Index: head/lib/libc/stdio/flags.c =================================================================== --- head/lib/libc/stdio/flags.c (revision 281844) +++ head/lib/libc/stdio/flags.c (revision 281845) @@ -1,113 +1,117 @@ /*- * Copyright (c) 1990, 1993 * The Regents of the University of California. All rights reserved. * * This code is derived from software contributed to Berkeley by * Chris Torek. * * 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 defined(LIBC_SCCS) && !defined(lint) static char sccsid[] = "@(#)flags.c 8.1 (Berkeley) 6/4/93"; #endif /* LIBC_SCCS and not lint */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include "local.h" /* * Return the (stdio) flags for a given mode. Store the flags * to be passed to an _open() syscall through *optr. * Return 0 on error. */ int __sflags(const char *mode, int *optr) { int ret, m, o, known; switch (*mode++) { case 'r': /* open for reading */ ret = __SRD; m = O_RDONLY; o = 0; break; case 'w': /* open for writing */ ret = __SWR; m = O_WRONLY; o = O_CREAT | O_TRUNC; break; case 'a': /* open for appending */ ret = __SWR; m = O_WRONLY; o = O_CREAT | O_APPEND; break; default: /* illegal mode */ errno = EINVAL; return (0); } do { known = 1; switch (*mode++) { case 'b': /* 'b' (binary) is ignored */ break; case '+': /* [rwa][b]\+ means read and write */ ret = __SRW; m = O_RDWR; break; case 'x': /* 'x' means exclusive (fail if the file exists) */ o |= O_EXCL; break; case 'e': /* set close-on-exec */ o |= O_CLOEXEC; break; + case 'v': + /* verify */ + o |= O_VERIFY; + break; default: known = 0; break; } } while (known); if ((o & O_EXCL) != 0 && m == O_RDONLY) { errno = EINVAL; return (0); } *optr = m | o; return (ret); } Index: head/libexec/rtld-elf/rtld.c =================================================================== --- head/libexec/rtld-elf/rtld.c (revision 281844) +++ head/libexec/rtld-elf/rtld.c (revision 281845) @@ -1,5083 +1,5083 @@ /*- * Copyright 1996, 1997, 1998, 1999, 2000 John D. Polstra. * Copyright 2003 Alexander Kabaev . * Copyright 2009-2012 Konstantin Belousov . * Copyright 2012 John Marino . * 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. * * $FreeBSD$ */ /* * Dynamic linker for ELF. * * John Polstra . */ #ifndef __GNUC__ #error "GCC is needed to compile this file" #endif #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "debug.h" #include "rtld.h" #include "libmap.h" #include "rtld_tls.h" #include "rtld_printf.h" #include "notes.h" #ifndef COMPAT_32BIT #define PATH_RTLD "/libexec/ld-elf.so.1" #else #define PATH_RTLD "/libexec/ld-elf32.so.1" #endif /* Types. */ typedef void (*func_ptr_type)(); typedef void * (*path_enum_proc) (const char *path, size_t len, void *arg); /* * Function declarations. */ static const char *basename(const char *); static void digest_dynamic1(Obj_Entry *, int, const Elf_Dyn **, const Elf_Dyn **, const Elf_Dyn **); static void digest_dynamic2(Obj_Entry *, const Elf_Dyn *, const Elf_Dyn *, const Elf_Dyn *); static void digest_dynamic(Obj_Entry *, int); static Obj_Entry *digest_phdr(const Elf_Phdr *, int, caddr_t, const char *); static Obj_Entry *dlcheck(void *); static Obj_Entry *dlopen_object(const char *name, int fd, Obj_Entry *refobj, int lo_flags, int mode, RtldLockState *lockstate); static Obj_Entry *do_load_object(int, const char *, char *, struct stat *, int); static int do_search_info(const Obj_Entry *obj, int, struct dl_serinfo *); static bool donelist_check(DoneList *, const Obj_Entry *); static void errmsg_restore(char *); static char *errmsg_save(void); static void *fill_search_info(const char *, size_t, void *); static char *find_library(const char *, const Obj_Entry *, int *); static const char *gethints(bool); static void init_dag(Obj_Entry *); static void init_pagesizes(Elf_Auxinfo **aux_info); static void init_rtld(caddr_t, Elf_Auxinfo **); static void initlist_add_neededs(Needed_Entry *, Objlist *); static void initlist_add_objects(Obj_Entry *, Obj_Entry **, Objlist *); static void linkmap_add(Obj_Entry *); static void linkmap_delete(Obj_Entry *); static void load_filtees(Obj_Entry *, int flags, RtldLockState *); static void unload_filtees(Obj_Entry *); static int load_needed_objects(Obj_Entry *, int); static int load_preload_objects(void); static Obj_Entry *load_object(const char *, int fd, const Obj_Entry *, int); static void map_stacks_exec(RtldLockState *); static Obj_Entry *obj_from_addr(const void *); static void objlist_call_fini(Objlist *, Obj_Entry *, RtldLockState *); static void objlist_call_init(Objlist *, RtldLockState *); static void objlist_clear(Objlist *); static Objlist_Entry *objlist_find(Objlist *, const Obj_Entry *); static void objlist_init(Objlist *); static void objlist_push_head(Objlist *, Obj_Entry *); static void objlist_push_tail(Objlist *, Obj_Entry *); static void objlist_put_after(Objlist *, Obj_Entry *, Obj_Entry *); static void objlist_remove(Objlist *, Obj_Entry *); static int parse_libdir(const char *); static void *path_enumerate(const char *, path_enum_proc, void *); static int relocate_object_dag(Obj_Entry *root, bool bind_now, Obj_Entry *rtldobj, int flags, RtldLockState *lockstate); static int relocate_object(Obj_Entry *obj, bool bind_now, Obj_Entry *rtldobj, int flags, RtldLockState *lockstate); static int relocate_objects(Obj_Entry *, bool, Obj_Entry *, int, RtldLockState *); static int resolve_objects_ifunc(Obj_Entry *first, bool bind_now, int flags, RtldLockState *lockstate); static int rtld_dirname(const char *, char *); static int rtld_dirname_abs(const char *, char *); static void *rtld_dlopen(const char *name, int fd, int mode); static void rtld_exit(void); static char *search_library_path(const char *, const char *); static char *search_library_pathfds(const char *, const char *, int *); static const void **get_program_var_addr(const char *, RtldLockState *); static void set_program_var(const char *, const void *); static int symlook_default(SymLook *, const Obj_Entry *refobj); static int symlook_global(SymLook *, DoneList *); static void symlook_init_from_req(SymLook *, const SymLook *); static int symlook_list(SymLook *, const Objlist *, DoneList *); static int symlook_needed(SymLook *, const Needed_Entry *, DoneList *); static int symlook_obj1_sysv(SymLook *, const Obj_Entry *); static int symlook_obj1_gnu(SymLook *, const Obj_Entry *); static void trace_loaded_objects(Obj_Entry *); static void unlink_object(Obj_Entry *); static void unload_object(Obj_Entry *); static void unref_dag(Obj_Entry *); static void ref_dag(Obj_Entry *); static char *origin_subst_one(char *, const char *, const char *, bool); static char *origin_subst(char *, const char *); static void preinit_main(void); static int rtld_verify_versions(const Objlist *); static int rtld_verify_object_versions(Obj_Entry *); static void object_add_name(Obj_Entry *, const char *); static int object_match_name(const Obj_Entry *, const char *); static void ld_utrace_log(int, void *, void *, size_t, int, const char *); static void rtld_fill_dl_phdr_info(const Obj_Entry *obj, struct dl_phdr_info *phdr_info); static uint32_t gnu_hash(const char *); static bool matched_symbol(SymLook *, const Obj_Entry *, Sym_Match_Result *, const unsigned long); void r_debug_state(struct r_debug *, struct link_map *) __noinline __exported; void _r_debug_postinit(struct link_map *) __noinline __exported; int __sys_openat(int, const char *, int, ...); /* * Data declarations. */ static char *error_message; /* Message for dlerror(), or NULL */ struct r_debug r_debug __exported; /* for GDB; */ static bool libmap_disable; /* Disable libmap */ static bool ld_loadfltr; /* Immediate filters processing */ static char *libmap_override; /* Maps to use in addition to libmap.conf */ static bool trust; /* False for setuid and setgid programs */ static bool dangerous_ld_env; /* True if environment variables have been used to affect the libraries loaded */ static char *ld_bind_now; /* Environment variable for immediate binding */ static char *ld_debug; /* Environment variable for debugging */ static char *ld_library_path; /* Environment variable for search path */ static char *ld_library_dirs; /* Environment variable for library descriptors */ static char *ld_preload; /* Environment variable for libraries to load first */ static char *ld_elf_hints_path; /* Environment variable for alternative hints path */ static char *ld_tracing; /* Called from ldd to print libs */ static char *ld_utrace; /* Use utrace() to log events. */ static Obj_Entry *obj_list; /* Head of linked list of shared objects */ static Obj_Entry **obj_tail; /* Link field of last object in list */ static Obj_Entry *obj_main; /* The main program shared object */ static Obj_Entry obj_rtld; /* The dynamic linker shared object */ static unsigned int obj_count; /* Number of objects in obj_list */ static unsigned int obj_loads; /* Number of objects in obj_list */ static Objlist list_global = /* Objects dlopened with RTLD_GLOBAL */ STAILQ_HEAD_INITIALIZER(list_global); static Objlist list_main = /* Objects loaded at program startup */ STAILQ_HEAD_INITIALIZER(list_main); static Objlist list_fini = /* Objects needing fini() calls */ STAILQ_HEAD_INITIALIZER(list_fini); Elf_Sym sym_zero; /* For resolving undefined weak refs. */ #define GDB_STATE(s,m) r_debug.r_state = s; r_debug_state(&r_debug,m); extern Elf_Dyn _DYNAMIC; #pragma weak _DYNAMIC #ifndef RTLD_IS_DYNAMIC #define RTLD_IS_DYNAMIC() (&_DYNAMIC != NULL) #endif int dlclose(void *) __exported; char *dlerror(void) __exported; void *dlopen(const char *, int) __exported; void *fdlopen(int, int) __exported; void *dlsym(void *, const char *) __exported; dlfunc_t dlfunc(void *, const char *) __exported; void *dlvsym(void *, const char *, const char *) __exported; int dladdr(const void *, Dl_info *) __exported; void dllockinit(void *, void *(*)(void *), void (*)(void *), void (*)(void *), void (*)(void *), void (*)(void *), void (*)(void *)) __exported; int dlinfo(void *, int , void *) __exported; int dl_iterate_phdr(__dl_iterate_hdr_callback, void *) __exported; int _rtld_addr_phdr(const void *, struct dl_phdr_info *) __exported; int _rtld_get_stack_prot(void) __exported; int _rtld_is_dlopened(void *) __exported; void _rtld_error(const char *, ...) __exported; int npagesizes, osreldate; size_t *pagesizes; long __stack_chk_guard[8] = {0, 0, 0, 0, 0, 0, 0, 0}; static int stack_prot = PROT_READ | PROT_WRITE | RTLD_DEFAULT_STACK_EXEC; static int max_stack_flags; /* * Global declarations normally provided by crt1. The dynamic linker is * not built with crt1, so we have to provide them ourselves. */ char *__progname; char **environ; /* * Used to pass argc, argv to init functions. */ int main_argc; char **main_argv; /* * Globals to control TLS allocation. */ size_t tls_last_offset; /* Static TLS offset of last module */ size_t tls_last_size; /* Static TLS size of last module */ size_t tls_static_space; /* Static TLS space