_t pa) { vm_page_t m; #ifdef VM_PHYSSEG_SPARSE m = vm_phys_paddr_to_vm_page(pa); if (m == NULL) m = vm_phys_fictitious_to_vm_page(pa); return (m); #elif defined(VM_PHYSSEG_DENSE) long pi; pi = atop(pa); if (pi >= first_page && (pi - first_page) < vm_page_array_size) { m = &vm_page_array[pi - first_page]; return (m); } return (vm_phys_fictitious_to_vm_page(pa)); #else #error "Either VM_PHYSSEG_DENSE or VM_PHYSSEG_SPARSE must be defined." #endif } /* * vm_page_getfake: * * Create a fictitious page with the specified physical address and * memory attribute. The memory attribute is the only the machine- * dependent aspect of a fictitious page that must be initialized. */ vm_page_t vm_page_getfake(vm_paddr_t paddr, vm_memattr_t memattr) { vm_page_t m; m = uma_zalloc(fakepg_zone, M_WAITOK | M_ZERO); vm_page_initfake(m, paddr, memattr); return (m); } void vm_page_initfake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr) { if ((m->flags & PG_FICTITIOUS) != 0) { /* * The page's memattr might have changed since the * previous initialization. Update the pmap to the * new memattr. */ goto memattr; } m->phys_addr = paddr; m->a.queue = PQ_NONE; /* Fictitious pages don't use "segind". */ m->flags = PG_FICTITIOUS; /* Fictitious pages don't use "order" or "pool". */ m->oflags = VPO_UNMANAGED; m->busy_lock = VPB_CURTHREAD_EXCLUSIVE; /* Fictitious pages are unevictable. */ m->ref_count = 1; pmap_page_init(m); memattr: pmap_page_set_memattr(m, memattr); } /* * vm_page_putfake: * * Release a fictitious page. */ void vm_page_putfake(vm_page_t m) { KASSERT((m->oflags & VPO_UNMANAGED) != 0, ("managed %p", m)); KASSERT((m->flags & PG_FICTITIOUS) != 0, ("vm_page_putfake: bad page %p", m)); vm_page_assert_xbusied(m); vm_page_busy_free(m); uma_zfree(fakepg_zone, m); } /* * vm_page_updatefake: * * Update the given fictitious page to the specified physical address and * memory attribute. */ void vm_page_updatefake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr) { KASSERT((m->flags & PG_FICTITIOUS) != 0, ("vm_page_updatefake: bad page %p", m)); m->phys_addr = paddr; pmap_page_set_memattr(m, memattr); } /* * vm_page_free: * * Free a page. */ void vm_page_free(vm_page_t m) { m->flags &= ~PG_ZERO; vm_page_free_toq(m); } /* * vm_page_free_zero: * * Free a page to the zerod-pages queue */ void vm_page_free_zero(vm_page_t m) { m->flags |= PG_ZERO; vm_page_free_toq(m); } /* * Unbusy and handle the page queueing for a page from a getpages request that * was optionally read ahead or behind. */ void vm_page_readahead_finish(vm_page_t m) { /* We shouldn't put invalid pages on queues. */ KASSERT(!vm_page_none_valid(m), ("%s: %p is invalid", __func__, m)); /* * Since the page is not the actually needed one, whether it should * be activated or deactivated is not obvious. Empirical results * have shown that deactivating the page is usually the best choice, * unless the page is wanted by another thread. */ if ((m->busy_lock & VPB_BIT_WAITERS) != 0) vm_page_activate(m); else vm_page_deactivate(m); vm_page_xunbusy_unchecked(m); } /* * vm_page_sleep_if_busy: * * Sleep and release the object lock if the page is busied. * Returns TRUE if the thread slept. * * The given page must be unlocked and object containing it must * be locked. */ int vm_page_sleep_if_busy(vm_page_t m, const char *wmesg) { vm_object_t obj; vm_page_lock_assert(m, MA_NOTOWNED); VM_OBJECT_ASSERT_WLOCKED(m->object); /* * The page-specific object must be cached because page * identity can change during the sleep, causing the * re-lock of a different object. * It is assumed that a reference to the object is already * held by the callers. */ obj = m->object; if (_vm_page_busy_sleep(obj, m, m->pindex, wmesg, 0, true)) { VM_OBJECT_WLOCK(obj); return (TRUE); } return (FALSE); } /* * vm_page_sleep_if_xbusy: * * Sleep and release the object lock if the page is xbusied. * Returns TRUE if the thread slept. * * The given page must be unlocked and object containing it must * be locked. */ int vm_page_sleep_if_xbusy(vm_page_t m, const char *wmesg) { vm_object_t obj; vm_page_lock_assert(m, MA_NOTOWNED); VM_OBJECT_ASSERT_WLOCKED(m->object); /* * The page-specific object must be cached because page * identity can change during the sleep, causing the * re-lock of a different object. * It is assumed that a reference to the object is already * held by the callers. */ obj = m->object; if (_vm_page_busy_sleep(obj, m, m->pindex, wmesg, VM_ALLOC_SBUSY, true)) { VM_OBJECT_WLOCK(obj); return (TRUE); } return (FALSE); } /* * vm_page_dirty_KBI: [ internal use only ] * * Set all bits in the page's dirty field. * * The object containing the specified page must be locked if the * call is made from the machine-independent layer. * * See vm_page_clear_dirty_mask(). * * This function should only be called by vm_page_dirty(). */ void vm_page_dirty_KBI(vm_page_t m) { /* Refer to this operation by its public name. */ KASSERT(vm_page_all_valid(m), ("vm_page_dirty: page is invalid!")); m->dirty = VM_PAGE_BITS_ALL; } /* * vm_page_insert: [ internal use only ] * * Inserts the given mem entry into the object and object list. * * The object must be locked. */ int vm_page_insert(vm_page_t m, vm_object_t object, vm_pindex_t pindex) { vm_page_t mpred; VM_OBJECT_ASSERT_WLOCKED(object); mpred = vm_radix_lookup_le(&object->rtree, pindex); return (vm_page_insert_after(m, object, pindex, mpred)); } /* * vm_page_insert_after: * * Inserts the page "m" into the specified object at offset "pindex". * * The page "mpred" must immediately precede the offset "pindex" within * the specified object. * * The object must be locked. */ static int vm_page_insert_after(vm_page_t m, vm_object_t object, vm_pindex_t pindex, vm_page_t mpred) { vm_page_t msucc; VM_OBJECT_ASSERT_WLOCKED(object); KASSERT(m->object == NULL, ("vm_page_insert_after: page already inserted")); if (mpred != NULL) { KASSERT(mpred->object == object, ("vm_page_insert_after: object doesn't contain mpred")); KASSERT(mpred->pindex < pindex, ("vm_page_insert_after: mpred doesn't precede pindex")); msucc = TAILQ_NEXT(mpred, listq); } else msucc = TAILQ_FIRST(&object->memq); if (msucc != NULL) KASSERT(msucc->pindex > pindex, ("vm_page_insert_after: msucc doesn't succeed pindex")); /* * Record the object/offset pair in this page. */ m->object = object; m->pindex = pindex; m->ref_count |= VPRC_OBJREF; /* * Now link into the object's ordered list of backed pages. */ if (vm_radix_insert(&object->rtree, m)) { m->object = NULL; m->pindex = 0; m->ref_count &= ~VPRC_OBJREF; return (1); } vm_page_insert_radixdone(m, object, mpred); return (0); } /* * vm_page_insert_radixdone: * * Complete page "m" insertion into the specified object after the * radix trie hooking. * * The page "mpred" must precede the offset "m->pindex" within the * specified object. * * The object must be locked. */ static void vm_page_insert_radixdone(vm_page_t m, vm_object_t object, vm_page_t mpred) { VM_OBJECT_ASSERT_WLOCKED(object); KASSERT(object != NULL && m->object == object, ("vm_page_insert_radixdone: page %p has inconsistent object", m)); KASSERT((m->ref_count & VPRC_OBJREF) != 0, ("vm_page_insert_radixdone: page %p is missing object ref", m)); if (mpred != NULL) { KASSERT(mpred->object == object, ("vm_page_insert_radixdone: object doesn't contain mpred")); KASSERT(mpred->pindex < m->pindex, ("vm_page_insert_radixdone: mpred doesn't precede pindex")); } if (mpred != NULL) TAILQ_INSERT_AFTER(&object->memq, mpred, m, listq); else TAILQ_INSERT_HEAD(&object->memq, m, listq); /* * Show that the object has one more resident page. */ object->resident_page_count++; /* * Hold the vnode until the last page is released. */ if (object->resident_page_count == 1 && object->type == OBJT_VNODE) vhold(object->handle); /* * Since we are inserting a new and possibly dirty page, * update the object's generation count. */ if (pmap_page_is_write_mapped(m)) vm_object_set_writeable_dirty(object); } /* * Do the work to remove a page from its object. The caller is responsible for * updating the page's fields to reflect this removal. */ static void vm_page_object_remove(vm_page_t m) { vm_object_t object; vm_page_t mrem; vm_page_assert_xbusied(m); object = m->object; VM_OBJECT_ASSERT_WLOCKED(object); KASSERT((m->ref_count & VPRC_OBJREF) != 0, ("page %p is missing its object ref", m)); /* Deferred free of swap space. */ if ((m->a.flags & PGA_SWAP_FREE) != 0) vm_pager_page_unswapped(m); m->object = NULL; mrem = vm_radix_remove(&object->rtree, m->pindex); KASSERT(mrem == m, ("removed page %p, expected page %p", mrem, m)); /* * Now remove from the object's list of backed pages. */ TAILQ_REMOVE(&object->memq, m, listq); /* * And show that the object has one fewer resident page. */ object->resident_page_count--; /* * The vnode may now be recycled. */ if (object->resident_page_count == 0 && object->type == OBJT_VNODE) vdrop(object->handle); } /* * vm_page_remove: * * Removes the specified page from its containing object, but does not * invalidate any backing storage. Returns true if the object's reference * was the last reference to the page, and false otherwise. * * The object must be locked and the page must be exclusively busied. * The exclusive busy will be released on return. If this is not the * final ref and the caller does not hold a wire reference it may not * continue to access the page. */ bool vm_page_remove(vm_page_t m) { bool dropped; dropped = vm_page_remove_xbusy(m); vm_page_xunbusy(m); return (dropped); } /* * vm_page_remove_xbusy * * Removes the page but leaves the xbusy held. Returns true if this * removed the final ref and false otherwise. */ bool vm_page_remove_xbusy(vm_page_t m) { vm_page_object_remove(m); return (vm_page_drop(m, VPRC_OBJREF) == VPRC_OBJREF); } /* * vm_page_lookup: * * Returns the page associated with the object/offset * pair specified; if none is found, NULL is returned. * * The object must be locked. */ vm_page_t vm_page_lookup(vm_object_t object, vm_pindex_t pindex) { VM_OBJECT_ASSERT_LOCKED(object); return (vm_radix_lookup(&object->rtree, pindex)); } /* * vm_page_relookup: * * Returns a page that must already have been busied by * the caller. Used for bogus page replacement. */ vm_page_t vm_page_relookup(vm_object_t object, vm_pindex_t pindex) { vm_page_t m; m = vm_radix_lookup_unlocked(&object->rtree, pindex); KASSERT(m != NULL && vm_page_busied(m) && m->object == object && m->pindex == pindex, ("vm_page_relookup: Invalid page %p", m)); return (m); } /* * This should only be used by lockless functions for releasing transient * incorrect acquires. The page may have been freed after we acquired a * busy lock. In this case busy_lock == VPB_FREED and we have nothing * further to do. */ static void vm_page_busy_release(vm_page_t m) { u_int x; x = atomic_load_int(&m->busy_lock); for (;;) { if (x == VPB_FREED) break; if ((x & VPB_BIT_SHARED) != 0 && VPB_SHARERS(x) > 1) { if (atomic_fcmpset_int(&m->busy_lock, &x, x - VPB_ONE_SHARER)) break; continue; } KASSERT((x & VPB_BIT_SHARED) != 0 || (x & ~VPB_BIT_WAITERS) == VPB_CURTHREAD_EXCLUSIVE, ("vm_page_busy_release: %p xbusy not owned.", m)); if (!atomic_fcmpset_rel_int(&m->busy_lock, &x, VPB_UNBUSIED)) continue; if ((x & VPB_BIT_WAITERS) != 0) wakeup(m); break; } } /* * vm_page_find_least: * * Returns the page associated with the object with least pindex * greater than or equal to the parameter pindex, or NULL. * * The object must be locked. */ vm_page_t vm_page_find_least(vm_object_t object, vm_pindex_t pindex) { vm_page_t m; VM_OBJECT_ASSERT_LOCKED(object); if ((m = TAILQ_FIRST(&object->memq)) != NULL && m->pindex < pindex) m = vm_radix_lookup_ge(&object->rtree, pindex); return (m); } /* * Returns the given page's successor (by pindex) within the object if it is * resident; if none is found, NULL is returned. * * The object must be locked. */ vm_page_t vm_page_next(vm_page_t m) { vm_page_t next; VM_OBJECT_ASSERT_LOCKED(m->object); if ((next = TAILQ_NEXT(m, listq)) != NULL) { MPASS(next->object == m->object); if (next->pindex != m->pindex + 1) next = NULL; } return (next); } /* * Returns the given page's predecessor (by pindex) within the object if it is * resident; if none is found, NULL is returned. * * The object must be locked. */ vm_page_t vm_page_prev(vm_page_t m) { vm_page_t prev; VM_OBJECT_ASSERT_LOCKED(m->object); if ((prev = TAILQ_PREV(m, pglist, listq)) != NULL) { MPASS(prev->object == m->object); if (prev->pindex != m->pindex - 1) prev = NULL; } return (prev); } /* * Uses the page mnew as a replacement for an existing page at index * pindex which must be already present in the object. * * Both pages must be exclusively busied on enter. The old page is * unbusied on exit. * * A return value of true means mold is now free. If this is not the * final ref and the caller does not hold a wire reference it may not * continue to access the page. */ static bool vm_page_replace_hold(vm_page_t mnew, vm_object_t object, vm_pindex_t pindex, vm_page_t mold) { vm_page_t mret; bool dropped; VM_OBJECT_ASSERT_WLOCKED(object); vm_page_assert_xbusied(mold); KASSERT(mnew->object == NULL && (mnew->ref_count & VPRC_OBJREF) == 0, ("vm_page_replace: page %p already in object", mnew)); /* * This function mostly follows vm_page_insert() and * vm_page_remove() without the radix, object count and vnode * dance. Double check such functions for more comments. */ mnew->object = object; mnew->pindex = pindex; atomic_set_int(&mnew->ref_count, VPRC_OBJREF); mret = vm_radix_replace(&object->rtree, mnew); KASSERT(mret == mold, ("invalid page replacement, mold=%p, mret=%p", mold, mret)); KASSERT((mold->oflags & VPO_UNMANAGED) == (mnew->oflags & VPO_UNMANAGED), ("vm_page_replace: mismatched VPO_UNMANAGED")); /* Keep the resident page list in sorted order. */ TAILQ_INSERT_AFTER(&object->memq, mold, mnew, listq); TAILQ_REMOVE(&object->memq, mold, listq); mold->object = NULL; /* * The object's resident_page_count does not change because we have * swapped one page for another, but the generation count should * change if the page is dirty. */ if (pmap_page_is_write_mapped(mnew)) vm_object_set_writeable_dirty(object); dropped = vm_page_drop(mold, VPRC_OBJREF) == VPRC_OBJREF; vm_page_xunbusy(mold); return (dropped); } void vm_page_replace(vm_page_t mnew, vm_object_t object, vm_pindex_t pindex, vm_page_t mold) { vm_page_assert_xbusied(mnew); if (vm_page_replace_hold(mnew, object, pindex, mold)) vm_page_free(mold); } /* * vm_page_rename: * * Move the given memory entry from its * current object to the specified target object/offset. * * Note: swap associated with the page must be invalidated by the move. We * have to do this for several reasons: (1) we aren't freeing the * page, (2) we are dirtying the page, (3) the VM system is probably * moving the page from object A to B, and will then later move * the backing store from A to B and we can't have a conflict. * * Note: we *always* dirty the page. It is necessary both for the * fact that we moved it, and because we may be invalidating * swap. * * The objects must be locked. */ int vm_page_rename(vm_page_t m, vm_object_t new_object, vm_pindex_t new_pindex) { vm_page_t mpred; vm_pindex_t opidx; VM_OBJECT_ASSERT_WLOCKED(new_object); KASSERT(m->ref_count != 0, ("vm_page_rename: page %p has no refs", m)); mpred = vm_radix_lookup_le(&new_object->rtree, new_pindex); KASSERT(mpred == NULL || mpred->pindex != new_pindex, ("vm_page_rename: pindex already renamed")); /* * Create a custom version of vm_page_insert() which does not depend * by m_prev and can cheat on the implementation aspects of the * function. */ opidx = m->pindex; m->pindex = new_pindex; if (vm_radix_insert(&new_object->rtree, m)) { m->pindex = opidx; return (1); } /* * The operation cannot fail anymore. The removal must happen before * the listq iterator is tainted. */ m->pindex = opidx; vm_page_object_remove(m); /* Return back to the new pindex to complete vm_page_insert(). */ m->pindex = new_pindex; m->object = new_object; vm_page_insert_radixdone(m, new_object, mpred); vm_page_dirty(m); return (0); } /* * vm_page_alloc: * * Allocate and return a page that is associated with the specified * object and offset pair. By default, this page is exclusive busied. * * The caller must always specify an allocation class. * * allocation classes: * VM_ALLOC_NORMAL normal process request * VM_ALLOC_SYSTEM system *really* needs a page * VM_ALLOC_INTERRUPT interrupt time request * * optional allocation flags: * VM_ALLOC_COUNT(number) the number of additional pages that the caller * intends to allocate * VM_ALLOC_NOBUSY do not exclusive busy the page * VM_ALLOC_NODUMP do not include the page in a kernel core dump * VM_ALLOC_NOOBJ page is not associated with an object and * should not be exclusive busy * VM_ALLOC_SBUSY shared busy the allocated page * VM_ALLOC_WIRED wire the allocated page * VM_ALLOC_ZERO prefer a zeroed page */ vm_page_t vm_page_alloc(vm_object_t object, vm_pindex_t pindex, int req) { return (vm_page_alloc_after(object, pindex, req, object != NULL ? vm_radix_lookup_le(&object->rtree, pindex) : NULL)); } vm_page_t vm_page_alloc_domain(vm_object_t object, vm_pindex_t pindex, int domain, int req) { return (vm_page_alloc_domain_after(object, pindex, domain, req, object != NULL ? vm_radix_lookup_le(&object->rtree, pindex) : NULL)); } /* * Allocate a page in the specified object with the given page index. To * optimize insertion of the page into the object, the caller must also specifiy * the resident page in the object with largest index smaller than the given * page index, or NULL if no such page exists. */ vm_page_t vm_page_alloc_after(vm_object_t object, vm_pindex_t pindex, int req, vm_page_t mpred) { struct vm_domainset_iter di; vm_page_t m; int domain; vm_domainset_iter_page_init(&di, object, pindex, &domain, &req); do { m = vm_page_alloc_domain_after(object, pindex, domain, req, mpred); if (m != NULL) break; } while (vm_domainset_iter_page(&di, object, &domain) == 0); return (m); } /* * Returns true if the number of free pages exceeds the minimum * for the request class and false otherwise. */ static int _vm_domain_allocate(struct vm_domain *vmd, int req_class, int npages) { u_int limit, old, new; if (req_class == VM_ALLOC_INTERRUPT) limit = 0; else if (req_class == VM_ALLOC_SYSTEM) limit = vmd->vmd_interrupt_free_min; else limit = vmd->vmd_free_reserved; /* * Attempt to reserve the pages. Fail if we're below the limit. */ limit += npages; old = vmd->vmd_free_count; do { if (old < limit) return (0); new = old - npages; } while (atomic_fcmpset_int(&vmd->vmd_free_count, &old, new) == 0); /* Wake the page daemon if we've crossed the threshold. */ if (vm_paging_needed(vmd, new) && !vm_paging_needed(vmd, old)) pagedaemon_wakeup(vmd->vmd_domain); /* Only update bitsets on transitions. */ if ((old >= vmd->vmd_free_min && new < vmd->vmd_free_min) || (old >= vmd->vmd_free_severe && new < vmd->vmd_free_severe)) vm_domain_set(vmd); return (1); } int vm_domain_allocate(struct vm_domain *vmd, int req, int npages) { int req_class; /* * The page daemon is allowed to dig deeper into the free page list. */ req_class = req & VM_ALLOC_CLASS_MASK; if (curproc == pageproc && req_class != VM_ALLOC_INTERRUPT) req_class = VM_ALLOC_SYSTEM; return (_vm_domain_allocate(vmd, req_class, npages)); } vm_page_t vm_page_alloc_domain_after(vm_object_t object, vm_pindex_t pindex, int domain, int req, vm_page_t mpred) { struct vm_domain *vmd; vm_page_t m; int flags, pool; KASSERT((object != NULL) == ((req & VM_ALLOC_NOOBJ) == 0) && (object != NULL || (req & VM_ALLOC_SBUSY) == 0) && ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) != (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)), ("inconsistent object(%p)/req(%x)", object, req)); KASSERT(object == NULL || (req & VM_ALLOC_WAITOK) == 0, ("Can't sleep and retry object insertion.")); KASSERT(mpred == NULL || mpred->pindex < pindex, ("mpred %p doesn't precede pindex 0x%jx", mpred, (uintmax_t)pindex)); if (object != NULL) VM_OBJECT_ASSERT_WLOCKED(object); flags = 0; m = NULL; pool = object != NULL ? VM_FREEPOOL_DEFAULT : VM_FREEPOOL_DIRECT; again: #if VM_NRESERVLEVEL > 0 /* * Can we allocate the page from a reservation? */ if (vm_object_reserv(object) && (m = vm_reserv_alloc_page(object, pindex, domain, req, mpred)) != NULL) { domain = vm_phys_domain(m); vmd = VM_DOMAIN(domain); goto found; } #endif vmd = VM_DOMAIN(domain); if (vmd->vmd_pgcache[pool].zone != NULL) { m = uma_zalloc(vmd->vmd_pgcache[pool].zone, M_NOWAIT | M_NOVM); if (m != NULL) { flags |= PG_PCPU_CACHE; goto found; } } if (vm_domain_allocate(vmd, req, 1)) { /* * If not, allocate it from the free page queues. */ vm_domain_free_lock(vmd); m = vm_phys_alloc_pages(domain, pool, 0); vm_domain_free_unlock(vmd); if (m == NULL) { vm_domain_freecnt_inc(vmd, 1); #if VM_NRESERVLEVEL > 0 if (vm_reserv_reclaim_inactive(domain)) goto again; #endif } } if (m == NULL) { /* * Not allocatable, give up. */ if (vm_domain_alloc_fail(vmd, object, req)) goto again; return (NULL); } /* * At this point we had better have found a good page. */ found: vm_page_dequeue(m); vm_page_alloc_check(m); /* * Initialize the page. Only the PG_ZERO flag is inherited. */ if ((req & VM_ALLOC_ZERO) != 0) flags |= (m->flags & PG_ZERO); if ((req & VM_ALLOC_NODUMP) != 0) flags |= PG_NODUMP; m->flags = flags; m->a.flags = 0; m->oflags = object == NULL || (object->flags & OBJ_UNMANAGED) != 0 ? VPO_UNMANAGED : 0; if ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_NOOBJ | VM_ALLOC_SBUSY)) == 0) m->busy_lock = VPB_CURTHREAD_EXCLUSIVE; else if ((req & VM_ALLOC_SBUSY) != 0) m->busy_lock = VPB_SHARERS_WORD(1); else m->busy_lock = VPB_UNBUSIED; if (req & VM_ALLOC_WIRED) { vm_wire_add(1); m->ref_count = 1; } m->a.act_count = 0; if (object != NULL) { if (vm_page_insert_after(m, object, pindex, mpred)) { if (req & VM_ALLOC_WIRED) { vm_wire_sub(1); m->ref_count = 0; } KASSERT(m->object == NULL, ("page %p has object", m)); m->oflags = VPO_UNMANAGED; m->busy_lock = VPB_UNBUSIED; /* Don't change PG_ZERO. */ vm_page_free_toq(m); if (req & VM_ALLOC_WAITFAIL) { VM_OBJECT_WUNLOCK(object); vm_radix_wait(); VM_OBJECT_WLOCK(object); } return (NULL); } /* Ignore device objects; the pager sets "memattr" for them. */ if (object->memattr != VM_MEMATTR_DEFAULT && (object->flags & OBJ_FICTITIOUS) == 0) pmap_page_set_memattr(m, object->memattr); } else m->pindex = pindex; return (m); } /* * vm_page_alloc_contig: * * Allocate a contiguous set of physical pages of the given size "npages" * from the free lists. All of the physical pages must be at or above * the given physical address "low" and below the given physical address * "high". The given value "alignment" determines the alignment of the * first physical page in the set. If the given value "boundary" is * non-zero, then the set of physical pages cannot cross any physical * address boundary that is a multiple of that value. Both "alignment" * and "boundary" must be a power of two. * * If the specified memory attribute, "memattr", is VM_MEMATTR_DEFAULT, * then the memory attribute setting for the physical pages is configured * to the object's memory attribute setting. Otherwise, the memory * attribute setting for the physical pages is configured to "memattr", * overriding the object's memory attribute setting. However, if the * object's memory attribute setting is not VM_MEMATTR_DEFAULT, then the * memory attribute setting for the physical pages cannot be configured * to VM_MEMATTR_DEFAULT. * * The specified object may not contain fictitious pages. * * The caller must always specify an allocation class. * * allocation classes: * VM_ALLOC_NORMAL normal process request * VM_ALLOC_SYSTEM system *really* needs a page * VM_ALLOC_INTERRUPT interrupt time request * * optional allocation flags: * VM_ALLOC_NOBUSY do not exclusive busy the page * VM_ALLOC_NODUMP do not include the page in a kernel core dump * VM_ALLOC_NOOBJ page is not associated with an object and * should not be exclusive busy * VM_ALLOC_SBUSY shared busy the allocated page * VM_ALLOC_WIRED wire the allocated page * VM_ALLOC_ZERO prefer a zeroed page */ vm_page_t vm_page_alloc_contig(vm_object_t object, vm_pindex_t pindex, int req, u_long npages, vm_paddr_t low, vm_paddr_t high, u_long alignment, vm_paddr_t boundary, vm_memattr_t memattr) { struct vm_domainset_iter di; vm_page_t m; int domain; vm_domainset_iter_page_init(&di, object, pindex, &domain, &req); do { m = vm_page_alloc_contig_domain(object, pindex, domain, req, npages, low, high, alignment, boundary, memattr); if (m != NULL) break; } while (vm_domainset_iter_page(&di, object, &domain) == 0); return (m); } vm_page_t vm_page_alloc_contig_domain(vm_object_t object, vm_pindex_t pindex, int domain, int req, u_long npages, vm_paddr_t low, vm_paddr_t high, u_long alignment, vm_paddr_t boundary, vm_memattr_t memattr) { struct vm_domain *vmd; vm_page_t m, m_ret, mpred; u_int busy_lock, flags, oflags; mpred = NULL; /* XXX: pacify gcc */ KASSERT((object != NULL) == ((req & VM_ALLOC_NOOBJ) == 0) && (object != NULL || (req & VM_ALLOC_SBUSY) == 0) && ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) != (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)), ("vm_page_alloc_contig: inconsistent object(%p)/req(%x)", object, req)); KASSERT(object == NULL || (req & VM_ALLOC_WAITOK) == 0, ("Can't sleep and retry object insertion.")); if (object != NULL) { VM_OBJECT_ASSERT_WLOCKED(object); KASSERT((object->flags & OBJ_FICTITIOUS) == 0, ("vm_page_alloc_contig: object %p has fictitious pages", object)); } KASSERT(npages > 0, ("vm_page_alloc_contig: npages is zero")); if (object != NULL) { mpred = vm_radix_lookup_le(&object->rtree, pindex); KASSERT(mpred == NULL || mpred->pindex != pindex, ("vm_page_alloc_contig: pindex already allocated")); } /* * Can we allocate the pages without the number of free pages falling * below the lower bound for the allocation class? */ m_ret = NULL; again: #if VM_NRESERVLEVEL > 0 /* * Can we allocate the pages from a reservation? */ if (vm_object_reserv(object) && (m_ret = vm_reserv_alloc_contig(object, pindex, domain, req, mpred, npages, low, high, alignment, boundary)) != NULL) { domain = vm_phys_domain(m_ret); vmd = VM_DOMAIN(domain); goto found; } #endif vmd = VM_DOMAIN(domain); if (vm_domain_allocate(vmd, req, npages)) { /* * allocate them from the free page queues. */ vm_domain_free_lock(vmd); m_ret = vm_phys_alloc_contig(domain, npages, low, high, alignment, boundary); vm_domain_free_unlock(vmd); if (m_ret == NULL) { vm_domain_freecnt_inc(vmd, npages); #if VM_NRESERVLEVEL > 0 if (vm_reserv_reclaim_contig(domain, npages, low, high, alignment, boundary)) goto again; #endif } } if (m_ret == NULL) { if (vm_domain_alloc_fail(vmd, object, req)) goto again; return (NULL); } #if VM_NRESERVLEVEL > 0 found: #endif for (m = m_ret; m < &m_ret[npages]; m++) { vm_page_dequeue(m); vm_page_alloc_check(m); } /* * Initialize the pages. Only the PG_ZERO flag is inherited. */ flags = 0; if ((req & VM_ALLOC_ZERO) != 0) flags = PG_ZERO; if ((req & VM_ALLOC_NODUMP) != 0) flags |= PG_NODUMP; oflags = object == NULL || (object->flags & OBJ_UNMANAGED) != 0 ? VPO_UNMANAGED : 0; if ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_NOOBJ | VM_ALLOC_SBUSY)) == 0) busy_lock = VPB_CURTHREAD_EXCLUSIVE; else if ((req & VM_ALLOC_SBUSY) != 0) busy_lock = VPB_SHARERS_WORD(1); else busy_lock = VPB_UNBUSIED; if ((req & VM_ALLOC_WIRED) != 0) vm_wire_add(npages); if (object != NULL) { if (object->memattr != VM_MEMATTR_DEFAULT && memattr == VM_MEMATTR_DEFAULT) memattr = object->memattr; } for (m = m_ret; m < &m_ret[npages]; m++) { m->a.flags = 0; m->flags = (m->flags | PG_NODUMP) & flags; m->busy_lock = busy_lock; if ((req & VM_ALLOC_WIRED) != 0) m->ref_count = 1; m->a.act_count = 0; m->oflags = oflags; if (object != NULL) { if (vm_page_insert_after(m, object, pindex, mpred)) { if ((req & VM_ALLOC_WIRED) != 0) vm_wire_sub(npages); KASSERT(m->object == NULL, ("page %p has object", m)); mpred = m; for (m = m_ret; m < &m_ret[npages]; m++) { if (m <= mpred && (req & VM_ALLOC_WIRED) != 0) m->ref_count = 0; m->oflags = VPO_UNMANAGED; m->busy_lock = VPB_UNBUSIED; /* Don't change PG_ZERO. */ vm_page_free_toq(m); } if (req & VM_ALLOC_WAITFAIL) { VM_OBJECT_WUNLOCK(object); vm_radix_wait(); VM_OBJECT_WLOCK(object); } return (NULL); } mpred = m; } else m->pindex = pindex; if (memattr != VM_MEMATTR_DEFAULT) pmap_page_set_memattr(m, memattr); pindex++; } return (m_ret); } /* * Check a page that has been freshly dequeued from a freelist. */ static void vm_page_alloc_check(vm_page_t m) { KASSERT(m->object == NULL, ("page %p has object", m)); KASSERT(m->a.queue == PQ_NONE && (m->a.flags & PGA_QUEUE_STATE_MASK) == 0, ("page %p has unexpected queue %d, flags %#x", m, m->a.queue, (m->a.flags & PGA_QUEUE_STATE_MASK))); KASSERT(m->ref_count == 0, ("page %p has references", m)); KASSERT(vm_page_busy_freed(m), ("page %p is not freed", m)); KASSERT(m->dirty == 0, ("page %p is dirty", m)); KASSERT(pmap_page_get_memattr(m) == VM_MEMATTR_DEFAULT, ("page %p has unexpected memattr %d", m, pmap_page_get_memattr(m))); KASSERT(m->valid == 0, ("free page %p is valid", m)); } /* * vm_page_alloc_freelist: * * Allocate a physical page from the specified free page list. * * The caller must always specify an allocation class. * * allocation classes: * VM_ALLOC_NORMAL normal process request * VM_ALLOC_SYSTEM system *really* needs a page * VM_ALLOC_INTERRUPT interrupt time request * * optional allocation flags: * VM_ALLOC_COUNT(number) the number of additional pages that the caller * intends to allocate * VM_ALLOC_WIRED wire the allocated page * VM_ALLOC_ZERO prefer a zeroed page */ vm_page_t vm_page_alloc_freelist(int freelist, int req) { struct vm_domainset_iter di; vm_page_t m; int domain; vm_domainset_iter_page_init(&di, NULL, 0, &domain, &req); do { m = vm_page_alloc_freelist_domain(domain, freelist, req); if (m != NULL) break; } while (vm_domainset_iter_page(&di, NULL, &domain) == 0); return (m); } vm_page_t vm_page_alloc_freelist_domain(int domain, int freelist, int req) { struct vm_domain *vmd; vm_page_t m; u_int flags; m = NULL; vmd = VM_DOMAIN(domain); again: if (vm_domain_allocate(vmd, req, 1)) { vm_domain_free_lock(vmd); m = vm_phys_alloc_freelist_pages(domain, freelist, VM_FREEPOOL_DIRECT, 0); vm_domain_free_unlock(vmd); if (m == NULL) vm_domain_freecnt_inc(vmd, 1); } if (m == NULL) { if (vm_domain_alloc_fail(vmd, NULL, req)) goto again; return (NULL); } vm_page_dequeue(m); vm_page_alloc_check(m); /* * Initialize the page. Only the PG_ZERO flag is inherited. */ m->a.flags = 0; flags = 0; if ((req & VM_ALLOC_ZERO) != 0) flags = PG_ZERO; m->flags &= flags; if ((req & VM_ALLOC_WIRED) != 0) { vm_wire_add(1); m->ref_count = 1; } /* Unmanaged pages don't use "act_count". */ m->oflags = VPO_UNMANAGED; return (m); } static int vm_page_zone_import(void *arg, void **store, int cnt, int domain, int flags) { struct vm_domain *vmd; struct vm_pgcache *pgcache; int i; pgcache = arg; vmd = VM_DOMAIN(pgcache->domain); /* * The page daemon should avoid creating extra memory pressure since its * main purpose is to replenish the store of free pages. */ if (vmd->vmd_severeset || curproc == pageproc || !_vm_domain_allocate(vmd, VM_ALLOC_NORMAL, cnt)) return (0); domain = vmd->vmd_domain; vm_domain_free_lock(vmd); i = vm_phys_alloc_npages(domain, pgcache->pool, cnt, (vm_page_t *)store); vm_domain_free_unlock(vmd); if (cnt != i) vm_domain_freecnt_inc(vmd, cnt - i); return (i); } static void vm_page_zone_release(void *arg, void **store, int cnt) { struct vm_domain *vmd; struct vm_pgcache *pgcache; vm_page_t m; int i; pgcache = arg; vmd = VM_DOMAIN(pgcache->domain); vm_domain_free_lock(vmd); for (i = 0; i < cnt; i++) { m = (vm_page_t)store[i]; vm_phys_free_pages(m, 0); } vm_domain_free_unlock(vmd); vm_domain_freecnt_inc(vmd, cnt); } #define VPSC_ANY 0 /* No restrictions. */ #define VPSC_NORESERV 1 /* Skip reservations; implies VPSC_NOSUPER. */ #define VPSC_NOSUPER 2 /* Skip superpages. */ /* * vm_page_scan_contig: * * Scan vm_page_array[] between the specified entries "m_start" and * "m_end" for a run of contiguous physical pages that satisfy the * specified conditions, and return the lowest page in the run. The * specified "alignment" determines the alignment of the lowest physical * page in the run. If the specified "boundary" is non-zero, then the * run of physical pages cannot span a physical address that is a * multiple of "boundary". * * "m_end" is never dereferenced, so it need not point to a vm_page * structure within vm_page_array[]. * * "npages" must be greater than zero. "m_start" and "m_end" must not * span a hole (or discontiguity) in the physical address space. Both * "alignment" and "boundary" must be a power of two. */ vm_page_t vm_page_scan_contig(u_long npages, vm_page_t m_start, vm_page_t m_end, u_long alignment, vm_paddr_t boundary, int options) { vm_object_t object; vm_paddr_t pa; vm_page_t m, m_run; #if VM_NRESERVLEVEL > 0 int level; #endif int m_inc, order, run_ext, run_len; KASSERT(npages > 0, ("npages is 0")); KASSERT(powerof2(alignment), ("alignment is not a power of 2")); KASSERT(powerof2(boundary), ("boundary is not a power of 2")); m_run = NULL; run_len = 0; for (m = m_start; m < m_end && run_len < npages; m += m_inc) { KASSERT((m->flags & PG_MARKER) == 0, ("page %p is PG_MARKER", m)); KASSERT((m->flags & PG_FICTITIOUS) == 0 || m->ref_count >= 1, ("fictitious page %p has invalid ref count", m)); /* * If the current page would be the start of a run, check its * physical address against the end, alignment, and boundary * conditions. If it doesn't satisfy these conditions, either * terminate the scan or advance to the next page that * satisfies the failed condition. */ if (run_len == 0) { KASSERT(m_run == NULL, ("m_run != NULL")); if (m + npages > m_end) break; pa = VM_PAGE_TO_PHYS(m); if ((pa & (alignment - 1)) != 0) { m_inc = atop(roundup2(pa, alignment) - pa); continue; } if (rounddown2(pa ^ (pa + ptoa(npages) - 1), boundary) != 0) { m_inc = atop(roundup2(pa, boundary) - pa); continue; } } else KASSERT(m_run != NULL, ("m_run == NULL")); retry: m_inc = 1; if (vm_page_wired(m)) run_ext = 0; #if VM_NRESERVLEVEL > 0 else if ((level = vm_reserv_level(m)) >= 0 && (options & VPSC_NORESERV) != 0) { run_ext = 0; /* Advance to the end of the reservation. */ pa = VM_PAGE_TO_PHYS(m); m_inc = atop(roundup2(pa + 1, vm_reserv_size(level)) - pa); } #endif else if ((object = atomic_load_ptr(&m->object)) != NULL) { /* * The page is considered eligible for relocation if * and only if it could be laundered or reclaimed by * the page daemon. */ VM_OBJECT_RLOCK(object); if (object != m->object) { VM_OBJECT_RUNLOCK(object); goto retry; } /* Don't care: PG_NODUMP, PG_ZERO. */ if (object->type != OBJT_DEFAULT && object->type != OBJT_SWAP && object->type != OBJT_VNODE) { run_ext = 0; #if VM_NRESERVLEVEL > 0 } else if ((options & VPSC_NOSUPER) != 0 && (level = vm_reserv_level_iffullpop(m)) >= 0) { run_ext = 0; /* Advance to the end of the superpage. */ pa = VM_PAGE_TO_PHYS(m); m_inc = atop(roundup2(pa + 1, vm_reserv_size(level)) - pa); #endif } else if (object->memattr == VM_MEMATTR_DEFAULT && vm_page_queue(m) != PQ_NONE && !vm_page_busied(m)) { /* * The page is allocated but eligible for * relocation. Extend the current run by one * page. */ KASSERT(pmap_page_get_memattr(m) == VM_MEMATTR_DEFAULT, ("page %p has an unexpected memattr", m)); KASSERT((m->oflags & (VPO_SWAPINPROG | VPO_SWAPSLEEP | VPO_UNMANAGED)) == 0, ("page %p has unexpected oflags", m)); /* Don't care: PGA_NOSYNC. */ run_ext = 1; } else run_ext = 0; VM_OBJECT_RUNLOCK(object); #if VM_NRESERVLEVEL > 0 } else if (level >= 0) { /* * The page is reserved but not yet allocated. In * other words, it is still free. Extend the current * run by one page. */ run_ext = 1; #endif } else if ((order = m->order) < VM_NFREEORDER) { /* * The page is enqueued in the physical memory * allocator's free page queues. Moreover, it is the * first page in a power-of-two-sized run of * contiguous free pages. Add these pages to the end * of the current run, and jump ahead. */ run_ext = 1 << order; m_inc = 1 << order; } else { /* * Skip the page for one of the following reasons: (1) * It is enqueued in the physical memory allocator's * free page queues. However, it is not the first * page in a run of contiguous free pages. (This case * rarely occurs because the scan is performed in * ascending order.) (2) It is not reserved, and it is * transitioning from free to allocated. (Conversely, * the transition from allocated to free for managed * pages is blocked by the page lock.) (3) It is * allocated but not contained by an object and not * wired, e.g., allocated by Xen's balloon driver. */ run_ext = 0; } /* * Extend or reset the current run of pages. */ if (run_ext > 0) { if (run_len == 0) m_run = m; run_len += run_ext; } else { if (run_len > 0) { m_run = NULL; run_len = 0; } } } if (run_len >= npages) return (m_run); return (NULL); } /* * vm_page_reclaim_run: * * Try to relocate each of the allocated virtual pages within the * specified run of physical pages to a new physical address. Free the * physical pages underlying the relocated virtual pages. A virtual page * is relocatable if and only if it could be laundered or reclaimed by * the page daemon. Whenever possible, a virtual page is relocated to a * physical address above "high". * * Returns 0 if every physical page within the run was already free or * just freed by a successful relocation. Otherwise, returns a non-zero * value indicating why the last attempt to relocate a virtual page was * unsuccessful. * * "req_class" must be an allocation class. */ static int vm_page_reclaim_run(int req_class, int domain, u_long npages, vm_page_t m_run, vm_paddr_t high) { struct vm_domain *vmd; struct spglist free; vm_object_t object; vm_paddr_t pa; vm_page_t m, m_end, m_new; int error, order, req; KASSERT((req_class & VM_ALLOC_CLASS_MASK) == req_class, ("req_class is not an allocation class")); SLIST_INIT(&free); error = 0; m = m_run; m_end = m_run + npages; for (; error == 0 && m < m_end; m++) { KASSERT((m->flags & (PG_FICTITIOUS | PG_MARKER)) == 0, ("page %p is PG_FICTITIOUS or PG_MARKER", m)); /* * Racily check for wirings. Races are handled once the object * lock is held and the page is unmapped. */ if (vm_page_wired(m)) error = EBUSY; else if ((object = atomic_load_ptr(&m->object)) != NULL) { /* * The page is relocated if and only if it could be * laundered or reclaimed by the page daemon. */ VM_OBJECT_WLOCK(object); /* Don't care: PG_NODUMP, PG_ZERO. */ if (m->object != object || (object->type != OBJT_DEFAULT && object->type != OBJT_SWAP && object->type != OBJT_VNODE)) error = EINVAL; else if (object->memattr != VM_MEMATTR_DEFAULT) error = EINVAL; else if (vm_page_queue(m) != PQ_NONE && vm_page_tryxbusy(m) != 0) { if (vm_page_wired(m)) { vm_page_xunbusy(m); error = EBUSY; goto unlock; } KASSERT(pmap_page_get_memattr(m) == VM_MEMATTR_DEFAULT, ("page %p has an unexpected memattr", m)); KASSERT(m->oflags == 0, ("page %p has unexpected oflags", m)); /* Don't care: PGA_NOSYNC. */ if (!vm_page_none_valid(m)) { /* * First, try to allocate a new page * that is above "high". Failing * that, try to allocate a new page * that is below "m_run". Allocate * the new page between the end of * "m_run" and "high" only as a last * resort. */ req = req_class | VM_ALLOC_NOOBJ; if ((m->flags & PG_NODUMP) != 0) req |= VM_ALLOC_NODUMP; if (trunc_page(high) != ~(vm_paddr_t)PAGE_MASK) { m_new = vm_page_alloc_contig( NULL, 0, req, 1, round_page(high), ~(vm_paddr_t)0, PAGE_SIZE, 0, VM_MEMATTR_DEFAULT); } else m_new = NULL; if (m_new == NULL) { pa = VM_PAGE_TO_PHYS(m_run); m_new = vm_page_alloc_contig( NULL, 0, req, 1, 0, pa - 1, PAGE_SIZE, 0, VM_MEMATTR_DEFAULT); } if (m_new == NULL) { pa += ptoa(npages); m_new = vm_page_alloc_contig( NULL, 0, req, 1, pa, high, PAGE_SIZE, 0, VM_MEMATTR_DEFAULT); } if (m_new == NULL) { vm_page_xunbusy(m); error = ENOMEM; goto unlock; } /* * Unmap the page and check for new * wirings that may have been acquired * through a pmap lookup. */ if (object->ref_count != 0 && !vm_page_try_remove_all(m)) { vm_page_xunbusy(m); vm_page_free(m_new); error = EBUSY; goto unlock; } /* * Replace "m" with the new page. For * vm_page_replace(), "m" must be busy * and dequeued. Finally, change "m" * as if vm_page_free() was called. */ m_new->a.flags = m->a.flags & ~PGA_QUEUE_STATE_MASK; KASSERT(m_new->oflags == VPO_UNMANAGED, ("page %p is managed", m_new)); m_new->oflags = 0; pmap_copy_page(m, m_new); m_new->valid = m->valid; m_new->dirty = m->dirty; m->flags &= ~PG_ZERO; vm_page_dequeue(m); if (vm_page_replace_hold(m_new, object, m->pindex, m) && vm_page_free_prep(m)) SLIST_INSERT_HEAD(&free, m, plinks.s.ss); /* * The new page must be deactivated * before the object is unlocked. */ vm_page_deactivate(m_new); } else { m->flags &= ~PG_ZERO; vm_page_dequeue(m); if (vm_page_free_prep(m)) SLIST_INSERT_HEAD(&free, m, plinks.s.ss); KASSERT(m->dirty == 0, ("page %p is dirty", m)); } } else error = EBUSY; unlock: VM_OBJECT_WUNLOCK(object); } else { MPASS(vm_phys_domain(m) == domain); vmd = VM_DOMAIN(domain); vm_domain_free_lock(vmd); order = m->order; if (order < VM_NFREEORDER) { /* * The page is enqueued in the physical memory * allocator's free page queues. Moreover, it * is the first page in a power-of-two-sized * run of contiguous free pages. Jump ahead * to the last page within that run, and * continue from there. */ m += (1 << order) - 1; } #if VM_NRESERVLEVEL > 0 else if (vm_reserv_is_page_free(m)) order = 0; #endif vm_domain_free_unlock(vmd); if (order == VM_NFREEORDER) error = EINVAL; } } if ((m = SLIST_FIRST(&free)) != NULL) { int cnt; vmd = VM_DOMAIN(domain); cnt = 0; vm_domain_free_lock(vmd); do { MPASS(vm_phys_domain(m) == domain); SLIST_REMOVE_HEAD(&free, plinks.s.ss); vm_phys_free_pages(m, 0); cnt++; } while ((m = SLIST_FIRST(&free)) != NULL); vm_domain_free_unlock(vmd); vm_domain_freecnt_inc(vmd, cnt); } return (error); } #define NRUNS 16 CTASSERT(powerof2(NRUNS)); #define RUN_INDEX(count) ((count) & (NRUNS - 1)) #define MIN_RECLAIM 8 /* * vm_page_reclaim_contig: * * Reclaim allocated, contiguous physical memory satisfying the specified * conditions by relocating the virtual pages using that physical memory. * Returns true if reclamation is successful and false otherwise. Since * relocation requires the allocation of physical pages, reclamation may * fail due to a shortage of free pages. When reclamation fails, callers * are expected to perform vm_wait() before retrying a failed allocation * operation, e.g., vm_page_alloc_contig(). * * The caller must always specify an allocation class through "req". * * allocation classes: * VM_ALLOC_NORMAL normal process request * VM_ALLOC_SYSTEM system *really* needs a page * VM_ALLOC_INTERRUPT interrupt time request * * The optional allocation flags are ignored. * * "npages" must be greater than zero. Both "alignment" and "boundary" * must be a power of two. */ bool vm_page_reclaim_contig_domain(int domain, int req, u_long npages, vm_paddr_t low, vm_paddr_t high, u_long alignment, vm_paddr_t boundary) { struct vm_domain *vmd; vm_paddr_t curr_low; vm_page_t m_run, m_runs[NRUNS]; u_long count, reclaimed; int error, i, options, req_class; KASSERT(npages > 0, ("npages is 0")); KASSERT(powerof2(alignment), ("alignment is not a power of 2")); KASSERT(powerof2(boundary), ("boundary is not a power of 2")); req_class = req & VM_ALLOC_CLASS_MASK; /* * The page daemon is allowed to dig deeper into the free page list. */ if (curproc == pageproc && req_class != VM_ALLOC_INTERRUPT) req_class = VM_ALLOC_SYSTEM; /* * Return if the number of free pages cannot satisfy the requested * allocation. */ vmd = VM_DOMAIN(domain); count = vmd->vmd_free_count; if (count < npages + vmd->vmd_free_reserved || (count < npages + vmd->vmd_interrupt_free_min && req_class == VM_ALLOC_SYSTEM) || (count < npages && req_class == VM_ALLOC_INTERRUPT)) return (false); /* * Scan up to three times, relaxing the restrictions ("options") on * the reclamation of reservations and superpages each time. */ for (options = VPSC_NORESERV;;) { /* * Find the highest runs that satisfy the given constraints * and restrictions, and record them in "m_runs". */ curr_low = low; count = 0; for (;;) { m_run = vm_phys_scan_contig(domain, npages, curr_low, high, alignment, boundary, options); if (m_run == NULL) break; curr_low = VM_PAGE_TO_PHYS(m_run) + ptoa(npages); m_runs[RUN_INDEX(count)] = m_run; count++; } /* * Reclaim the highest runs in LIFO (descending) order until * the number of reclaimed pages, "reclaimed", is at least * MIN_RECLAIM. Reset "reclaimed" each time because each * reclamation is idempotent, and runs will (likely) recur * from one scan to the next as restrictions are relaxed. */ reclaimed = 0; for (i = 0; count > 0 && i < NRUNS; i++) { count--; m_run = m_runs[RUN_INDEX(count)]; error = vm_page_reclaim_run(req_class, domain, npages, m_run, high); if (error == 0) { reclaimed += npages; if (reclaimed >= MIN_RECLAIM) return (true); } } /* * Either relax the restrictions on the next scan or return if * the last scan had no restrictions. */ if (options == VPSC_NORESERV) options = VPSC_NOSUPER; else if (options == VPSC_NOSUPER) options = VPSC_ANY; else if (options == VPSC_ANY) return (reclaimed != 0); } } bool vm_page_reclaim_contig(int req, u_long npages, vm_paddr_t low, vm_paddr_t high, u_long alignment, vm_paddr_t boundary) { struct vm_domainset_iter di; int domain; bool ret; vm_domainset_iter_page_init(&di, NULL, 0, &domain, &req); do { ret = vm_page_reclaim_contig_domain(domain, req, npages, low, high, alignment, boundary); if (ret) break; } while (vm_domainset_iter_page(&di, NULL, &domain) == 0); return (ret); } /* * Set the domain in the appropriate page level domainset. */ void vm_domain_set(struct vm_domain *vmd) { mtx_lock(&vm_domainset_lock); if (!vmd->vmd_minset && vm_paging_min(vmd)) { vmd->vmd_minset = 1; DOMAINSET_SET(vmd->vmd_domain, &vm_min_domains); } if (!vmd->vmd_severeset && vm_paging_severe(vmd)) { vmd->vmd_severeset = 1; DOMAINSET_SET(vmd->vmd_domain, &vm_severe_domains); } mtx_unlock(&vm_domainset_lock); } /* * Clear the domain from the appropriate page level domainset. */ void vm_domain_clear(struct vm_domain *vmd) { mtx_lock(&vm_domainset_lock); if (vmd->vmd_minset && !vm_paging_min(vmd)) { vmd->vmd_minset = 0; DOMAINSET_CLR(vmd->vmd_domain, &vm_min_domains); if (vm_min_waiters != 0) { vm_min_waiters = 0; wakeup(&vm_min_domains); } } if (vmd->vmd_severeset && !vm_paging_severe(vmd)) { vmd->vmd_severeset = 0; DOMAINSET_CLR(vmd->vmd_domain, &vm_severe_domains); if (vm_severe_waiters != 0) { vm_severe_waiters = 0; wakeup(&vm_severe_domains); } } /* * If pageout daemon needs pages, then tell it that there are * some free. */ if (vmd->vmd_pageout_pages_needed && vmd->vmd_free_count >= vmd->vmd_pageout_free_min) { wakeup(&vmd->vmd_pageout_pages_needed); vmd->vmd_pageout_pages_needed = 0; } /* See comments in vm_wait_doms(). */ if (vm_pageproc_waiters) { vm_pageproc_waiters = 0; wakeup(&vm_pageproc_waiters); } mtx_unlock(&vm_domainset_lock); } /* * Wait for free pages to exceed the min threshold globally. */ void vm_wait_min(void) { mtx_lock(&vm_domainset_lock); while (vm_page_count_min()) { vm_min_waiters++; msleep(&vm_min_domains, &vm_domainset_lock, PVM, "vmwait", 0); } mtx_unlock(&vm_domainset_lock); } /* * Wait for free pages to exceed the severe threshold globally. */ void vm_wait_severe(void) { mtx_lock(&vm_domainset_lock); while (vm_page_count_severe()) { vm_severe_waiters++; msleep(&vm_severe_domains, &vm_domainset_lock, PVM, "vmwait", 0); } mtx_unlock(&vm_domainset_lock); } u_int vm_wait_count(void) { return (vm_severe_waiters + vm_min_waiters + vm_pageproc_waiters); } void vm_wait_doms(const domainset_t *wdoms) { /* * We use racey wakeup synchronization to avoid expensive global * locking for the pageproc when sleeping with a non-specific vm_wait. * To handle this, we only sleep for one tick in this instance. It * is expected that most allocations for the pageproc will come from * kmem or vm_page_grab* which will use the more specific and * race-free vm_wait_domain(). */ if (curproc == pageproc) { mtx_lock(&vm_domainset_lock); vm_pageproc_waiters++; msleep(&vm_pageproc_waiters, &vm_domainset_lock, PVM | PDROP, "pageprocwait", 1); } else { /* * XXX Ideally we would wait only until the allocation could * be satisfied. This condition can cause new allocators to * consume all freed pages while old allocators wait. */ mtx_lock(&vm_domainset_lock); if (vm_page_count_min_set(wdoms)) { vm_min_waiters++; msleep(&vm_min_domains, &vm_domainset_lock, PVM | PDROP, "vmwait", 0); } else mtx_unlock(&vm_domainset_lock); } } /* * vm_wait_domain: * * Sleep until free pages are available for allocation. * - Called in various places after failed memory allocations. */ void vm_wait_domain(int domain) { struct vm_domain *vmd; domainset_t wdom; vmd = VM_DOMAIN(domain); vm_domain_free_assert_unlocked(vmd); if (curproc == pageproc) { mtx_lock(&vm_domainset_lock); if (vmd->vmd_free_count < vmd->vmd_pageout_free_min) { vmd->vmd_pageout_pages_needed = 1; msleep(&vmd->vmd_pageout_pages_needed, &vm_domainset_lock, PDROP | PSWP, "VMWait", 0); } else mtx_unlock(&vm_domainset_lock); } else { if (pageproc == NULL) panic("vm_wait in early boot"); DOMAINSET_ZERO(&wdom); DOMAINSET_SET(vmd->vmd_domain, &wdom); vm_wait_doms(&wdom); } } /* * vm_wait: * * Sleep until free pages are available for allocation in the * affinity domains of the obj. If obj is NULL, the domain set * for the calling thread is used. * Called in various places after failed memory allocations. */ void vm_wait(vm_object_t obj) { struct domainset *d; d = NULL; /* * Carefully fetch pointers only once: the struct domainset * itself is ummutable but the pointer might change. */ if (obj != NULL) d = obj->domain.dr_policy; if (d == NULL) d = curthread->td_domain.dr_policy; vm_wait_doms(&d->ds_mask); } /* * vm_domain_alloc_fail: * * Called when a page allocation function fails. Informs the * pagedaemon and performs the requested wait. Requires the * domain_free and object lock on entry. Returns with the * object lock held and free lock released. Returns an error when * retry is necessary. * */ static int vm_domain_alloc_fail(struct vm_domain *vmd, vm_object_t object, int req) { vm_domain_free_assert_unlocked(vmd); atomic_add_int(&vmd->vmd_pageout_deficit, max((u_int)req >> VM_ALLOC_COUNT_SHIFT, 1)); if (req & (VM_ALLOC_WAITOK | VM_ALLOC_WAITFAIL)) { if (object != NULL) VM_OBJECT_WUNLOCK(object); vm_wait_domain(vmd->vmd_domain); if (object != NULL) VM_OBJECT_WLOCK(object); if (req & VM_ALLOC_WAITOK) return (EAGAIN); } return (0); } /* * vm_waitpfault: * * Sleep until free pages are available for allocation. * - Called only in vm_fault so that processes page faulting * can be easily tracked. * - Sleeps at a lower priority than vm_wait() so that vm_wait()ing * processes will be able to grab memory first. Do not change * this balance without careful testing first. */ void vm_waitpfault(struct domainset *dset, int timo) { /* * XXX Ideally we would wait only until the allocation could * be satisfied. This condition can cause new allocators to * consume all freed pages while old allocators wait. */ mtx_lock(&vm_domainset_lock); if (vm_page_count_min_set(&dset->ds_mask)) { vm_min_waiters++; msleep(&vm_min_domains, &vm_domainset_lock, PUSER | PDROP, "pfault", timo); } else mtx_unlock(&vm_domainset_lock); } static struct vm_pagequeue * _vm_page_pagequeue(vm_page_t m, uint8_t queue) { return (&vm_pagequeue_domain(m)->vmd_pagequeues[queue]); } #ifdef INVARIANTS static struct vm_pagequeue * vm_page_pagequeue(vm_page_t m) { return (_vm_page_pagequeue(m, vm_page_astate_load(m).queue)); } #endif static __always_inline bool vm_page_pqstate_fcmpset(vm_page_t m, vm_page_astate_t *old, vm_page_astate_t new) { vm_page_astate_t tmp; tmp = *old; do { if (__predict_true(vm_page_astate_fcmpset(m, old, new))) return (true); counter_u64_add(pqstate_commit_retries, 1); } while (old->_bits == tmp._bits); return (false); } /* * Do the work of committing a queue state update that moves the page out of * its current queue. */ static bool _vm_page_pqstate_commit_dequeue(struct vm_pagequeue *pq, vm_page_t m, vm_page_astate_t *old, vm_page_astate_t new) { vm_page_t next; vm_pagequeue_assert_locked(pq); KASSERT(vm_page_pagequeue(m) == pq, ("%s: queue %p does not match page %p", __func__, pq, m)); KASSERT(old->queue != PQ_NONE && new.queue != old->queue, ("%s: invalid queue indices %d %d", __func__, old->queue, new.queue)); /* * Once the queue index of the page changes there is nothing * synchronizing with further updates to the page's physical * queue state. Therefore we must speculatively remove the page * from the queue now and be prepared to roll back if the queue * state update fails. If the page is not physically enqueued then * we just update its queue index. */ if ((old->flags & PGA_ENQUEUED) != 0) { new.flags &= ~PGA_ENQUEUED; next = TAILQ_NEXT(m, plinks.q); TAILQ_REMOVE(&pq->pq_pl, m, plinks.q); vm_pagequeue_cnt_dec(pq); if (!vm_page_pqstate_fcmpset(m, old, new)) { if (next == NULL) TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q); else TAILQ_INSERT_BEFORE(next, m, plinks.q); vm_pagequeue_cnt_inc(pq); return (false); } else { return (true); } } else { return (vm_page_pqstate_fcmpset(m, old, new)); } } static bool vm_page_pqstate_commit_dequeue(vm_page_t m, vm_page_astate_t *old, vm_page_astate_t new) { struct vm_pagequeue *pq; vm_page_astate_t as; bool ret; pq = _vm_page_pagequeue(m, old->queue); /* * The queue field and PGA_ENQUEUED flag are stable only so long as the * corresponding page queue lock is held. */ vm_pagequeue_lock(pq); as = vm_page_astate_load(m); if (__predict_false(as._bits != old->_bits)) { *old = as; ret = false; } else { ret = _vm_page_pqstate_commit_dequeue(pq, m, old, new); } vm_pagequeue_unlock(pq); return (ret); } /* * Commit a queue state update that enqueues or requeues a page. */ static bool _vm_page_pqstate_commit_requeue(struct vm_pagequeue *pq, vm_page_t m, vm_page_astate_t *old, vm_page_astate_t new) { struct vm_domain *vmd; vm_pagequeue_assert_locked(pq); KASSERT(old->queue != PQ_NONE && new.queue == old->queue, ("%s: invalid queue indices %d %d", __func__, old->queue, new.queue)); new.flags |= PGA_ENQUEUED; if (!vm_page_pqstate_fcmpset(m, old, new)) return (false); if ((old->flags & PGA_ENQUEUED) != 0) TAILQ_REMOVE(&pq->pq_pl, m, plinks.q); else vm_pagequeue_cnt_inc(pq); /* * Give PGA_REQUEUE_HEAD precedence over PGA_REQUEUE. In particular, if * both flags are set in close succession, only PGA_REQUEUE_HEAD will be * applied, even if it was set first. */ if ((old->flags & PGA_REQUEUE_HEAD) != 0) { vmd = vm_pagequeue_domain(m); KASSERT(pq == &vmd->vmd_pagequeues[PQ_INACTIVE], ("%s: invalid page queue for page %p", __func__, m)); TAILQ_INSERT_BEFORE(&vmd->vmd_inacthead, m, plinks.q); } else { TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q); } return (true); } /* * Commit a queue state update that encodes a request for a deferred queue * operation. */ static bool vm_page_pqstate_commit_request(vm_page_t m, vm_page_astate_t *old, vm_page_astate_t new) { KASSERT(old->queue == new.queue || new.queue != PQ_NONE, ("%s: invalid state, queue %d flags %x", __func__, new.queue, new.flags)); if (old->_bits != new._bits && !vm_page_pqstate_fcmpset(m, old, new)) return (false); vm_page_pqbatch_submit(m, new.queue); return (true); } /* * A generic queue state update function. This handles more cases than the * specialized functions above. */ bool vm_page_pqstate_commit(vm_page_t m, vm_page_astate_t *old, vm_page_astate_t new) { if (old->_bits == new._bits) return (true); if (old->queue != PQ_NONE && new.queue != old->queue) { if (!vm_page_pqstate_commit_dequeue(m, old, new)) return (false); if (new.queue != PQ_NONE) vm_page_pqbatch_submit(m, new.queue); } else { if (!vm_page_pqstate_fcmpset(m, old, new)) return (false); if (new.queue != PQ_NONE && ((new.flags & ~old->flags) & PGA_QUEUE_OP_MASK) != 0) vm_page_pqbatch_submit(m, new.queue); } return (true); } /* * Apply deferred queue state updates to a page. */ static inline void vm_pqbatch_process_page(struct vm_pagequeue *pq, vm_page_t m, uint8_t queue) { vm_page_astate_t new, old; CRITICAL_ASSERT(curthread); vm_pagequeue_assert_locked(pq); KASSERT(queue < PQ_COUNT, ("%s: invalid queue index %d", __func__, queue)); KASSERT(pq == _vm_page_pagequeue(m, queue), ("%s: page %p does not belong to queue %p", __func__, m, pq)); for (old = vm_page_astate_load(m);;) { if (__predict_false(old.queue != queue || (old.flags & PGA_QUEUE_OP_MASK) == 0)) { counter_u64_add(queue_nops, 1); break; } KASSERT(old.queue != PQ_NONE || (old.flags & PGA_QUEUE_STATE_MASK) == 0, ("%s: page %p has unexpected queue state", __func__, m)); new = old; if ((old.flags & PGA_DEQUEUE) != 0) { new.flags &= ~PGA_QUEUE_OP_MASK; new.queue = PQ_NONE; if (__predict_true(_vm_page_pqstate_commit_dequeue(pq, m, &old, new))) { counter_u64_add(queue_ops, 1); break; } } else { new.flags &= ~(PGA_REQUEUE | PGA_REQUEUE_HEAD); if (__predict_true(_vm_page_pqstate_commit_requeue(pq, m, &old, new))) { counter_u64_add(queue_ops, 1); break; } } } } static void vm_pqbatch_process(struct vm_pagequeue *pq, struct vm_batchqueue *bq, uint8_t queue) { int i; for (i = 0; i < bq->bq_cnt; i++) vm_pqbatch_process_page(pq, bq->bq_pa[i], queue); vm_batchqueue_init(bq); } /* * vm_page_pqbatch_submit: [ internal use only ] * * Enqueue a page in the specified page queue's batched work queue. * The caller must have encoded the requested operation in the page * structure's a.flags field. */ void vm_page_pqbatch_submit(vm_page_t m, uint8_t queue) { struct vm_batchqueue *bq; struct vm_pagequeue *pq; int domain; KASSERT((m->oflags & VPO_UNMANAGED) == 0, ("page %p is unmanaged", m)); KASSERT(queue < PQ_COUNT, ("invalid queue %d", queue)); domain = vm_phys_domain(m); pq = &vm_pagequeue_domain(m)->vmd_pagequeues[queue]; critical_enter(); bq = DPCPU_PTR(pqbatch[domain][queue]); if (vm_batchqueue_insert(bq, m)) { critical_exit(); return; } critical_exit(); vm_pagequeue_lock(pq); critical_enter(); bq = DPCPU_PTR(pqbatch[domain][queue]); vm_pqbatch_process(pq, bq, queue); vm_pqbatch_process_page(pq, m, queue); vm_pagequeue_unlock(pq); critical_exit(); } /* * vm_page_pqbatch_drain: [ internal use only ] * * Force all per-CPU page queue batch queues to be drained. This is * intended for use in severe memory shortages, to ensure that pages * do not remain stuck in the batch queues. */ void vm_page_pqbatch_drain(void) { struct thread *td; struct vm_domain *vmd; struct vm_pagequeue *pq; int cpu, domain, queue; td = curthread; CPU_FOREACH(cpu) { thread_lock(td); sched_bind(td, cpu); thread_unlock(td); for (domain = 0; domain < vm_ndomains; domain++) { vmd = VM_DOMAIN(domain); for (queue = 0; queue < PQ_COUNT; queue++) { pq = &vmd->vmd_pagequeues[queue]; vm_pagequeue_lock(pq); critical_enter(); vm_pqbatch_process(pq, DPCPU_PTR(pqbatch[domain][queue]), queue); critical_exit(); vm_pagequeue_unlock(pq); } } } thread_lock(td); sched_unbind(td); thread_unlock(td); } /* * vm_page_dequeue_deferred: [ internal use only ] * * Request removal of the given page from its current page * queue. Physical removal from the queue may be deferred * indefinitely. * * The page must be locked. */ void vm_page_dequeue_deferred(vm_page_t m) { vm_page_astate_t new, old; old = vm_page_astate_load(m); do { if (old.queue == PQ_NONE) { KASSERT((old.flags & PGA_QUEUE_STATE_MASK) == 0, ("%s: page %p has unexpected queue state", __func__, m)); break; } new = old; new.flags |= PGA_DEQUEUE; } while (!vm_page_pqstate_commit_request(m, &old, new)); } /* * vm_page_dequeue: * * Remove the page from whichever page queue it's in, if any, before * returning. */ void vm_page_dequeue(vm_page_t m) { vm_page_astate_t new, old; old = vm_page_astate_load(m); do { if (old.queue == PQ_NONE) { KASSERT((old.flags & PGA_QUEUE_STATE_MASK) == 0, ("%s: page %p has unexpected queue state", __func__, m)); break; } new = old; new.flags &= ~PGA_QUEUE_OP_MASK; new.queue = PQ_NONE; } while (!vm_page_pqstate_commit_dequeue(m, &old, new)); } /* * Schedule the given page for insertion into the specified page queue. * Physical insertion of the page may be deferred indefinitely. */ static void vm_page_enqueue(vm_page_t m, uint8_t queue) { KASSERT(m->a.queue == PQ_NONE && (m->a.flags & PGA_QUEUE_STATE_MASK) == 0, ("%s: page %p is already enqueued", __func__, m)); KASSERT(m->ref_count > 0, ("%s: page %p does not carry any references", __func__, m)); m->a.queue = queue; if ((m->a.flags & PGA_REQUEUE) == 0) vm_page_aflag_set(m, PGA_REQUEUE); vm_page_pqbatch_submit(m, queue); } /* * vm_page_free_prep: * * Prepares the given page to be put on the free list, * disassociating it from any VM object. The caller may return * the page to the free list only if this function returns true. * * The object must be locked. The page must be locked if it is * managed. */ static bool vm_page_free_prep(vm_page_t m) { /* * Synchronize with threads that have dropped a reference to this * page. */ atomic_thread_fence_acq(); #if defined(DIAGNOSTIC) && defined(PHYS_TO_DMAP) if (PMAP_HAS_DMAP && (m->flags & PG_ZERO) != 0) { uint64_t *p; int i; p = (uint64_t *)PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m)); for (i = 0; i < PAGE_SIZE / sizeof(uint64_t); i++, p++) KASSERT(*p == 0, ("vm_page_free_prep %p PG_ZERO %d %jx", m, i, (uintmax_t)*p)); } #endif if ((m->oflags & VPO_UNMANAGED) == 0) { KASSERT(!pmap_page_is_mapped(m), ("vm_page_free_prep: freeing mapped page %p", m)); KASSERT((m->a.flags & (PGA_EXECUTABLE | PGA_WRITEABLE)) == 0, ("vm_page_free_prep: mapping flags set in page %p", m)); } else { KASSERT(m->a.queue == PQ_NONE, ("vm_page_free_prep: unmanaged page %p is queued", m)); } VM_CNT_INC(v_tfree); if (m->object != NULL) { KASSERT(((m->oflags & VPO_UNMANAGED) != 0) == ((m->object->flags & OBJ_UNMANAGED) != 0), ("vm_page_free_prep: managed flag mismatch for page %p", m)); vm_page_assert_xbusied(m); /* * The object reference can be released without an atomic * operation. */ KASSERT((m->flags & PG_FICTITIOUS) != 0 || m->ref_count == VPRC_OBJREF, ("vm_page_free_prep: page %p has unexpected ref_count %u", m, m->ref_count)); vm_page_object_remove(m); m->ref_count -= VPRC_OBJREF; } else vm_page_assert_unbusied(m); vm_page_busy_free(m); /* * If fictitious remove object association and * return. */ if ((m->flags & PG_FICTITIOUS) != 0) { KASSERT(m->ref_count == 1, ("fictitious page %p is referenced", m)); KASSERT(m->a.queue == PQ_NONE, ("fictitious page %p is queued", m)); return (false); } /* * Pages need not be dequeued before they are returned to the physical * memory allocator, but they must at least be marked for a deferred * dequeue. */ if ((m->oflags & VPO_UNMANAGED) == 0) vm_page_dequeue_deferred(m); m->valid = 0; vm_page_undirty(m); if (m->ref_count != 0) panic("vm_page_free_prep: page %p has references", m); /* * Restore the default memory attribute to the page. */ if (pmap_page_get_memattr(m) != VM_MEMATTR_DEFAULT) pmap_page_set_memattr(m, VM_MEMATTR_DEFAULT); #if VM_NRESERVLEVEL > 0 /* * Determine whether the page belongs to a reservation. If the page was * allocated from a per-CPU cache, it cannot belong to a reservation, so * as an optimization, we avoid the check in that case. */ if ((m->flags & PG_PCPU_CACHE) == 0 && vm_reserv_free_page(m)) return (false); #endif return (true); } /* * vm_page_free_toq: * * Returns the given page to the free list, disassociating it * from any VM object. * * The object must be locked. The page must be locked if it is * managed. */ static void vm_page_free_toq(vm_page_t m) { struct vm_domain *vmd; uma_zone_t zone; if (!vm_page_free_prep(m)) return; vmd = vm_pagequeue_domain(m); zone = vmd->vmd_pgcache[m->pool].zone; if ((m->flags & PG_PCPU_CACHE) != 0 && zone != NULL) { uma_zfree(zone, m); return; } vm_domain_free_lock(vmd); vm_phys_free_pages(m, 0); vm_domain_free_unlock(vmd); vm_domain_freecnt_inc(vmd, 1); } /* * vm_page_free_pages_toq: * * Returns a list of pages to the free list, disassociating it * from any VM object. In other words, this is equivalent to * calling vm_page_free_toq() for each page of a list of VM objects. * * The objects must be locked. The pages must be locked if it is * managed. */ void vm_page_free_pages_toq(struct spglist *free, bool update_wire_count) { vm_page_t m; int count; if (SLIST_EMPTY(free)) return; count = 0; while ((m = SLIST_FIRST(free)) != NULL) { count++; SLIST_REMOVE_HEAD(free, plinks.s.ss); vm_page_free_toq(m); } if (update_wire_count) vm_wire_sub(count); } /* * Mark this page as wired down, preventing reclamation by the page daemon * or when the containing object is destroyed. */ void vm_page_wire(vm_page_t m) { u_int old; KASSERT(m->object != NULL, ("vm_page_wire: page %p does not belong to an object", m)); if (!vm_page_busied(m) && !vm_object_busied(m->object)) VM_OBJECT_ASSERT_LOCKED(m->object); KASSERT((m->flags & PG_FICTITIOUS) == 0 || VPRC_WIRE_COUNT(m->ref_count) >= 1, ("vm_page_wire: fictitious page %p has zero wirings", m)); old = atomic_fetchadd_int(&m->ref_count, 1); KASSERT(VPRC_WIRE_COUNT(old) != VPRC_WIRE_COUNT_MAX, ("vm_page_wire: counter overflow for page %p", m)); if (VPRC_WIRE_COUNT(old) == 0) { if ((m->oflags & VPO_UNMANAGED) == 0) vm_page_aflag_set(m, PGA_DEQUEUE); vm_wire_add(1); } } /* * Attempt to wire a mapped page following a pmap lookup of that page. * This may fail if a thread is concurrently tearing down mappings of the page. * The transient failure is acceptable because it translates to the * failure of the caller pmap_extract_and_hold(), which should be then * followed by the vm_fault() fallback, see e.g. vm_fault_quick_hold_pages(). */ bool vm_page_wire_mapped(vm_page_t m) { u_int old; old = m->ref_count; do { KASSERT(old > 0, ("vm_page_wire_mapped: wiring unreferenced page %p", m)); if ((old & VPRC_BLOCKED) != 0) return (false); } while (!atomic_fcmpset_int(&m->ref_count, &old, old + 1)); if (VPRC_WIRE_COUNT(old) == 0) { if ((m->oflags & VPO_UNMANAGED) == 0) vm_page_aflag_set(m, PGA_DEQUEUE); vm_wire_add(1); } return (true); } /* * Release a wiring reference to a managed page. If the page still belongs to * an object, update its position in the page queues to reflect the reference. * If the wiring was the last reference to the page, free the page. */ static void vm_page_unwire_managed(vm_page_t m, uint8_t nqueue, bool noreuse) { u_int old; KASSERT((m->oflags & VPO_UNMANAGED) == 0, ("%s: page %p is unmanaged", __func__, m)); /* * Update LRU state before releasing the wiring reference. * Use a release store when updating the reference count to * synchronize with vm_page_free_prep(). */ old = m->ref_count; do { KASSERT(VPRC_WIRE_COUNT(old) > 0, ("vm_page_unwire: wire count underflow for page %p", m)); if (old > VPRC_OBJREF + 1) { /* * The page has at least one other wiring reference. An * earlier iteration of this loop may have called * vm_page_release_toq() and cleared PGA_DEQUEUE, so * re-set it if necessary. */ if ((vm_page_astate_load(m).flags & PGA_DEQUEUE) == 0) vm_page_aflag_set(m, PGA_DEQUEUE); } else if (old == VPRC_OBJREF + 1) { /* * This is the last wiring. Clear PGA_DEQUEUE and * update the page's queue state to reflect the * reference. If the page does not belong to an object * (i.e., the VPRC_OBJREF bit is clear), we only need to * clear leftover queue state. */ vm_page_release_toq(m, nqueue, false); } else if (old == 1) { vm_page_aflag_clear(m, PGA_DEQUEUE); } } while (!atomic_fcmpset_rel_int(&m->ref_count, &old, old - 1)); if (VPRC_WIRE_COUNT(old) == 1) { vm_wire_sub(1); if (old == 1) vm_page_free(m); } } /* * Release one wiring of the specified page, potentially allowing it to be * paged out. * * Only managed pages belonging to an object can be paged out. If the number * of wirings transitions to zero and the page is eligible for page out, then * the page is added to the specified paging queue. If the released wiring * represented the last reference to the page, the page is freed. * * A managed page must be locked. */ void vm_page_unwire(vm_page_t m, uint8_t nqueue) { KASSERT(nqueue < PQ_COUNT, ("vm_page_unwire: invalid queue %u request for page %p", nqueue, m)); if ((m->oflags & VPO_UNMANAGED) != 0) { if (vm_page_unwire_noq(m) && m->ref_count == 0) vm_page_free(m); return; } vm_page_unwire_managed(m, nqueue, false); } /* * Unwire a page without (re-)inserting it into a page queue. It is up * to the caller to enqueue, requeue, or free the page as appropriate. * In most cases involving managed pages, vm_page_unwire() should be used * instead. */ bool vm_page_unwire_noq(vm_page_t m) { u_int old; old = vm_page_drop(m, 1); KASSERT(VPRC_WIRE_COUNT(old) != 0, ("vm_page_unref: counter underflow for page %p", m)); KASSERT((m->flags & PG_FICTITIOUS) == 0 || VPRC_WIRE_COUNT(old) > 1, ("vm_page_unref: missing ref on fictitious page %p", m)); if (VPRC_WIRE_COUNT(old) > 1) return (false); if ((m->oflags & VPO_UNMANAGED) == 0) vm_page_aflag_clear(m, PGA_DEQUEUE); vm_wire_sub(1); return (true); } /* * Ensure that the page ends up in the specified page queue. If the page is * active or being moved to the active queue, ensure that its act_count is * at least ACT_INIT but do not otherwise mess with it. * * A managed page must be locked. */ static __always_inline void vm_page_mvqueue(vm_page_t m, const uint8_t nqueue, const uint16_t nflag) { vm_page_astate_t old, new; KASSERT(m->ref_count > 0, ("%s: page %p does not carry any references", __func__, m)); KASSERT(nflag == PGA_REQUEUE || nflag == PGA_REQUEUE_HEAD, ("%s: invalid flags %x", __func__, nflag)); if ((m->oflags & VPO_UNMANAGED) != 0 || vm_page_wired(m)) return; old = vm_page_astate_load(m); do { if ((old.flags & PGA_DEQUEUE) != 0) break; new = old; new.flags &= ~PGA_QUEUE_OP_MASK; if (nqueue == PQ_ACTIVE) new.act_count = max(old.act_count, ACT_INIT); if (old.queue == nqueue) { if (nqueue != PQ_ACTIVE) new.flags |= nflag; } else { new.flags |= nflag; new.queue = nqueue; } } while (!vm_page_pqstate_commit(m, &old, new)); } /* * Put the specified page on the active list (if appropriate). */ void vm_page_activate(vm_page_t m) { vm_page_mvqueue(m, PQ_ACTIVE, PGA_REQUEUE); } /* * Move the specified page to the tail of the inactive queue, or requeue * the page if it is already in the inactive queue. */ void vm_page_deactivate(vm_page_t m) { vm_page_mvqueue(m, PQ_INACTIVE, PGA_REQUEUE); } void vm_page_deactivate_noreuse(vm_page_t m) { vm_page_mvqueue(m, PQ_INACTIVE, PGA_REQUEUE_HEAD); } /* * Put a page in the laundry, or requeue it if it is already there. */ void vm_page_launder(vm_page_t m) { vm_page_mvqueue(m, PQ_LAUNDRY, PGA_REQUEUE); } /* * Put a page in the PQ_UNSWAPPABLE holding queue. */ void vm_page_unswappable(vm_page_t m) { KASSERT(!vm_page_wired(m) && (m->oflags & VPO_UNMANAGED) == 0, ("page %p already unswappable", m)); vm_page_dequeue(m); vm_page_enqueue(m, PQ_UNSWAPPABLE); } /* * Release a page back to the page queues in preparation for unwiring. */ static void vm_page_release_toq(vm_page_t m, uint8_t nqueue, const bool noreuse) { vm_page_astate_t old, new; uint16_t nflag; /* * Use a check of the valid bits to determine whether we should * accelerate reclamation of the page. The object lock might not be * held here, in which case the check is racy. At worst we will either * accelerate reclamation of a valid page and violate LRU, or * unnecessarily defer reclamation of an invalid page. * * If we were asked to not cache the page, place it near the head of the * inactive queue so that is reclaimed sooner. */ if (noreuse || m->valid == 0) { nqueue = PQ_INACTIVE; nflag = PGA_REQUEUE_HEAD; } else { nflag = PGA_REQUEUE; } old = vm_page_astate_load(m); do { new = old; /* * If the page is already in the active queue and we are not * trying to accelerate reclamation, simply mark it as * referenced and avoid any queue operations. */ new.flags &= ~PGA_QUEUE_OP_MASK; if (nflag != PGA_REQUEUE_HEAD && old.queue == PQ_ACTIVE) new.flags |= PGA_REFERENCED; else { new.flags |= nflag; new.queue = nqueue; } } while (!vm_page_pqstate_commit(m, &old, new)); } /* * Unwire a page and either attempt to free it or re-add it to the page queues. */ void vm_page_release(vm_page_t m, int flags) { vm_object_t object; KASSERT((m->oflags & VPO_UNMANAGED) == 0, ("vm_page_release: page %p is unmanaged", m)); if ((flags & VPR_TRYFREE) != 0) { for (;;) { object = atomic_load_ptr(&m->object); if (object == NULL) break; /* Depends on type-stability. */ if (vm_page_busied(m) || !VM_OBJECT_TRYWLOCK(object)) break; if (object == m->object) { vm_page_release_locked(m, flags); VM_OBJECT_WUNLOCK(object); return; } VM_OBJECT_WUNLOCK(object); } } vm_page_unwire_managed(m, PQ_INACTIVE, flags != 0); } /* See vm_page_release(). */ void vm_page_release_locked(vm_page_t m, int flags) { VM_OBJECT_ASSERT_WLOCKED(m->object); KASSERT((m->oflags & VPO_UNMANAGED) == 0, ("vm_page_release_locked: page %p is unmanaged", m)); if (vm_page_unwire_noq(m)) { if ((flags & VPR_TRYFREE) != 0 && (m->object->ref_count == 0 || !pmap_page_is_mapped(m)) && m->dirty == 0 && vm_page_tryxbusy(m)) { vm_page_free(m); } else { vm_page_release_toq(m, PQ_INACTIVE, flags != 0); } } } static bool vm_page_try_blocked_op(vm_page_t m, void (*op)(vm_page_t)) { u_int old; KASSERT(m->object != NULL && (m->oflags & VPO_UNMANAGED) == 0, ("vm_page_try_blocked_op: page %p has no object", m)); KASSERT(vm_page_busied(m), ("vm_page_try_blocked_op: page %p is not busy", m)); VM_OBJECT_ASSERT_LOCKED(m->object); old = m->ref_count; do { KASSERT(old != 0, ("vm_page_try_blocked_op: page %p has no references", m)); if (VPRC_WIRE_COUNT(old) != 0) return (false); } while (!atomic_fcmpset_int(&m->ref_count, &old, old | VPRC_BLOCKED)); (op)(m); /* * If the object is read-locked, new wirings may be created via an * object lookup. */ old = vm_page_drop(m, VPRC_BLOCKED); KASSERT(!VM_OBJECT_WOWNED(m->object) || old == (VPRC_BLOCKED | VPRC_OBJREF), ("vm_page_try_blocked_op: unexpected refcount value %u for %p", old, m)); return (true); } /* * Atomically check for wirings and remove all mappings of the page. */ bool vm_page_try_remove_all(vm_page_t m) { return (vm_page_try_blocked_op(m, pmap_remove_all)); } /* * Atomically check for wirings and remove all writeable mappings of the page. */ bool vm_page_try_remove_write(vm_page_t m) { return (vm_page_try_blocked_op(m, pmap_remove_write)); } /* * vm_page_advise * * Apply the specified advice to the given page. * * The object and page must be locked. */ void vm_page_advise(vm_page_t m, int advice) { VM_OBJECT_ASSERT_WLOCKED(m->object); if (advice == MADV_FREE) /* * Mark the page clean. This will allow the page to be freed * without first paging it out. MADV_FREE pages are often * quickly reused by malloc(3), so we do not do anything that * would result in a page fault on a later access. */ vm_page_undirty(m); else if (advice != MADV_DONTNEED) { if (advice == MADV_WILLNEED) vm_page_activate(m); return; } if (advice != MADV_FREE && m->dirty == 0 && pmap_is_modified(m)) vm_page_dirty(m); /* * Clear any references to the page. Otherwise, the page daemon will * immediately reactivate the page. */ vm_page_aflag_clear(m, PGA_REFERENCED); /* * Place clean pages near the head of the inactive queue rather than * the tail, thus defeating the queue's LRU operation and ensuring that * the page will be reused quickly. Dirty pages not already in the * laundry are moved there. */ if (m->dirty == 0) vm_page_deactivate_noreuse(m); else if (!vm_page_in_laundry(m)) vm_page_launder(m); } /* * vm_page_grab_release * * Helper routine for grab functions to release busy on return. */ static inline void vm_page_grab_release(vm_page_t m, int allocflags) { if ((allocflags & VM_ALLOC_NOBUSY) != 0) { if ((allocflags & VM_ALLOC_IGN_SBUSY) != 0) vm_page_sunbusy(m); else vm_page_xunbusy(m); } } /* * vm_page_grab_sleep * * Sleep for busy according to VM_ALLOC_ parameters. Returns true * if the caller should retry and false otherwise. * * If the object is locked on entry the object will be unlocked with * false returns and still locked but possibly having been dropped * with true returns. */ static bool vm_page_grab_sleep(vm_object_t object, vm_page_t m, vm_pindex_t pindex, const char *wmesg, int allocflags, bool locked) { if ((allocflags & VM_ALLOC_NOWAIT) != 0) return (false); /* * Reference the page before unlocking and sleeping so that * the page daemon is less likely to reclaim it. */ if (locked && (allocflags & VM_ALLOC_NOCREAT) == 0) vm_page_reference(m); if (_vm_page_busy_sleep(object, m, m->pindex, wmesg, allocflags, locked) && locked) VM_OBJECT_WLOCK(object); if ((allocflags & VM_ALLOC_WAITFAIL) != 0) return (false); return (true); } /* * Assert that the grab flags are valid. */ static inline void vm_page_grab_check(int allocflags) { KASSERT((allocflags & VM_ALLOC_NOBUSY) == 0 || (allocflags & VM_ALLOC_WIRED) != 0, ("vm_page_grab*: the pages must be busied or wired")); KASSERT((allocflags & VM_ALLOC_SBUSY) == 0 || (allocflags & VM_ALLOC_IGN_SBUSY) != 0, ("vm_page_grab*: VM_ALLOC_SBUSY/VM_ALLOC_IGN_SBUSY mismatch")); } /* * Calculate the page allocation flags for grab. */ static inline int vm_page_grab_pflags(int allocflags) { int pflags; pflags = allocflags & ~(VM_ALLOC_NOWAIT | VM_ALLOC_WAITOK | VM_ALLOC_WAITFAIL | VM_ALLOC_NOBUSY); if ((allocflags & VM_ALLOC_NOWAIT) == 0) pflags |= VM_ALLOC_WAITFAIL; if ((allocflags & VM_ALLOC_IGN_SBUSY) != 0) pflags |= VM_ALLOC_SBUSY; return (pflags); } /* * Grab a page, waiting until we are waken up due to the page * changing state. We keep on waiting, if the page continues * to be in the object. If the page doesn't exist, first allocate it * and then conditionally zero it. * * This routine may sleep. * * The object must be locked on entry. The lock will, however, be released * and reacquired if the routine sleeps. */ vm_page_t vm_page_grab(vm_object_t object, vm_pindex_t pindex, int allocflags) { vm_page_t m; VM_OBJECT_ASSERT_WLOCKED(object); vm_page_grab_check(allocflags); retrylookup: if ((m = vm_page_lookup(object, pindex)) != NULL) { if (!vm_page_tryacquire(m, allocflags)) { if (vm_page_grab_sleep(object, m, pindex, "pgrbwt", allocflags, true)) goto retrylookup; return (NULL); } goto out; } if ((allocflags & VM_ALLOC_NOCREAT) != 0) return (NULL); m = vm_page_alloc(object, pindex, vm_page_grab_pflags(allocflags)); if (m == NULL) { if ((allocflags & (VM_ALLOC_NOWAIT | VM_ALLOC_WAITFAIL)) != 0) return (NULL); goto retrylookup; } if (allocflags & VM_ALLOC_ZERO && (m->flags & PG_ZERO) == 0) pmap_zero_page(m); out: vm_page_grab_release(m, allocflags); return (m); } /* * Locklessly attempt to acquire a page given a (object, pindex) tuple * and an optional previous page to avoid the radix lookup. The resulting * page will be validated against the identity tuple and busied or wired * as requested. A NULL *mp return guarantees that the page was not in * radix at the time of the call but callers must perform higher level * synchronization or retry the operation under a lock if they require * an atomic answer. This is the only lock free validation routine, * other routines can depend on the resulting page state. * * The return value indicates whether the operation failed due to caller * flags. The return is tri-state with mp: * * (true, *mp != NULL) - The operation was successful. * (true, *mp == NULL) - The page was not found in tree. * (false, *mp == NULL) - WAITFAIL or NOWAIT prevented acquisition. */ static bool vm_page_acquire_unlocked(vm_object_t object, vm_pindex_t pindex, vm_page_t prev, vm_page_t *mp, int allocflags) { vm_page_t m; vm_page_grab_check(allocflags); MPASS(prev == NULL || vm_page_busied(prev) || vm_page_wired(prev)); *mp = NULL; for (;;) { /* * We may see a false NULL here because the previous page * has been removed or just inserted and the list is loaded * without barriers. Switch to radix to verify. */ if (prev == NULL || (m = TAILQ_NEXT(prev, listq)) == NULL || m->pindex != pindex || atomic_load_ptr(&m->object) != object) { prev = NULL; /* * This guarantees the result is instantaneously * correct. */ m = vm_radix_lookup_unlocked(&object->rtree, pindex); } if (m == NULL) return (true); if (vm_page_trybusy(m, allocflags)) { if (m->object == object && m->pindex == pindex) break; /* relookup. */ vm_page_busy_release(m); cpu_spinwait(); continue; } if (!vm_page_grab_sleep(object, m, pindex, "pgnslp", allocflags, false)) return (false); } if ((allocflags & VM_ALLOC_WIRED) != 0) vm_page_wire(m); vm_page_grab_release(m, allocflags); *mp = m; return (true); } /* * Try to locklessly grab a page and fall back to the object lock if NOCREAT * is not set. */ vm_page_t vm_page_grab_unlocked(vm_object_t object, vm_pindex_t pindex, int allocflags) { vm_page_t m; vm_page_grab_check(allocflags); if (!vm_page_acquire_unlocked(object, pindex, NULL, &m, allocflags)) return (NULL); if (m != NULL) return (m); /* * The radix lockless lookup should never return a false negative * errors. If the user specifies NOCREAT they are guaranteed there * was no page present at the instant of the call. A NOCREAT caller * must handle create races gracefully. */ if ((allocflags & VM_ALLOC_NOCREAT) != 0) return (NULL); VM_OBJECT_WLOCK(object); m = vm_page_grab(object, pindex, allocflags); VM_OBJECT_WUNLOCK(object); return (m); } /* * Grab a page and make it valid, paging in if necessary. Pages missing from * their pager are zero filled and validated. If a VM_ALLOC_COUNT is supplied * and the page is not valid as many as VM_INITIAL_PAGEIN pages can be brought * in simultaneously. Additional pages will be left on a paging queue but * will neither be wired nor busy regardless of allocflags. */ int vm_page_grab_valid(vm_page_t *mp, vm_object_t object, vm_pindex_t pindex, int allocflags) { vm_page_t m; vm_page_t ma[VM_INITIAL_PAGEIN]; int after, i, pflags, rv; KASSERT((allocflags & VM_ALLOC_SBUSY) == 0 || (allocflags & VM_ALLOC_IGN_SBUSY) != 0, ("vm_page_grab_valid: VM_ALLOC_SBUSY/VM_ALLOC_IGN_SBUSY mismatch")); KASSERT((allocflags & (VM_ALLOC_NOWAIT | VM_ALLOC_WAITFAIL | VM_ALLOC_ZERO)) == 0, ("vm_page_grab_valid: Invalid flags 0x%X", allocflags)); VM_OBJECT_ASSERT_WLOCKED(object); pflags = allocflags & ~(VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY | VM_ALLOC_WIRED); pflags |= VM_ALLOC_WAITFAIL; retrylookup: if ((m = vm_page_lookup(object, pindex)) != NULL) { /* * If the page is fully valid it can only become invalid * with the object lock held. If it is not valid it can * become valid with the busy lock held. Therefore, we * may unnecessarily lock the exclusive busy here if we * race with I/O completion not using the object lock. * However, we will not end up with an invalid page and a * shared lock. */ if (!vm_page_trybusy(m, vm_page_all_valid(m) ? allocflags : 0)) { (void)vm_page_grab_sleep(object, m, pindex, "pgrbwt", allocflags, true); goto retrylookup; } if (vm_page_all_valid(m)) goto out; if ((allocflags & VM_ALLOC_NOCREAT) != 0) { vm_page_busy_release(m); *mp = NULL; return (VM_PAGER_FAIL); } } else if ((allocflags & VM_ALLOC_NOCREAT) != 0) { *mp = NULL; return (VM_PAGER_FAIL); } else if ((m = vm_page_alloc(object, pindex, pflags)) == NULL) { goto retrylookup; } vm_page_assert_xbusied(m); if (vm_pager_has_page(object, pindex, NULL, &after)) { after = MIN(after, VM_INITIAL_PAGEIN); after = MIN(after, allocflags >> VM_ALLOC_COUNT_SHIFT); after = MAX(after, 1); ma[0] = m; for (i = 1; i < after; i++) { if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) { if (ma[i]->valid || !vm_page_tryxbusy(ma[i])) break; } else { ma[i] = vm_page_alloc(object, m->pindex + i, VM_ALLOC_NORMAL); if (ma[i] == NULL) break; } } after = i; vm_object_pip_add(object, after); VM_OBJECT_WUNLOCK(object); rv = vm_pager_get_pages(object, ma, after, NULL, NULL); VM_OBJECT_WLOCK(object); vm_object_pip_wakeupn(object, after); /* Pager may have replaced a page. */ m = ma[0]; if (rv != VM_PAGER_OK) { for (i = 0; i < after; i++) { if (!vm_page_wired(ma[i])) vm_page_free(ma[i]); else vm_page_xunbusy(ma[i]); } *mp = NULL; return (rv); } for (i = 1; i < after; i++) vm_page_readahead_finish(ma[i]); MPASS(vm_page_all_valid(m)); } else { vm_page_zero_invalid(m, TRUE); } out: if ((allocflags & VM_ALLOC_WIRED) != 0) vm_page_wire(m); if ((allocflags & VM_ALLOC_SBUSY) != 0 && vm_page_xbusied(m)) vm_page_busy_downgrade(m); else if ((allocflags & VM_ALLOC_NOBUSY) != 0) vm_page_busy_release(m); *mp = m; return (VM_PAGER_OK); } /* * Locklessly grab a valid page. If the page is not valid or not yet * allocated this will fall back to the object lock method. */ int vm_page_grab_valid_unlocked(vm_page_t *mp, vm_object_t object, vm_pindex_t pindex, int allocflags) { vm_page_t m; int flags; int error; KASSERT((allocflags & VM_ALLOC_SBUSY) == 0 || (allocflags & VM_ALLOC_IGN_SBUSY) != 0, ("vm_page_grab_valid_unlocked: VM_ALLOC_SBUSY/VM_ALLOC_IGN_SBUSY " "mismatch")); KASSERT((allocflags & (VM_ALLOC_NOWAIT | VM_ALLOC_WAITFAIL | VM_ALLOC_ZERO)) == 0, ("vm_page_grab_valid_unlocked: Invalid flags 0x%X", allocflags)); /* * Attempt a lockless lookup and busy. We need at least an sbusy * before we can inspect the valid field and return a wired page. */ flags = allocflags & ~(VM_ALLOC_NOBUSY | VM_ALLOC_WIRED); if (!vm_page_acquire_unlocked(object, pindex, NULL, mp, flags)) return (VM_PAGER_FAIL); if ((m = *mp) != NULL) { if (vm_page_all_valid(m)) { if ((allocflags & VM_ALLOC_WIRED) != 0) vm_page_wire(m); vm_page_grab_release(m, allocflags); return (VM_PAGER_OK); } vm_page_busy_release(m); } if ((allocflags & VM_ALLOC_NOCREAT) != 0) { *mp = NULL; return (VM_PAGER_FAIL); } VM_OBJECT_WLOCK(object); error = vm_page_grab_valid(mp, object, pindex, allocflags); VM_OBJECT_WUNLOCK(object); return (error); } /* * Return the specified range of pages from the given object. For each * page offset within the range, if a page already exists within the object * at that offset and it is busy, then wait for it to change state. If, * instead, the page doesn't exist, then allocate it. * * The caller must always specify an allocation class. * * allocation classes: * VM_ALLOC_NORMAL normal process request * VM_ALLOC_SYSTEM system *really* needs the pages * * The caller must always specify that the pages are to be busied and/or * wired. * * optional allocation flags: * VM_ALLOC_IGN_SBUSY do not sleep on soft busy pages * VM_ALLOC_NOBUSY do not exclusive busy the page * VM_ALLOC_NOWAIT do not sleep * VM_ALLOC_SBUSY set page to sbusy state * VM_ALLOC_WIRED wire the pages * VM_ALLOC_ZERO zero and validate any invalid pages * * If VM_ALLOC_NOWAIT is not specified, this routine may sleep. Otherwise, it * may return a partial prefix of the requested range. */ int vm_page_grab_pages(vm_object_t object, vm_pindex_t pindex, int allocflags, vm_page_t *ma, int count) { vm_page_t m, mpred; int pflags; int i; VM_OBJECT_ASSERT_WLOCKED(object); KASSERT(((u_int)allocflags >> VM_ALLOC_COUNT_SHIFT) == 0, ("vm_page_grap_pages: VM_ALLOC_COUNT() is not allowed")); vm_page_grab_check(allocflags); pflags = vm_page_grab_pflags(allocflags); if (count == 0) return (0); i = 0; retrylookup: m = vm_radix_lookup_le(&object->rtree, pindex + i); if (m == NULL || m->pindex != pindex + i) { mpred = m; m = NULL; } else mpred = TAILQ_PREV(m, pglist, listq); for (; i < count; i++) { if (m != NULL) { if (!vm_page_tryacquire(m, allocflags)) { if (vm_page_grab_sleep(object, m, pindex, "grbmaw", allocflags, true)) goto retrylookup; break; } } else { if ((allocflags & VM_ALLOC_NOCREAT) != 0) break; m = vm_page_alloc_after(object, pindex + i, pflags | VM_ALLOC_COUNT(count - i), mpred); if (m == NULL) { if ((allocflags & (VM_ALLOC_NOWAIT | VM_ALLOC_WAITFAIL)) != 0) break; goto retrylookup; } } if (vm_page_none_valid(m) && (allocflags & VM_ALLOC_ZERO) != 0) { if ((m->flags & PG_ZERO) == 0) pmap_zero_page(m); vm_page_valid(m); } vm_page_grab_release(m, allocflags); ma[i] = mpred = m; m = vm_page_next(m); } return (i); } /* * Unlocked variant of vm_page_grab_pages(). This accepts the same flags * and will fall back to the locked variant to handle allocation. */ int vm_page_grab_pages_unlocked(vm_object_t object, vm_pindex_t pindex, int allocflags, vm_page_t *ma, int count) { vm_page_t m, pred; int flags; int i; vm_page_grab_check(allocflags); /* * Modify flags for lockless acquire to hold the page until we * set it valid if necessary. */ flags = allocflags & ~VM_ALLOC_NOBUSY; pred = NULL; for (i = 0; i < count; i++, pindex++) { if (!vm_page_acquire_unlocked(object, pindex, pred, &m, flags)) return (i); if (m == NULL) break; if ((flags & VM_ALLOC_ZERO) != 0 && vm_page_none_valid(m)) { if ((m->flags & PG_ZERO) == 0) pmap_zero_page(m); vm_page_valid(m); } /* m will still be wired or busy according to flags. */ vm_page_grab_release(m, allocflags); pred = ma[i] = m; } if ((allocflags & VM_ALLOC_NOCREAT) != 0) return (i); count -= i; VM_OBJECT_WLOCK(object); i += vm_page_grab_pages(object, pindex, allocflags, &ma[i], count); VM_OBJECT_WUNLOCK(object); return (i); } /* * Mapping function for valid or dirty bits in a page. * * Inputs are required to range within a page. */ vm_page_bits_t vm_page_bits(int base, int size) { int first_bit; int last_bit; KASSERT( base + size <= PAGE_SIZE, ("vm_page_bits: illegal base/size %d/%d", base, size) ); if (size == 0) /* handle degenerate case */ return (0); first_bit = base >> DEV_BSHIFT; last_bit = (base + size - 1) >> DEV_BSHIFT; return (((vm_page_bits_t)2 << last_bit) - ((vm_page_bits_t)1 << first_bit)); } void vm_page_bits_set(vm_page_t m, vm_page_bits_t *bits, vm_page_bits_t set) { #if PAGE_SIZE == 32768 atomic_set_64((uint64_t *)bits, set); #elif PAGE_SIZE == 16384 atomic_set_32((uint32_t *)bits, set); #elif (PAGE_SIZE == 8192) && defined(atomic_set_16) atomic_set_16((uint16_t *)bits, set); #elif (PAGE_SIZE == 4096) && defined(atomic_set_8) atomic_set_8((uint8_t *)bits, set); #else /* PAGE_SIZE <= 8192 */ uintptr_t addr; int shift; addr = (uintptr_t)bits; /* * Use a trick to perform a 32-bit atomic on the * containing aligned word, to not depend on the existence * of atomic_{set, clear}_{8, 16}. */ shift = addr & (sizeof(uint32_t) - 1); #if BYTE_ORDER == BIG_ENDIAN shift = (sizeof(uint32_t) - sizeof(vm_page_bits_t) - shift) * NBBY; #else shift *= NBBY; #endif addr &= ~(sizeof(uint32_t) - 1); atomic_set_32((uint32_t *)addr, set << shift); #endif /* PAGE_SIZE */ } static inline void vm_page_bits_clear(vm_page_t m, vm_page_bits_t *bits, vm_page_bits_t clear) { #if PAGE_SIZE == 32768 atomic_clear_64((uint64_t *)bits, clear); #elif PAGE_SIZE == 16384 atomic_clear_32((uint32_t *)bits, clear); #elif (PAGE_SIZE == 8192) && defined(atomic_clear_16) atomic_clear_16((uint16_t *)bits, clear); #elif (PAGE_SIZE == 4096) && defined(atomic_clear_8) atomic_clear_8((uint8_t *)bits, clear); #else /* PAGE_SIZE <= 8192 */ uintptr_t addr; int shift; addr = (uintptr_t)bits; /* * Use a trick to perform a 32-bit atomic on the * containing aligned word, to not depend on the existence * of atomic_{set, clear}_{8, 16}. */ shift = addr & (sizeof(uint32_t) - 1); #if BYTE_ORDER == BIG_ENDIAN shift = (sizeof(uint32_t) - sizeof(vm_page_bits_t) - shift) * NBBY; #else shift *= NBBY; #endif addr &= ~(sizeof(uint32_t) - 1); atomic_clear_32((uint32_t *)addr, clear << shift); #endif /* PAGE_SIZE */ } static inline vm_page_bits_t vm_page_bits_swap(vm_page_t m, vm_page_bits_t *bits, vm_page_bits_t newbits) { #if PAGE_SIZE == 32768 uint64_t old; old = *bits; while (atomic_fcmpset_64(bits, &old, newbits) == 0); return (old); #elif PAGE_SIZE == 16384 uint32_t old; old = *bits; while (atomic_fcmpset_32(bits, &old, newbits) == 0); return (old); #elif (PAGE_SIZE == 8192) && defined(atomic_fcmpset_16) uint16_t old; old = *bits; while (atomic_fcmpset_16(bits, &old, newbits) == 0); return (old); #elif (PAGE_SIZE == 4096) && defined(atomic_fcmpset_8) uint8_t old; old = *bits; while (atomic_fcmpset_8(bits, &old, newbits) == 0); return (old); #else /* PAGE_SIZE <= 4096*/ uintptr_t addr; uint32_t old, new, mask; int shift; addr = (uintptr_t)bits; /* * Use a trick to perform a 32-bit atomic on the * containing aligned word, to not depend on the existence * of atomic_{set, swap, clear}_{8, 16}. */ shift = addr & (sizeof(uint32_t) - 1); #if BYTE_ORDER == BIG_ENDIAN shift = (sizeof(uint32_t) - sizeof(vm_page_bits_t) - shift) * NBBY; #else shift *= NBBY; #endif addr &= ~(sizeof(uint32_t) - 1); mask = VM_PAGE_BITS_ALL << shift; old = *bits; do { new = old & ~mask; new |= newbits << shift; } while (atomic_fcmpset_32((uint32_t *)addr, &old, new) == 0); return (old >> shift); #endif /* PAGE_SIZE */ } /* * vm_page_set_valid_range: * * Sets portions of a page valid. The arguments are expected * to be DEV_BSIZE aligned but if they aren't the bitmap is inclusive * of any partial chunks touched by the range. The invalid portion of * such chunks will be zeroed. * * (base + size) must be less then or equal to PAGE_SIZE. */ void vm_page_set_valid_range(vm_page_t m, int base, int size) { int endoff, frag; vm_page_bits_t pagebits; vm_page_assert_busied(m); if (size == 0) /* handle degenerate case */ return; /* * If the base is not DEV_BSIZE aligned and the valid * bit is clear, we have to zero out a portion of the * first block. */ if ((frag = rounddown2(base, DEV_BSIZE)) != base && (m->valid & (1 << (base >> DEV_BSHIFT))) == 0) pmap_zero_page_area(m, frag, base - frag); /* * If the ending offset is not DEV_BSIZE aligned and the * valid bit is clear, we have to zero out a portion of * the last block. */ endoff = base + size; if ((frag = rounddown2(endoff, DEV_BSIZE)) != endoff && (m->valid & (1 << (endoff >> DEV_BSHIFT))) == 0) pmap_zero_page_area(m, endoff, DEV_BSIZE - (endoff & (DEV_BSIZE - 1))); /* * Assert that no previously invalid block that is now being validated * is already dirty. */ KASSERT((~m->valid & vm_page_bits(base, size) & m->dirty) == 0, ("vm_page_set_valid_range: page %p is dirty", m)); /* * Set valid bits inclusive of any overlap. */ pagebits = vm_page_bits(base, size); if (vm_page_xbusied(m)) m->valid |= pagebits; else vm_page_bits_set(m, &m->valid, pagebits); } /* * Set the page dirty bits and free the invalid swap space if * present. Returns the previous dirty bits. */ vm_page_bits_t vm_page_set_dirty(vm_page_t m) { vm_page_bits_t old; VM_PAGE_OBJECT_BUSY_ASSERT(m); if (vm_page_xbusied(m) && !pmap_page_is_write_mapped(m)) { old = m->dirty; m->dirty = VM_PAGE_BITS_ALL; } else old = vm_page_bits_swap(m, &m->dirty, VM_PAGE_BITS_ALL); if (old == 0 && (m->a.flags & PGA_SWAP_SPACE) != 0) vm_pager_page_unswapped(m); return (old); } /* * Clear the given bits from the specified page's dirty field. */ static __inline void vm_page_clear_dirty_mask(vm_page_t m, vm_page_bits_t pagebits) { vm_page_assert_busied(m); /* * If the page is xbusied and not write mapped we are the * only thread that can modify dirty bits. Otherwise, The pmap * layer can call vm_page_dirty() without holding a distinguished * lock. The combination of page busy and atomic operations * suffice to guarantee consistency of the page dirty field. */ if (vm_page_xbusied(m) && !pmap_page_is_write_mapped(m)) m->dirty &= ~pagebits; else vm_page_bits_clear(m, &m->dirty, pagebits); } /* * vm_page_set_validclean: * * Sets portions of a page valid and clean. The arguments are expected * to be DEV_BSIZE aligned but if they aren't the bitmap is inclusive * of any partial chunks touched by the range. The invalid portion of * such chunks will be zero'd. * * (base + size) must be less then or equal to PAGE_SIZE. */ void vm_page_set_validclean(vm_page_t m, int base, int size) { vm_page_bits_t oldvalid, pagebits; int endoff, frag; vm_page_assert_busied(m); if (size == 0) /* handle degenerate case */ return; /* * If the base is not DEV_BSIZE aligned and the valid * bit is clear, we have to zero out a portion of the * first block. */ if ((frag = rounddown2(base, DEV_BSIZE)) != base && (m->valid & ((vm_page_bits_t)1 << (base >> DEV_BSHIFT))) == 0) pmap_zero_page_area(m, frag, base - frag); /* * If the ending offset is not DEV_BSIZE aligned and the * valid bit is clear, we have to zero out a portion of * the last block. */ endoff = base + size; if ((frag = rounddown2(endoff, DEV_BSIZE)) != endoff && (m->valid & ((vm_page_bits_t)1 << (endoff >> DEV_BSHIFT))) == 0) pmap_zero_page_area(m, endoff, DEV_BSIZE - (endoff & (DEV_BSIZE - 1))); /* * Set valid, clear dirty bits. If validating the entire * page we can safely clear the pmap modify bit. We also * use this opportunity to clear the PGA_NOSYNC flag. If a process * takes a write fault on a MAP_NOSYNC memory area the flag will * be set again. * * We set valid bits inclusive of any overlap, but we can only * clear dirty bits for DEV_BSIZE chunks that are fully within * the range. */ oldvalid = m->valid; pagebits = vm_page_bits(base, size); if (vm_page_xbusied(m)) m->valid |= pagebits; else vm_page_bits_set(m, &m->valid, pagebits); #if 0 /* NOT YET */ if ((frag = base & (DEV_BSIZE - 1)) != 0) { frag = DEV_BSIZE - frag; base += frag; size -= frag; if (size < 0) size = 0; } pagebits = vm_page_bits(base, size & (DEV_BSIZE - 1)); #endif if (base == 0 && size == PAGE_SIZE) { /* * The page can only be modified within the pmap if it is * mapped, and it can only be mapped if it was previously * fully valid. */ if (oldvalid == VM_PAGE_BITS_ALL) /* * Perform the pmap_clear_modify() first. Otherwise, * a concurrent pmap operation, such as * pmap_protect(), could clear a modification in the * pmap and set the dirty field on the page before * pmap_clear_modify() had begun and after the dirty * field was cleared here. */ pmap_clear_modify(m); m->dirty = 0; vm_page_aflag_clear(m, PGA_NOSYNC); } else if (oldvalid != VM_PAGE_BITS_ALL && vm_page_xbusied(m)) m->dirty &= ~pagebits; else vm_page_clear_dirty_mask(m, pagebits); } void vm_page_clear_dirty(vm_page_t m, int base, int size) { vm_page_clear_dirty_mask(m, vm_page_bits(base, size)); } /* * vm_page_set_invalid: * * Invalidates DEV_BSIZE'd chunks within a page. Both the * valid and dirty bits for the effected areas are cleared. */ void vm_page_set_invalid(vm_page_t m, int base, int size) { vm_page_bits_t bits; vm_object_t object; /* * The object lock is required so that pages can't be mapped * read-only while we're in the process of invalidating them. */ object = m->object; VM_OBJECT_ASSERT_WLOCKED(object); vm_page_assert_busied(m); if (object->type == OBJT_VNODE && base == 0 && IDX_TO_OFF(m->pindex) + size >= object->un_pager.vnp.vnp_size) bits = VM_PAGE_BITS_ALL; else bits = vm_page_bits(base, size); if (object->ref_count != 0 && vm_page_all_valid(m) && bits != 0) pmap_remove_all(m); KASSERT((bits == 0 && vm_page_all_valid(m)) || !pmap_page_is_mapped(m), ("vm_page_set_invalid: page %p is mapped", m)); if (vm_page_xbusied(m)) { m->valid &= ~bits; m->dirty &= ~bits; } else { vm_page_bits_clear(m, &m->valid, bits); vm_page_bits_clear(m, &m->dirty, bits); } } /* * vm_page_invalid: * * Invalidates the entire page. The page must be busy, unmapped, and * the enclosing object must be locked. The object locks protects * against concurrent read-only pmap enter which is done without * busy. */ void vm_page_invalid(vm_page_t m) { vm_page_assert_busied(m); VM_OBJECT_ASSERT_LOCKED(m->object); MPASS(!pmap_page_is_mapped(m)); if (vm_page_xbusied(m)) m->valid = 0; else vm_page_bits_clear(m, &m->valid, VM_PAGE_BITS_ALL); } /* * vm_page_zero_invalid() * * The kernel assumes that the invalid portions of a page contain * garbage, but such pages can be mapped into memory by user code. * When this occurs, we must zero out the non-valid portions of the * page so user code sees what it expects. * * Pages are most often semi-valid when the end of a file is mapped * into memory and the file's size is not page aligned. */ void vm_page_zero_invalid(vm_page_t m, boolean_t setvalid) { int b; int i; /* * Scan the valid bits looking for invalid sections that * must be zeroed. Invalid sub-DEV_BSIZE'd areas ( where the * valid bit may be set ) have already been zeroed by * vm_page_set_validclean(). */ for (b = i = 0; i <= PAGE_SIZE / DEV_BSIZE; ++i) { if (i == (PAGE_SIZE / DEV_BSIZE) || (m->valid & ((vm_page_bits_t)1 << i))) { if (i > b) { pmap_zero_page_area(m, b << DEV_BSHIFT, (i - b) << DEV_BSHIFT); } b = i + 1; } } /* * setvalid is TRUE when we can safely set the zero'd areas * as being valid. We can do this if there are no cache consistancy * issues. e.g. it is ok to do with UFS, but not ok to do with NFS. */ if (setvalid) vm_page_valid(m); } /* * vm_page_is_valid: * * Is (partial) page valid? Note that the case where size == 0 * will return FALSE in the degenerate case where the page is * entirely invalid, and TRUE otherwise. * * Some callers envoke this routine without the busy lock held and * handle races via higher level locks. Typical callers should * hold a busy lock to prevent invalidation. */ int vm_page_is_valid(vm_page_t m, int base, int size) { vm_page_bits_t bits; bits = vm_page_bits(base, size); return (m->valid != 0 && (m->valid & bits) == bits); } /* * Returns true if all of the specified predicates are true for the entire * (super)page and false otherwise. */ bool vm_page_ps_test(vm_page_t m, int flags, vm_page_t skip_m) { vm_object_t object; int i, npages; object = m->object; if (skip_m != NULL && skip_m->object != object) return (false); VM_OBJECT_ASSERT_LOCKED(object); npages = atop(pagesizes[m->psind]); /* * The physically contiguous pages that make up a superpage, i.e., a * page with a page size index ("psind") greater than zero, will * occupy adjacent entries in vm_page_array[]. */ for (i = 0; i < npages; i++) { /* Always test object consistency, including "skip_m". */ if (m[i].object != object) return (false); if (&m[i] == skip_m) continue; if ((flags & PS_NONE_BUSY) != 0 && vm_page_busied(&m[i])) return (false); if ((flags & PS_ALL_DIRTY) != 0) { /* * Calling vm_page_test_dirty() or pmap_is_modified() * might stop this case from spuriously returning * "false". However, that would require a write lock * on the object containing "m[i]". */ if (m[i].dirty != VM_PAGE_BITS_ALL) return (false); } if ((flags & PS_ALL_VALID) != 0 && m[i].valid != VM_PAGE_BITS_ALL) return (false); } return (true); } /* * Set the page's dirty bits if the page is modified. */ void vm_page_test_dirty(vm_page_t m) { vm_page_assert_busied(m); if (m->dirty != VM_PAGE_BITS_ALL && pmap_is_modified(m)) vm_page_dirty(m); } void vm_page_valid(vm_page_t m) { vm_page_assert_busied(m); if (vm_page_xbusied(m)) m->valid = VM_PAGE_BITS_ALL; else vm_page_bits_set(m, &m->valid, VM_PAGE_BITS_ALL); } void vm_page_lock_KBI(vm_page_t m, const char *file, int line) { mtx_lock_flags_(vm_page_lockptr(m), 0, file, line); } void vm_page_unlock_KBI(vm_page_t m, const char *file, int line) { mtx_unlock_flags_(vm_page_lockptr(m), 0, file, line); } int vm_page_trylock_KBI(vm_page_t m, const char *file, int line) { return (mtx_trylock_flags_(vm_page_lockptr(m), 0, file, line)); } #if defined(INVARIANTS) || defined(INVARIANT_SUPPORT) void vm_page_assert_locked_KBI(vm_page_t m, const char *file, int line) { vm_page_lock_assert_KBI(m, MA_OWNED, file, line); } void vm_page_lock_assert_KBI(vm_page_t m, int a, const char *file, int line) { mtx_assert_(vm_page_lockptr(m), a, file, line); } #endif #ifdef INVARIANTS void vm_page_object_busy_assert(vm_page_t m) { /* * Certain of the page's fields may only be modified by the * holder of a page or object busy. */ if (m->object != NULL && !vm_page_busied(m)) VM_OBJECT_ASSERT_BUSY(m->object); } void vm_page_assert_pga_writeable(vm_page_t m, uint16_t bits) { if ((bits & PGA_WRITEABLE) == 0) return; /* * The PGA_WRITEABLE flag can only be set if the page is * managed, is exclusively busied or the object is locked. * Currently, this flag is only set by pmap_enter(). */ KASSERT((m->oflags & VPO_UNMANAGED) == 0, ("PGA_WRITEABLE on unmanaged page")); if (!vm_page_xbusied(m)) VM_OBJECT_ASSERT_BUSY(m->object); } #endif #include "opt_ddb.h" #ifdef DDB #include #include DB_SHOW_COMMAND(page, vm_page_print_page_info) { db_printf("vm_cnt.v_free_count: %d\n", vm_free_count()); db_printf("vm_cnt.v_inactive_count: %d\n", vm_inactive_count()); db_printf("vm_cnt.v_active_count: %d\n", vm_active_count()); db_printf("vm_cnt.v_laundry_count: %d\n", vm_laundry_count()); db_printf("vm_cnt.v_wire_count: %d\n", vm_wire_count()); db_printf("vm_cnt.v_free_reserved: %d\n", vm_cnt.v_free_reserved); db_printf("vm_cnt.v_free_min: %d\n", vm_cnt.v_free_min); db_printf("vm_cnt.v_free_target: %d\n", vm_cnt.v_free_target); db_printf("vm_cnt.v_inactive_target: %d\n", vm_cnt.v_inactive_target); } DB_SHOW_COMMAND(pageq, vm_page_print_pageq_info) { int dom; db_printf("pq_free %d\n", vm_free_count()); for (dom = 0; dom < vm_ndomains; dom++) { db_printf( "dom %d page_cnt %d free %d pq_act %d pq_inact %d pq_laund %d pq_unsw %d\n", dom, vm_dom[dom].vmd_page_count, vm_dom[dom].vmd_free_count, vm_dom[dom].vmd_pagequeues[PQ_ACTIVE].pq_cnt, vm_dom[dom].vmd_pagequeues[PQ_INACTIVE].pq_cnt, vm_dom[dom].vmd_pagequeues[PQ_LAUNDRY].pq_cnt, vm_dom[dom].vmd_pagequeues[PQ_UNSWAPPABLE].pq_cnt); } } DB_SHOW_COMMAND(pginfo, vm_page_print_pginfo) { vm_page_t m; boolean_t phys, virt; if (!have_addr) { db_printf("show pginfo addr\n"); return; } phys = strchr(modif, 'p') != NULL; virt = strchr(modif, 'v') != NULL; if (virt) m = PHYS_TO_VM_PAGE(pmap_kextract(addr)); else if (phys) m = PHYS_TO_VM_PAGE(addr); else m = (vm_page_t)addr; db_printf( "page %p obj %p pidx 0x%jx phys 0x%jx q %d ref %u\n" " af 0x%x of 0x%x f 0x%x act %d busy %x valid 0x%x dirty 0x%x\n", m, m->object, (uintmax_t)m->pindex, (uintmax_t)m->phys_addr, m->a.queue, m->ref_count, m->a.flags, m->oflags, m->flags, m->a.act_count, m->busy_lock, m->valid, m->dirty); } #endif /* DDB */ diff --git a/sys/vm/vm_radix.c b/sys/vm/vm_radix.c index 40c4558ae017..ce9edad13c1e 100644 --- a/sys/vm/vm_radix.c +++ b/sys/vm/vm_radix.c @@ -1,910 +1,911 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2013 EMC Corp. * Copyright (c) 2011 Jeffrey Roberson * Copyright (c) 2008 Mayur Shardul * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * */ /* * Path-compressed radix trie implementation. * The following code is not generalized into a general purpose library * because there are way too many parameters embedded that should really * be decided by the library consumers. At the same time, consumers * of this code must achieve highest possible performance. * * The implementation takes into account the following rationale: * - Size of the nodes should be as small as possible but still big enough * to avoid a large maximum depth for the trie. This is a balance * between the necessity to not wire too much physical memory for the nodes * and the necessity to avoid too much cache pollution during the trie * operations. * - There is not a huge bias toward the number of lookup operations over * the number of insert and remove operations. This basically implies * that optimizations supposedly helping one operation but hurting the * other might be carefully evaluated. * - On average not many nodes are expected to be fully populated, hence * level compression may just complicate things. */ #include __FBSDID("$FreeBSD$"); #include "opt_ddb.h" #include #include #include #include #include #include +#include #include #include #include #include #include #include #ifdef DDB #include #endif /* * These widths should allow the pointers to a node's children to fit within * a single cache line. The extra levels from a narrow width should not be * a problem thanks to path compression. */ #ifdef __LP64__ #define VM_RADIX_WIDTH 4 #else #define VM_RADIX_WIDTH 3 #endif #define VM_RADIX_COUNT (1 << VM_RADIX_WIDTH) #define VM_RADIX_MASK (VM_RADIX_COUNT - 1) #define VM_RADIX_LIMIT \ (howmany(sizeof(vm_pindex_t) * NBBY, VM_RADIX_WIDTH) - 1) /* Flag bits stored in node pointers. */ #define VM_RADIX_ISLEAF 0x1 #define VM_RADIX_FLAGS 0x1 #define VM_RADIX_PAD VM_RADIX_FLAGS /* Returns one unit associated with specified level. */ #define VM_RADIX_UNITLEVEL(lev) \ ((vm_pindex_t)1 << ((lev) * VM_RADIX_WIDTH)) enum vm_radix_access { SMR, LOCKED, UNSERIALIZED }; struct vm_radix_node; -SMR_TYPE_DECLARE(smrnode_t, struct vm_radix_node *); +typedef SMR_POINTER(struct vm_radix_node *) smrnode_t; struct vm_radix_node { vm_pindex_t rn_owner; /* Owner of record. */ uint16_t rn_count; /* Valid children. */ uint8_t rn_clev; /* Current level. */ int8_t rn_last; /* zero last ptr. */ smrnode_t rn_child[VM_RADIX_COUNT]; /* Child nodes. */ }; static uma_zone_t vm_radix_node_zone; static smr_t vm_radix_smr; static void vm_radix_node_store(smrnode_t *p, struct vm_radix_node *v, enum vm_radix_access access); /* * Allocate a radix node. */ static struct vm_radix_node * vm_radix_node_get(vm_pindex_t owner, uint16_t count, uint16_t clevel) { struct vm_radix_node *rnode; rnode = uma_zalloc_smr(vm_radix_node_zone, M_NOWAIT); if (rnode == NULL) return (NULL); /* * We want to clear the last child pointer after the final section * has exited so lookup can not return false negatives. It is done * here because it will be cache-cold in the dtor callback. */ if (rnode->rn_last != 0) { vm_radix_node_store(&rnode->rn_child[rnode->rn_last - 1], NULL, UNSERIALIZED); rnode->rn_last = 0; } rnode->rn_owner = owner; rnode->rn_count = count; rnode->rn_clev = clevel; return (rnode); } /* * Free radix node. */ static __inline void vm_radix_node_put(struct vm_radix_node *rnode, int8_t last) { #ifdef INVARIANTS int slot; KASSERT(rnode->rn_count == 0, ("vm_radix_node_put: rnode %p has %d children", rnode, rnode->rn_count)); for (slot = 0; slot < VM_RADIX_COUNT; slot++) { if (slot == last) continue; KASSERT(smr_unserialized_load(&rnode->rn_child[slot], true) == NULL, ("vm_radix_node_put: rnode %p has a child", rnode)); } #endif /* Off by one so a freshly zero'd node is not assigned to. */ rnode->rn_last = last + 1; uma_zfree_smr(vm_radix_node_zone, rnode); } /* * Return the position in the array for a given level. */ static __inline int vm_radix_slot(vm_pindex_t index, uint16_t level) { return ((index >> (level * VM_RADIX_WIDTH)) & VM_RADIX_MASK); } /* Trims the key after the specified level. */ static __inline vm_pindex_t vm_radix_trimkey(vm_pindex_t index, uint16_t level) { vm_pindex_t ret; ret = index; if (level > 0) { ret >>= level * VM_RADIX_WIDTH; ret <<= level * VM_RADIX_WIDTH; } return (ret); } /* * Fetch a node pointer from a slot in another node. */ static __inline struct vm_radix_node * vm_radix_node_load(smrnode_t *p, enum vm_radix_access access) { switch (access) { case UNSERIALIZED: return (smr_unserialized_load(p, true)); case LOCKED: return (smr_serialized_load(p, true)); case SMR: return (smr_entered_load(p, vm_radix_smr)); } __unreachable(); } static __inline void vm_radix_node_store(smrnode_t *p, struct vm_radix_node *v, enum vm_radix_access access) { switch (access) { case UNSERIALIZED: smr_unserialized_store(p, v, true); break; case LOCKED: smr_serialized_store(p, v, true); break; case SMR: panic("vm_radix_node_store: Not supported in smr section."); } } /* * Get the root node for a radix tree. */ static __inline struct vm_radix_node * vm_radix_root_load(struct vm_radix *rtree, enum vm_radix_access access) { return (vm_radix_node_load((smrnode_t *)&rtree->rt_root, access)); } /* * Set the root node for a radix tree. */ static __inline void vm_radix_root_store(struct vm_radix *rtree, struct vm_radix_node *rnode, enum vm_radix_access access) { vm_radix_node_store((smrnode_t *)&rtree->rt_root, rnode, access); } /* * Returns TRUE if the specified radix node is a leaf and FALSE otherwise. */ static __inline boolean_t vm_radix_isleaf(struct vm_radix_node *rnode) { return (((uintptr_t)rnode & VM_RADIX_ISLEAF) != 0); } /* * Returns the associated page extracted from rnode. */ static __inline vm_page_t vm_radix_topage(struct vm_radix_node *rnode) { return ((vm_page_t)((uintptr_t)rnode & ~VM_RADIX_FLAGS)); } /* * Adds the page as a child of the provided node. */ static __inline void vm_radix_addpage(struct vm_radix_node *rnode, vm_pindex_t index, uint16_t clev, vm_page_t page, enum vm_radix_access access) { int slot; slot = vm_radix_slot(index, clev); vm_radix_node_store(&rnode->rn_child[slot], (struct vm_radix_node *)((uintptr_t)page | VM_RADIX_ISLEAF), access); } /* * Returns the slot where two keys differ. * It cannot accept 2 equal keys. */ static __inline uint16_t vm_radix_keydiff(vm_pindex_t index1, vm_pindex_t index2) { uint16_t clev; KASSERT(index1 != index2, ("%s: passing the same key value %jx", __func__, (uintmax_t)index1)); index1 ^= index2; for (clev = VM_RADIX_LIMIT;; clev--) if (vm_radix_slot(index1, clev) != 0) return (clev); } /* * Returns TRUE if it can be determined that key does not belong to the * specified rnode. Otherwise, returns FALSE. */ static __inline boolean_t vm_radix_keybarr(struct vm_radix_node *rnode, vm_pindex_t idx) { if (rnode->rn_clev < VM_RADIX_LIMIT) { idx = vm_radix_trimkey(idx, rnode->rn_clev + 1); return (idx != rnode->rn_owner); } return (FALSE); } /* * Internal helper for vm_radix_reclaim_allnodes(). * This function is recursive. */ static void vm_radix_reclaim_allnodes_int(struct vm_radix_node *rnode) { struct vm_radix_node *child; int slot; KASSERT(rnode->rn_count <= VM_RADIX_COUNT, ("vm_radix_reclaim_allnodes_int: bad count in rnode %p", rnode)); for (slot = 0; rnode->rn_count != 0; slot++) { child = vm_radix_node_load(&rnode->rn_child[slot], UNSERIALIZED); if (child == NULL) continue; if (!vm_radix_isleaf(child)) vm_radix_reclaim_allnodes_int(child); vm_radix_node_store(&rnode->rn_child[slot], NULL, UNSERIALIZED); rnode->rn_count--; } vm_radix_node_put(rnode, -1); } #ifndef UMA_MD_SMALL_ALLOC void vm_radix_reserve_kva(void); /* * Reserve the KVA necessary to satisfy the node allocation. * This is mandatory in architectures not supporting direct * mapping as they will need otherwise to carve into the kernel maps for * every node allocation, resulting into deadlocks for consumers already * working with kernel maps. */ void vm_radix_reserve_kva(void) { /* * Calculate the number of reserved nodes, discounting the pages that * are needed to store them. */ if (!uma_zone_reserve_kva(vm_radix_node_zone, ((vm_paddr_t)vm_cnt.v_page_count * PAGE_SIZE) / (PAGE_SIZE + sizeof(struct vm_radix_node)))) panic("%s: unable to reserve KVA", __func__); } #endif /* * Initialize the UMA slab zone. */ void vm_radix_zinit(void) { vm_radix_node_zone = uma_zcreate("RADIX NODE", sizeof(struct vm_radix_node), NULL, NULL, NULL, NULL, VM_RADIX_PAD, UMA_ZONE_VM | UMA_ZONE_SMR | UMA_ZONE_ZINIT); vm_radix_smr = uma_zone_get_smr(vm_radix_node_zone); } /* * Inserts the key-value pair into the trie. * Panics if the key already exists. */ int vm_radix_insert(struct vm_radix *rtree, vm_page_t page) { vm_pindex_t index, newind; struct vm_radix_node *rnode, *tmp; smrnode_t *parentp; vm_page_t m; int slot; uint16_t clev; index = page->pindex; /* * The owner of record for root is not really important because it * will never be used. */ rnode = vm_radix_root_load(rtree, LOCKED); if (rnode == NULL) { rtree->rt_root = (uintptr_t)page | VM_RADIX_ISLEAF; return (0); } parentp = (smrnode_t *)&rtree->rt_root; for (;;) { if (vm_radix_isleaf(rnode)) { m = vm_radix_topage(rnode); if (m->pindex == index) panic("%s: key %jx is already present", __func__, (uintmax_t)index); clev = vm_radix_keydiff(m->pindex, index); tmp = vm_radix_node_get(vm_radix_trimkey(index, clev + 1), 2, clev); if (tmp == NULL) return (ENOMEM); /* These writes are not yet visible due to ordering. */ vm_radix_addpage(tmp, index, clev, page, UNSERIALIZED); vm_radix_addpage(tmp, m->pindex, clev, m, UNSERIALIZED); /* Synchronize to make leaf visible. */ vm_radix_node_store(parentp, tmp, LOCKED); return (0); } else if (vm_radix_keybarr(rnode, index)) break; slot = vm_radix_slot(index, rnode->rn_clev); parentp = &rnode->rn_child[slot]; tmp = vm_radix_node_load(parentp, LOCKED); if (tmp == NULL) { rnode->rn_count++; vm_radix_addpage(rnode, index, rnode->rn_clev, page, LOCKED); return (0); } rnode = tmp; } /* * A new node is needed because the right insertion level is reached. * Setup the new intermediate node and add the 2 children: the * new object and the older edge. */ newind = rnode->rn_owner; clev = vm_radix_keydiff(newind, index); tmp = vm_radix_node_get(vm_radix_trimkey(index, clev + 1), 2, clev); if (tmp == NULL) return (ENOMEM); slot = vm_radix_slot(newind, clev); /* These writes are not yet visible due to ordering. */ vm_radix_addpage(tmp, index, clev, page, UNSERIALIZED); vm_radix_node_store(&tmp->rn_child[slot], rnode, UNSERIALIZED); /* Serializing write to make the above visible. */ vm_radix_node_store(parentp, tmp, LOCKED); return (0); } /* * Returns TRUE if the specified radix tree contains a single leaf and FALSE * otherwise. */ boolean_t vm_radix_is_singleton(struct vm_radix *rtree) { struct vm_radix_node *rnode; rnode = vm_radix_root_load(rtree, LOCKED); if (rnode == NULL) return (FALSE); return (vm_radix_isleaf(rnode)); } /* * Returns the value stored at the index. If the index is not present, * NULL is returned. */ static __always_inline vm_page_t _vm_radix_lookup(struct vm_radix *rtree, vm_pindex_t index, enum vm_radix_access access) { struct vm_radix_node *rnode; vm_page_t m; int slot; rnode = vm_radix_root_load(rtree, access); while (rnode != NULL) { if (vm_radix_isleaf(rnode)) { m = vm_radix_topage(rnode); if (m->pindex == index) return (m); break; } if (vm_radix_keybarr(rnode, index)) break; slot = vm_radix_slot(index, rnode->rn_clev); rnode = vm_radix_node_load(&rnode->rn_child[slot], access); } return (NULL); } /* * Returns the value stored at the index assuming there is an external lock. * * If the index is not present, NULL is returned. */ vm_page_t vm_radix_lookup(struct vm_radix *rtree, vm_pindex_t index) { return _vm_radix_lookup(rtree, index, LOCKED); } /* * Returns the value stored at the index without requiring an external lock. * * If the index is not present, NULL is returned. */ vm_page_t vm_radix_lookup_unlocked(struct vm_radix *rtree, vm_pindex_t index) { vm_page_t m; smr_enter(vm_radix_smr); m = _vm_radix_lookup(rtree, index, SMR); smr_exit(vm_radix_smr); return (m); } /* * Look up the nearest entry at a position greater than or equal to index. */ vm_page_t vm_radix_lookup_ge(struct vm_radix *rtree, vm_pindex_t index) { struct vm_radix_node *stack[VM_RADIX_LIMIT]; vm_pindex_t inc; vm_page_t m; struct vm_radix_node *child, *rnode; #ifdef INVARIANTS int loops = 0; #endif int slot, tos; rnode = vm_radix_root_load(rtree, LOCKED); if (rnode == NULL) return (NULL); else if (vm_radix_isleaf(rnode)) { m = vm_radix_topage(rnode); if (m->pindex >= index) return (m); else return (NULL); } tos = 0; for (;;) { /* * If the keys differ before the current bisection node, * then the search key might rollback to the earliest * available bisection node or to the smallest key * in the current node (if the owner is greater than the * search key). */ if (vm_radix_keybarr(rnode, index)) { if (index > rnode->rn_owner) { ascend: KASSERT(++loops < 1000, ("vm_radix_lookup_ge: too many loops")); /* * Pop nodes from the stack until either the * stack is empty or a node that could have a * matching descendant is found. */ do { if (tos == 0) return (NULL); rnode = stack[--tos]; } while (vm_radix_slot(index, rnode->rn_clev) == (VM_RADIX_COUNT - 1)); /* * The following computation cannot overflow * because index's slot at the current level * is less than VM_RADIX_COUNT - 1. */ index = vm_radix_trimkey(index, rnode->rn_clev); index += VM_RADIX_UNITLEVEL(rnode->rn_clev); } else index = rnode->rn_owner; KASSERT(!vm_radix_keybarr(rnode, index), ("vm_radix_lookup_ge: keybarr failed")); } slot = vm_radix_slot(index, rnode->rn_clev); child = vm_radix_node_load(&rnode->rn_child[slot], LOCKED); if (vm_radix_isleaf(child)) { m = vm_radix_topage(child); if (m->pindex >= index) return (m); } else if (child != NULL) goto descend; /* * Look for an available edge or page within the current * bisection node. */ if (slot < (VM_RADIX_COUNT - 1)) { inc = VM_RADIX_UNITLEVEL(rnode->rn_clev); index = vm_radix_trimkey(index, rnode->rn_clev); do { index += inc; slot++; child = vm_radix_node_load(&rnode->rn_child[slot], LOCKED); if (vm_radix_isleaf(child)) { m = vm_radix_topage(child); if (m->pindex >= index) return (m); } else if (child != NULL) goto descend; } while (slot < (VM_RADIX_COUNT - 1)); } KASSERT(child == NULL || vm_radix_isleaf(child), ("vm_radix_lookup_ge: child is radix node")); /* * If a page or edge greater than the search slot is not found * in the current node, ascend to the next higher-level node. */ goto ascend; descend: KASSERT(rnode->rn_clev > 0, ("vm_radix_lookup_ge: pushing leaf's parent")); KASSERT(tos < VM_RADIX_LIMIT, ("vm_radix_lookup_ge: stack overflow")); stack[tos++] = rnode; rnode = child; } } /* * Look up the nearest entry at a position less than or equal to index. */ vm_page_t vm_radix_lookup_le(struct vm_radix *rtree, vm_pindex_t index) { struct vm_radix_node *stack[VM_RADIX_LIMIT]; vm_pindex_t inc; vm_page_t m; struct vm_radix_node *child, *rnode; #ifdef INVARIANTS int loops = 0; #endif int slot, tos; rnode = vm_radix_root_load(rtree, LOCKED); if (rnode == NULL) return (NULL); else if (vm_radix_isleaf(rnode)) { m = vm_radix_topage(rnode); if (m->pindex <= index) return (m); else return (NULL); } tos = 0; for (;;) { /* * If the keys differ before the current bisection node, * then the search key might rollback to the earliest * available bisection node or to the largest key * in the current node (if the owner is smaller than the * search key). */ if (vm_radix_keybarr(rnode, index)) { if (index > rnode->rn_owner) { index = rnode->rn_owner + VM_RADIX_COUNT * VM_RADIX_UNITLEVEL(rnode->rn_clev); } else { ascend: KASSERT(++loops < 1000, ("vm_radix_lookup_le: too many loops")); /* * Pop nodes from the stack until either the * stack is empty or a node that could have a * matching descendant is found. */ do { if (tos == 0) return (NULL); rnode = stack[--tos]; } while (vm_radix_slot(index, rnode->rn_clev) == 0); /* * The following computation cannot overflow * because index's slot at the current level * is greater than 0. */ index = vm_radix_trimkey(index, rnode->rn_clev); } index--; KASSERT(!vm_radix_keybarr(rnode, index), ("vm_radix_lookup_le: keybarr failed")); } slot = vm_radix_slot(index, rnode->rn_clev); child = vm_radix_node_load(&rnode->rn_child[slot], LOCKED); if (vm_radix_isleaf(child)) { m = vm_radix_topage(child); if (m->pindex <= index) return (m); } else if (child != NULL) goto descend; /* * Look for an available edge or page within the current * bisection node. */ if (slot > 0) { inc = VM_RADIX_UNITLEVEL(rnode->rn_clev); index |= inc - 1; do { index -= inc; slot--; child = vm_radix_node_load(&rnode->rn_child[slot], LOCKED); if (vm_radix_isleaf(child)) { m = vm_radix_topage(child); if (m->pindex <= index) return (m); } else if (child != NULL) goto descend; } while (slot > 0); } KASSERT(child == NULL || vm_radix_isleaf(child), ("vm_radix_lookup_le: child is radix node")); /* * If a page or edge smaller than the search slot is not found * in the current node, ascend to the next higher-level node. */ goto ascend; descend: KASSERT(rnode->rn_clev > 0, ("vm_radix_lookup_le: pushing leaf's parent")); KASSERT(tos < VM_RADIX_LIMIT, ("vm_radix_lookup_le: stack overflow")); stack[tos++] = rnode; rnode = child; } } /* * Remove the specified index from the trie, and return the value stored at * that index. If the index is not present, return NULL. */ vm_page_t vm_radix_remove(struct vm_radix *rtree, vm_pindex_t index) { struct vm_radix_node *rnode, *parent, *tmp; vm_page_t m; int i, slot; rnode = vm_radix_root_load(rtree, LOCKED); if (vm_radix_isleaf(rnode)) { m = vm_radix_topage(rnode); if (m->pindex != index) return (NULL); vm_radix_root_store(rtree, NULL, LOCKED); return (m); } parent = NULL; for (;;) { if (rnode == NULL) return (NULL); slot = vm_radix_slot(index, rnode->rn_clev); tmp = vm_radix_node_load(&rnode->rn_child[slot], LOCKED); if (vm_radix_isleaf(tmp)) { m = vm_radix_topage(tmp); if (m->pindex != index) return (NULL); vm_radix_node_store(&rnode->rn_child[slot], NULL, LOCKED); rnode->rn_count--; if (rnode->rn_count > 1) return (m); for (i = 0; i < VM_RADIX_COUNT; i++) if (vm_radix_node_load(&rnode->rn_child[i], LOCKED) != NULL) break; KASSERT(i != VM_RADIX_COUNT, ("%s: invalid node configuration", __func__)); tmp = vm_radix_node_load(&rnode->rn_child[i], LOCKED); if (parent == NULL) vm_radix_root_store(rtree, tmp, LOCKED); else { slot = vm_radix_slot(index, parent->rn_clev); KASSERT(vm_radix_node_load( &parent->rn_child[slot], LOCKED) == rnode, ("%s: invalid child value", __func__)); vm_radix_node_store(&parent->rn_child[slot], tmp, LOCKED); } /* * The child is still valid and we can not zero the * pointer until all smr references are gone. */ rnode->rn_count--; vm_radix_node_put(rnode, i); return (m); } parent = rnode; rnode = tmp; } } /* * Remove and free all the nodes from the radix tree. * This function is recursive but there is a tight control on it as the * maximum depth of the tree is fixed. */ void vm_radix_reclaim_allnodes(struct vm_radix *rtree) { struct vm_radix_node *root; root = vm_radix_root_load(rtree, LOCKED); if (root == NULL) return; vm_radix_root_store(rtree, NULL, UNSERIALIZED); if (!vm_radix_isleaf(root)) vm_radix_reclaim_allnodes_int(root); } /* * Replace an existing page in the trie with another one. * Panics if there is not an old page in the trie at the new page's index. */ vm_page_t vm_radix_replace(struct vm_radix *rtree, vm_page_t newpage) { struct vm_radix_node *rnode, *tmp; vm_page_t m; vm_pindex_t index; int slot; index = newpage->pindex; rnode = vm_radix_root_load(rtree, LOCKED); if (rnode == NULL) panic("%s: replacing page on an empty trie", __func__); if (vm_radix_isleaf(rnode)) { m = vm_radix_topage(rnode); if (m->pindex != index) panic("%s: original replacing root key not found", __func__); rtree->rt_root = (uintptr_t)newpage | VM_RADIX_ISLEAF; return (m); } for (;;) { slot = vm_radix_slot(index, rnode->rn_clev); tmp = vm_radix_node_load(&rnode->rn_child[slot], LOCKED); if (vm_radix_isleaf(tmp)) { m = vm_radix_topage(tmp); if (m->pindex == index) { vm_radix_node_store(&rnode->rn_child[slot], (struct vm_radix_node *)((uintptr_t)newpage | VM_RADIX_ISLEAF), LOCKED); return (m); } else break; } else if (tmp == NULL || vm_radix_keybarr(tmp, index)) break; rnode = tmp; } panic("%s: original replacing page not found", __func__); } void vm_radix_wait(void) { uma_zwait(vm_radix_node_zone); } #ifdef DDB /* * Show details about the given radix node. */ DB_SHOW_COMMAND(radixnode, db_show_radixnode) { struct vm_radix_node *rnode, *tmp; int i; if (!have_addr) return; rnode = (struct vm_radix_node *)addr; db_printf("radixnode %p, owner %jx, children count %u, level %u:\n", (void *)rnode, (uintmax_t)rnode->rn_owner, rnode->rn_count, rnode->rn_clev); for (i = 0; i < VM_RADIX_COUNT; i++) { tmp = vm_radix_node_load(&rnode->rn_child[i], UNSERIALIZED); if (tmp != NULL) db_printf("slot: %d, val: %p, page: %p, clev: %d\n", i, (void *)tmp, vm_radix_isleaf(tmp) ? vm_radix_topage(tmp) : NULL, rnode->rn_clev); } } #endif /* DDB */ diff --git a/sys/vm/vm_reserv.c b/sys/vm/vm_reserv.c index 6023f886a034..c08895164171 100644 --- a/sys/vm/vm_reserv.c +++ b/sys/vm/vm_reserv.c @@ -1,1476 +1,1461 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (c) 2002-2006 Rice University * Copyright (c) 2007-2011 Alan L. Cox * All rights reserved. * * This software was developed for the FreeBSD Project by Alan L. Cox, * Olivier Crameri, Peter Druschel, Sitaram Iyer, and Juan Navarro. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT * HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS * OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY * WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE * POSSIBILITY OF SUCH DAMAGE. */ /* * Superpage reservation management module * * Any external functions defined by this module are only to be used by the * virtual memory system. */ #include __FBSDID("$FreeBSD$"); #include "opt_vm.