allocated */ size_t tls_static_max_align; int tls_dtv_generation = 1; /* Used to detect when dtv size changes */ int tls_max_index = 1; /* Largest module index allocated */ bool ld_library_path_rpath = false; /* * Fill in a DoneList with an allocation large enough to hold all of * the currently-loaded objects. Keep this as a macro since it calls * alloca and we want that to occur within the scope of the caller. */ #define donelist_init(dlp) \ ((dlp)->objs = alloca(obj_count * sizeof (dlp)->objs[0]), \ assert((dlp)->objs != NULL), \ (dlp)->num_alloc = obj_count, \ (dlp)->num_used = 0) #define UTRACE_DLOPEN_START 1 #define UTRACE_DLOPEN_STOP 2 #define UTRACE_DLCLOSE_START 3 #define UTRACE_DLCLOSE_STOP 4 #define UTRACE_LOAD_OBJECT 5 #define UTRACE_UNLOAD_OBJECT 6 #define UTRACE_ADD_RUNDEP 7 #define UTRACE_PRELOAD_FINISHED 8 #define UTRACE_INIT_CALL 9 #define UTRACE_FINI_CALL 10 #define UTRACE_DLSYM_START 11 #define UTRACE_DLSYM_STOP 12 struct utrace_rtld { char sig[4]; /* 'RTLD' */ int event; void *handle; void *mapbase; /* Used for 'parent' and 'init/fini' */ size_t mapsize; int refcnt; /* Used for 'mode' */ char name[MAXPATHLEN]; }; #define LD_UTRACE(e, h, mb, ms, r, n) do { \ if (ld_utrace != NULL) \ ld_utrace_log(e, h, mb, ms, r, n); \ } while (0) static void ld_utrace_log(int event, void *handle, void *mapbase, size_t mapsize, int refcnt, const char *name) { struct utrace_rtld ut; ut.sig[0] = 'R'; ut.sig[1] = 'T'; ut.sig[2] = 'L'; ut.sig[3] = 'D'; ut.event = event; ut.handle = handle; ut.mapbase = mapbase; ut.mapsize = mapsize; ut.refcnt = refcnt; bzero(ut.name, sizeof(ut.name)); if (name) strlcpy(ut.name, name, sizeof(ut.name)); utrace(&ut, sizeof(ut)); } /* * Main entry point for dynamic linking. The first argument is the * stack pointer. The stack is expected to be laid out as described * in the SVR4 ABI specification, Intel 386 Processor Supplement. * Specifically, the stack pointer points to a word containing * ARGC. Following that in the stack is a null-terminated sequence * of pointers to argument strings. Then comes a null-terminated * sequence of pointers to environment strings. Finally, there is a * sequence of "auxiliary vector" entries. * * The second argument points to a place to store the dynamic linker's * exit procedure pointer and the third to a place to store the main * program's object. * * The return value is the main program's entry point. */ func_ptr_type _rtld(Elf_Addr *sp, func_ptr_type *exit_proc, Obj_Entry **objp) { Elf_Auxinfo *aux_info[AT_COUNT]; int i; int argc; char **argv; char **env; Elf_Auxinfo *aux; Elf_Auxinfo *auxp; const char *argv0; Objlist_Entry *entry; Obj_Entry *obj; Obj_Entry **preload_tail; Obj_Entry *last_interposer; Objlist initlist; RtldLockState lockstate; char *library_path_rpath; int mib[2]; size_t len; /* * On entry, the dynamic linker itself has not been relocated yet. * Be very careful not to reference any global data until after * init_rtld has returned. It is OK to reference file-scope statics * and string constants, and to call static and global functions. */ /* Find the auxiliary vector on the stack. */ argc = *sp++; argv = (char **) sp; sp += argc + 1; /* Skip over arguments and NULL terminator */ env = (char **) sp; while (*sp++ != 0) /* Skip over environment, and NULL terminator */ ; aux = (Elf_Auxinfo *) sp; /* Digest the auxiliary vector. */ for (i = 0; i < AT_COUNT; i++) aux_info[i] = NULL; for (auxp = aux; auxp->a_type != AT_NULL; auxp++) { if (auxp->a_type < AT_COUNT) aux_info[auxp->a_type] = auxp; } /* Initialize and relocate ourselves. */ assert(aux_info[AT_BASE] != NULL); init_rtld((caddr_t) aux_info[AT_BASE]->a_un.a_ptr, aux_info); __progname = obj_rtld.path; argv0 = argv[0] != NULL ? argv[0] : "(null)"; environ = env; main_argc = argc; main_argv = argv; if (aux_info[AT_CANARY] != NULL && aux_info[AT_CANARY]->a_un.a_ptr != NULL) { i = aux_info[AT_CANARYLEN]->a_un.a_val; if (i > sizeof(__stack_chk_guard)) i = sizeof(__stack_chk_guard); memcpy(__stack_chk_guard, aux_info[AT_CANARY]->a_un.a_ptr, i); } else { mib[0] = CTL_KERN; mib[1] = KERN_ARND; len = sizeof(__stack_chk_guard); if (sysctl(mib, 2, __stack_chk_guard, &len, NULL, 0) == -1 || len != sizeof(__stack_chk_guard)) { /* If sysctl was unsuccessful, use the "terminator canary". */ ((unsigned char *)(void *)__stack_chk_guard)[0] = 0; ((unsigned char *)(void *)__stack_chk_guard)[1] = 0; ((unsigned char *)(void *)__stack_chk_guard)[2] = '\n'; ((unsigned char *)(void *)__stack_chk_guard)[3] = 255; } } trust = !issetugid(); ld_bind_now = getenv(LD_ "BIND_NOW"); /* * If the process is tainted, then we un-set the dangerous environment * variables. The process will be marked as tainted until setuid(2) * is called. If any child process calls setuid(2) we do not want any * future processes to honor the potentially un-safe variables. */ if (!trust) { if (unsetenv(LD_ "PRELOAD") || unsetenv(LD_ "LIBMAP") || unsetenv(LD_ "LIBRARY_PATH") || unsetenv(LD_ "LIBRARY_PATH_FDS") || unsetenv(LD_ "LIBMAP_DISABLE") || unsetenv(LD_ "DEBUG") || unsetenv(LD_ "ELF_HINTS_PATH") || unsetenv(LD_ "LOADFLTR") || unsetenv(LD_ "LIBRARY_PATH_RPATH")) { _rtld_error("environment corrupt; aborting"); rtld_die(); } } ld_debug = getenv(LD_ "DEBUG"); libmap_disable = getenv(LD_ "LIBMAP_DISABLE") != NULL; libmap_override = getenv(LD_ "LIBMAP"); ld_library_path = getenv(LD_ "LIBRARY_PATH"); ld_library_dirs = getenv(LD_ "LIBRARY_PATH_FDS"); ld_preload = getenv(LD_ "PRELOAD"); ld_elf_hints_path = getenv(LD_ "ELF_HINTS_PATH"); ld_loadfltr = getenv(LD_ "LOADFLTR") != NULL; library_path_rpath = getenv(LD_ "LIBRARY_PATH_RPATH"); if (library_path_rpath != NULL) { if (library_path_rpath[0] == 'y' || library_path_rpath[0] == 'Y' || library_path_rpath[0] == '1') ld_library_path_rpath = true; else ld_library_path_rpath = false; } dangerous_ld_env = libmap_disable || (libmap_override != NULL) || (ld_library_path != NULL) || (ld_preload != NULL) || (ld_elf_hints_path != NULL) || ld_loadfltr; ld_tracing = getenv(LD_ "TRACE_LOADED_OBJECTS"); ld_utrace = getenv(LD_ "UTRACE"); if ((ld_elf_hints_path == NULL) || strlen(ld_elf_hints_path) == 0) ld_elf_hints_path = _PATH_ELF_HINTS; if (ld_debug != NULL && *ld_debug != '\0') debug = 1; dbg("%s is initialized, base address = %p", __progname, (caddr_t) aux_info[AT_BASE]->a_un.a_ptr); dbg("RTLD dynamic = %p", obj_rtld.dynamic); dbg("RTLD pltgot = %p", obj_rtld.pltgot); dbg("initializing thread locks"); lockdflt_init(); /* * Load the main program, or process its program header if it is * already loaded. */ if (aux_info[AT_EXECFD] != NULL) { /* Load the main program. */ int fd = aux_info[AT_EXECFD]->a_un.a_val; dbg("loading main program"); obj_main = map_object(fd, argv0, NULL); close(fd); if (obj_main == NULL) rtld_die(); max_stack_flags = obj->stack_flags; } else { /* Main program already loaded. */ const Elf_Phdr *phdr; int phnum; caddr_t entry; dbg("processing main program's program header"); assert(aux_info[AT_PHDR] != NULL); phdr = (const Elf_Phdr *) aux_info[AT_PHDR]->a_un.a_ptr; assert(aux_info[AT_PHNUM] != NULL); phnum = aux_info[AT_PHNUM]->a_un.a_val; assert(aux_info[AT_PHENT] != NULL); assert(aux_info[AT_PHENT]->a_un.a_val == sizeof(Elf_Phdr)); assert(aux_info[AT_ENTRY] != NULL); entry = (caddr_t) aux_info[AT_ENTRY]->a_un.a_ptr; if ((obj_main = digest_phdr(phdr, phnum, entry, argv0)) == NULL) rtld_die(); } if (aux_info[AT_EXECPATH] != 0) { char *kexecpath; char buf[MAXPATHLEN]; kexecpath = aux_info[AT_EXECPATH]->a_un.a_ptr; dbg("AT_EXECPATH %p %s", kexecpath, kexecpath); if (kexecpath[0] == '/') obj_main->path = kexecpath; else if (getcwd(buf, sizeof(buf)) == NULL || strlcat(buf, "/", sizeof(buf)) >= sizeof(buf) || strlcat(buf, kexecpath, sizeof(buf)) >= sizeof(buf)) obj_main->path = xstrdup(argv0); else obj_main->path = xstrdup(buf); } else { dbg("No AT_EXECPATH"); obj_main->path = xstrdup(argv0); } dbg("obj_main path %s", obj_main->path); obj_main->mainprog = true; if (aux_info[AT_STACKPROT] != NULL && aux_info[AT_STACKPROT]->a_un.a_val != 0) stack_prot = aux_info[AT_STACKPROT]->a_un.a_val; #ifndef COMPAT_32BIT /* * Get the actual dynamic linker pathname from the executable if * possible. (It should always be possible.) That ensures that * gdb will find the right dynamic linker even if a non-standard * one is being used. */ if (obj_main->interp != NULL && strcmp(obj_main->interp, obj_rtld.path) != 0) { free(obj_rtld.path); obj_rtld.path = xstrdup(obj_main->interp); __progname = obj_rtld.path; } #endif digest_dynamic(obj_main, 0); dbg("%s valid_hash_sysv %d valid_hash_gnu %d dynsymcount %d", obj_main->path, obj_main->valid_hash_sysv, obj_main->valid_hash_gnu, obj_main->dynsymcount); linkmap_add(obj_main); linkmap_add(&obj_rtld); /* Link the main program into the list of objects. */ *obj_tail = obj_main; obj_tail = &obj_main->next; obj_count++; obj_loads++; /* Initialize a fake symbol for resolving undefined weak references. */ sym_zero.st_info = ELF_ST_INFO(STB_GLOBAL, STT_NOTYPE); sym_zero.st_shndx = SHN_UNDEF; sym_zero.st_value = -(uintptr_t)obj_main->relocbase; if (!libmap_disable) libmap_disable = (bool)lm_init(libmap_override); dbg("loading LD_PRELOAD libraries"); if (load_preload_objects() == -1) rtld_die(); preload_tail = obj_tail; dbg("loading needed objects"); if (load_needed_objects(obj_main, 0) == -1) rtld_die(); /* Make a list of all objects loaded at startup. */ last_interposer = obj_main; for (obj = obj_list; obj != NULL; obj = obj->next) { if (obj->z_interpose && obj != obj_main) { objlist_put_after(&list_main, last_interposer, obj); last_interposer = obj; } else { objlist_push_tail(&list_main, obj); } obj->refcount++; } dbg("checking for required versions"); if (rtld_verify_versions(&list_main) == -1 && !ld_tracing) rtld_die(); if (ld_tracing) { /* We're done */ trace_loaded_objects(obj_main); exit(0); } if (getenv(LD_ "DUMP_REL_PRE") != NULL) { dump_relocations(obj_main); exit (0); } /* * Processing tls relocations requires having the tls offsets * initialized. Prepare offsets before starting initial * relocation processing. */ dbg("initializing initial thread local storage offsets"); STAILQ_FOREACH(entry, &list_main, link) { /* * Allocate all the initial objects out of the static TLS * block even if they didn't ask for it. */ allocate_tls_offset(entry->obj); } if (relocate_objects(obj_main, ld_bind_now != NULL && *ld_bind_now != '\0', &obj_rtld, SYMLOOK_EARLY, NULL) == -1) rtld_die(); dbg("doing copy relocations"); if (do_copy_relocations(obj_main) == -1) rtld_die(); if (getenv(LD_ "DUMP_REL_POST") != NULL) { dump_relocations(obj_main); exit (0); } /* * Setup TLS for main thread. This must be done after the * relocations are processed, since tls initialization section * might be the subject for relocations. */ dbg("initializing initial thread local storage"); allocate_initial_tls(obj_list); dbg("initializing key program variables"); set_program_var("__progname", argv[0] != NULL ? basename(argv[0]) : ""); set_program_var("environ", env); set_program_var("__elf_aux_vector", aux); /* Make a list of init functions to call. */ objlist_init(&initlist); initlist_add_objects(obj_list, preload_tail, &initlist); r_debug_state(NULL, &obj_main->linkmap); /* say hello to gdb! */ map_stacks_exec(NULL); dbg("resolving ifuncs"); if (resolve_objects_ifunc(obj_main, ld_bind_now != NULL && *ld_bind_now != '\0', SYMLOOK_EARLY, NULL) == -1) rtld_die(); if (!obj_main->crt_no_init) { /* * Make sure we don't call the main program's init and fini * functions for binaries linked with old crt1 which calls * _init itself. */ obj_main->init = obj_main->fini = (Elf_Addr)NULL; obj_main->preinit_array = obj_main->init_array = obj_main->fini_array = (Elf_Addr)NULL; } wlock_acquire(rtld_bind_lock, &lockstate); if (obj_main->crt_no_init) preinit_main(); objlist_call_init(&initlist, &lockstate); _r_debug_postinit(&obj_main->linkmap); objlist_clear(&initlist); dbg("loading filtees"); for (obj = obj_list->next; obj != NULL; obj = obj->next) { if (ld_loadfltr || obj->z_loadfltr) load_filtees(obj, 0, &lockstate); } lock_release(rtld_bind_lock, &lockstate); dbg("transferring control to program entry point = %p", obj_main->entry); /* Return the exit procedure and the program entry point. */ *exit_proc = rtld_exit; *objp = obj_main; return (func_ptr_type) obj_main->entry; } void * rtld_resolve_ifunc(const Obj_Entry *obj, const Elf_Sym *def) { void *ptr; Elf_Addr target; ptr = (void *)make_function_pointer(def, obj); target = ((Elf_Addr (*)(void))ptr)(); return ((void *)target); } Elf_Addr _rtld_bind(Obj_Entry *obj, Elf_Size reloff) { const Elf_Rel *rel; const Elf_Sym *def; const Obj_Entry *defobj; Elf_Addr *where; Elf_Addr target; RtldLockState lockstate; rlock_acquire(rtld_bind_lock, &lockstate); if (sigsetjmp(lockstate.env, 0) != 0) lock_upgrade(rtld_bind_lock, &lockstate); if (obj->pltrel) rel = (const Elf_Rel *) ((caddr_t) obj->pltrel + reloff); else rel = (const Elf_Rel *) ((caddr_t) obj->pltrela + reloff); where = (Elf_Addr *) (obj->relocbase + rel->r_offset); def = find_symdef(ELF_R_SYM(rel->r_info), obj, &defobj, true, NULL, &lockstate); if (def == NULL) rtld_die(); if (ELF_ST_TYPE(def->st_info) == STT_GNU_IFUNC) target = (Elf_Addr)rtld_resolve_ifunc(defobj, def); else target = (Elf_Addr)(defobj->relocbase + def->st_value); dbg("\"%s\" in \"%s\" ==> %p in \"%s\"", defobj->strtab + def->st_name, basename(obj->path), (void *)target, basename(defobj->path)); /* * Write the new contents for the jmpslot. Note that depending on * architecture, the value which we need to return back to the * lazy binding trampoline may or may not be the target * address. The value returned from reloc_jmpslot() is the value * that the trampoline needs. */ target = reloc_jmpslot(where, target, defobj, obj, rel); lock_release(rtld_bind_lock, &lockstate); return target; } /* * Error reporting function. Use it like printf. If formats the message * into a buffer, and sets things up so that the next call to dlerror() * will return the message. */ void _rtld_error(const char *fmt, ...) { static char buf[512]; va_list ap; va_start(ap, fmt); rtld_vsnprintf(buf, sizeof buf, fmt, ap); error_message = buf; va_end(ap); } /* * Return a dynamically-allocated copy of the current error message, if any. */ static char * errmsg_save(void) { return error_message == NULL ? NULL : xstrdup(error_message); } /* * Restore the current error message from a copy which was previously saved * by errmsg_save(). The copy is freed. */ static void errmsg_restore(char *saved_msg) { if (saved_msg == NULL) error_message = NULL; else { _rtld_error("%s", saved_msg); free(saved_msg); } } static const char * basename(const char *name) { const char *p = strrchr(name, '/'); return p != NULL ? p + 1 : name; } static struct utsname uts; static char * origin_subst_one(char *real, const char *kw, const char *subst, bool may_free) { char *p, *p1, *res, *resp; int subst_len, kw_len, subst_count, old_len, new_len; kw_len = strlen(kw); /* * First, count the number of the keyword occurences, to * preallocate the final string. */ for (p = real, subst_count = 0;; p = p1 + kw_len, subst_count++) { p1 = strstr(p, kw); if (p1 == NULL) break; } /* * If the keyword is not found, just return. */ if (subst_count == 0) return (may_free ? real : xstrdup(real)); /* * There is indeed something to substitute. Calculate the * length of the resulting string, and allocate it. */ subst_len = strlen(subst); old_len = strlen(real); new_len = old_len + (subst_len - kw_len) * subst_count; res = xmalloc(new_len + 1); /* * Now, execute the substitution loop. */ for (p = real, resp = res, *resp = '\0';;) { p1 = strstr(p, kw); if (p1 != NULL) { /* Copy the prefix before keyword. */ memcpy(resp, p, p1 - p); resp += p1 - p; /* Keyword replacement. */ memcpy(resp, subst, subst_len); resp += subst_len; *resp = '\0'; p = p1 + kw_len; } else break; } /* Copy to the end of string and finish. */ strcat(resp, p); if (may_free) free(real); return (res); } static char * origin_subst(char *real, const char *origin_path) { char *res1, *res2, *res3, *res4; if (uts.sysname[0] == '\0') { if (uname(&uts) != 0) { _rtld_error("utsname failed: %d", errno); return (NULL); } } res1 = origin_subst_one(real, "$ORIGIN", origin_path, false); res2 = origin_subst_one(res1, "$OSNAME", uts.sysname, true); res3 = origin_subst_one(res2, "$OSREL", uts.release, true); res4 = origin_subst_one(res3, "$PLATFORM", uts.machine, true); return (res4); } void rtld_die(void) { const char *msg = dlerror(); if (msg == NULL) msg = "Fatal error"; rtld_fdputstr(STDERR_FILENO, msg); rtld_fdputchar(STDERR_FILENO, '\n'); _exit(1); } /* * Process a shared object's DYNAMIC section, and save the important * information in its Obj_Entry structure. */ static void digest_dynamic1(Obj_Entry *obj, int early, const Elf_Dyn **dyn_rpath, const Elf_Dyn **dyn_soname, const Elf_Dyn **dyn_runpath) { const Elf_Dyn *dynp; Needed_Entry **needed_tail = &obj->needed; Needed_Entry **needed_filtees_tail = &obj->needed_filtees; Needed_Entry **needed_aux_filtees_tail = &obj->needed_aux_filtees; const Elf_Hashelt *hashtab; const Elf32_Word *hashval; Elf32_Word bkt, nmaskwords; int bloom_size32; int plttype = DT_REL; *dyn_rpath = NULL; *dyn_soname = NULL; *dyn_runpath = NULL; obj->bind_now = false; for (dynp = obj->dynamic; dynp->d_tag != DT_NULL; dynp++) { switch (dynp->d_tag) { case DT_REL: obj->rel = (const Elf_Rel *) (obj->relocbase + dynp->d_un.d_ptr); break; case DT_RELSZ: obj->relsize = dynp->d_un.d_val; break; case DT_RELENT: assert(dynp->d_un.d_val == sizeof(Elf_Rel)); break; case DT_JMPREL: obj->pltrel = (const Elf_Rel *) (obj->relocbase + dynp->d_un.d_ptr); break; case DT_PLTRELSZ: obj->pltrelsize = dynp->d_un.d_val; break; case DT_RELA: obj->rela = (const Elf_Rela *) (obj->relocbase + dynp->d_un.d_ptr); break; case DT_RELASZ: obj->relasize = dynp->d_un.d_val; break; case DT_RELAENT: assert(dynp->d_un.d_val == sizeof(Elf_Rela)); break; case DT_PLTREL: plttype = dynp->d_un.d_val; assert(dynp->d_un.d_val == DT_REL || plttype == DT_RELA); break; case DT_SYMTAB: obj->symtab = (const Elf_Sym *) (obj->relocbase + dynp->d_un.d_ptr); break; case DT_SYMENT: assert(dynp->d_un.d_val == sizeof(Elf_Sym)); break; case DT_STRTAB: obj->strtab = (const char *) (obj->relocbase + dynp->d_un.d_ptr); break; case DT_STRSZ: obj->strsize = dynp->d_un.d_val; break; case DT_VERNEED: obj->verneed = (const Elf_Verneed *) (obj->relocbase + dynp->d_un.d_val); break; case DT_VERNEEDNUM: obj->verneednum = dynp->d_un.d_val; break; case DT_VERDEF: obj->verdef = (const Elf_Verdef *) (obj->relocbase + dynp->d_un.d_val); break; case DT_VERDEFNUM: obj->verdefnum = dynp->d_un.d_val; break; case DT_VERSYM: obj->versyms = (const Elf_Versym *)(obj->relocbase + dynp->d_un.d_val); break; case DT_HASH: { hashtab = (const Elf_Hashelt *)(obj->relocbase + dynp->d_un.d_ptr); obj->nbuckets = hashtab[0]; obj->nchains = hashtab[1]; obj->buckets = hashtab + 2; obj->chains = obj->buckets + obj->nbuckets; obj->valid_hash_sysv = obj->nbuckets > 0 && obj->nchains > 0 && obj->buckets != NULL; } break; case DT_GNU_HASH: { hashtab = (const Elf_Hashelt *)(obj->relocbase + dynp->d_un.d_ptr); obj->nbuckets_gnu = hashtab[0]; obj->symndx_gnu = hashtab[1]; nmaskwords = hashtab[2]; bloom_size32 = (__ELF_WORD_SIZE / 32) * nmaskwords; obj->maskwords_bm_gnu = nmaskwords - 1; obj->shift2_gnu = hashtab[3]; obj->bloom_gnu = (Elf_Addr *) (hashtab + 4); obj->buckets_gnu = hashtab + 4 + bloom_size32; obj->chain_zero_gnu = obj->buckets_gnu + obj->nbuckets_gnu - obj->symndx_gnu; /* Number of bitmask words is required to be power of 2 */ obj->valid_hash_gnu = powerof2(nmaskwords) && obj->nbuckets_gnu > 0 && obj->buckets_gnu != NULL; } break; case DT_NEEDED: if (!obj->rtld) { Needed_Entry *nep = NEW(Needed_Entry); nep->name = dynp->d_un.d_val; nep->obj = NULL; nep->next = NULL; *needed_tail = nep; needed_tail = &nep->next; } break; case DT_FILTER: if (!obj->rtld) { Needed_Entry *nep = NEW(Needed_Entry); nep->name = dynp->d_un.d_val; nep->obj = NULL; nep->next = NULL; *needed_filtees_tail = nep; needed_filtees_tail = &nep->next; } break; case DT_AUXILIARY: if (!obj->rtld) { Needed_Entry *nep = NEW(Needed_Entry); nep->name = dynp->d_un.d_val; nep->obj = NULL; nep->next = NULL; *needed_aux_filtees_tail = nep; needed_aux_filtees_tail = &nep->next; } break; case DT_PLTGOT: obj->pltgot = (Elf_Addr *) (obj->relocbase + dynp->d_un.d_ptr); break; case DT_TEXTREL: obj->textrel = true; break; case DT_SYMBOLIC: obj->symbolic = true; break; case DT_RPATH: /* * We have to wait until later to process this, because we * might not have gotten the address of the string table yet. */ *dyn_rpath = dynp; break; case DT_SONAME: *dyn_soname = dynp; break; case DT_RUNPATH: *dyn_runpath = dynp; break; case DT_INIT: obj->init = (Elf_Addr) (obj->relocbase + dynp->d_un.d_ptr); break; case DT_PREINIT_ARRAY: obj->preinit_array = (Elf_Addr)(obj->relocbase + dynp->d_un.d_ptr); break; case DT_PREINIT_ARRAYSZ: obj->preinit_array_num = dynp->d_un.d_val / sizeof(Elf_Addr); break; case DT_INIT_ARRAY: obj->init_array = (Elf_Addr)(obj->relocbase + dynp->d_un.d_ptr); break; case DT_INIT_ARRAYSZ: obj->init_array_num = dynp->d_un.d_val / sizeof(Elf_Addr); break; case DT_FINI: obj->fini = (Elf_Addr) (obj->relocbase + dynp->d_un.d_ptr); break; case DT_FINI_ARRAY: obj->fini_array = (Elf_Addr)(obj->relocbase + dynp->d_un.d_ptr); break; case DT_FINI_ARRAYSZ: obj->fini_array_num = dynp->d_un.d_val / sizeof(Elf_Addr); break; /* * Don't process DT_DEBUG on MIPS as the dynamic section * is mapped read-only. DT_MIPS_RLD_MAP is used instead. */ #ifndef __mips__ case DT_DEBUG: /* XXX - not implemented yet */ if (!early) dbg("Filling in DT_DEBUG entry"); ((Elf_Dyn*)dynp)->d_un.d_ptr = (Elf_Addr) &r_debug; break; #endif case DT_FLAGS: if ((dynp->d_un.d_val & DF_ORIGIN) && trust) obj->z_origin = true; if (dynp->d_un.d_val & DF_SYMBOLIC) obj->symbolic = true; if (dynp->d_un.d_val & DF_TEXTREL) obj->textrel = true; if (dynp->d_un.d_val & DF_BIND_NOW) obj->bind_now = true; /*if (dynp->d_un.d_val & DF_STATIC_TLS) ;*/ break; #ifdef __mips__ case DT_MIPS_LOCAL_GOTNO: obj->local_gotno = dynp->d_un.d_val; break; case DT_MIPS_SYMTABNO: obj->symtabno = dynp->d_un.d_val; break; case DT_MIPS_GOTSYM: obj->gotsym = dynp->d_un.d_val; break; case DT_MIPS_RLD_MAP: *((Elf_Addr *)(dynp->d_un.d_ptr)) = (Elf_Addr) &r_debug; break; #endif case DT_FLAGS_1: if (dynp->d_un.d_val & DF_1_NOOPEN) obj->z_noopen = true; if ((dynp->d_un.d_val & DF_1_ORIGIN) && trust) obj->z_origin = true; if (dynp->d_un.d_val & DF_1_GLOBAL) obj->z_global = true; if (dynp->d_un.d_val & DF_1_BIND_NOW) obj->bind_now = true; if (dynp->d_un.d_val & DF_1_NODELETE) obj->z_nodelete = true; if (dynp->d_un.d_val & DF_1_LOADFLTR) obj->z_loadfltr = true; if (dynp->d_un.d_val & DF_1_INTERPOSE) obj->z_interpose = true; if (dynp->d_un.d_val & DF_1_NODEFLIB) obj->z_nodeflib = true; break; default: if (!early) { dbg("Ignoring d_tag %ld = %#lx", (long)dynp->d_tag, (long)dynp->d_tag); } break; } } obj->traced = false; if (plttype == DT_RELA) { obj->pltrela = (const Elf_Rela *) obj->pltrel; obj->pltrel = NULL; obj->pltrelasize = obj->pltrelsize; obj->pltrelsize = 0; } /* Determine size of dynsym table (equal to nchains of sysv hash) */ if (obj->valid_hash_sysv) obj->dynsymcount = obj->nchains; else if (obj->valid_hash_gnu) { obj->dynsymcount = 0; for (bkt = 0; bkt < obj->nbuckets_gnu; bkt++) { if (obj->buckets_gnu[bkt] == 0) continue; hashval = &obj->chain_zero_gnu[obj->buckets_gnu[bkt]]; do obj->dynsymcount++; while ((*hashval++ & 1u) == 0); } obj->dynsymcount += obj->symndx_gnu; } } static void digest_dynamic2(Obj_Entry *obj, const Elf_Dyn *dyn_rpath, const Elf_Dyn *dyn_soname, const Elf_Dyn *dyn_runpath) { if (obj->z_origin && obj->origin_path == NULL) { obj->origin_path = xmalloc(PATH_MAX); if (rtld_dirname_abs(obj->path, obj->origin_path) == -1) rtld_die(); } if (dyn_runpath != NULL) { obj->runpath = (char *)obj->strtab + dyn_runpath->d_un.d_val; if (obj->z_origin) obj->runpath = origin_subst(obj->runpath, obj->origin_path); } else if (dyn_rpath != NULL) { obj->rpath = (char *)obj->strtab + dyn_rpath->d_un.d_val; if (obj->z_origin) obj->rpath = origin_subst(obj->rpath, obj->origin_path); } if (dyn_soname != NULL) object_add_name(obj, obj->strtab + dyn_soname->d_un.d_val); } static void digest_dynamic(Obj_Entry *obj, int early) { const Elf_Dyn *dyn_rpath; const Elf_Dyn *dyn_soname; const Elf_Dyn *dyn_runpath; digest_dynamic1(obj, early, &dyn_rpath, &dyn_soname, &dyn_runpath); digest_dynamic2(obj, dyn_rpath, dyn_soname, dyn_runpath); } /* * Process a shared object's program header. This is used only for the * main program, when the kernel has already loaded the main program * into memory before calling the dynamic linker. It creates and * returns an Obj_Entry structure. */ static Obj_Entry * digest_phdr(const Elf_Phdr *phdr, int phnum, caddr_t entry, const char *path) { Obj_Entry *obj; const Elf_Phdr *phlimit = phdr + phnum; const Elf_Phdr *ph; Elf_Addr note_start, note_end; int nsegs = 0; obj = obj_new(); for (ph = phdr; ph < phlimit; ph++) { if (ph->p_type != PT_PHDR) continue; obj->phdr = phdr; obj->phsize = ph->p_memsz; obj->relocbase = (caddr_t)phdr - ph->p_vaddr; break; } obj->stack_flags = PF_X | PF_R | PF_W; for (ph = phdr; ph < phlimit; ph++) { switch (ph->p_type) { case PT_INTERP: obj->interp = (const char *)(ph->p_vaddr + obj->relocbase); break; case PT_LOAD: if (nsegs == 0) { /* First load segment */ obj->vaddrbase = trunc_page(ph->p_vaddr); obj->mapbase = obj->vaddrbase + obj->relocbase; obj->textsize = round_page(ph->p_vaddr + ph->p_memsz) - obj->vaddrbase; } else { /* Last load segment */ obj->mapsize = round_page(ph->p_vaddr + ph->p_memsz) - obj->vaddrbase; } nsegs++; break; case PT_DYNAMIC: obj->dynamic = (const Elf_Dyn *)(ph->p_vaddr + obj->relocbase); break; case PT_TLS: obj->tlsindex = 1; obj->tlssize = ph->p_memsz; obj->tlsalign = ph->p_align; obj->tlsinitsize = ph->p_filesz; obj->tlsinit = (void*)(ph->p_vaddr + obj->relocbase); break; case PT_GNU_STACK: obj->stack_flags = ph->p_flags; break; case PT_GNU_RELRO: obj->relro_page = obj->relocbase + trunc_page(ph->p_vaddr); obj->relro_size = round_page(ph->p_memsz); break; case PT_NOTE: note_start = (Elf_Addr)obj->relocbase + ph->p_vaddr; note_end = note_start + ph->p_filesz; digest_notes(obj, note_start, note_end); break; } } if (nsegs < 1) { _rtld_error("%s: too few PT_LOAD segments", path); return NULL; } obj->entry = entry; return obj; } void digest_notes(Obj_Entry *obj, Elf_Addr note_start, Elf_Addr note_end) { const Elf_Note *note; const char *note_name; uintptr_t p; for (note = (const Elf_Note *)note_start; (Elf_Addr)note < note_end; note = (const Elf_Note *)((const char *)(note + 1) + roundup2(note->n_namesz, sizeof(Elf32_Addr)) + roundup2(note->n_descsz, sizeof(Elf32_Addr)))) { if (note->n_namesz != sizeof(NOTE_FREEBSD_VENDOR) || note->n_descsz != sizeof(int32_t)) continue; if (note->n_type != ABI_NOTETYPE && note->n_type != CRT_NOINIT_NOTETYPE) continue; note_name = (const char *)(note + 1); if (strncmp(NOTE_FREEBSD_VENDOR, note_name, sizeof(NOTE_FREEBSD_VENDOR)) != 0) continue; switch (note->n_type) { case ABI_NOTETYPE: /* FreeBSD osrel note */ p = (uintptr_t)(note + 1); p += roundup2(note->n_namesz, sizeof(Elf32_Addr)); obj->osrel = *(const int32_t *)(p); dbg("note osrel %d", obj->osrel); break; case CRT_NOINIT_NOTETYPE: /* FreeBSD 'crt does not call init' note */ obj->crt_no_init = true; dbg("note crt_no_init"); break; } } } static Obj_Entry * dlcheck(void *handle) { Obj_Entry *obj; for (obj = obj_list; obj != NULL; obj = obj->next) if (obj == (Obj_Entry *) handle) break; if (obj == NULL || obj->refcount == 0 || obj->dl_refcount == 0) { _rtld_error("Invalid shared object handle %p", handle); return NULL; } return obj; } /* * If the given object is already in the donelist, return true. Otherwise * add the object to the list and return false. */ static bool donelist_check(DoneList *dlp, const Obj_Entry *obj) { unsigned int i; for (i = 0; i < dlp->num_used; i++) if (dlp->objs[i] == obj) return true; /* * Our donelist allocation should always be sufficient. But if * our threads locking isn't working properly, more shared objects * could have been loaded since we allocated the list. That should * never happen, but we'll handle it properly just in case it does. */ if (dlp->num_used < dlp->num_alloc) dlp->objs[dlp->num_used++] = obj; return false; } /* * Hash function for symbol table lookup. Don't even think about changing * this. It is specified by the System V ABI. */ unsigned long elf_hash(const char *name) { const unsigned char *p = (const unsigned char *) name; unsigned long h = 0; unsigned long g; while (*p != '\0') { h = (h << 4) + *p++; if ((g = h & 0xf0000000) != 0) h ^= g >> 24; h &= ~g; } return h; } /* * The GNU hash function is the Daniel J. Bernstein hash clipped to 32 bits * unsigned in case it's implemented with a wider type. */ static uint32_t gnu_hash(const char *s) { uint32_t h; unsigned char c; h = 5381; for (c = *s; c != '\0'; c = *++s) h = h * 33 + c; return (h & 0xffffffff); } /* * Find the library with the given name, and return its full pathname. * The returned string is dynamically allocated. Generates an error * message and returns NULL if the library cannot be found. * * If the second argument is non-NULL, then it refers to an already- * loaded shared object, whose library search path will be searched. * * If a library is successfully located via LD_LIBRARY_PATH_FDS, its * descriptor (which is close-on-exec) will be passed out via the third * argument. * * The search order is: * DT_RPATH in the referencing file _unless_ DT_RUNPATH is present (1) * DT_RPATH of the main object if DSO without defined DT_RUNPATH (1) * LD_LIBRARY_PATH * DT_RUNPATH in the referencing file * ldconfig hints (if -z nodefaultlib, filter out default library directories * from list) * /lib:/usr/lib _unless_ the referencing file is linked with -z nodefaultlib * * (1) Handled in digest_dynamic2 - rpath left NULL if runpath defined. */ static char * find_library(const char *xname, const Obj_Entry *refobj, int *fdp) { char *pathname; char *name; bool nodeflib, objgiven; objgiven = refobj != NULL; if (strchr(xname, '/') != NULL) { /* Hard coded pathname */ if (xname[0] != '/' && !trust) { _rtld_error("Absolute pathname required for shared object \"%s\"", xname); return NULL; } if (objgiven && refobj->z_origin) { return (origin_subst(__DECONST(char *, xname), refobj->origin_path)); } else { return (xstrdup(xname)); } } if (libmap_disable || !objgiven || (name = lm_find(refobj->path, xname)) == NULL) name = (char *)xname; dbg(" Searching for \"%s\"", name); /* * If refobj->rpath != NULL, then refobj->runpath is NULL. Fall * back to pre-conforming behaviour if user requested so with * LD_LIBRARY_PATH_RPATH environment variable and ignore -z * nodeflib. */ if (objgiven && refobj->rpath != NULL && ld_library_path_rpath) { if ((pathname = search_library_path(name, ld_library_path)) != NULL || (refobj != NULL && (pathname = search_library_path(name, refobj->rpath)) != NULL) || (pathname = search_library_pathfds(name, ld_library_dirs, fdp)) != NULL || (pathname = search_library_path(name, gethints(false))) != NULL || (pathname = search_library_path(name, STANDARD_LIBRARY_PATH)) != NULL) return (pathname); } else { nodeflib = objgiven ? refobj->z_nodeflib : false; if ((objgiven && (pathname = search_library_path(name, refobj->rpath)) != NULL) || (objgiven && refobj->runpath == NULL && refobj != obj_main && (pathname = search_library_path(name, obj_main->rpath)) != NULL) || (pathname = search_library_path(name, ld_library_path)) != NULL || (objgiven && (pathname = search_library_path(name, refobj->runpath)) != NULL) || (pathname = search_library_pathfds(name, ld_library_dirs, fdp)) != NULL || (pathname = search_library_path(name, gethints(nodeflib))) != NULL || (objgiven && !nodeflib && (pathname = search_library_path(name, STANDARD_LIBRARY_PATH)) != NULL)) return (pathname); } if (objgiven && refobj->path != NULL) { _rtld_error("Shared object \"%s\" not found, required by \"%s\"", name, basename(refobj->path)); } else { _rtld_error("Shared object \"%s\" not found", name); } return NULL; } /* * Given a symbol number in a referencing object, find the corresponding * definition of the symbol. Returns a pointer to the symbol, or NULL if * no definition was found. Returns a pointer to the Obj_Entry of the * defining object via the reference parameter DEFOBJ_OUT. */ const Elf_Sym * find_symdef(unsigned long symnum, const Obj_Entry *refobj, const Obj_Entry **defobj_out, int flags, SymCache *cache, RtldLockState *lockstate) { const Elf_Sym *ref; const Elf_Sym *def; const Obj_Entry *defobj; SymLook req; const char *name; int res; /* * If we have already found this symbol, get the