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include /* * The reservation system supports the speculative allocation of large physical * pages ("superpages"). Speculative allocation enables the fully automatic * utilization of superpages by the virtual memory system. In other words, no * programmatic directives are required to use superpages. */ #if VM_NRESERVLEVEL > 0 #ifndef VM_LEVEL_0_ORDER_MAX #define VM_LEVEL_0_ORDER_MAX VM_LEVEL_0_ORDER #endif /* * The number of small pages that are contained in a level 0 reservation */ #define VM_LEVEL_0_NPAGES (1 << VM_LEVEL_0_ORDER) #define VM_LEVEL_0_NPAGES_MAX (1 << VM_LEVEL_0_ORDER_MAX) /* * The number of bits by which a physical address is shifted to obtain the * reservation number */ #define VM_LEVEL_0_SHIFT (VM_LEVEL_0_ORDER + PAGE_SHIFT) /* * The size of a level 0 reservation in bytes */ #define VM_LEVEL_0_SIZE (1 << VM_LEVEL_0_SHIFT) /* * Computes the index of the small page underlying the given (object, pindex) * within the reservation's array of small pages. */ #define VM_RESERV_INDEX(object, pindex) \ (((object)->pg_color + (pindex)) & (VM_LEVEL_0_NPAGES - 1)) /* * The size of a population map entry */ typedef u_long popmap_t; /* * The number of bits in a population map entry */ #define NBPOPMAP (NBBY * sizeof(popmap_t)) /* * The number of population map entries in a reservation */ #define NPOPMAP howmany(VM_LEVEL_0_NPAGES, NBPOPMAP) #define NPOPMAP_MAX howmany(VM_LEVEL_0_NPAGES_MAX, NBPOPMAP) /* * Number of elapsed ticks before we update the LRU queue position. Used * to reduce contention and churn on the list. */ #define PARTPOPSLOP 1 /* * Clear a bit in the population map. */ static __inline void popmap_clear(popmap_t popmap[], int i) { popmap[i / NBPOPMAP] &= ~(1UL << (i % NBPOPMAP)); } /* * Set a bit in the population map. */ static __inline void popmap_set(popmap_t popmap[], int i) { popmap[i / NBPOPMAP] |= 1UL << (i % NBPOPMAP); } /* * Is a bit in the population map clear? */ static __inline boolean_t popmap_is_clear(popmap_t popmap[], int i) { return ((popmap[i / NBPOPMAP] & (1UL << (i % NBPOPMAP))) == 0); } /* * Is a bit in the population map set? */ static __inline boolean_t popmap_is_set(popmap_t popmap[], int i) { return ((popmap[i / NBPOPMAP] & (1UL << (i % NBPOPMAP))) != 0); } /* * The reservation structure * * A reservation structure is constructed whenever a large physical page is * speculatively allocated to an object. The reservation provides the small * physical pages for the range [pindex, pindex + VM_LEVEL_0_NPAGES) of offsets * within that object. The reservation's "popcnt" tracks the number of these * small physical pages that are in use at any given time. When and if the * reservation is not fully utilized, it appears in the queue of partially * populated reservations. The reservation always appears on the containing * object's list of reservations. * * A partially populated reservation can be broken and reclaimed at any time. * * c - constant after boot * d - vm_reserv_domain_lock * o - vm_reserv_object_lock * r - vm_reserv_lock * s - vm_reserv_domain_scan_lock */ struct vm_reserv { struct mtx lock; /* reservation lock. */ TAILQ_ENTRY(vm_reserv) partpopq; /* (d, r) per-domain queue. */ LIST_ENTRY(vm_reserv) objq; /* (o, r) object queue */ vm_object_t object; /* (o, r) containing object */ vm_pindex_t pindex; /* (o, r) offset in object */ vm_page_t pages; /* (c) first page */ uint16_t popcnt; /* (r) # of pages in use */ uint8_t domain; /* (c) NUMA domain. */ char inpartpopq; /* (d, r) */ int lasttick; /* (r) last pop update tick. */ popmap_t popmap[NPOPMAP_MAX]; /* (r) bit vector, used pages */ }; TAILQ_HEAD(vm_reserv_queue, vm_reserv); #define vm_reserv_lockptr(rv) (&(rv)->lock) #define vm_reserv_assert_locked(rv) \ mtx_assert(vm_reserv_lockptr(rv), MA_OWNED) #define vm_reserv_lock(rv) mtx_lock(vm_reserv_lockptr(rv)) #define vm_reserv_trylock(rv) mtx_trylock(vm_reserv_lockptr(rv)) #define vm_reserv_unlock(rv) mtx_unlock(vm_reserv_lockptr(rv)) /* * The reservation array * * This array is analoguous in function to vm_page_array. It differs in the * respect that it may contain a greater number of useful reservation * structures than there are (physical) superpages. These "invalid" * reservation structures exist to trade-off space for time in the * implementation of vm_reserv_from_page(). Invalid reservation structures are * distinguishable from "valid" reservation structures by inspecting the * reservation's "pages" field. Invalid reservation structures have a NULL * "pages" field. * * vm_reserv_from_page() maps a small (physical) page to an element of this * array by computing a physical reservation number from the page's physical * address. The physical reservation number is used as the array index. * * An "active" reservation is a valid reservation structure that has a non-NULL * "object" field and a non-zero "popcnt" field. In other words, every active * reservation belongs to a particular object. Moreover, every active * reservation has an entry in the containing object's list of reservations. */ static vm_reserv_t vm_reserv_array; /* * The per-domain partially populated reservation queues * * These queues enable the fast recovery of an unused free small page from a * partially populated reservation. The reservation at the head of a queue * is the least recently changed, partially populated reservation. * * Access to this queue is synchronized by the per-domain reservation lock. * Threads reclaiming free pages from the queue must hold the per-domain scan * lock. */ struct vm_reserv_domain { struct mtx lock; struct vm_reserv_queue partpop; /* (d) */ struct vm_reserv marker; /* (d, s) scan marker/lock */ } __aligned(CACHE_LINE_SIZE); static struct vm_reserv_domain vm_rvd[MAXMEMDOM]; #define vm_reserv_domain_lockptr(d) (&vm_rvd[(d)].lock) #define vm_reserv_domain_assert_locked(d) \ mtx_assert(vm_reserv_domain_lockptr(d), MA_OWNED) #define vm_reserv_domain_lock(d) mtx_lock(vm_reserv_domain_lockptr(d)) #define vm_reserv_domain_unlock(d) mtx_unlock(vm_reserv_domain_lockptr(d)) #define vm_reserv_domain_scan_lock(d) mtx_lock(&vm_rvd[(d)].marker.lock) #define vm_reserv_domain_scan_unlock(d) mtx_unlock(&vm_rvd[(d)].marker.lock) static SYSCTL_NODE(_vm, OID_AUTO, reserv, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "Reservation Info"); -static counter_u64_t vm_reserv_broken = EARLY_COUNTER; +static COUNTER_U64_DEFINE_EARLY(vm_reserv_broken); SYSCTL_COUNTER_U64(_vm_reserv, OID_AUTO, broken, CTLFLAG_RD, &vm_reserv_broken, "Cumulative number of broken reservations"); -static counter_u64_t vm_reserv_freed = EARLY_COUNTER; +static COUNTER_U64_DEFINE_EARLY(vm_reserv_freed); SYSCTL_COUNTER_U64(_vm_reserv, OID_AUTO, freed, CTLFLAG_RD, &vm_reserv_freed, "Cumulative number of freed reservations"); static int sysctl_vm_reserv_fullpop(SYSCTL_HANDLER_ARGS); SYSCTL_PROC(_vm_reserv, OID_AUTO, fullpop, CTLTYPE_INT | CTLFLAG_MPSAFE | CTLFLAG_RD, NULL, 0, sysctl_vm_reserv_fullpop, "I", "Current number of full reservations"); static int sysctl_vm_reserv_partpopq(SYSCTL_HANDLER_ARGS); SYSCTL_OID(_vm_reserv, OID_AUTO, partpopq, CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_NEEDGIANT, NULL, 0, sysctl_vm_reserv_partpopq, "A", "Partially populated reservation queues"); -static counter_u64_t vm_reserv_reclaimed = EARLY_COUNTER; +static COUNTER_U64_DEFINE_EARLY(vm_reserv_reclaimed); SYSCTL_COUNTER_U64(_vm_reserv, OID_AUTO, reclaimed, CTLFLAG_RD, &vm_reserv_reclaimed, "Cumulative number of reclaimed reservations"); /* * The object lock pool is used to synchronize the rvq. We can not use a * pool mutex because it is required before malloc works. * * The "hash" function could be made faster without divide and modulo. */ #define VM_RESERV_OBJ_LOCK_COUNT MAXCPU struct mtx_padalign vm_reserv_object_mtx[VM_RESERV_OBJ_LOCK_COUNT]; #define vm_reserv_object_lock_idx(object) \ (((uintptr_t)object / sizeof(*object)) % VM_RESERV_OBJ_LOCK_COUNT) #define vm_reserv_object_lock_ptr(object) \ &vm_reserv_object_mtx[vm_reserv_object_lock_idx((object))] #define vm_reserv_object_lock(object) \ mtx_lock(vm_reserv_object_lock_ptr((object))) #define vm_reserv_object_unlock(object) \ mtx_unlock(vm_reserv_object_lock_ptr((object))) static void vm_reserv_break(vm_reserv_t rv); static void vm_reserv_depopulate(vm_reserv_t rv, int index); static vm_reserv_t vm_reserv_from_page(vm_page_t m); static boolean_t vm_reserv_has_pindex(vm_reserv_t rv, vm_pindex_t pindex); static void vm_reserv_populate(vm_reserv_t rv, int index); static void vm_reserv_reclaim(vm_reserv_t rv); /* * Returns the current number of full reservations. * * Since the number of full reservations is computed without acquiring any * locks, the returned value is inexact. */ static int sysctl_vm_reserv_fullpop(SYSCTL_HANDLER_ARGS) { vm_paddr_t paddr; struct vm_phys_seg *seg; vm_reserv_t rv; int fullpop, segind; fullpop = 0; for (segind = 0; segind < vm_phys_nsegs; segind++) { seg = &vm_phys_segs[segind]; paddr = roundup2(seg->start, VM_LEVEL_0_SIZE); while (paddr + VM_LEVEL_0_SIZE > paddr && paddr + VM_LEVEL_0_SIZE <= seg->end) { rv = &vm_reserv_array[paddr >> VM_LEVEL_0_SHIFT]; fullpop += rv->popcnt == VM_LEVEL_0_NPAGES; paddr += VM_LEVEL_0_SIZE; } } return (sysctl_handle_int(oidp, &fullpop, 0, req)); } /* * Describes the current state of the partially populated reservation queue. */ static int sysctl_vm_reserv_partpopq(SYSCTL_HANDLER_ARGS) { struct sbuf sbuf; vm_reserv_t rv; int counter, error, domain, level, unused_pages; error = sysctl_wire_old_buffer(req, 0); if (error != 0) return (error); sbuf_new_for_sysctl(&sbuf, NULL, 128, req); sbuf_printf(&sbuf, "\nDOMAIN LEVEL SIZE NUMBER\n\n"); for (domain = 0; domain < vm_ndomains; domain++) { for (level = -1; level <= VM_NRESERVLEVEL - 2; level++) { counter = 0; unused_pages = 0; vm_reserv_domain_lock(domain); TAILQ_FOREACH(rv, &vm_rvd[domain].partpop, partpopq) { if (rv == &vm_rvd[domain].marker) continue; counter++; unused_pages += VM_LEVEL_0_NPAGES - rv->popcnt; } vm_reserv_domain_unlock(domain); sbuf_printf(&sbuf, "%6d, %7d, %6dK, %6d\n", domain, level, unused_pages * ((int)PAGE_SIZE / 1024), counter); } } error = sbuf_finish(&sbuf); sbuf_delete(&sbuf); return (error); } /* * Remove a reservation from the object's objq. */ static void vm_reserv_remove(vm_reserv_t rv) { vm_object_t object; vm_reserv_assert_locked(rv); CTR5(KTR_VM, "%s: rv %p object %p popcnt %d inpartpop %d", __FUNCTION__, rv, rv->object, rv->popcnt, rv->inpartpopq); KASSERT(rv->object != NULL, ("vm_reserv_remove: reserv %p is free", rv)); KASSERT(!rv->inpartpopq, ("vm_reserv_remove: reserv %p's inpartpopq is TRUE", rv)); object = rv->object; vm_reserv_object_lock(object); LIST_REMOVE(rv, objq); rv->object = NULL; vm_reserv_object_unlock(object); } /* * Insert a new reservation into the object's objq. */ static void vm_reserv_insert(vm_reserv_t rv, vm_object_t object, vm_pindex_t pindex) { int i; vm_reserv_assert_locked(rv); CTR6(KTR_VM, "%s: rv %p(%p) object %p new %p popcnt %d", __FUNCTION__, rv, rv->pages, rv->object, object, rv->popcnt); KASSERT(rv->object == NULL, ("vm_reserv_insert: reserv %p isn't free", rv)); KASSERT(rv->popcnt == 0, ("vm_reserv_insert: reserv %p's popcnt is corrupted", rv)); KASSERT(!rv->inpartpopq, ("vm_reserv_insert: reserv %p's inpartpopq is TRUE", rv)); for (i = 0; i < NPOPMAP; i++) KASSERT(rv->popmap[i] == 0, ("vm_reserv_insert: reserv %p's popmap is corrupted", rv)); vm_reserv_object_lock(object); rv->pindex = pindex; rv->object = object; rv->lasttick = ticks; LIST_INSERT_HEAD(&object->rvq, rv, objq); vm_reserv_object_unlock(object); } /* * Reduces the given reservation's population count. If the population count * becomes zero, the reservation is destroyed. Additionally, moves the * reservation to the tail of the partially populated reservation queue if the * population count is non-zero. */ static void vm_reserv_depopulate(vm_reserv_t rv, int index) { struct vm_domain *vmd; vm_reserv_assert_locked(rv); CTR5(KTR_VM, "%s: rv %p object %p popcnt %d inpartpop %d", __FUNCTION__, rv, rv->object, rv->popcnt, rv->inpartpopq); KASSERT(rv->object != NULL, ("vm_reserv_depopulate: reserv %p is free", rv)); KASSERT(popmap_is_set(rv->popmap, index), ("vm_reserv_depopulate: reserv %p's popmap[%d] is clear", rv, index)); KASSERT(rv->popcnt > 0, ("vm_reserv_depopulate: reserv %p's popcnt is corrupted", rv)); KASSERT(rv->domain < vm_ndomains, ("vm_reserv_depopulate: reserv %p's domain is corrupted %d", rv, rv->domain)); if (rv->popcnt == VM_LEVEL_0_NPAGES) { KASSERT(rv->pages->psind == 1, ("vm_reserv_depopulate: reserv %p is already demoted", rv)); rv->pages->psind = 0; } popmap_clear(rv->popmap, index); rv->popcnt--; if ((unsigned)(ticks - rv->lasttick) >= PARTPOPSLOP || rv->popcnt == 0) { vm_reserv_domain_lock(rv->domain); if (rv->inpartpopq) { TAILQ_REMOVE(&vm_rvd[rv->domain].partpop, rv, partpopq); rv->inpartpopq = FALSE; } if (rv->popcnt != 0) { rv->inpartpopq = TRUE; TAILQ_INSERT_TAIL(&vm_rvd[rv->domain].partpop, rv, partpopq); } vm_reserv_domain_unlock(rv->domain); rv->lasttick = ticks; } vmd = VM_DOMAIN(rv->domain); if (rv->popcnt == 0) { vm_reserv_remove(rv); vm_domain_free_lock(vmd); vm_phys_free_pages(rv->pages, VM_LEVEL_0_ORDER); vm_domain_free_unlock(vmd); counter_u64_add(vm_reserv_freed, 1); } vm_domain_freecnt_inc(vmd, 1); } /* * Returns the reservation to which the given page might belong. */ static __inline vm_reserv_t vm_reserv_from_page(vm_page_t m) { return (&vm_reserv_array[VM_PAGE_TO_PHYS(m) >> VM_LEVEL_0_SHIFT]); } /* * Returns an existing reservation or NULL and initialized successor pointer. */ static vm_reserv_t vm_reserv_from_object(vm_object_t object, vm_pindex_t pindex, vm_page_t mpred, vm_page_t *msuccp) { vm_reserv_t rv; vm_page_t msucc; msucc = NULL; if (mpred != NULL) { KASSERT(mpred->object == object, ("vm_reserv_from_object: object doesn't contain mpred")); KASSERT(mpred->pindex < pindex, ("vm_reserv_from_object: mpred doesn't precede pindex")); rv = vm_reserv_from_page(mpred); if (rv->object == object && vm_reserv_has_pindex(rv, pindex)) goto found; msucc = TAILQ_NEXT(mpred, listq); } else msucc = TAILQ_FIRST(&object->memq); if (msucc != NULL) { KASSERT(msucc->pindex > pindex, ("vm_reserv_from_object: msucc doesn't succeed pindex")); rv = vm_reserv_from_page(msucc); if (rv->object == object && vm_reserv_has_pindex(rv, pindex)) goto found; } rv = NULL; found: *msuccp = msucc; return (rv); } /* * Returns TRUE if the given reservation contains the given page index and * FALSE otherwise. */ static __inline boolean_t vm_reserv_has_pindex(vm_reserv_t rv, vm_pindex_t pindex) { return (((pindex - rv->pindex) & ~(VM_LEVEL_0_NPAGES - 1)) == 0); } /* * Increases the given reservation's population count. Moves the reservation * to the tail of the partially populated reservation queue. */ static void vm_reserv_populate(vm_reserv_t rv, int index) { vm_reserv_assert_locked(rv); CTR5(KTR_VM, "%s: rv %p object %p popcnt %d inpartpop %d", __FUNCTION__, rv, rv->object, rv->popcnt, rv->inpartpopq); KASSERT(rv->object != NULL, ("vm_reserv_populate: reserv %p is free", rv)); KASSERT(popmap_is_clear(rv->popmap, index), ("vm_reserv_populate: reserv %p's popmap[%d] is set", rv, index)); KASSERT(rv->popcnt < VM_LEVEL_0_NPAGES, ("vm_reserv_populate: reserv %p is already full", rv)); KASSERT(rv->pages->psind == 0, ("vm_reserv_populate: reserv %p is already promoted", rv)); KASSERT(rv->domain < vm_ndomains, ("vm_reserv_populate: reserv %p's domain is corrupted %d", rv, rv->domain)); popmap_set(rv->popmap, index); rv->popcnt++; if ((unsigned)(ticks - rv->lasttick) < PARTPOPSLOP && rv->inpartpopq && rv->popcnt != VM_LEVEL_0_NPAGES) return; rv->lasttick = ticks; vm_reserv_domain_lock(rv->domain); if (rv->inpartpopq) { TAILQ_REMOVE(&vm_rvd[rv->domain].partpop, rv, partpopq); rv->inpartpopq = FALSE; } if (rv->popcnt < VM_LEVEL_0_NPAGES) { rv->inpartpopq = TRUE; TAILQ_INSERT_TAIL(&vm_rvd[rv->domain].partpop, rv, partpopq); } else { KASSERT(rv->pages->psind == 0, ("vm_reserv_populate: reserv %p is already promoted", rv)); rv->pages->psind = 1; } vm_reserv_domain_unlock(rv->domain); } /* * Allocates a contiguous set of physical pages of the given size "npages" * from existing or newly created reservations. All of the physical pages * must be at or above the given physical address "low" and below the given * physical address "high". The given value "alignment" determines the * alignment of the first physical page in the set. If the given value * "boundary" is non-zero, then the set of physical pages cannot cross any * physical address boundary that is a multiple of that value. Both * "alignment" and "boundary" must be a power of two. * * The page "mpred" must immediately precede the offset "pindex" within the * specified object. * * The object must be locked. */ vm_page_t vm_reserv_alloc_contig(vm_object_t object, vm_pindex_t pindex, int domain, int req, vm_page_t mpred, u_long npages, vm_paddr_t low, vm_paddr_t high, u_long alignment, vm_paddr_t boundary) { struct vm_domain *vmd; vm_paddr_t pa, size; vm_page_t m, m_ret, msucc; vm_pindex_t first, leftcap, rightcap; vm_reserv_t rv; u_long allocpages, maxpages, minpages; int i, index, n; VM_OBJECT_ASSERT_WLOCKED(object); KASSERT(npages != 0, ("vm_reserv_alloc_contig: npages is 0")); /* * Is a reservation fundamentally impossible? */ if (pindex < VM_RESERV_INDEX(object, pindex) || pindex + npages > object->size) return (NULL); /* * All reservations of a particular size have the same alignment. * Assuming that the first page is allocated from a reservation, the * least significant bits of its physical address can be determined * from its offset from the beginning of the reservation and the size * of the reservation. * * Could the specified index within a reservation of the smallest * possible size satisfy the alignment and boundary requirements? */ pa = VM_RESERV_INDEX(object, pindex) << PAGE_SHIFT; if ((pa & (alignment - 1)) != 0) return (NULL); size = npages << PAGE_SHIFT; if (((pa ^ (pa + size - 1)) & ~(boundary - 1)) != 0) return (NULL); /* * Look for an existing reservation. */ rv = vm_reserv_from_object(object, pindex, mpred, &msucc); if (rv != NULL) { KASSERT(object != kernel_object || rv->domain == domain, ("vm_reserv_alloc_contig: domain mismatch")); index = VM_RESERV_INDEX(object, pindex); /* Does the allocation fit within the reservation? */ if (index + npages > VM_LEVEL_0_NPAGES) return (NULL); domain = rv->domain; vmd = VM_DOMAIN(domain); vm_reserv_lock(rv); /* Handle reclaim race. */ if (rv->object != object) goto out; m = &rv->pages[index]; pa = VM_PAGE_TO_PHYS(m); if (pa < low || pa + size > high || (pa & (alignment - 1)) != 0 || ((pa ^ (pa + size - 1)) & ~(boundary - 1)) != 0) goto out; /* Handle vm_page_rename(m, new_object, ...). */ for (i = 0; i < npages; i++) if (popmap_is_set(rv->popmap, index + i)) goto out; if (!vm_domain_allocate(vmd, req, npages)) goto out; for (i = 0; i < npages; i++) vm_reserv_populate(rv, index + i); vm_reserv_unlock(rv); return (m); out: vm_reserv_unlock(rv); return (NULL); } /* * Could at least one reservation fit between the first index to the * left that can be used ("leftcap") and the first index to the right * that cannot be used ("rightcap")? * * We must synchronize with the reserv object lock to protect the * pindex/object of the resulting reservations against rename while * we are inspecting. */ first = pindex - VM_RESERV_INDEX(object, pindex); minpages = VM_RESERV_INDEX(object, pindex) + npages; maxpages = roundup2(minpages, VM_LEVEL_0_NPAGES); allocpages = maxpages; vm_reserv_object_lock(object); if (mpred != NULL) { if ((rv = vm_reserv_from_page(mpred))->object != object) leftcap = mpred->pindex + 1; else leftcap = rv->pindex + VM_LEVEL_0_NPAGES; if (leftcap > first) { vm_reserv_object_unlock(object); return (NULL); } } if (msucc != NULL) { if ((rv = vm_reserv_from_page(msucc))->object != object) rightcap = msucc->pindex; else rightcap = rv->pindex; if (first + maxpages > rightcap) { if (maxpages == VM_LEVEL_0_NPAGES) { vm_reserv_object_unlock(object); return (NULL); } /* * At least one reservation will fit between "leftcap" * and "rightcap". However, a reservation for the * last of the requested pages will not fit. Reduce * the size of the upcoming allocation accordingly. */ allocpages = minpages; } } vm_reserv_object_unlock(object); /* * Would the last new reservation extend past the end of the object? * * If the object is unlikely to grow don't allocate a reservation for * the tail. */ if ((object->flags & OBJ_ANON) == 0 && first + maxpages > object->size) { if (maxpages == VM_LEVEL_0_NPAGES) return (NULL); allocpages = minpages; } /* * Allocate the physical pages. The alignment and boundary specified * for this allocation may be different from the alignment and * boundary specified for the requested pages. For instance, the * specified index may not be the first page within the first new * reservation. */ m = NULL; vmd = VM_DOMAIN(domain); if (vm_domain_allocate(vmd, req, npages)) { vm_domain_free_lock(vmd); m = vm_phys_alloc_contig(domain, allocpages, low, high, ulmax(alignment, VM_LEVEL_0_SIZE), boundary > VM_LEVEL_0_SIZE ? boundary : 0); vm_domain_free_unlock(vmd); if (m == NULL) { vm_domain_freecnt_inc(vmd, npages); return (NULL); } } else return (NULL); KASSERT(vm_phys_domain(m) == domain, ("vm_reserv_alloc_contig: Page domain does not match requested.")); /* * The allocated physical pages always begin at a reservation * boundary, but they do not always end at a reservation boundary. * Initialize every reservation that is completely covered by the * allocated physical pages. */ m_ret = NULL; index = VM_RESERV_INDEX(object, pindex); do { rv = vm_reserv_from_page(m); KASSERT(rv->pages == m, ("vm_reserv_alloc_contig: reserv %p's pages is corrupted", rv)); vm_reserv_lock(rv); vm_reserv_insert(rv, object, first); n = ulmin(VM_LEVEL_0_NPAGES - index, npages); for (i = 0; i < n; i++) vm_reserv_populate(rv, index + i); npages -= n; if (m_ret == NULL) { m_ret = &rv->pages[index]; index = 0; } vm_reserv_unlock(rv); m += VM_LEVEL_0_NPAGES; first += VM_LEVEL_0_NPAGES; allocpages -= VM_LEVEL_0_NPAGES; } while (allocpages >= VM_LEVEL_0_NPAGES); return (m_ret); } /* * Allocate a physical page from an existing or newly created reservation. * * The page "mpred" must immediately precede the offset "pindex" within the * specified object. * * The object must be locked. */ vm_page_t vm_reserv_alloc_page(vm_object_t object, vm_pindex_t pindex, int domain, int req, vm_page_t mpred) { struct vm_domain *vmd; vm_page_t m, msucc; vm_pindex_t first, leftcap, rightcap; vm_reserv_t rv; int index; VM_OBJECT_ASSERT_WLOCKED(object); /* * Is a reservation fundamentally impossible? */ if (pindex < VM_RESERV_INDEX(object, pindex) || pindex >= object->size) return (NULL); /* * Look for an existing reservation. */ rv = vm_reserv_from_object(object, pindex, mpred, &msucc); if (rv != NULL) { KASSERT(object != kernel_object || rv->domain == domain, ("vm_reserv_alloc_page: domain mismatch")); domain = rv->domain; vmd = VM_DOMAIN(domain); index = VM_RESERV_INDEX(object, pindex); m = &rv->pages[index]; vm_reserv_lock(rv); /* Handle reclaim race. */ if (rv->object != object || /* Handle vm_page_rename(m, new_object, ...). */ popmap_is_set(rv->popmap, index)) { m = NULL; goto out; } if (vm_domain_allocate(vmd, req, 1) == 0) m = NULL; else vm_reserv_populate(rv, index); out: vm_reserv_unlock(rv); return (m); } /* * Could a reservation fit between the first index to the left that * can be used and the first index to the right that cannot be used? * * We must synchronize with the reserv object lock to protect the * pindex/object of the resulting reservations against rename while * we are inspecting. */ first = pindex - VM_RESERV_INDEX(object, pindex); vm_reserv_object_lock(object); if (mpred != NULL) { if ((rv = vm_reserv_from_page(mpred))->object != object) leftcap = mpred->pindex + 1; else leftcap = rv->pindex + VM_LEVEL_0_NPAGES; if (leftcap > first) { vm_reserv_object_unlock(object); return (NULL); } } if (msucc != NULL) { if ((rv = vm_reserv_from_page(msucc))->object != object) rightcap = msucc->pindex; else rightcap = rv->pindex; if (first + VM_LEVEL_0_NPAGES > rightcap) { vm_reserv_object_unlock(object); return (NULL); } } vm_reserv_object_unlock(object); /* * Would the last new reservation extend past the end of the object? * * If the object is unlikely to grow don't allocate a reservation for * the tail. */ if ((object->flags & OBJ_ANON) == 0 && first + VM_LEVEL_0_NPAGES > object->size) return (NULL); /* * Allocate and populate the new reservation. */ m = NULL; vmd = VM_DOMAIN(domain); if (vm_domain_allocate(vmd, req, 1)) { vm_domain_free_lock(vmd); m = vm_phys_alloc_pages(domain, VM_FREEPOOL_DEFAULT, VM_LEVEL_0_ORDER); vm_domain_free_unlock(vmd); if (m == NULL) { vm_domain_freecnt_inc(vmd, 1); return (NULL); } } else return (NULL); rv = vm_reserv_from_page(m); vm_reserv_lock(rv); KASSERT(rv->pages == m, ("vm_reserv_alloc_page: reserv %p's pages is corrupted", rv)); vm_reserv_insert(rv, object, first); index = VM_RESERV_INDEX(object, pindex); vm_reserv_populate(rv, index); vm_reserv_unlock(rv); return (&rv->pages[index]); } /* * Breaks the given reservation. All free pages in the reservation * are returned to the physical memory allocator. The reservation's * population count and map are reset to their initial state. * * The given reservation must not be in the partially populated reservation * queue. */ static void vm_reserv_break(vm_reserv_t rv) { u_long changes; int bitpos, hi, i, lo; vm_reserv_assert_locked(rv); CTR5(KTR_VM, "%s: rv %p object %p popcnt %d inpartpop %d", __FUNCTION__, rv, rv->object, rv->popcnt, rv->inpartpopq); vm_reserv_remove(rv); rv->pages->psind = 0; hi = lo = -1; for (i = 0; i <= NPOPMAP; i++) { /* * "changes" is a bitmask that marks where a new sequence of * 0s or 1s begins in popmap[i], with last bit in popmap[i-1] * considered to be 1 if and only if lo == hi. The bits of * popmap[-1] and popmap[NPOPMAP] are considered all 1s. */ if (i == NPOPMAP) changes = lo != hi; else { changes = rv->popmap[i]; changes ^= (changes << 1) | (lo == hi); rv->popmap[i] = 0; } while (changes != 0) { /* * If the next change marked begins a run of 0s, set * lo to mark that position. Otherwise set hi and * free pages from lo up to hi. */ bitpos = ffsl(changes) - 1; changes ^= 1UL << bitpos; if (lo == hi) lo = NBPOPMAP * i + bitpos; else { hi = NBPOPMAP * i + bitpos; vm_domain_free_lock(VM_DOMAIN(rv->domain)); vm_phys_enqueue_contig(&rv->pages[lo], hi - lo); vm_domain_free_unlock(VM_DOMAIN(rv->domain)); lo = hi; } } } rv->popcnt = 0; counter_u64_add(vm_reserv_broken, 1); } /* * Breaks all reservations belonging to the given object. */ void vm_reserv_break_all(vm_object_t object) { vm_reserv_t rv; /* * This access of object->rvq is unsynchronized so that the * object rvq lock can nest after the domain_free lock. We * must check for races in the results. However, the object * lock prevents new additions, so we are guaranteed that when * it returns NULL the object is properly empty. */ while ((rv = LIST_FIRST(&object->rvq)) != NULL) { vm_reserv_lock(rv); /* Reclaim race. */ if (rv->object != object) { vm_reserv_unlock(rv); continue; } vm_reserv_domain_lock(rv->domain); if (rv->inpartpopq) { TAILQ_REMOVE(&vm_rvd[rv->domain].partpop, rv, partpopq); rv->inpartpopq = FALSE; } vm_reserv_domain_unlock(rv->domain); vm_reserv_break(rv); vm_reserv_unlock(rv); } } /* * Frees the given page if it belongs to a reservation. Returns TRUE if the * page is freed and FALSE otherwise. */ boolean_t vm_reserv_free_page(vm_page_t m) { vm_reserv_t rv; boolean_t ret; rv = vm_reserv_from_page(m); if (rv->object == NULL) return (FALSE); vm_reserv_lock(rv); /* Re-validate after lock. */ if (rv->object != NULL) { vm_reserv_depopulate(rv, m - rv->pages); ret = TRUE; } else ret = FALSE; vm_reserv_unlock(rv); return (ret); } /* * Initializes the reservation management system. Specifically, initializes * the reservation array. * * Requires that vm_page_array and first_page are initialized! */ void vm_reserv_init(void) { vm_paddr_t paddr; struct vm_phys_seg *seg; struct vm_reserv *rv; struct vm_reserv_domain *rvd; int i, j, segind; /* * Initialize the reservation array. Specifically, initialize the * "pages" field for every element that has an underlying superpage. */ for (segind = 0; segind < vm_phys_nsegs; segind++) { seg = &vm_phys_segs[segind]; paddr = roundup2(seg->start, VM_LEVEL_0_SIZE); while (paddr + VM_LEVEL_0_SIZE > paddr && paddr + VM_LEVEL_0_SIZE <= seg->end) { rv = &vm_reserv_array[paddr >> VM_LEVEL_0_SHIFT]; rv->pages = PHYS_TO_VM_PAGE(paddr); rv->domain = seg->domain; mtx_init(&rv->lock, "vm reserv", NULL, MTX_DEF); paddr += VM_LEVEL_0_SIZE; } } for (i = 0; i < MAXMEMDOM; i++) { rvd = &vm_rvd[i]; mtx_init(&rvd->lock, "vm reserv domain", NULL, MTX_DEF); TAILQ_INIT(&rvd->partpop); mtx_init(&rvd->marker.lock, "vm reserv marker", NULL, MTX_DEF); /* * Fully populated reservations should never be present in the * partially populated reservation queues. */ rvd->marker.popcnt = VM_LEVEL_0_NPAGES; for (j = 0; j < NBPOPMAP; j++) popmap_set(rvd->marker.popmap, j); } for (i = 0; i < VM_RESERV_OBJ_LOCK_COUNT; i++) mtx_init(&vm_reserv_object_mtx[i], "resv obj lock", NULL, MTX_DEF); } /* * Returns true if the given page belongs to a reservation and that page is * free. Otherwise, returns false. */ bool vm_reserv_is_page_free(vm_page_t m) { vm_reserv_t rv; rv = vm_reserv_from_page(m); if (rv->object == NULL) return (false); return (popmap_is_clear(rv->popmap, m - rv->pages)); } /* * If the given page belongs to a reservation, returns the level of that * reservation. Otherwise, returns -1. */ int vm_reserv_level(vm_page_t m) { vm_reserv_t rv; rv = vm_reserv_from_page(m); return (rv->object != NULL ? 