information from * the cache. */ if (symnum >= refobj->dynsymcount) return NULL; /* Bad object */ if (cache != NULL && cache[symnum].sym != NULL) { *defobj_out = cache[symnum].obj; return cache[symnum].sym; } ref = refobj->symtab + symnum; name = refobj->strtab + ref->st_name; def = NULL; defobj = NULL; /* * We don't have to do a full scale lookup if the symbol is local. * We know it will bind to the instance in this load module; to * which we already have a pointer (ie ref). By not doing a lookup, * we not only improve performance, but it also avoids unresolvable * symbols when local symbols are not in the hash table. This has * been seen with the ia64 toolchain. */ if (ELF_ST_BIND(ref->st_info) != STB_LOCAL) { if (ELF_ST_TYPE(ref->st_info) == STT_SECTION) { _rtld_error("%s: Bogus symbol table entry %lu", refobj->path, symnum); } symlook_init(&req, name); req.flags = flags; req.ventry = fetch_ventry(refobj, symnum); req.lockstate = lockstate; res = symlook_default(&req, refobj); if (res == 0) { def = req.sym_out; defobj = req.defobj_out; } } else { def = ref; defobj = refobj; } /* * If we found no definition and the reference is weak, treat the * symbol as having the value zero. */ if (def == NULL && ELF_ST_BIND(ref->st_info) == STB_WEAK) { def = &sym_zero; defobj = obj_main; } if (def != NULL) { *defobj_out = defobj; /* Record the information in the cache to avoid subsequent lookups. */ if (cache != NULL) { cache[symnum].sym = def; cache[symnum].obj = defobj; } } else { if (refobj != &obj_rtld) _rtld_error("%s: Undefined symbol \"%s\"", refobj->path, name); } return def; } /* * Return the search path from the ldconfig hints file, reading it if * necessary. If nostdlib is true, then the default search paths are * not added to result. * * Returns NULL if there are problems with the hints file, * or if the search path there is empty. */ static const char * gethints(bool nostdlib) { static char *hints, *filtered_path; struct elfhints_hdr hdr; struct fill_search_info_args sargs, hargs; struct dl_serinfo smeta, hmeta, *SLPinfo, *hintinfo; struct dl_serpath *SLPpath, *hintpath; char *p; unsigned int SLPndx, hintndx, fndx, fcount; int fd; size_t flen; bool skip; /* First call, read the hints file */ if (hints == NULL) { /* Keep from trying again in case the hints file is bad. */ hints = ""; if ((fd = open(ld_elf_hints_path, O_RDONLY | O_CLOEXEC)) == -1) return (NULL); if (read(fd, &hdr, sizeof hdr) != sizeof hdr || hdr.magic != ELFHINTS_MAGIC || hdr.version != 1) { close(fd); return (NULL); } p = xmalloc(hdr.dirlistlen + 1); if (lseek(fd, hdr.strtab + hdr.dirlist, SEEK_SET) == -1 || read(fd, p, hdr.dirlistlen + 1) != (ssize_t)hdr.dirlistlen + 1) { free(p); close(fd); return (NULL); } hints = p; close(fd); } /* * If caller agreed to receive list which includes the default * paths, we are done. Otherwise, if we still did not * calculated filtered result, do it now. */ if (!nostdlib) return (hints[0] != '\0' ? hints : NULL); if (filtered_path != NULL) goto filt_ret; /* * Obtain the list of all configured search paths, and the * list of the default paths. * * First estimate the size of the results. */ smeta.dls_size = __offsetof(struct dl_serinfo, dls_serpath); smeta.dls_cnt = 0; hmeta.dls_size = __offsetof(struct dl_serinfo, dls_serpath); hmeta.dls_cnt = 0; sargs.request = RTLD_DI_SERINFOSIZE; sargs.serinfo = &smeta; hargs.request = RTLD_DI_SERINFOSIZE; hargs.serinfo = &hmeta; path_enumerate(STANDARD_LIBRARY_PATH, fill_search_info, &sargs); path_enumerate(p, fill_search_info, &hargs); SLPinfo = xmalloc(smeta.dls_size); hintinfo = xmalloc(hmeta.dls_size); /* * Next fetch both sets of paths. */ sargs.request = RTLD_DI_SERINFO; sargs.serinfo = SLPinfo; sargs.serpath = &SLPinfo->dls_serpath[0]; sargs.strspace = (char *)&SLPinfo->dls_serpath[smeta.dls_cnt]; hargs.request = RTLD_DI_SERINFO; hargs.serinfo = hintinfo; hargs.serpath = &hintinfo->dls_serpath[0]; hargs.strspace = (char *)&hintinfo->dls_serpath[hmeta.dls_cnt]; path_enumerate(STANDARD_LIBRARY_PATH, fill_search_info, &sargs); path_enumerate(p, fill_search_info, &hargs); /* * Now calculate the difference between two sets, by excluding * standard paths from the full set. */ fndx = 0; fcount = 0; filtered_path = xmalloc(hdr.dirlistlen + 1); hintpath = &hintinfo->dls_serpath[0]; for (hintndx = 0; hintndx < hmeta.dls_cnt; hintndx++, hintpath++) { skip = false; SLPpath = &SLPinfo->dls_serpath[0]; /* * Check each standard path against current. */ for (SLPndx = 0; SLPndx < smeta.dls_cnt; SLPndx++, SLPpath++) { /* matched, skip the path */ if (!strcmp(hintpath->dls_name, SLPpath->dls_name)) { skip = true; break; } } if (skip) continue; /* * Not matched against any standard path, add the path * to result. Separate consequtive paths with ':'. */ if (fcount > 0) { filtered_path[fndx] = ':'; fndx++; } fcount++; flen = strlen(hintpath->dls_name); strncpy((filtered_path + fndx), hintpath->dls_name, flen); fndx += flen; } filtered_path[fndx] = '\0'; free(SLPinfo); free(hintinfo); filt_ret: return (filtered_path[0] != '\0' ? filtered_path : NULL); } static void init_dag(Obj_Entry *root) { const Needed_Entry *needed; const Objlist_Entry *elm; DoneList donelist; if (root->dag_inited) return; donelist_init(&donelist); /* Root object belongs to own DAG. */ objlist_push_tail(&root->dldags, root); objlist_push_tail(&root->dagmembers, root); donelist_check(&donelist, root); /* * Add dependencies of root object to DAG in breadth order * by exploiting the fact that each new object get added * to the tail of the dagmembers list. */ STAILQ_FOREACH(elm, &root->dagmembers, link) { for (needed = elm->obj->needed; needed != NULL; needed = needed->next) { if (needed->obj == NULL || donelist_check(&donelist, needed->obj)) continue; objlist_push_tail(&needed->obj->dldags, root); objlist_push_tail(&root->dagmembers, needed->obj); } } root->dag_inited = true; } static void process_z(Obj_Entry *root) { const Objlist_Entry *elm; Obj_Entry *obj; /* * Walk over object DAG and process every dependent object * that is marked as DF_1_NODELETE or DF_1_GLOBAL. They need * to grow their own DAG. * * For DF_1_GLOBAL, DAG is required for symbol lookups in * symlook_global() to work. * * For DF_1_NODELETE, the DAG should have its reference upped. */ STAILQ_FOREACH(elm, &root->dagmembers, link) { obj = elm->obj; if (obj == NULL) continue; if (obj->z_nodelete && !obj->ref_nodel) { dbg("obj %s -z nodelete", obj->path); init_dag(obj); ref_dag(obj); obj->ref_nodel = true; } if (obj->z_global && objlist_find(&list_global, obj) == NULL) { dbg("obj %s -z global", obj->path); objlist_push_tail(&list_global, obj); init_dag(obj); } } } /* * Initialize the dynamic linker. The argument is the address at which * the dynamic linker has been mapped into memory. The primary task of * this function is to relocate the dynamic linker. */ static void init_rtld(caddr_t mapbase, Elf_Auxinfo **aux_info) { Obj_Entry objtmp; /* Temporary rtld object */ const Elf_Dyn *dyn_rpath; const Elf_Dyn *dyn_soname; const Elf_Dyn *dyn_runpath; #ifdef RTLD_INIT_PAGESIZES_EARLY /* The page size is required by the dynamic memory allocator. */ init_pagesizes(aux_info); #endif /* * Conjure up an Obj_Entry structure for the dynamic linker. * * The "path" member can't be initialized yet because string constants * cannot yet be accessed. Below we will set it correctly. */ memset(&objtmp, 0, sizeof(objtmp)); objtmp.path = NULL; objtmp.rtld = true; objtmp.mapbase = mapbase; #ifdef PIC objtmp.relocbase = mapbase; #endif if (RTLD_IS_DYNAMIC()) { objtmp.dynamic = rtld_dynamic(&objtmp); digest_dynamic1(&objtmp, 1, &dyn_rpath, &dyn_soname, &dyn_runpath); assert(objtmp.needed == NULL); #if !defined(__mips__) /* MIPS has a bogus DT_TEXTREL. */ assert(!objtmp.textrel); #endif /* * Temporarily put the dynamic linker entry into the object list, so * that symbols can be found. */ relocate_objects(&objtmp, true, &objtmp, 0, NULL); } /* Initialize the object list. */ obj_tail = &obj_list; /* Now that non-local variables can be accesses, copy out obj_rtld. */ memcpy(&obj_rtld, &objtmp, sizeof(obj_rtld)); #ifndef RTLD_INIT_PAGESIZES_EARLY /* The page size is required by the dynamic memory allocator. */ init_pagesizes(aux_info); #endif if (aux_info[AT_OSRELDATE] != NULL) osreldate = aux_info[AT_OSRELDATE]->a_un.a_val; digest_dynamic2(&obj_rtld, dyn_rpath, dyn_soname, dyn_runpath); /* Replace the path with a dynamically allocated copy. */ obj_rtld.path = xstrdup(PATH_RTLD); r_debug.r_brk = r_debug_state; r_debug.r_state = RT_CONSISTENT; } /* * Retrieve the array of supported page sizes. The kernel provides the page * sizes in increasing order. */ static void init_pagesizes(Elf_Auxinfo **aux_info) { static size_t psa[MAXPAGESIZES]; int mib[2]; size_t len, size; if (aux_info[AT_PAGESIZES] != NULL && aux_info[AT_PAGESIZESLEN] != NULL) { size = aux_info[AT_PAGESIZESLEN]->a_un.a_val; pagesizes = aux_info[AT_PAGESIZES]->a_un.a_ptr; } else { len = 2; if (sysctlnametomib("hw.pagesizes", mib, &len) == 0) size = sizeof(psa); else { /* As a fallback, retrieve the base page size. */ size = sizeof(psa[0]); if (aux_info[AT_PAGESZ] != NULL) { psa[0] = aux_info[AT_PAGESZ]->a_un.a_val; goto psa_filled; } else { mib[0] = CTL_HW; mib[1] = HW_PAGESIZE; len = 2; } } if (sysctl(mib, len, psa, &size, NULL, 0) == -1) { _rtld_error("sysctl for hw.pagesize(s) failed"); rtld_die(); } psa_filled: pagesizes = psa; } npagesizes = size / sizeof(pagesizes[0]); /* Discard any invalid entries at the end of the array. */ while (npagesizes > 0 && pagesizes[npagesizes - 1] == 0) npagesizes--; } /* * Add the init functions from a needed object list (and its recursive * needed objects) to "list". This is not used directly; it is a helper * function for initlist_add_objects(). The write lock must be held * when this function is called. */ static void initlist_add_neededs(Needed_Entry *needed, Objlist *list) { /* Recursively process the successor needed objects. */ if (needed->next != NULL) initlist_add_neededs(needed->next, list); /* Process the current needed object. */ if (needed->obj != NULL) initlist_add_objects(needed->obj, &needed->obj->next, list); } /* * Scan all of the DAGs rooted in the range of objects from "obj" to * "tail" and add their init functions to "list". This recurses over * the DAGs and ensure the proper init ordering such that each object's * needed libraries are initialized before the object itself. At the * same time, this function adds the objects to the global finalization * list "list_fini" in the opposite order. The write lock must be * held when this function is called. */ static void initlist_add_objects(Obj_Entry *obj, Obj_Entry **tail, Objlist *list) { if (obj->init_scanned || obj->init_done) return; obj->init_scanned = true; /* Recursively process the successor objects. */ if (&obj->next != tail) initlist_add_objects(obj->next, tail, list); /* Recursively process the needed objects. */ if (obj->needed != NULL) initlist_add_neededs(obj->needed, list); if (obj->needed_filtees != NULL) initlist_add_neededs(obj->needed_filtees, list); if (obj->needed_aux_filtees != NULL) initlist_add_neededs(obj->needed_aux_filtees, list); /* Add the object to the init list. */ if (obj->preinit_array != (Elf_Addr)NULL || obj->init != (Elf_Addr)NULL || obj->init_array != (Elf_Addr)NULL) objlist_push_tail(list, obj); /* Add the object to the global fini list in the reverse order. */ if ((obj->fini != (Elf_Addr)NULL || obj->fini_array != (Elf_Addr)NULL) && !obj->on_fini_list) { objlist_push_head(&list_fini, obj); obj->on_fini_list = true; } } #ifndef FPTR_TARGET #define FPTR_TARGET(f) ((Elf_Addr) (f)) #endif static void free_needed_filtees(Needed_Entry *n) { Needed_Entry *needed, *needed1; for (needed = n; needed != NULL; needed = needed->next) { if (needed->obj != NULL) { dlclose(needed->obj); needed->obj = NULL; } } for (needed = n; needed != NULL; needed = needed1) { needed1 = needed->next; free(needed); } } static void unload_filtees(Obj_Entry *obj) { free_needed_filtees(obj->needed_filtees); obj->needed_filtees = NULL; free_needed_filtees(obj->needed_aux_filtees); obj->needed_aux_filtees = NULL; obj->filtees_loaded = false; } static void load_filtee1(Obj_Entry *obj, Needed_Entry *needed, int flags, RtldLockState *lockstate) { for (; needed != NULL; needed = needed->next) { needed->obj = dlopen_object(obj->strtab + needed->name, -1, obj, flags, ((ld_loadfltr || obj->z_loadfltr) ? RTLD_NOW : RTLD_LAZY) | RTLD_LOCAL, lockstate); } } static void load_filtees(Obj_Entry *obj, int flags, RtldLockState *lockstate) { lock_restart_for_upgrade(lockstate); if (!obj->filtees_loaded) { load_filtee1(obj, obj->needed_filtees, flags, lockstate); load_filtee1(obj, obj->needed_aux_filtees, flags, lockstate); obj->filtees_loaded = true; } } static int process_needed(Obj_Entry *obj, Needed_Entry *needed, int flags) { Obj_Entry *obj1; for (; needed != NULL; needed = needed->next) { obj1 = needed->obj = load_object(obj->strtab + needed->name, -1, obj, flags & ~RTLD_LO_NOLOAD); if (obj1 == NULL && !ld_tracing && (flags & RTLD_LO_FILTEES) == 0) return (-1); } return (0); } /* * Given a shared object, traverse its list of needed objects, and load * each of them. Returns 0 on success. Generates an error message and * returns -1 on failure. */ static int load_needed_objects(Obj_Entry *first, int flags) { Obj_Entry *obj; for (obj = first; obj != NULL; obj = obj->next) { if (process_needed(obj, obj->needed, flags) == -1) return (-1); } return (0); } static int load_preload_objects(void) { char *p = ld_preload; Obj_Entry *obj; static const char delim[] = " \t:;"; if (p == NULL) return 0; p += strspn(p, delim); while (*p != '\0') { size_t len = strcspn(p, delim); char savech; savech = p[len]; p[len] = '\0'; obj = load_object(p, -1, NULL, 0); if (obj == NULL) return -1; /* XXX - cleanup */ obj->z_interpose = true; p[len] = savech; p += len; p += strspn(p, delim); } LD_UTRACE(UTRACE_PRELOAD_FINISHED, NULL, NULL, 0, 0, NULL); return 0; } static const char * printable_path(const char *path) { return (path == NULL ? "" : path); } /* * Load a shared object into memory, if it is not already loaded. The * object may be specified by name or by user-supplied file descriptor * fd_u. In the later case, the fd_u descriptor is not closed, but its * duplicate is. * * Returns a pointer to the Obj_Entry for the object. Returns NULL * on failure. */ static Obj_Entry * load_object(const char *name, int fd_u, const Obj_Entry *refobj, int flags) { Obj_Entry *obj; int fd; struct stat sb; char *path; fd = -1; if (name != NULL) { for (obj = obj_list->next; obj != NULL; obj = obj->next) { if (object_match_name(obj, name)) return (obj); } path = find_library(name, refobj, &fd); if (path == NULL) return (NULL); } else path = NULL; if (fd >= 0) { /* * search_library_pathfds() opens a fresh file descriptor for the * library, so there is no need to dup(). */ } else if (fd_u == -1) { /* * If we didn't find a match by pathname, or the name is not * supplied, open the file and check again by device and inode. * This avoids false mismatches caused by multiple links or ".." * in pathnames. * * To avoid a race, we open the file and use fstat() rather than * using stat(). */ - if ((fd = open(path, O_RDONLY | O_CLOEXEC)) == -1) { + if ((fd = open(path, O_RDONLY | O_CLOEXEC | O_VERIFY)) == -1) { _rtld_error("Cannot open \"%s\"", path); free(path); return (NULL); } } else { fd = fcntl(fd_u, F_DUPFD_CLOEXEC, 0); if (fd == -1) { _rtld_error("Cannot dup fd"); free(path); return (NULL); } } if (fstat(fd, &sb) == -1) { _rtld_error("Cannot fstat \"%s\"", printable_path(path)); close(fd); free(path); return NULL; } for (obj = obj_list->next; obj != NULL; obj = obj->next) if (obj->ino == sb.st_ino && obj->dev == sb.st_dev) break; if (obj != NULL && name != NULL) { object_add_name(obj, name); free(path); close(fd); return obj; } if (flags & RTLD_LO_NOLOAD) { free(path); close(fd); return (NULL); } /* First use of this object, so we must map it in */ obj = do_load_object(fd, name, path, &sb, flags); if (obj == NULL) free(path); close(fd); return obj; } static Obj_Entry * do_load_object(int fd, const char *name, char *path, struct stat *sbp, int flags) { Obj_Entry *obj; struct statfs fs; /* * but first, make sure that environment variables haven't been * used to circumvent the noexec flag on a filesystem. */ if (dangerous_ld_env) { if (fstatfs(fd, &fs) != 0) { _rtld_error("Cannot fstatfs \"%s\"", printable_path(path)); return NULL; } if (fs.f_flags & MNT_NOEXEC) { _rtld_error("Cannot execute objects on %s\n", fs.f_mntonname); return NULL; } } dbg("loading \"%s\"", printable_path(path)); obj = map_object(fd, printable_path(path), sbp); if (obj == NULL) return NULL; /* * If DT_SONAME is present in the object, digest_dynamic2 already * added it to the object names. */ if (name != NULL) object_add_name(obj, name); obj->path = path; digest_dynamic(obj, 0); dbg("%s valid_hash_sysv %d valid_hash_gnu %d dynsymcount %d", obj->path, obj->valid_hash_sysv, obj->valid_hash_gnu, obj->dynsymcount); if (obj->z_noopen && (flags & (RTLD_LO_DLOPEN | RTLD_LO_TRACE)) == RTLD_LO_DLOPEN) { dbg("refusing to load non-loadable \"%s\"", obj->path); _rtld_error("Cannot dlopen non-loadable %s", obj->path); munmap(obj->mapbase, obj->mapsize); obj_free(obj); return (NULL); } obj->dlopened = (flags & RTLD_LO_DLOPEN) != 0; *obj_tail = obj; obj_tail = &obj->next; obj_count++; obj_loads++; linkmap_add(obj); /* for GDB & dlinfo() */ max_stack_flags |= obj->stack_flags; dbg(" %p .. %p: %s", obj->mapbase, obj->mapbase + obj->mapsize - 1, obj->path); if (obj->textrel) dbg(" WARNING: %s has impure text", obj->path); LD_UTRACE(UTRACE_LOAD_OBJECT, obj, obj->mapbase, obj->mapsize, 0, obj->path); return obj; } static Obj_Entry * obj_from_addr(const void *addr) { Obj_Entry *obj; for (obj = obj_list; obj != NULL; obj = obj->next) { if (addr < (void *) obj->mapbase) continue; if (addr < (void *) (obj->mapbase + obj->mapsize)) return obj; } return NULL; } static void preinit_main(void) { Elf_Addr *preinit_addr; int index; preinit_addr = (Elf_Addr *)obj_main->preinit_array; if (preinit_addr == NULL) return; for (index = 0; index < obj_main->preinit_array_num; index++) { if (preinit_addr[index] != 0 && preinit_addr[index] != 1) { dbg("calling preinit function for %s at %p", obj_main->path, (void *)preinit_addr[index]); LD_UTRACE(UTRACE_INIT_CALL, obj_main, (void *)preinit_addr[index], 0, 0, obj_main->path); call_init_pointer(obj_main, preinit_addr[index]); } } } /* * Call the finalization functions for each of the objects in "list" * belonging to the DAG of "root" and referenced once. If NULL "root" * is specified, every finalization function will be called regardless * of the reference count and the list elements won't be freed. All of * the objects are expected to have non-NULL fini functions. */ static void objlist_call_fini(Objlist *list, Obj_Entry *root, RtldLockState *lockstate) { Objlist_Entry *elm; char *saved_msg; Elf_Addr *fini_addr; int index; assert(root == NULL || root->refcount == 1); /* * Preserve the current error message since a fini function might * call into the dynamic linker and overwrite it. */ saved_msg = errmsg_save(); do { STAILQ_FOREACH(elm, list, link) { if (root != NULL && (elm->obj->refcount != 1 || objlist_find(&root->dagmembers, elm->obj) == NULL)) continue; /* Remove object from fini list to prevent recursive invocation. */ STAILQ_REMOVE(list, elm, Struct_Objlist_Entry, link); /* * XXX: If a dlopen() call references an object while the * fini function is in progress, we might end up trying to * unload the referenced object in dlclose() or the object * won't be unloaded although its fini function has been * called. */ lock_release(rtld_bind_lock, lockstate); /* * It is legal to have both DT_FINI and DT_FINI_ARRAY defined. * When this happens, DT_FINI_ARRAY is processed first. */ fini_addr = (Elf_Addr *)elm->obj->fini_array; if (fini_addr != NULL && elm->obj->fini_array_num > 0) { for (index = elm->obj->fini_array_num - 1; index >= 0; index--) { if (fini_addr[index] != 0 && fini_addr[index] != 1) { dbg("calling fini function for %s at %p", elm->obj->path, (void *)fini_addr[index]); LD_UTRACE(UTRACE_FINI_CALL, elm->obj, (void *)fini_addr[index], 0, 0, elm->obj->path); call_initfini_pointer(elm->obj, fini_addr[index]); } } } if (elm->obj->fini != (Elf_Addr)NULL) { dbg("calling fini function for %s at %p", elm->obj->path, (void *)elm->obj->fini); LD_UTRACE(UTRACE_FINI_CALL, elm->obj, (void *)elm->obj->fini, 0, 0, elm->obj->path); call_initfini_pointer(elm->obj, elm->obj->fini); } wlock_acquire(rtld_bind_lock, lockstate); /* No need to free anything if process is going down. */ if (root != NULL) free(elm); /* * We must restart the list traversal after every fini call * because a dlclose() call from the fini function or from * another thread might have modified the reference counts. */ break; } } while (elm != NULL); errmsg_restore(saved_msg); } /* * Call the initialization functions for each of the objects in * "list". All of the objects are expected to have non-NULL init * functions. */ static void objlist_call_init(Objlist *list, RtldLockState *lockstate) { Objlist_Entry *elm; Obj_Entry *obj; char *saved_msg; Elf_Addr *init_addr; int index; /* * Clean init_scanned flag so that objects can be rechecked and * possibly initialized earlier if any of vectors called below * cause the change by using dlopen. */ for (obj = obj_list; obj != NULL; obj = obj->next) obj->init_scanned = false; /* * Preserve the current error message since an init function might * call into the dynamic linker and overwrite it. */ saved_msg = errmsg_save(); STAILQ_FOREACH(elm, list, link) { if (elm->obj->init_done) /* Initialized early. */ continue; /* * Race: other thread might try to use this object before current * one completes the initilization. Not much can be done here * without better locking. */ elm->obj->init_done = true; lock_release(rtld_bind_lock, lockstate); /* * It is legal to have both DT_INIT and DT_INIT_ARRAY defined. * When this happens, DT_INIT is processed first. */ if (elm->obj->init != (Elf_Addr)NULL) { dbg("calling init function for %s at %p", elm->obj->path, (void *)elm->obj->init); LD_UTRACE(UTRACE_INIT_CALL, elm->obj, (void *)elm->obj->init, 0, 0, elm->obj->path); call_initfini_pointer(elm->obj, elm->obj->init); } init_addr = (Elf_Addr *)elm->obj->init_array; if (init_addr != NULL) { for (index = 0; index < elm->obj->init_array_num; index++) { if (init_addr[index] != 0 && init_addr[index] != 1) { dbg("calling init function for %s at %p", elm->obj->path, (void *)init_addr[index]); LD_UTRACE(UTRACE_INIT_CALL, elm->obj, (void *)init_addr[index], 0, 0, elm->obj->path); call_init_pointer(elm->obj, init_addr[index]); } } } wlock_acquire(rtld_bind_lock, lockstate); } errmsg_restore(saved_msg); } static void objlist_clear(Objlist *list) { Objlist_Entry *elm; while (!STAILQ_EMPTY(list)) { elm = STAILQ_FIRST(list); STAILQ_REMOVE_HEAD(list, link); free(elm); } } static Objlist_Entry * objlist_find(Objlist *list, const Obj_Entry *obj) { Objlist_Entry *elm; STAILQ_FOREACH(elm, list, link) if (elm->obj == obj) return elm; return NULL; } static void objlist_init(Objlist *list) { STAILQ_INIT(list); } static void objlist_push_head(Objlist *list, Obj_Entry *obj) { Objlist_Entry *elm; elm = NEW(Objlist_Entry); elm->obj = obj; STAILQ_INSERT_HEAD(list, elm, link); } static void objlist_push_tail(Objlist *list, Obj_Entry *obj) { Objlist_Entry *elm; elm = NEW(Objlist_Entry); elm->obj = obj; STAILQ_INSERT_TAIL(list, elm, link); } static void objlist_put_after(Objlist *list, Obj_Entry *listobj, Obj_Entry *obj) { Objlist_Entry *elm, *listelm; STAILQ_FOREACH(listelm, list, link) { if (listelm->obj == listobj) break; } elm = NEW(Objlist_Entry); elm->obj = obj; if (listelm != NULL) STAILQ_INSERT_AFTER(list, listelm, elm, link); else STAILQ_INSERT_TAIL(list, elm, link); } static void objlist_remove(Objlist *list, Obj_Entry *obj) { Objlist_Entry *elm; if ((elm = objlist_find(list, obj)) != NULL) { STAILQ_REMOVE(list, elm, Struct_Objlist_Entry, link); free(elm); } } /* * Relocate dag rooted in the specified object. * Returns 0 on success, or -1 on failure. */ static int relocate_object_dag(Obj_Entry *root, bool bind_now, Obj_Entry *rtldobj, int flags, RtldLockState *lockstate) { Objlist_Entry *elm; int error; error = 0; STAILQ_FOREACH(elm, &root->dagmembers, link) { error = relocate_object(elm->obj, bind_now, rtldobj, flags, lockstate); if (error == -1) break; } return (error); } /* * Relocate single object. * Returns 0 on success, or -1 on failure. */ static int relocate_object(Obj_Entry *obj, bool bind_now, Obj_Entry *rtldobj, int flags, RtldLockState *lockstate) { if (obj->relocated) return (0); obj->relocated = true; if (obj != rtldobj) dbg("relocating \"%s\"", obj->path); if (obj->symtab == NULL || obj->strtab == NULL || !(obj->valid_hash_sysv || obj->valid_hash_gnu)) { _rtld_error("%s: Shared object has no run-time symbol table", obj->path); return (-1); } if (obj->textrel) { /* There are relocations to the write-protected text segment. */ if (mprotect(obj->mapbase, obj->textsize, PROT_READ|PROT_WRITE|PROT_EXEC) == -1) { _rtld_error("%s: Cannot write-enable text segment: %s", obj->path, rtld_strerror(errno)); return (-1); } } /* Process the non-PLT non-IFUNC relocations. */ if (reloc_non_plt(obj, rtldobj, flags, lockstate)) return (-1); if (obj->textrel) { /* Re-protected the text segment. */ if (mprotect(obj->mapbase, obj->textsize, PROT_READ|PROT_EXEC) == -1) { _rtld_error("%s: Cannot write-protect text segment: %s", obj->path, rtld_strerror(errno)); return (-1); } } /* Set the special PLT or GOT entries. */ init_pltgot(obj); /* Process the PLT relocations. */ if (reloc_plt(obj) == -1) return (-1); /* Relocate the jump slots if we are doing immediate binding. */ if (obj->bind_now || bind_now) if (reloc_jmpslots(obj, flags, lockstate) == -1) return (-1); /* * Process the non-PLT IFUNC relocations. The relocations are * processed in two phases, because IFUNC resolvers may * reference other symbols, which must be readily processed * before resolvers are called. */ if (obj->non_plt_gnu_ifunc && reloc_non_plt(obj, rtldobj, flags | SYMLOOK_IFUNC, lockstate)) return (-1); if (obj->relro_size > 0) { if (mprotect(obj->relro_page, obj->relro_size, PROT_READ) == -1) { _rtld_error("%s: Cannot enforce relro protection: %s", obj->path, rtld_strerror(errno)); return (-1); } } /* * Set up the magic number and version in the Obj_Entry. These * were checked in the crt1.o from the original ElfKit, so we * set them for backward compatibility. */ obj->magic = RTLD_MAGIC; obj->version = RTLD_VERSION; return (0); } /* * Relocate newly-loaded shared objects. The argument is a pointer to * the Obj_Entry for the first such object. All objects from the first * to the end of the list of objects are relocated. Returns 0 on success, * or -1 on failure. */ static int relocate_objects(Obj_Entry *first, bool bind_now, Obj_Entry *rtldobj, int flags, RtldLockState *lockstate) { Obj_Entry *obj; int error; for (error = 0, obj = first; obj != NULL; obj = obj->next) { error = relocate_object(obj, bind_now, rtldobj, flags, lockstate); if (error == -1) break; } return (error); } /* * The handling of R_MACHINE_IRELATIVE relocations and jumpslots * referencing STT_GNU_IFUNC symbols is postponed till the other * relocations are done. The indirect functions specified as * ifunc are allowed to call other symbols, so we need to have * objects relocated before asking for resolution from indirects. * * The R_MACHINE_IRELATIVE slots are resolved in greedy fashion, * instead of the usual lazy handling of PLT slots. It is * consistent with how GNU does it. */ static int resolve_object_ifunc(Obj_Entry *obj, bool bind_now, int flags, RtldLockState *lockstate) { if (obj->irelative && reloc_iresolve(obj, lockstate) == -1) return (-1); if ((obj->bind_now || bind_now) && obj->gnu_ifunc && reloc_gnu_ifunc(obj, flags, lockstate) == -1) return (-1); return (0); } static int resolve_objects_ifunc(Obj_Entry *first, bool bind_now, int flags, RtldLockState *lockstate) { Obj_Entry *obj; for (obj = first; obj != NULL; obj = obj->next) { if (resolve_object_ifunc(obj, bind_now, flags, lockstate) == -1) return (-1); } return (0); } static int initlist_objects_ifunc(Objlist *list, bool bind_now, int flags, RtldLockState *lockstate) { Objlist_Entry *elm; STAILQ_FOREACH(elm, list, link) { if (resolve_object_ifunc(elm->obj, bind_now, flags, lockstate) == -1) return (-1); } return (0); } /* * Cleanup procedure. It will be called (by the atexit mechanism) just * before the process exits. */ static void rtld_exit(void) { RtldLockState lockstate; wlock_acquire(rtld_bind_lock, &lockstate); dbg("rtld_exit()"); objlist_call_fini(&list_fini, NULL, &lockstate); /* No need to remove the items from the list, since we are exiting. */ if (!libmap_disable) lm_fini(); lock_release(rtld_bind_lock, &lockstate); } /* * Iterate over a search path, translate each element, and invoke the * callback on the result. */ static void * path_enumerate(const char *path, path_enum_proc callback, void *arg) { const char *trans; if (path == NULL) return (NULL); path += strspn(path, ":;"); while (*path != '\0') { size_t len; char *res; len = strcspn(path, ":;"); trans = lm_findn(NULL, path, len); if (trans) res = callback(trans, strlen(trans), arg); else res = callback(path, len, arg); if (res != NULL) return (res); path += len; path += strspn(path, ":;"); } return (NULL); } struct try_library_args { const char *name; size_t namelen; char *buffer; size_t buflen; }; static void * try_library_path(const char *dir, size_t dirlen, void *param) { struct try_library_args *arg; arg = param; if (*dir == '/' || trust) { char *pathname; if (dirlen + 1 + arg->namelen + 1 > arg->buflen) return (NULL); pathname = arg->buffer; strncpy(pathname, dir, dirlen); pathname[dirlen] = '/'; strcpy(pathname + dirlen + 1, arg->name); dbg(" Trying \"%s\"", pathname); if (access(pathname, F_OK) == 0) { /* We found it */ pathname = xmalloc(dirlen + 1 + arg->namelen + 1); strcpy(pathname, arg->buffer); return (pathname); } } return (NULL); } static char * search_library_path(const char *name, const char *path) { char *p; struct try_library_args arg; if (path == NULL) return NULL; arg.name = name; arg.namelen = strlen(name); arg.buffer = xmalloc(PATH_MAX); arg.buflen = PATH_MAX; p = path_enumerate(path, try_library_path, &arg); free(arg.buffer); return (p); } /* * Finds the library with the given name using the directory descriptors * listed in the LD_LIBRARY_PATH_FDS environment variable. * * Returns a freshly-opened close-on-exec file descriptor for the library, * or -1 if the library cannot be found. */ static char * search_library_pathfds(const char *name, const char *path, int *fdp) { char *envcopy, *fdstr, *found, *last_token; size_t len; int dirfd, fd; dbg("%s('%s', '%s', fdp)", __func__, name, path); /* Don't load from user-specified libdirs into setuid binaries. */ if (!trust) return (NULL); /* We can't do anything if LD_LIBRARY_PATH_FDS isn't set. */ if (path == NULL) return (NULL); /* LD_LIBRARY_PATH_FDS only works with relative paths. */ if (name[0] == '/') { dbg("Absolute path (%s) passed to %s", name, __func__); return (NULL); } /* * Use strtok_r() to walk the FD:FD:FD list. This requires a local * copy of the path, as strtok_r rewrites separator tokens * with '\0'. */ found = NULL; envcopy = xstrdup(path); for (fdstr = strtok_r(envcopy, ":", &last_token); fdstr != NULL; fdstr = strtok_r(NULL, ":", &last_token)) { dirfd = parse_libdir(fdstr); if (dirfd < 0) break; - fd = __sys_openat(dirfd, name, O_RDONLY | O_CLOEXEC); + fd = __sys_openat(dirfd, name, O_RDONLY | O_CLOEXEC | O_VERIFY); if (fd >= 0) { *fdp = fd; len = strlen(fdstr) + strlen(name) + 3; found = xmalloc(len); if (rtld_snprintf(found, len, "#%d/%s", dirfd, name) < 0) { _rtld_error("error generating '%d/%s'", dirfd, name); rtld_die(); } dbg("open('%s') => %d", found, fd); break; } } free(envcopy); return (found); } int dlclose(void *handle) { Obj_Entry *root; RtldLockState lockstate; wlock_acquire(rtld_bind_lock, &lockstate); root = dlcheck(handle); if (root == NULL) { lock_release(rtld_bind_lock, &lockstate); return -1; } LD_UTRACE(UTRACE_DLCLOSE_START, handle, NULL, 0, root->dl_refcount, root->path); /* Unreference the object and its dependencies. */ root->dl_refcount--; if (root->refcount == 1) { /* * The object will be no longer referenced, so we must unload it. * First, call the fini functions. */ objlist_call_fini(&list_fini, root, &lockstate); unref_dag(root); /* Finish cleaning up the newly-unreferenced objects. */ GDB_STATE(RT_DELETE,&root->linkmap); unload_object(root); GDB_STATE(RT_CONSISTENT,NULL); } else unref_dag(root); LD_UTRACE(UTRACE_DLCLOSE_STOP, handle, NULL, 