0 : -1); } /* * Returns a reservation level if the given page belongs to a fully populated * reservation and -1 otherwise. */ int vm_reserv_level_iffullpop(vm_page_t m) { vm_reserv_t rv; rv = vm_reserv_from_page(m); return (rv->popcnt == VM_LEVEL_0_NPAGES ? 0 : -1); } /* * Remove a partially populated reservation from the queue. */ static void vm_reserv_dequeue(vm_reserv_t rv) { vm_reserv_domain_assert_locked(rv->domain); vm_reserv_assert_locked(rv); CTR5(KTR_VM, "%s: rv %p object %p popcnt %d inpartpop %d", __FUNCTION__, rv, rv->object, rv->popcnt, rv->inpartpopq); KASSERT(rv->inpartpopq, ("vm_reserv_reclaim: reserv %p's inpartpopq is FALSE", rv)); TAILQ_REMOVE(&vm_rvd[rv->domain].partpop, rv, partpopq); rv->inpartpopq = FALSE; } /* * Breaks the given partially populated reservation, releasing its free pages * to the physical memory allocator. */ static void vm_reserv_reclaim(vm_reserv_t rv) { vm_reserv_assert_locked(rv); CTR5(KTR_VM, "%s: rv %p object %p popcnt %d inpartpop %d", __FUNCTION__, rv, rv->object, rv->popcnt, rv->inpartpopq); if (rv->inpartpopq) { vm_reserv_domain_lock(rv->domain); vm_reserv_dequeue(rv); vm_reserv_domain_unlock(rv->domain); } vm_reserv_break(rv); counter_u64_add(vm_reserv_reclaimed, 1); } /* * Breaks a reservation near the head of the partially populated reservation * queue, releasing its free pages to the physical memory allocator. Returns * TRUE if a reservation is broken and FALSE otherwise. */ bool vm_reserv_reclaim_inactive(int domain) { vm_reserv_t rv; vm_reserv_domain_lock(domain); TAILQ_FOREACH(rv, &vm_rvd[domain].partpop, partpopq) { /* * A locked reservation is likely being updated or reclaimed, * so just skip ahead. */ if (rv != &vm_rvd[domain].marker && vm_reserv_trylock(rv)) { vm_reserv_dequeue(rv); break; } } vm_reserv_domain_unlock(domain); if (rv != NULL) { vm_reserv_reclaim(rv); vm_reserv_unlock(rv); return (true); } return (false); } /* * Determine whether this reservation has free pages that satisfy the given * request for contiguous physical memory. Start searching from the lower * bound, defined by low_index. */ static bool vm_reserv_test_contig(vm_reserv_t rv, u_long npages, vm_paddr_t low, vm_paddr_t high, u_long alignment, vm_paddr_t boundary) { vm_paddr_t pa, size; u_long changes; int bitpos, bits_left, i, hi, lo, n; vm_reserv_assert_locked(rv); size = npages << PAGE_SHIFT; pa = VM_PAGE_TO_PHYS(&rv->pages[0]); lo = (pa < low) ? ((low + PAGE_MASK - pa) >> PAGE_SHIFT) : 0; i = lo / NBPOPMAP; changes = rv->popmap[i] | ((1UL << (lo % NBPOPMAP)) - 1); hi = (pa + VM_LEVEL_0_SIZE > high) ? ((high + PAGE_MASK - pa) >> PAGE_SHIFT) : VM_LEVEL_0_NPAGES; n = hi / NBPOPMAP; bits_left = hi % NBPOPMAP; hi = lo = -1; for (;;) { /* * "changes" is a bitmask that marks where a new sequence of * 0s or 1s begins in popmap[i], with last bit in popmap[i-1] * considered to be 1 if and only if lo == hi. The bits of * popmap[-1] and popmap[NPOPMAP] are considered all 1s. */ changes ^= (changes << 1) | (lo == hi); while (changes != 0) { /* * If the next change marked begins a run of 0s, set * lo to mark that position. Otherwise set hi and * look for a satisfactory first page from lo up to hi. */ bitpos = ffsl(changes) - 1; changes ^= 1UL << bitpos; if (lo == hi) { lo = NBPOPMAP * i + bitpos; continue; } hi = NBPOPMAP * i + bitpos; pa = VM_PAGE_TO_PHYS(&rv->pages[lo]); if ((pa & (alignment - 1)) != 0) { /* Skip to next aligned page. */ lo += (((pa - 1) | (alignment - 1)) + 1) >> PAGE_SHIFT; if (lo >= VM_LEVEL_0_NPAGES) return (false); pa = VM_PAGE_TO_PHYS(&rv->pages[lo]); } if (((pa ^ (pa + size - 1)) & ~(boundary - 1)) != 0) { /* Skip to next boundary-matching page. */ lo += (((pa - 1) | (boundary - 1)) + 1) >> PAGE_SHIFT; if (lo >= VM_LEVEL_0_NPAGES) return (false); pa = VM_PAGE_TO_PHYS(&rv->pages[lo]); } if (lo * PAGE_SIZE + size <= hi * PAGE_SIZE) return (true); lo = hi; } if (++i < n) changes = rv->popmap[i]; else if (i == n) changes = bits_left == 0 ? -1UL : (rv->popmap[n] | (-1UL << bits_left)); else return (false); } } /* * Searches the partially populated reservation queue for the least recently * changed reservation with free pages that satisfy the given request for * contiguous physical memory. If a satisfactory reservation is found, it is * broken. Returns true if a reservation is broken and false otherwise. */ bool vm_reserv_reclaim_contig(int domain, u_long npages, vm_paddr_t low, vm_paddr_t high, u_long alignment, vm_paddr_t boundary) { struct vm_reserv_queue *queue; vm_paddr_t pa, size; vm_reserv_t marker, rv, rvn; if (npages > VM_LEVEL_0_NPAGES - 1) return (false); marker = &vm_rvd[domain].marker; queue = &vm_rvd[domain].partpop; size = npages << PAGE_SHIFT; vm_reserv_domain_scan_lock(domain); vm_reserv_domain_lock(domain); TAILQ_FOREACH_SAFE(rv, queue, partpopq, rvn) { pa = VM_PAGE_TO_PHYS(&rv->pages[0]); if (pa + VM_LEVEL_0_SIZE - size < low) { /* This entire reservation is too low; go to next. */ continue; } if (pa + size > high) { /* This entire reservation is too high; go to next. */ continue; } if (vm_reserv_trylock(rv) == 0) { TAILQ_INSERT_AFTER(queue, rv, marker, partpopq); vm_reserv_domain_unlock(domain); vm_reserv_lock(rv); if (!rv->inpartpopq || TAILQ_NEXT(rv, partpopq) != marker) { vm_reserv_unlock(rv); vm_reserv_domain_lock(domain); rvn = TAILQ_NEXT(marker, partpopq); TAILQ_REMOVE(queue, marker, partpopq); continue; } vm_reserv_domain_lock(domain); TAILQ_REMOVE(queue, marker, partpopq); } vm_reserv_domain_unlock(domain); if (vm_reserv_test_contig(rv, npages, low, high, alignment, boundary)) { vm_reserv_domain_scan_unlock(domain); vm_reserv_reclaim(rv); vm_reserv_unlock(rv); return (true); } vm_reserv_unlock(rv); vm_reserv_domain_lock(domain); } vm_reserv_domain_unlock(domain); vm_reserv_domain_scan_unlock(domain); return (false); } /* * Transfers the reservation underlying the given page to a new object. * * The object must be locked. */ void vm_reserv_rename(vm_page_t m, vm_object_t new_object, vm_object_t old_object, vm_pindex_t old_object_offset) { vm_reserv_t rv; VM_OBJECT_ASSERT_WLOCKED(new_object); rv = vm_reserv_from_page(m); if (rv->object == old_object) { vm_reserv_lock(rv); CTR6(KTR_VM, "%s: rv %p object %p new %p popcnt %d inpartpop %d", __FUNCTION__, rv, rv->object, new_object, rv->popcnt, rv->inpartpopq); if (rv->object == old_object) { vm_reserv_object_lock(old_object); rv->object = NULL; LIST_REMOVE(rv, objq); vm_reserv_object_unlock(old_object); vm_reserv_object_lock(new_object); rv->object = new_object; rv->pindex -= old_object_offset; LIST_INSERT_HEAD(&new_object->rvq, rv, objq); vm_reserv_object_unlock(new_object); } vm_reserv_unlock(rv); } } /* * Returns the size (in bytes) of a reservation of the specified level. */ int vm_reserv_size(int level) { switch (level) { case 0: return (VM_LEVEL_0_SIZE); case -1: return (PAGE_SIZE); default: return (0); } } /* * Allocates the virtual and physical memory required by the reservation * management system's data structures, in particular, the reservation array. */ vm_paddr_t vm_reserv_startup(vm_offset_t *vaddr, vm_paddr_t end) { vm_paddr_t new_end, high_water; size_t size; int i; high_water = phys_avail[1]; for (i = 0; i < vm_phys_nsegs; i++) { if (vm_phys_segs[i].end > high_water) high_water = vm_phys_segs[i].end; } /* Skip the first chunk. It is already accounted for. */ for (i = 2; phys_avail[i + 1] != 0; i += 2) { if (phys_avail[i + 1] > high_water) high_water = phys_avail[i + 1]; } /* * Calculate the size (in bytes) of the reservation array. Round up * from "high_water" because every small page is mapped to an element * in the reservation array based on its physical address. Thus, the * number of elements in the reservation array can be greater than the * number of superpages. */ size = howmany(high_water, VM_LEVEL_0_SIZE) * sizeof(struct vm_reserv); /* * Allocate and map the physical memory for the reservation array. The * next available virtual address is returned by reference. */ new_end = end - round_page(size); vm_reserv_array = (void *)(uintptr_t)pmap_map(vaddr, new_end, end, VM_PROT_READ | VM_PROT_WRITE); bzero(vm_reserv_array, size); /* * Return the next available physical address. */ return (new_end); } -/* - * Initializes the reservation management system. Specifically, initializes - * the reservation counters. - */ -static void -vm_reserv_counter_init(void *unused) -{ - - vm_reserv_freed = counter_u64_alloc(M_WAITOK); - vm_reserv_broken = counter_u64_alloc(M_WAITOK); - vm_reserv_reclaimed = counter_u64_alloc(M_WAITOK); -} -SYSINIT(vm_reserv_counter_init, SI_SUB_CPU, SI_ORDER_ANY, - vm_reserv_counter_init, NULL); - /* * Returns the superpage containing the given page. */ vm_page_t vm_reserv_to_superpage(vm_page_t m) { vm_reserv_t rv; VM_OBJECT_ASSERT_LOCKED(m->object); rv = vm_reserv_from_page(m); if (rv->object == m->object && rv->popcnt == VM_LEVEL_0_NPAGES) m = rv->pages; else m = NULL; return (m); } #endif /* VM_NRESERVLEVEL > 0 */ diff --git a/usr.bin/netstat/inet.c b/usr.bin/netstat/inet.c index 6712cd66e578..d4f9b23c10b7 100644 --- a/usr.bin/netstat/inet.c +++ b/usr.bin/netstat/inet.c @@ -1,1428 +1,1444 @@ /*- * Copyright (c) 1983, 1988, 1993, 1995 * The Regents of the University of California. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * 3. Neither the name of the University nor the names of its contributors * may be used to endorse or promote products derived from this software * without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #if 0 #ifndef lint static char sccsid[] = "@(#)inet.c 8.5 (Berkeley) 5/24/95"; #endif /* not lint */ #endif #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #define _WANT_SOCKET #include #include #include #include #include #include #include #include #ifdef INET6 #include #endif /* INET6 */ #include #include #include #include #include #include #include #include #include #define TCPSTATES #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "netstat.h" #include "nl_defs.h" #ifdef INET static void inetprint(const char *, struct in_addr *, int, const char *, int, const int); #endif #ifdef INET6 static int udp_done, tcp_done, sdp_done; #endif /* INET6 */ static int pcblist_sysctl(int proto, const char *name, char **bufp) { const char *mibvar; char *buf; size_t len; switch (proto) { case IPPROTO_TCP: mibvar = "net.inet.tcp.pcblist"; break; case IPPROTO_UDP: mibvar = "net.inet.udp.pcblist"; break; case IPPROTO_DIVERT: mibvar = "net.inet.divert.pcblist"; break; default: mibvar = "net.inet.raw.pcblist"; break; } if (strncmp(name, "sdp", 3) == 0) mibvar = "net.inet.sdp.pcblist"; len = 0; if (sysctlbyname(mibvar, 0, &len, 0, 0) < 0) { if (errno != ENOENT) xo_warn("sysctl: %s", mibvar); return (0); } if ((buf = malloc(len)) == NULL) { xo_warnx("malloc %lu bytes", (u_long)len); return (0); } if (sysctlbyname(mibvar, buf, &len, 0, 0) < 0) { xo_warn("sysctl: %s", mibvar); free(buf); return (0); } *bufp = buf; return (1); } /* * Copied directly from uipc_socket2.c. We leave out some fields that are in * nested structures that aren't used to avoid extra work. */ static void sbtoxsockbuf(struct sockbuf *sb, struct xsockbuf *xsb) { xsb->sb_cc = sb->sb_ccc; xsb->sb_hiwat = sb->sb_hiwat; xsb->sb_mbcnt = sb->sb_mbcnt; xsb->sb_mcnt = sb->sb_mcnt; xsb->sb_ccnt = sb->sb_ccnt; xsb->sb_mbmax = sb->sb_mbmax; xsb->sb_lowat = sb->sb_lowat; xsb->sb_flags = sb->sb_flags; xsb->sb_timeo = sb->sb_timeo; } int sotoxsocket(struct socket *so, struct xsocket *xso) { struct protosw proto; struct domain domain; bzero(xso, sizeof *xso); xso->xso_len = sizeof *xso; xso->xso_so = (uintptr_t)so; xso->so_type = so->so_type; xso->so_options = so->so_options; xso->so_linger = so->so_linger; xso->so_state = so->so_state; xso->so_pcb = (uintptr_t)so->so_pcb; if (kread((uintptr_t)so->so_proto, &proto, sizeof(proto)) != 0) return (-1); xso->xso_protocol = proto.pr_protocol; if (kread((uintptr_t)proto.pr_domain, &domain, sizeof(domain)) != 0) return (-1); xso->xso_family = domain.dom_family; xso->so_timeo = so->so_timeo; xso->so_error = so->so_error; if ((so->so_options & SO_ACCEPTCONN) != 0) { xso->so_qlen = so->sol_qlen; xso->so_incqlen = so->sol_incqlen; xso->so_qlimit = so->sol_qlimit; } else { sbtoxsockbuf(&so->so_snd, &xso->so_snd); sbtoxsockbuf(&so->so_rcv, &xso->so_rcv); xso->so_oobmark = so->so_oobmark; } return (0); } /* * Print a summary of connections related to an Internet * protocol. For TCP, also give state of connection. * Listening processes (aflag) are suppressed unless the * -a (all) flag is specified. */ void protopr(u_long off, const char *name, int af1, int proto) { static int first = 1; int istcp; char *buf; const char *vchar; struct xtcpcb *tp; struct xinpcb *inp; struct xinpgen *xig, *oxig; struct xsocket *so; istcp = 0; switch (proto) { case IPPROTO_TCP: #ifdef INET6 if (strncmp(name, "sdp", 3) != 0) { if (tcp_done != 0) return; else tcp_done = 1; } else { if (sdp_done != 0) return; else sdp_done = 1; } #endif istcp = 1; break; case IPPROTO_UDP: #ifdef INET6 if (udp_done != 0) return; else udp_done = 1; #endif break; } if (!pcblist_sysctl(proto, name, &buf)) return; oxig = xig = (struct xinpgen *)buf; for (xig = (struct xinpgen *)((char *)xig + xig->xig_len); xig->xig_len > sizeof(struct xinpgen); xig = (struct xinpgen *)((char *)xig + xig->xig_len)) { if (istcp) { tp = (struct xtcpcb *)xig; inp = &tp->xt_inp; } else { inp = (struct xinpcb *)xig; } so = &inp->xi_socket; /* Ignore sockets for protocols other than the desired one. */ if (so->xso_protocol != proto) continue; /* Ignore PCBs which were freed during copyout. */ if (inp->inp_gencnt > oxig->xig_gen) continue; if ((af1 == AF_INET && (inp->inp_vflag & INP_IPV4) == 0) #ifdef INET6 || (af1 == AF_INET6 && (inp->inp_vflag & INP_IPV6) == 0) #endif /* INET6 */ || (af1 == AF_UNSPEC && ((inp->inp_vflag & INP_IPV4) == 0 #ifdef INET6 && (inp->inp_vflag & INP_IPV6) == 0 #endif /* INET6 */ )) ) continue; if (!aflag && ( (istcp && tp->t_state == TCPS_LISTEN) || (af1 == AF_INET && inet_lnaof(inp->inp_laddr) == INADDR_ANY) #ifdef INET6 || (af1 == AF_INET6 && IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) #endif /* INET6 */ || (af1 == AF_UNSPEC && (((inp->inp_vflag & INP_IPV4) != 0 && inet_lnaof(inp->inp_laddr) == INADDR_ANY) #ifdef INET6 || ((inp->inp_vflag & INP_IPV6) != 0 && IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)) #endif )) )) continue; if (first) { if (!Lflag) { xo_emit("Active Internet connections"); if (aflag) xo_emit(" (including servers)"); } else xo_emit( "Current listen queue sizes (qlen/incqlen/maxqlen)"); xo_emit("\n"); if (Aflag) xo_emit("{T:/%-*s} ", 2 * (int)sizeof(void *), "Tcpcb"); if (Lflag) xo_emit((Aflag && !Wflag) ? "{T:/%-5.5s} {T:/%-32.32s} {T:/%-18.18s}" : ((!Wflag || af1 == AF_INET) ? "{T:/%-5.5s} {T:/%-32.32s} {T:/%-22.22s}" : "{T:/%-5.5s} {T:/%-32.32s} {T:/%-45.45s}"), "Proto", "Listen", "Local Address"); else if (Tflag) xo_emit((Aflag && !Wflag) ? "{T:/%-5.5s} {T:/%-6.6s} {T:/%-6.6s} {T:/%-6.6s} {T:/%-18.18s} {T:/%s}" : ((!Wflag || af1 == AF_INET) ? "{T:/%-5.5s} {T:/%-6.6s} {T:/%-6.6s} {T:/%-6.6s} {T:/%-22.22s} {T:/%s}" : "{T:/%-5.5s} {T:/%-6.6s} {T:/%-6.6s} {T:/%-6.6s} {T:/%-45.45s} {T:/%s}"), "Proto", "Rexmit", "OOORcv", "0-win", "Local Address", "Foreign Address"); else { xo_emit((Aflag && !Wflag) ? "{T:/%-5.5s} {T:/%-6.6s} {T:/%-6.6s} {T:/%-18.18s} {T:/%-18.18s}" : ((!Wflag || af1 == AF_INET) ? "{T:/%-5.5s} {T:/%-6.6s} {T:/%-6.6s} {T:/%-22.22s} {T:/%-22.22s}" : "{T:/%-5.5s} {T:/%-6.6s} {T:/%-6.6s} {T:/%-45.45s} {T:/%-45.45s}"), "Proto", "Recv-Q", "Send-Q", "Local Address", "Foreign Address"); if (!xflag && !Rflag) xo_emit(" {T:/%-11.11s}", "(state)"); } if (xflag) { xo_emit(" {T:/%-6.6s} {T:/%-6.6s} {T:/%-6.6s} " "{T:/%-6.6s} {T:/%-6.6s} {T:/%-6.6s} " "{T:/%-6.6s} {T:/%-6.6s} {T:/%-6.6s} " "{T:/%-6.6s} {T:/%-6.6s} {T:/%-6.6s}", "R-MBUF", "S-MBUF", "R-CLUS", "S-CLUS", "R-HIWA", "S-HIWA", "R-LOWA", "S-LOWA", "R-BCNT", "S-BCNT", "R-BMAX", "S-BMAX"); xo_emit(" {T:/%7.7s} {T:/%7.7s} {T:/%7.7s} " "{T:/%7.7s} {T:/%7.7s} {T:/%7.7s}", "rexmt", "persist", "keep", "2msl", "delack", "rcvtime"); } else if (Rflag) { xo_emit(" {T:/%8.8s} {T:/%5.5s}", "flowid", "ftype"); } if (Pflag) xo_emit(" {T:/%s}", "Log ID"); xo_emit("\n"); first = 0; } if (Lflag && so->so_qlimit == 0) continue; xo_open_instance("socket"); if (Aflag) { if (istcp) xo_emit("{q:address/%*lx} ", 2 * (int)sizeof(void *), (u_long)inp->inp_ppcb); else xo_emit("{q:address/%*lx} ", 2 * (int)sizeof(void *), (u_long)so->so_pcb); } #ifdef INET6 if ((inp->inp_vflag & INP_IPV6) != 0) vchar = ((inp->inp_vflag & INP_IPV4) != 0) ? "46" : "6"; else #endif vchar = ((inp->inp_vflag & INP_IPV4) != 0) ? "4" : ""; if (istcp && (tp->t_flags & TF_TOE) != 0) xo_emit("{:protocol/%-3.3s%-2.2s/%s%s} ", "toe", vchar); else xo_emit("{:protocol/%-3.3s%-2.2s/%s%s} ", name, vchar); if (Lflag) { char buf1[33]; snprintf(buf1, sizeof buf1, "%u/%u/%u", so->so_qlen, so->so_incqlen, so->so_qlimit); xo_emit("{:listen-queue-sizes/%-32.32s} ", buf1); } else if (Tflag) { if (istcp) xo_emit("{:sent-retransmit-packets/%6u} " "{:received-out-of-order-packets/%6u} " "{:sent-zero-window/%6u} ", tp->t_sndrexmitpack, tp->t_rcvoopack, tp->t_sndzerowin); else xo_emit("{P:/%21s}", ""); } else { xo_emit("{:receive-bytes-waiting/%6u} " "{:send-bytes-waiting/%6u} ", so->so_rcv.sb_cc, so->so_snd.sb_cc); } if (numeric_port) { #ifdef INET if (inp->inp_vflag & INP_IPV4) { inetprint("local", &inp->inp_laddr, (int)inp->inp_lport, name, 1, af1); if (!Lflag) inetprint("remote", &inp->inp_faddr, (int)inp->inp_fport, name, 1, af1); } #endif #if defined(INET) && defined(INET6) else #endif #ifdef INET6 if (inp->inp_vflag & INP_IPV6) { inet6print("local", &inp->in6p_laddr, (int)inp->inp_lport, name, 1); if (!Lflag) inet6print("remote", &inp->in6p_faddr, (int)inp->inp_fport, name, 1); } /* else nothing printed now */ #endif /* INET6 */ } else if (inp->inp_flags & INP_ANONPORT) { #ifdef INET if (inp->inp_vflag & INP_IPV4) { inetprint("local", &inp->inp_laddr, (int)inp->inp_lport, name, 1, af1); if (!Lflag) inetprint("remote", &inp->inp_faddr, (int)inp->inp_fport, name, 0, af1); } #endif #if defined(INET) && defined(INET6) else #endif #ifdef INET6 if (inp->inp_vflag & INP_IPV6) { inet6print("local", &inp->in6p_laddr, (int)inp->inp_lport, name, 1); if (!Lflag) inet6print("remote", &inp->in6p_faddr, (int)inp->inp_fport, name, 0); } /* else nothing printed now */ #endif /* INET6 */ } else { #ifdef INET if (inp->inp_vflag & INP_IPV4) { inetprint("local", &inp->inp_laddr, (int)inp->inp_lport, name, 0, af1); if (!Lflag) inetprint("remote", &inp->inp_faddr, (int)inp->inp_fport, name, inp->inp_lport != inp->inp_fport, af1); } #endif #if defined(INET) && defined(INET6) else #endif #ifdef INET6 if (inp->inp_vflag & INP_IPV6) { inet6print("local", &inp->in6p_laddr, (int)inp->inp_lport, name, 0); if (!Lflag) inet6print("remote", &inp->in6p_faddr, (int)inp->inp_fport, name, inp->inp_lport != inp->inp_fport); } /* else nothing printed now */ #endif /* INET6 */ } if (xflag) { xo_emit("{:receive-mbufs/%6u} {:send-mbufs/%6u} " "{:receive-clusters/%6u} {:send-clusters/%6u} " "{:receive-high-water/%6u} {:send-high-water/%6u} " "{:receive-low-water/%6u} {:send-low-water/%6u} " "{:receive-mbuf-bytes/%6u} {:send-mbuf-bytes/%6u} " "{:receive-mbuf-bytes-max/%6u} " "{:send-mbuf-bytes-max/%6u}", so->so_rcv.sb_mcnt, so->so_snd.sb_mcnt, so->so_rcv.sb_ccnt, so->so_snd.sb_ccnt, so->so_rcv.sb_hiwat, so->so_snd.sb_hiwat, so->so_rcv.sb_lowat, so->so_snd.sb_lowat, so->so_rcv.sb_mbcnt, so->so_snd.sb_mbcnt, so->so_rcv.sb_mbmax, so->so_snd.sb_mbmax); if (istcp) xo_emit(" {:retransmit-timer/%4d.%02d} " "{:persist-timer/%4d.%02d} " "{:keepalive-timer/%4d.%02d} " "{:msl2-timer/%4d.%02d} " "{:delay-ack-timer/%4d.%02d} " "{:inactivity-timer/%4d.%02d}", tp->tt_rexmt / 1000, (tp->tt_rexmt % 1000) / 10, tp->tt_persist / 1000, (tp->tt_persist % 1000) / 10, tp->tt_keep / 1000, (tp->tt_keep % 1000) / 10, tp->tt_2msl / 1000, (tp->tt_2msl % 1000) / 10, tp->tt_delack / 1000, (tp->tt_delack % 1000) / 10, tp->t_rcvtime / 1000, (tp->t_rcvtime % 1000) / 10); } if (istcp && !Lflag && !xflag && !Tflag && !Rflag) { if (tp->t_state < 0 || tp->t_state >= TCP_NSTATES) xo_emit("{:tcp-state/%-11d}", tp->t_state); else { xo_emit("{:tcp-state/%-11s}", tcpstates[tp->t_state]); #if defined(TF_NEEDSYN) && defined(TF_NEEDFIN) /* Show T/TCP `hidden state' */ if (tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) xo_emit("{:need-syn-or-fin/*}"); #endif /* defined(TF_NEEDSYN) && defined(TF_NEEDFIN) */ } } if (Rflag) { /* XXX: is this right Alfred */ xo_emit(" {:flow-id/%08x} {:flow-type/%5d}", inp->inp_flowid, inp->inp_flowtype); } if (istcp && Pflag) xo_emit(" {:log-id/%s}", tp->xt_logid[0] == '\0' ? "-" : tp->xt_logid); xo_emit("\n"); xo_close_instance("socket"); } if (xig != oxig && xig->xig_gen != oxig->xig_gen) { if (oxig->xig_count > xig->xig_count) { xo_emit("Some {d:lost/%s} sockets may have been " "deleted.\n", name); } else if (oxig->xig_count < xig->xig_count) { xo_emit("Some {d:created/%s} sockets may have been " "created.\n", name); } else { xo_emit("Some {d:changed/%s} sockets may have been " "created or deleted.\n", name); } } free(buf); } /* * Dump TCP statistics structure. */ void tcp_stats(u_long off, const char *name, int af1 __unused, int proto __unused) { struct tcpstat tcpstat; uint64_t tcps_states[TCP_NSTATES]; #ifdef INET6 if (tcp_done != 0) return; else tcp_done = 1; #endif if (fetch_stats("net.inet.tcp.stats", off, &tcpstat, sizeof(tcpstat), kread_counters) != 0) return; if (fetch_stats_ro("net.inet.tcp.states", nl[N_TCPS_STATES].n_value, &tcps_states, sizeof(tcps_states), kread_counters) != 0) return; xo_open_container("tcp"); xo_emit("{T:/%s}:\n", name); #define p(f, m) if (tcpstat.f || sflag <= 1) \ xo_emit(m, (uintmax_t )tcpstat.f, plural(tcpstat.f)) #define p1a(f, m) if (tcpstat.f || sflag <= 1) \ xo_emit(m, (uintmax_t )tcpstat.f) #define p2(f1, f2, m) if (tcpstat.f1 || tcpstat.f2 || sflag <= 1) \ xo_emit(m, (uintmax_t )tcpstat.f1, plural(tcpstat.f1), \ (uintmax_t )tcpstat.f2, plural(tcpstat.f2)) #define p2a(f1, f2, m) if (tcpstat.f1 || tcpstat.f2 || sflag <= 1) \ xo_emit(m, (uintmax_t )tcpstat.f1, plural(tcpstat.f1), \ (uintmax_t )tcpstat.f2) #define p3(f, m) if (tcpstat.f || sflag <= 1) \ xo_emit(m, (uintmax_t )tcpstat.f, pluralies(tcpstat.f)) p(tcps_sndtotal, "\t{:sent-packets/%ju} {N:/packet%s sent}\n"); p2(tcps_sndpack,tcps_sndbyte, "\t\t{:sent-data-packets/%ju} " "{N:/data packet%s} ({:sent-data-bytes/%ju} {N:/byte%s})\n"); p2(tcps_sndrexmitpack, tcps_sndrexmitbyte, "\t\t" "{:sent-retransmitted-packets/%ju} {N:/data packet%s} " "({:sent-retransmitted-bytes/%ju} {N:/byte%s}) " "{N:retransmitted}\n"); p(tcps_sndrexmitbad, "\t\t" "{:sent-unnecessary-retransmitted-packets/%ju} " "{N:/data packet%s unnecessarily retransmitted}\n"); p(tcps_mturesent, "\t\t{:sent-resends-by-mtu-discovery/%ju} " "{N:/resend%s initiated by MTU discovery}\n"); p2a(tcps_sndacks, tcps_delack, "\t\t{:sent-ack-only-packets/%ju} " "{N:/ack-only packet%s/} ({:sent-packets-delayed/%ju} " "{N:delayed})\n"); p(tcps_sndurg, "\t\t{:sent-urg-only-packets/%ju} " "{N:/URG only packet%s}\n"); p(tcps_sndprobe, "\t\t{:sent-window-probe-packets/%ju} " "{N:/window probe packet%s}\n"); p(tcps_sndwinup, "\t\t{:sent-window-update-packets/%ju} " "{N:/window update packet%s}\n"); p(tcps_sndctrl, "\t\t{:sent-control-packets/%ju} " "{N:/control packet%s}\n"); p(tcps_rcvtotal, "\t{:received-packets/%ju} " "{N:/packet%s received}\n"); p2(tcps_rcvackpack, tcps_rcvackbyte, "\t\t" "{:received-ack-packets/%ju} {N:/ack%s} " "{N:(for} {:received-ack-bytes/%ju} {N:/byte%s})\n"); p(tcps_rcvdupack, "\t\t{:received-duplicate-acks/%ju} " "{N:/duplicate ack%s}\n"); p(tcps_rcvacktoomuch, "\t\t{:received-acks-for-unsent-data/%ju} " "{N:/ack%s for unsent data}\n"); p2(tcps_rcvpack, tcps_rcvbyte, "\t\t" "{:received-in-sequence-packets/%ju} {N:/packet%s} " "({:received-in-sequence-bytes/%ju} {N:/byte%s}) " "{N:received in-sequence}\n"); p2(tcps_rcvduppack, tcps_rcvdupbyte, "\t\t" "{:received-completely-duplicate-packets/%ju} " "{N:/completely duplicate packet%s} " "({:received-completely-duplicate-bytes/%ju} {N:/byte%s})\n"); p(tcps_pawsdrop, "\t\t{:received-old-duplicate-packets/%ju} " "{N:/old duplicate packet%s}\n"); p2(tcps_rcvpartduppack, tcps_rcvpartdupbyte, "\t\t" "{:received-some-duplicate-packets/%ju} " "{N:/packet%s with some dup. data} " "({:received-some-duplicate-bytes/%ju} {N:/byte%s duped/})\n"); p2(tcps_rcvoopack, tcps_rcvoobyte, "\t\t{:received-out-of-order/%ju} " "{N:/out-of-order packet%s} " "({:received-out-of-order-bytes/%ju} {N:/byte%s})\n"); p2(tcps_rcvpackafterwin, tcps_rcvbyteafterwin, "\t\t" "{:received-after-window-packets/%ju} {N:/packet%s} " "({:received-after-window-bytes/%ju} {N:/byte%s}) " "{N:of data after window}\n"); p(tcps_rcvwinprobe, "\t\t{:received-window-probes/%ju} " "{N:/window probe%s}\n"); p(tcps_rcvwinupd, "\t\t{:receive-window-update-packets/%ju} " "{N:/window update packet%s}\n"); p(tcps_rcvafterclose, "\t\t{:received-after-close-packets/%ju} " "{N:/packet%s received after close}\n"); p(tcps_rcvbadsum, "\t\t{:discard-bad-checksum/%ju} " "{N:/discarded for bad checksum%s}\n"); p(tcps_rcvbadoff, "\t\t{:discard-bad-header-offset/%ju} " "{N:/discarded for bad header offset field%s}\n"); p1a(tcps_rcvshort, "\t\t{:discard-too-short/%ju} " "{N:discarded because packet too short}\n"); p1a(tcps_rcvreassfull, "\t\t{:discard-reassembly-queue-full/%ju} " "{N:discarded due to full reassembly queue}\n"); p(tcps_connattempt, "\t{:connection-requests/%ju} " "{N:/connection request%s}\n"); p(tcps_accepts, "\t{:connections-accepts/%ju} " "{N:/connection accept%s}\n"); p(tcps_badsyn, "\t{:bad-connection-attempts/%ju} " "{N:/bad connection attempt%s}\n"); p(tcps_listendrop, "\t{:listen-queue-overflows/%ju} " "{N:/listen queue overflow%s}\n"); p(tcps_badrst, "\t{:ignored-in-window-resets/%ju} " "{N:/ignored RSTs in the window%s}\n"); p(tcps_connects, "\t{:connections-established/%ju} " "{N:/connection%s established (including accepts)}\n"); p(tcps_usedrtt, "\t\t{:connections-hostcache-rtt/%ju} " "{N:/time%s used RTT from hostcache}\n"); p(tcps_usedrttvar, "\t\t{:connections-hostcache-rttvar/%ju} " "{N:/time%s used RTT variance from hostcache}\n"); p(tcps_usedssthresh, "\t\t{:connections-hostcache-ssthresh/%ju} " "{N:/time%s used slow-start threshold from hostcache}\n"); p2(tcps_closed, tcps_drops, "\t{:connections-closed/%ju} " "{N:/connection%s closed (including} " "{:connection-drops/%ju} {N:/drop%s})\n"); p(tcps_cachedrtt, "\t\t{:connections-updated-rtt-on-close/%ju} " "{N:/connection%s updated cached RTT on close}\n"); p(tcps_cachedrttvar, "\t\t" "{:connections-updated-variance-on-close/%ju} " "{N:/connection%s updated cached RTT variance on close}\n"); p(tcps_cachedssthresh, "\t\t" "{:connections-updated-ssthresh-on-close/%ju} " "{N:/connection%s updated cached ssthresh on close}\n"); p(tcps_conndrops, "\t{:embryonic-connections-dropped/%ju} " "{N:/embryonic connection%s dropped}\n"); p2(tcps_rttupdated, tcps_segstimed, "\t{:segments-updated-rtt/%ju} " "{N:/segment%s updated rtt (of} " "{:segment-update-attempts/%ju} {N:/attempt%s})\n"); p(tcps_rexmttimeo, "\t{:retransmit-timeouts/%ju} " "{N:/retransmit timeout%s}\n"); p(tcps_timeoutdrop, "\t\t" "{:connections-dropped-by-retransmit-timeout/%ju} " "{N:/connection%s dropped by rexmit timeout}\n"); p(tcps_persisttimeo, "\t{:persist-timeout/%ju} " "{N:/persist timeout%s}\n"); p(tcps_persistdrop, "\t\t" "{:connections-dropped-by-persist-timeout/%ju} " "{N:/connection%s dropped by persist timeout}\n"); p(tcps_finwait2_drops, "\t" "{:connections-dropped-by-finwait2-timeout/%ju} " "{N:/Connection%s (fin_wait_2) dropped because of timeout}\n"); p(tcps_keeptimeo, "\t{:keepalive-timeout/%ju} " "{N:/keepalive timeout%s}\n"); p(tcps_keepprobe, "\t\t{:keepalive-probes/%ju} " "{N:/keepalive probe%s sent}\n"); p(tcps_keepdrops, "\t\t{:connections-dropped-by-keepalives/%ju} " "{N:/connection%s dropped by keepalive}\n"); p(tcps_predack, "\t{:ack-header-predictions/%ju} " "{N:/correct ACK header prediction%s}\n"); p(tcps_preddat, "\t{:data-packet-header-predictions/%ju} " "{N:/correct data packet header prediction%s}\n"); xo_open_container("syncache"); p3(tcps_sc_added, "\t{:entries-added/%ju} " "{N:/syncache entr%s added}\n"); p1a(tcps_sc_retransmitted, "\t\t{:retransmitted/%ju} " "{N:/retransmitted}\n"); p1a(tcps_sc_dupsyn, "\t\t{:duplicates/%ju} {N:/dupsyn}\n"); p1a(tcps_sc_dropped, "\t\t{:dropped/%ju} {N:/dropped}\n"); p1a(tcps_sc_completed, "\t\t{:completed/%ju} {N:/completed}\n"); p1a(tcps_sc_bucketoverflow, "\t\t{:bucket-overflow/%ju} " "{N:/bucket overflow}\n"); p1a(tcps_sc_cacheoverflow, "\t\t{:cache-overflow/%ju} " "{N:/cache overflow}\n"); p1a(tcps_sc_reset, "\t\t{:reset/%ju} {N:/reset}\n"); p1a(tcps_sc_stale, "\t\t{:stale/%ju} {N:/stale}\n"); p1a(tcps_sc_aborted, "\t\t{:aborted/%ju} {N:/aborted}\n"); p1a(tcps_sc_badack, "\t\t{:bad-ack/%ju} {N:/badack}\n"); p1a(tcps_sc_unreach, "\t\t{:unreachable/%ju} {N:/unreach}\n"); p(tcps_sc_zonefail, "\t\t{:zone-failures/%ju} {N:/zone failure%s}\n"); p(tcps_sc_sendcookie, "\t{:sent-cookies/%ju} {N:/cookie%s sent}\n"); p(tcps_sc_recvcookie, "\t{:receivd-cookies/%ju} " "{N:/cookie%s received}\n"); xo_close_container("syncache"); xo_open_container("hostcache"); p3(tcps_hc_added, "\t{:entries-added/%ju} " "{N:/hostcache entr%s added}\n"); p1a(tcps_hc_bucketoverflow, "\t\t{:buffer-overflows/%ju} " "{N:/bucket overflow}\n"); xo_close_container("hostcache"); xo_open_container("sack"); p(tcps_sack_recovery_episode, "\t{:recovery-episodes/%ju} " "{N:/SACK recovery episode%s}\n"); p(tcps_sack_rexmits, "\t{:segment-retransmits/%ju} " "{N:/segment rexmit%s in SACK recovery episodes}\n"); p(tcps_sack_rexmit_bytes, "\t{:byte-retransmits/%ju} " "{N:/byte rexmit%s in SACK recovery episodes}\n"); p(tcps_sack_rcv_blocks, "\t{:received-blocks/%ju} " "{N:/SACK option%s (SACK blocks) received}\n"); p(tcps_sack_send_blocks, "\t{:sent-option-blocks/%ju} " "{N:/SACK option%s (SACK blocks) sent}\n"); p1a(tcps_sack_sboverflow, "\t{:scoreboard-overflows/%ju} " "{N:/SACK scoreboard overflow}\n"); xo_close_container("sack"); xo_open_container("ecn"); p(tcps_ecn_ce, "\t{:ce-packets/%ju} " "{N:/packet%s with ECN CE bit set}\n"); p(tcps_ecn_ect0, "\t{:ect0-packets/%ju} " "{N:/packet%s with ECN ECT(0) bit set}\n"); p(tcps_ecn_ect1, "\t{:ect1-packets/%ju} " "{N:/packet%s with ECN ECT(1) bit set}\n"); p(tcps_ecn_shs, "\t{:handshakes/%ju} " "{N:/successful ECN handshake%s}\n"); p(tcps_ecn_rcwnd, "\t{:congestion-reductions/%ju} " "{N:/time%s ECN reduced the congestion window}\n"); xo_close_container("ecn"); xo_open_container("tcp-signature"); p(tcps_sig_rcvgoodsig, "\t{:received-good-signature/%ju} " "{N:/packet%s with matching signature received}\n"); p(tcps_sig_rcvbadsig, "\t{:received-bad-signature/%ju} " "{N:/packet%s with bad signature received}\n"); p(tcps_sig_err_buildsig, "\t{:failed-make-signature/%ju} " "{N:/time%s failed to make signature due to no SA}\n"); p(tcps_sig_err_sigopt, "\t{:no-signature-expected/%ju} " "{N:/time%s unexpected signature received}\n"); p(tcps_sig_err_nosigopt, "\t{:no-signature-provided/%ju} " "{N:/time%s no signature provided by segment}\n"); xo_close_container("tcp-signature"); xo_open_container("pmtud"); p(tcps_pmtud_blackhole_activated, "\t{:pmtud-activated/%ju} " "{N:/Path MTU discovery black hole detection activation%s}\n"); p(tcps_pmtud_blackhole_activated_min_mss, "\t{:pmtud-activated-min-mss/%ju} " "{N:/Path MTU discovery black hole detection min MSS activation%s}\n"); p(tcps_pmtud_blackhole_failed, "\t{:pmtud-failed/%ju} " "{N:/Path MTU discovery black hole detection failure%s}\n"); #undef p #undef p1a #undef p2 #undef p2a #undef p3 xo_close_container("pmtud"); xo_open_container("TCP connection count by state"); xo_emit("{T:/TCP connection count by state}:\n"); for (int i = 0; i < TCP_NSTATES; i++) { /* * XXXGL: is there a way in libxo to use %s * in the "content string" of a format * string? I failed to do that, that's why * a temporary buffer is used to construct * format string for xo_emit(). */ char fmtbuf[80]; if (sflag > 1 && tcps_states[i] == 0) continue; snprintf(fmtbuf, sizeof(fmtbuf), "\t{:%s/%%ju} " "{Np:/connection ,connections} in %s state\n", tcpstates[i], tcpstates[i]); xo_emit(fmtbuf, (uintmax_t )tcps_states[i]); } xo_close_container("TCP connection count by state"); xo_close_container("tcp"); } /* * Dump UDP statistics structure. */ void udp_stats(u_long off, const char *name, int af1 __unused, int proto __unused) { struct udpstat udpstat; uint64_t delivered; #ifdef INET6 if (udp_done != 0) return; else udp_done = 1; #endif if (fetch_stats("net.inet.udp.stats", off, &udpstat, sizeof(udpstat), kread_counters) != 0) return; xo_open_container("udp"); xo_emit("{T:/%s}:\n", name); #define p(f, m) if (udpstat.f || sflag <= 1) \ xo_emit("\t" m, (uintmax_t)udpstat.f, plural(udpstat.f)) #define p1a(f, m) if (udpstat.f || sflag <= 1) \ xo_emit("\t" m, (uintmax_t)udpstat.f) p(udps_ipackets, "{:received-datagrams/%ju} " "{N:/datagram%s received}\n"); p1a(udps_hdrops, "{:dropped-incomplete-headers/%ju} " "{N:/with incomplete header}\n"); p1a(udps_badlen, "{:dropped-bad-data-length/%ju} " "{N:/with bad data length field}\n"); p1a(udps_badsum, "{:dropped-bad-checksum/%ju} " "{N:/with bad checksum}\n"); p1a(udps_nosum, "{:dropped-no-checksum/%ju} " "{N:/with no checksum}\n"); p1a(udps_noport, "{:dropped-no-socket/%ju} " "{N:/dropped due to no socket}\n"); p(udps_noportbcast, "{:dropped-broadcast-multicast/%ju} " "{N:/broadcast\\/multicast datagram%s undelivered}\n"); p1a(udps_fullsock, "{:dropped-full-socket-buffer/%ju} " "{N:/dropped due to full socket buffers}\n"); p1a(udpps_pcbhashmiss, "{:not-for-hashed-pcb/%ju} " "{N:/not for hashed pcb}\n"); delivered = udpstat.udps_ipackets - udpstat.udps_hdrops - udpstat.udps_badlen - udpstat.udps_badsum - udpstat.udps_noport - udpstat.udps_noportbcast - udpstat.udps_fullsock; if (delivered || sflag <= 1) xo_emit("\t{:delivered-packets/%ju} {N:/delivered}\n", (uint64_t)delivered); p(udps_opackets, "{:output-packets/%ju} {N:/datagram%s output}\n"); /* the next statistic is cumulative in udps_noportbcast */ p(udps_filtermcast, "{:multicast-source-filter-matches/%ju} " "{N:/time%s multicast source filter matched}\n"); #undef p #undef p1a xo_close_container("udp"); } /* * Dump CARP statistics structure. */ void carp_stats(u_long off, const char *name, int af1 __unused, int proto __unused) { struct carpstats carpstat; if (fetch_stats("net.inet.carp.stats", off, &carpstat, sizeof(carpstat), kread_counters) != 0) return; xo_open_container(name); xo_emit("{T:/%s}:\n", name); #define p(f, m) if (carpstat.f || sflag <= 1) \ xo_emit(m, (uintmax_t)carpstat.f, plural(carpstat.f)) #define p2(f, m) if (carpstat.f || sflag <= 1) \ xo_emit(m, (uintmax_t)carpstat.f) p(carps_ipackets, "\t{:received-inet-packets/%ju} " "{N:/packet%s received (IPv4)}\n"); p(carps_ipackets6, "\t{:received-inet6-packets/%ju} " "{N:/packet%s received (IPv6)}\n"); p(carps_badttl, "\t\t{:dropped-wrong-ttl/%ju} " "{N:/packet%s discarded for wrong TTL}\n"); p(carps_hdrops, "\t\t{:dropped-short-header/%ju} " "{N:/packet%s shorter than header}\n"); p(carps_badsum, "\t\t{:dropped-bad-checksum/%ju} " "{N:/discarded for bad checksum%s}\n"); p(carps_badver, "\t\t{:dropped-bad-version/%ju} " "{N:/discarded packet%s with a bad version}\n"); p2(carps_badlen, "\t\t{:dropped-short-packet/%ju} " "{N:/discarded because packet too short}\n"); p2(carps_badauth, "\t\t{:dropped-bad-authentication/%ju} " "{N:/discarded for bad authentication}\n"); p2(carps_badvhid, "\t\t{:dropped-bad-vhid/%ju} " "{N:/discarded for bad vhid}\n"); p2(carps_badaddrs, "\t\t{:dropped-bad-address-list/%ju} " "{N:/discarded because of a bad address list}\n"); p(carps_opackets, "\t{:sent-inet-packets/%ju} " "{N:/packet%s sent (IPv4)}\n"); p(carps_opackets6, "\t{:sent-inet6-packets/%ju} " "{N:/packet%s sent (IPv6)}\n"); p2(carps_onomem, "\t\t{:send-failed-memory-error/%ju} " "{N:/send failed due to mbuf memory error}\n"); #if notyet p(carps_ostates, "\t\t{:send-state-updates/%s} " "{N:/state update%s sent}\n"); #endif #undef p #undef p2 xo_close_container(name); } /* * Dump IP statistics structure. */ void ip_stats(u_long off, const char *name, int af1 __unused, int proto __unused) { struct ipstat ipstat; if (fetch_stats("net.inet.ip.stats", off, &ipstat, sizeof(ipstat), kread_counters) != 0) return; xo_open_container(name); xo_emit("{T:/%s}:\n", name); #define p(f, m) if (ipstat.f || sflag <= 1) \ xo_emit(m, (uintmax_t )ipstat.f, plural(ipstat.f)) #define p1a(f, m) if (ipstat.f || sflag <= 1) \ xo_emit(m, (uintmax_t )ipstat.f) p(ips_total, "\t{:received-packets/%ju} " "{N:/total packet%s received}\n"); p(ips_badsum, "\t{:dropped-bad-checksum/%ju} " "{N:/bad header checksum%s}\n"); p1a(ips_toosmall, "\t{:dropped-below-minimum-size/%ju} " "{N:/with size smaller than minimum}\n"); p1a(ips_tooshort, "\t{:dropped-short-packets/%ju} " "{N:/with data size < data length}\n"); p1a(ips_toolong, "\t{:dropped-too-long/%ju} " "{N:/with ip length > max ip packet size}\n"); p1a(ips_badhlen, "\t{:dropped-short-header-length/%ju} " "{N:/with header length < data size}\n"); p1a(ips_badlen, "\t{:dropped-short-data/%ju} " "{N:/with data length < header length}\n"); p1a(ips_badoptions, "\t{:dropped-bad-options/%ju} " "{N:/with bad options}\n"); p1a(ips_badvers, "\t{:dropped-bad-version/%ju} " "{N:/with incorrect version number}\n"); p(ips_fragments, "\t{:received-fragments/%ju} " "{N:/fragment%s received}\n"); p(ips_fragdropped, "\t{:dropped-fragments/%ju} " "{N:/fragment%s dropped (dup or out of space)}\n"); p(ips_fragtimeout, "\t{:dropped-fragments-after-timeout/%ju} " "{N:/fragment%s dropped after timeout}\n"); p(ips_reassembled, "\t{:reassembled-packets/%ju} " "{N:/packet%s reassembled ok}\n"); p(ips_delivered, "\t{:received-local-packets/%ju} " "{N:/packet%s for this host}\n"); p(ips_noproto, "\t{:dropped-unknown-protocol/%ju} " "{N:/packet%s for unknown\\/unsupported protocol}\n"); p(ips_forward, "\t{:forwarded-packets/%ju} " "{N:/packet%s forwarded}"); p(ips_fastforward, " ({:fast-forwarded-packets/%ju} " "{N:/packet%s fast forwarded})"); if (ipstat.ips_forward || sflag <= 1) xo_emit("\n"); p(ips_cantforward, "\t{:packets-cannot-forward/%ju} " "{N:/packet%s not forwardable}\n"); p(ips_notmember, "\t{:received-unknown-multicast-group/%ju} " "{N:/packet%s received for unknown multicast group}\n"); p(ips_redirectsent, "\t{:redirects-sent/%ju} " "{N:/redirect%s sent}\n"); p(ips_localout, "\t{:sent-packets/%ju} " "{N:/packet%s sent from this host}\n"); p(ips_rawout, "\t{:send-packets-fabricated-header/%ju} " "{N:/packet%s sent with fabricated ip header}\n"); p(ips_odropped, "\t{:discard-no-mbufs/%ju} " "{N:/output packet%s dropped due to no bufs, etc.}\n"); p(ips_noroute, "\t{:discard-no-route/%ju} " "{N:/output packet%s discarded due to no route}\n"); p(ips_fragmented, "\t{:sent-fragments/%ju} " "{N:/output datagram%s fragmented}\n"); p(ips_ofragments, "\t{:fragments-created/%ju} " "{N:/fragment%s created}\n"); p(ips_cantfrag, "\t{:discard-cannot-fragment/%ju} " "{N:/datagram%s that can't be fragmented}\n"); p(ips_nogif, "\t{:discard-tunnel-no-gif/%ju} " "{N:/tunneling packet%s that can't find gif}\n"); p(ips_badaddr, "\t{:discard-bad-address/%ju} " "{N:/datagram%s with bad address in header}\n"); #undef p #undef p1a xo_close_container(name); } /* * Dump ARP statistics structure. */ void arp_stats(u_long off, const char *name, int af1 __unused, int proto __unused) { struct arpstat arpstat; if (fetch_stats("net.link.ether.arp.stats", off, &arpstat, sizeof(arpstat), kread_counters) != 0) return; xo_open_container(name); xo_emit("{T:/%s}:\n", name); #define p(f, m) if (arpstat.f || sflag <= 1) \ xo_emit("\t" m, (uintmax_t)arpstat.f, plural(arpstat.f)) #define p2(f, m) if (arpstat.f || sflag <= 1) \ xo_emit("\t" m, (uintmax_t)arpstat.f, pluralies(arpstat.f)) p(txrequests, "{:sent-requests/%ju} {N:/ARP request%s sent}\n"); p(txerrors, "{:sent-failures/%ju} {N:/ARP request%s failed to sent}\n"); p2(txreplies, "{:sent-replies/%ju} {N:/ARP repl%s sent}\n"); p(rxrequests, "{:received-requests/%ju} " "{N:/ARP request%s received}\n"); p2(rxreplies, "{:received-replies/%ju} " "{N:/ARP repl%s received}\n"); p(received, "{:received-packers/%ju} " "{N:/ARP packet%s received}\n"); p(dropped, "{:dropped-no-entry/%ju} " "{N:/total packet%s dropped due to no ARP entry}\n"); p(timeouts, "{:entries-timeout/%ju} " "{N:/ARP entry%s timed out}\n"); p(dupips, "{:dropped-duplicate-address/%ju} " "{N:/Duplicate IP%s seen}\n"); #undef p #undef p2 xo_close_container(name); } static const char *icmpnames[ICMP_MAXTYPE + 1] = { "echo reply", /* RFC 792 */ "#1", "#2", "destination unreachable", /* RFC 792 */ "source quench", /* RFC 792 */ "routing redirect", /* RFC 792 */ "#6", "#7", "echo", /* RFC 792 */ "router advertisement", /* RFC 1256 */ "router solicitation", /* RFC 1256 */ "time exceeded", /* RFC 792 */ "parameter problem", /* RFC 792 */ "time stamp", /* RFC 792 */ "time stamp reply", /* RFC 792 */ "information request", /* RFC 792 */ "information request reply", /* RFC 792 */ "address mask request", /* RFC 950 */ "address mask reply", /* RFC 950 */ "#19", "#20", "#21", "#22", "#23", "#24", "#25", "#26", "#27", "#28", "#29", "icmp traceroute", /* RFC 1393 */ "datagram conversion error", /* RFC 1475 */ "mobile host redirect", "IPv6 where-are-you", "IPv6 i-am-here", "mobile registration req", "mobile registration reply", "domain name request", /* RFC 1788 */ "domain name reply", /* RFC 1788 */ "icmp SKIP", "icmp photuris", /* RFC 2521 */ }; /* * Dump ICMP statistics. */ void icmp_stats(u_long off, const char *name, int af1 __unused, int proto __unused) { struct icmpstat icmpstat; size_t len; int i, first; if (fetch_stats("net.inet.icmp.stats", off, &icmpstat, sizeof(icmpstat), kread_counters) != 0) return; xo_open_container(name); xo_emit("{T:/%s}:\n", name); #define p(f, m) if (icmpstat.f || sflag <= 1) \ xo_emit(m, icmpstat.f, plural(icmpstat.f)) #define p1a(f, m) if (icmpstat.f || sflag <= 1) \ xo_emit(m, icmpstat.f) #define p2(f, m) if (icmpstat.f || sflag <= 1) \ xo_emit(m, icmpstat.f, plurales(icmpstat.f)) p(icps_error, "\t{:icmp-calls/%lu} " "{N:/call%s to icmp_error}\n"); p(icps_oldicmp, "\t{:errors-not-from-message/%lu} " "{N:/error%s not generated in response to an icmp message}\n"); for (first = 1, i = 0; i < ICMP_MAXTYPE + 1; i++) { if (icmpstat.icps_outhist[i] != 0) { if (first) { xo_open_list("output-histogram"); xo_emit("\tOutput histogram:\n"); first = 0; } xo_open_instance("output-histogram"); if (icmpnames[i] != NULL) xo_emit("\t\t{k:name/%s}: {:count/%lu}\n", icmpnames[i], icmpstat.icps_outhist[i]); else xo_emit("\t\tunknown ICMP #{k:name/%d}: " "{:count/%lu}\n", i, icmpstat.icps_outhist[i]); xo_close_instance("output-histogram"); } } if (!first) xo_close_list("output-histogram"); p(icps_badcode, "\t{:dropped-bad-code/%lu} " "{N:/message%s with bad code fields}\n"); p(icps_tooshort, "\t{:dropped-too-short/%lu} " "{N:/message%s less than the minimum length}\n"); p(icps_checksum, "\t{:dropped-bad-checksum/%lu} " "{N:/message%s with bad checksum}\n"); p(icps_badlen, "\t{:dropped-bad-length/%lu} " "{N:/message%s with bad length}\n"); p1a(icps_bmcastecho, "\t{:dropped-multicast-echo/%lu} " "{N:/multicast echo requests ignored}\n"); p1a(icps_bmcasttstamp, "\t{:dropped-multicast-timestamp/%lu} " "{N:/multicast timestamp requests ignored}\n"); for (first = 1, i = 0; i < ICMP_MAXTYPE + 1; i++) { if (icmpstat.icps_inhist[i] != 0) { if (first) { xo_open_list("input-histogram"); xo_emit("\tInput histogram:\n"); first = 0; } xo_open_instance("input-histogram"); if (icmpnames[i] != NULL) xo_emit("\t\t{k:name/%s}: {:count/%lu}\n", icmpnames[i], icmpstat.icps_inhist[i]); else xo_emit( "\t\tunknown ICMP #{k:name/%d}: {:count/%lu}\n", i, icmpstat.icps_inhist[i]); xo_close_instance("input-histogram"); } } if (!first) xo_close_list("input-histogram"); p(icps_reflect, "\t{:sent-packets/%lu} " "{N:/message response%s generated}\n"); p2(icps_badaddr, "\t{:discard-invalid-return-address/%lu} " "{N:/invalid return address%s}\n"); p(icps_noroute, "\t{:discard-no-route/%lu} " "{N:/no return route%s}\n"); #undef p #undef p1a #undef p2 if (live) { len = sizeof i; if (sysctlbyname("net.inet.icmp.maskrepl", &i, &len, NULL, 0) < 0) return; xo_emit("\tICMP address mask responses are " "{q:icmp-address-responses/%sabled}\n", i ? "en" : "dis"); } xo_close_container(name); } /* * Dump IGMP statistics structure. */ void igmp_stats(u_long off, const char *name, int af1 __unused, int proto __unused) { struct igmpstat igmpstat; + int error, zflag0; if (fetch_stats("net.inet.igmp.stats", 0, &igmpstat, sizeof(igmpstat), kread) != 0) return; + /* + * Reread net.inet.igmp.stats when zflag == 1. + * This is because this MIB contains version number and + * length of the structure which are not set when clearing + * the counters. + */ + zflag0 = zflag; + if (zflag) { + zflag = 0; + error = fetch_stats("net.inet.igmp.stats", 0, &igmpstat, + sizeof(igmpstat), kread); + zflag = zflag0; + if (error) + return; + } if (igmpstat.igps_version != IGPS_VERSION_3) { xo_warnx("%s: version mismatch (%d != %d)", __func__, igmpstat.igps_version, IGPS_VERSION_3); return; } if (igmpstat.igps_len != IGPS_VERSION3_LEN) { xo_warnx("%s: size mismatch (%d != %d)", __func__, igmpstat.igps_len, IGPS_VERSION3_LEN); return; } xo_open_container(name); xo_emit("{T:/%s}:\n", name); #define p64(f, m) if (igmpstat.f || sflag <= 1) \ xo_emit(m, (uintmax_t) igmpstat.f, plural(igmpstat.f)) #define py64(f, m) if (igmpstat.f || sflag <= 1) \ xo_emit(m, (uintmax_t) igmpstat.f, pluralies(igmpstat.f)) p64(igps_rcv_total, "\t{:received-messages/%ju} " "{N:/message%s received}\n"); p64(igps_rcv_tooshort, "\t{:dropped-too-short/%ju} " "{N:/message%s received with too few bytes}\n"); p64(igps_rcv_badttl, "\t{:dropped-wrong-ttl/%ju} " "{N:/message%s received with wrong TTL}\n"); p64(igps_rcv_badsum, "\t{:dropped-bad-checksum/%ju} " "{N:/message%s received with bad checksum}\n"); py64(igps_rcv_v1v2_queries, "\t{:received-membership-queries/%ju} " "{N:/V1\\/V2 membership quer%s received}\n"); py64(igps_rcv_v3_queries, "\t{:received-v3-membership-queries/%ju} " "{N:/V3 membership quer%s received}\n"); py64(igps_rcv_badqueries, "\t{:dropped-membership-queries/%ju} " "{N:/membership quer%s received with invalid field(s)}\n"); py64(igps_rcv_gen_queries, "\t{:received-general-queries/%ju} " "{N:/general quer%s received}\n"); py64(igps_rcv_group_queries, "\t{:received-group-queries/%ju} " "{N:/group quer%s received}\n"); py64(igps_rcv_gsr_queries, "\t{:received-group-source-queries/%ju} " "{N:/group-source quer%s received}\n"); py64(igps_drop_gsr_queries, "\t{:dropped-group-source-queries/%ju} " "{N:/group-source quer%s dropped}\n"); p64(igps_rcv_reports, "\t{:received-membership-requests/%ju} " "{N:/membership report%s received}\n"); p64(igps_rcv_badreports, "\t{:dropped-membership-reports/%ju} " "{N:/membership report%s received with invalid field(s)}\n"); p64(igps_rcv_ourreports, "\t" "{:received-membership-reports-matching/%ju} " "{N:/membership report%s received for groups to which we belong}" "\n"); p64(igps_rcv_nora, "\t{:received-v3-reports-no-router-alert/%ju} " "{N:/V3 report%s received without Router Alert}\n"); p64(igps_snd_reports, "\t{:sent-membership-reports/%ju} " "{N:/membership report%s sent}\n"); #undef p64 #undef py64 xo_close_container(name); } /* * Dump PIM statistics structure. */ void pim_stats(u_long off __unused, const char *name, int af1 __unused, int proto __unused) { struct pimstat pimstat; if (fetch_stats("net.inet.pim.stats", off, &pimstat, sizeof(pimstat), kread_counters) != 0) return; xo_open_container(name); xo_emit("{T:/%s}:\n", name); #define p(f, m) if (pimstat.f || sflag <= 1) \ xo_emit(m, (uintmax_t)pimstat.f, plural(pimstat.f)) #define py(f, m) if (pimstat.f || sflag <= 1) \ xo_emit(m, (uintmax_t)pimstat.f, pimstat.f != 1 ? "ies" : "y") p(pims_rcv_total_msgs, "\t{:received-messages/%ju} " "{N:/message%s received}\n"); p(pims_rcv_total_bytes, "\t{:received-bytes/%ju} " "{N:/byte%s received}\n"); p(pims_rcv_tooshort, "\t{:dropped-too-short/%ju} " "{N:/message%s received with too few bytes}\n"); p(pims_rcv_badsum, "\t{:dropped-bad-checksum/%ju} " "{N:/message%s received with bad checksum}\n"); p(pims_rcv_badversion, "\t{:dropped-bad-version/%ju} " "{N:/message%s received with bad version}\n"); p(pims_rcv_registers_msgs, "\t{:received-data-register-messages/%ju} " "{N:/data register message%s received}\n"); p(pims_rcv_registers_bytes, "\t{:received-data-register-bytes/%ju} " "{N:/data register byte%s received}\n"); p(pims_rcv_registers_wrongiif, "\t" "{:received-data-register-wrong-interface/%ju} " "{N:/data register message%s received on wrong iif}\n"); p(pims_rcv_badregisters, "\t{:received-bad-registers/%ju} " "{N:/bad register%s received}\n"); p(pims_snd_registers_msgs, "\t{:sent-data-register-messages/%ju} " "{N:/data register message%s sent}\n"); p(pims_snd_registers_bytes, "\t{:sent-data-register-bytes/%ju} " "{N:/data register byte%s sent}\n"); #undef p #undef py xo_close_container(name); } #ifdef INET /* * Pretty print an Internet address (net address + port). */ static void inetprint(const char *container, struct in_addr *in, int port, const char *proto, int num_port, const int af1) { struct servent *sp = 0; char line[80], *cp; int width; size_t alen, plen; if (container) xo_open_container(container); if (Wflag) snprintf(line, sizeof(line), "%s.", inetname(in)); else snprintf(line, sizeof(line), "%.*s.", (Aflag && !num_port) ? 12 : 16, inetname(in)); alen = strlen(line); cp = line + alen; if (!num_port && port) sp = getservbyport((int)port, proto); if (sp || port == 0) snprintf(cp, sizeof(line) - alen, "%.15s ", sp ? sp->s_name : "*"); else snprintf(cp, sizeof(line) - alen, "%d ", ntohs((u_short)port)); width = (Aflag && !Wflag) ? 18 : ((!Wflag || af1 == AF_INET) ? 22 : 45); if (Wflag) xo_emit("{d:target/%-*s} ", width, line); else xo_emit("{d:target/%-*.*s} ", width, width, line); plen = strlen(cp) - 1; alen--; xo_emit("{e:address/%*.*s}{e:port/%*.*s}", alen, alen, line, plen, plen, cp); if (container) xo_close_container(container); } /* * Construct an Internet address representation. * If numeric_addr has been supplied, give * numeric value, otherwise try for symbolic name. */ char * inetname(struct in_addr *inp) { char *cp; static char line[MAXHOSTNAMELEN]; struct hostent *hp; struct netent *np; cp = 0; if (!numeric_addr && inp->s_addr != INADDR_ANY) { int net = inet_netof(*inp); int lna = inet_lnaof(*inp); if (lna == INADDR_ANY) { np = getnetbyaddr(net, AF_INET); if (np) cp = np->n_name; } if (cp == NULL) { hp = gethostbyaddr((char *)inp, sizeof (*inp), AF_INET); if (hp) { cp = hp->h_name; trimdomain(cp, strlen(cp)); } } } if (inp->s_addr == INADDR_ANY) strcpy(line, "*"); else if (cp) { strlcpy(line, cp, sizeof(line)); } else { inp->s_addr = ntohl(inp->s_addr); #define C(x) ((u_int)((x) & 0xff)) snprintf(line, sizeof(line), "%u.%u.%u.%u", C(inp->s_addr >> 24), C(inp->s_addr >> 16), C(inp->s_addr >> 8), C(inp->s_addr)); } return (line); } #endif