0, 0, NULL); lock_release(rtld_bind_lock, &lockstate); return 0; } char * dlerror(void) { char *msg = error_message; error_message = NULL; return msg; } /* * This function is deprecated and has no effect. */ void dllockinit(void *context, void *(*lock_create)(void *context), void (*rlock_acquire)(void *lock), void (*wlock_acquire)(void *lock), void (*lock_release)(void *lock), void (*lock_destroy)(void *lock), void (*context_destroy)(void *context)) { static void *cur_context; static void (*cur_context_destroy)(void *); /* Just destroy the context from the previous call, if necessary. */ if (cur_context_destroy != NULL) cur_context_destroy(cur_context); cur_context = context; cur_context_destroy = context_destroy; } void * dlopen(const char *name, int mode) { return (rtld_dlopen(name, -1, mode)); } void * fdlopen(int fd, int mode) { return (rtld_dlopen(NULL, fd, mode)); } static void * rtld_dlopen(const char *name, int fd, int mode) { RtldLockState lockstate; int lo_flags; LD_UTRACE(UTRACE_DLOPEN_START, NULL, NULL, 0, mode, name); ld_tracing = (mode & RTLD_TRACE) == 0 ? NULL : "1"; if (ld_tracing != NULL) { rlock_acquire(rtld_bind_lock, &lockstate); if (sigsetjmp(lockstate.env, 0) != 0) lock_upgrade(rtld_bind_lock, &lockstate); environ = (char **)*get_program_var_addr("environ", &lockstate); lock_release(rtld_bind_lock, &lockstate); } lo_flags = RTLD_LO_DLOPEN; if (mode & RTLD_NODELETE) lo_flags |= RTLD_LO_NODELETE; if (mode & RTLD_NOLOAD) lo_flags |= RTLD_LO_NOLOAD; if (ld_tracing != NULL) lo_flags |= RTLD_LO_TRACE; return (dlopen_object(name, fd, obj_main, lo_flags, mode & (RTLD_MODEMASK | RTLD_GLOBAL), NULL)); } static void dlopen_cleanup(Obj_Entry *obj) { obj->dl_refcount--; unref_dag(obj); if (obj->refcount == 0) unload_object(obj); } static Obj_Entry * dlopen_object(const char *name, int fd, Obj_Entry *refobj, int lo_flags, int mode, RtldLockState *lockstate) { Obj_Entry **old_obj_tail; Obj_Entry *obj; Objlist initlist; RtldLockState mlockstate; int result; objlist_init(&initlist); if (lockstate == NULL && !(lo_flags & RTLD_LO_EARLY)) { wlock_acquire(rtld_bind_lock, &mlockstate); lockstate = &mlockstate; } GDB_STATE(RT_ADD,NULL); old_obj_tail = obj_tail; obj = NULL; if (name == NULL && fd == -1) { obj = obj_main; obj->refcount++; } else { obj = load_object(name, fd, refobj, lo_flags); } if (obj) { obj->dl_refcount++; if (mode & RTLD_GLOBAL && objlist_find(&list_global, obj) == NULL) objlist_push_tail(&list_global, obj); if (*old_obj_tail != NULL) { /* We loaded something new. */ assert(*old_obj_tail == obj); result = load_needed_objects(obj, lo_flags & (RTLD_LO_DLOPEN | RTLD_LO_EARLY)); init_dag(obj); ref_dag(obj); if (result != -1) result = rtld_verify_versions(&obj->dagmembers); if (result != -1 && ld_tracing) goto trace; if (result == -1 || relocate_object_dag(obj, (mode & RTLD_MODEMASK) == RTLD_NOW, &obj_rtld, (lo_flags & RTLD_LO_EARLY) ? SYMLOOK_EARLY : 0, lockstate) == -1) { dlopen_cleanup(obj); obj = NULL; } else if (lo_flags & RTLD_LO_EARLY) { /* * Do not call the init functions for early loaded * filtees. The image is still not initialized enough * for them to work. * * Our object is found by the global object list and * will be ordered among all init calls done right * before transferring control to main. */ } else { /* Make list of init functions to call. */ initlist_add_objects(obj, &obj->next, &initlist); } /* * Process all no_delete or global objects here, given * them own DAGs to prevent their dependencies from being * unloaded. This has to be done after we have loaded all * of the dependencies, so that we do not miss any. */ if (obj != NULL) process_z(obj); } else { /* * Bump the reference counts for objects on this DAG. If * this is the first dlopen() call for the object that was * already loaded as a dependency, initialize the dag * starting at it. */ init_dag(obj); ref_dag(obj); if ((lo_flags & RTLD_LO_TRACE) != 0) goto trace; } if (obj != NULL && ((lo_flags & RTLD_LO_NODELETE) != 0 || obj->z_nodelete) && !obj->ref_nodel) { dbg("obj %s nodelete", obj->path); ref_dag(obj); obj->z_nodelete = obj->ref_nodel = true; } } LD_UTRACE(UTRACE_DLOPEN_STOP, obj, NULL, 0, obj ? obj->dl_refcount : 0, name); GDB_STATE(RT_CONSISTENT,obj ? &obj->linkmap : NULL); if (!(lo_flags & RTLD_LO_EARLY)) { map_stacks_exec(lockstate); } if (initlist_objects_ifunc(&initlist, (mode & RTLD_MODEMASK) == RTLD_NOW, (lo_flags & RTLD_LO_EARLY) ? SYMLOOK_EARLY : 0, lockstate) == -1) { objlist_clear(&initlist); dlopen_cleanup(obj); if (lockstate == &mlockstate) lock_release(rtld_bind_lock, lockstate); return (NULL); } if (!(lo_flags & RTLD_LO_EARLY)) { /* Call the init functions. */ objlist_call_init(&initlist, lockstate); } objlist_clear(&initlist); if (lockstate == &mlockstate) lock_release(rtld_bind_lock, lockstate); return obj; trace: trace_loaded_objects(obj); if (lockstate == &mlockstate) lock_release(rtld_bind_lock, lockstate); exit(0); } static void * do_dlsym(void *handle, const char *name, void *retaddr, const Ver_Entry *ve, int flags) { DoneList donelist; const Obj_Entry *obj, *defobj; const Elf_Sym *def; SymLook req; RtldLockState lockstate; tls_index ti; void *sym; int res; def = NULL; defobj = NULL; symlook_init(&req, name); req.ventry = ve; req.flags = flags | SYMLOOK_IN_PLT; req.lockstate = &lockstate; LD_UTRACE(UTRACE_DLSYM_START, handle, NULL, 0, 0, name); rlock_acquire(rtld_bind_lock, &lockstate); if (sigsetjmp(lockstate.env, 0) != 0) lock_upgrade(rtld_bind_lock, &lockstate); if (handle == NULL || handle == RTLD_NEXT || handle == RTLD_DEFAULT || handle == RTLD_SELF) { if ((obj = obj_from_addr(retaddr)) == NULL) { _rtld_error("Cannot determine caller's shared object"); lock_release(rtld_bind_lock, &lockstate); LD_UTRACE(UTRACE_DLSYM_STOP, handle, NULL, 0, 0, name); return NULL; } if (handle == NULL) { /* Just the caller's shared object. */ res = symlook_obj(&req, obj); if (res == 0) { def = req.sym_out; defobj = req.defobj_out; } } else if (handle == RTLD_NEXT || /* Objects after caller's */ handle == RTLD_SELF) { /* ... caller included */ if (handle == RTLD_NEXT) obj = obj->next; for (; obj != NULL; obj = obj->next) { res = symlook_obj(&req, obj); if (res == 0) { if (def == NULL || ELF_ST_BIND(req.sym_out->st_info) != STB_WEAK) { def = req.sym_out; defobj = req.defobj_out; if (ELF_ST_BIND(def->st_info) != STB_WEAK) break; } } } /* * Search the dynamic linker itself, and possibly resolve the * symbol from there. This is how the application links to * dynamic linker services such as dlopen. */ if (def == NULL || ELF_ST_BIND(def->st_info) == STB_WEAK) { res = symlook_obj(&req, &obj_rtld); if (res == 0) { def = req.sym_out; defobj = req.defobj_out; } } } else { assert(handle == RTLD_DEFAULT); res = symlook_default(&req, obj); if (res == 0) { defobj = req.defobj_out; def = req.sym_out; } } } else { if ((obj = dlcheck(handle)) == NULL) { lock_release(rtld_bind_lock, &lockstate); LD_UTRACE(UTRACE_DLSYM_STOP, handle, NULL, 0, 0, name); return NULL; } donelist_init(&donelist); if (obj->mainprog) { /* Handle obtained by dlopen(NULL, ...) implies global scope. */ res = symlook_global(&req, &donelist); if (res == 0) { def = req.sym_out; defobj = req.defobj_out; } /* * Search the dynamic linker itself, and possibly resolve the * symbol from there. This is how the application links to * dynamic linker services such as dlopen. */ if (def == NULL || ELF_ST_BIND(def->st_info) == STB_WEAK) { res = symlook_obj(&req, &obj_rtld); if (res == 0) { def = req.sym_out; defobj = req.defobj_out; } } } else { /* Search the whole DAG rooted at the given object. */ res = symlook_list(&req, &obj->dagmembers, &donelist); if (res == 0) { def = req.sym_out; defobj = req.defobj_out; } } } if (def != NULL) { lock_release(rtld_bind_lock, &lockstate); /* * The value required by the caller is derived from the value * of the symbol. this is simply the relocated value of the * symbol. */ if (ELF_ST_TYPE(def->st_info) == STT_FUNC) sym = make_function_pointer(def, defobj); else if (ELF_ST_TYPE(def->st_info) == STT_GNU_IFUNC) sym = rtld_resolve_ifunc(defobj, def); else if (ELF_ST_TYPE(def->st_info) == STT_TLS) { ti.ti_module = defobj->tlsindex; ti.ti_offset = def->st_value; sym = __tls_get_addr(&ti); } else sym = defobj->relocbase + def->st_value; LD_UTRACE(UTRACE_DLSYM_STOP, handle, sym, 0, 0, name); return (sym); } _rtld_error("Undefined symbol \"%s\"", name); lock_release(rtld_bind_lock, &lockstate); LD_UTRACE(UTRACE_DLSYM_STOP, handle, NULL, 0, 0, name); return NULL; } void * dlsym(void *handle, const char *name) { return do_dlsym(handle, name, __builtin_return_address(0), NULL, SYMLOOK_DLSYM); } dlfunc_t dlfunc(void *handle, const char *name) { union { void *d; dlfunc_t f; } rv; rv.d = do_dlsym(handle, name, __builtin_return_address(0), NULL, SYMLOOK_DLSYM); return (rv.f); } void * dlvsym(void *handle, const char *name, const char *version) { Ver_Entry ventry; ventry.name = version; ventry.file = NULL; ventry.hash = elf_hash(version); ventry.flags= 0; return do_dlsym(handle, name, __builtin_return_address(0), &ventry, SYMLOOK_DLSYM); } int _rtld_addr_phdr(const void *addr, struct dl_phdr_info *phdr_info) { const Obj_Entry *obj; RtldLockState lockstate; rlock_acquire(rtld_bind_lock, &lockstate); obj = obj_from_addr(addr); if (obj == NULL) { _rtld_error("No shared object contains address"); lock_release(rtld_bind_lock, &lockstate); return (0); } rtld_fill_dl_phdr_info(obj, phdr_info); lock_release(rtld_bind_lock, &lockstate); return (1); } int dladdr(const void *addr, Dl_info *info) { const Obj_Entry *obj; const Elf_Sym *def; void *symbol_addr; unsigned long symoffset; RtldLockState lockstate; rlock_acquire(rtld_bind_lock, &lockstate); obj = obj_from_addr(addr); if (obj == NULL) { _rtld_error("No shared object contains address"); lock_release(rtld_bind_lock, &lockstate); return 0; } info->dli_fname = obj->path; info->dli_fbase = obj->mapbase; info->dli_saddr = (void *)0; info->dli_sname = NULL; /* * Walk the symbol list looking for the symbol whose address is * closest to the address sent in. */ for (symoffset = 0; symoffset < obj->dynsymcount; symoffset++) { def = obj->symtab + symoffset; /* * For skip the symbol if st_shndx is either SHN_UNDEF or * SHN_COMMON. */ if (def->st_shndx == SHN_UNDEF || def->st_shndx == SHN_COMMON) continue; /* * If the symbol is greater than the specified address, or if it * is further away from addr than the current nearest symbol, * then reject it. */ symbol_addr = obj->relocbase + def->st_value; if (symbol_addr > addr || symbol_addr < info->dli_saddr) continue; /* Update our idea of the nearest symbol. */ info->dli_sname = obj->strtab + def->st_name; info->dli_saddr = symbol_addr; /* Exact match? */ if (info->dli_saddr == addr) break; } lock_release(rtld_bind_lock, &lockstate); return 1; } int dlinfo(void *handle, int request, void *p) { const Obj_Entry *obj; RtldLockState lockstate; int error; rlock_acquire(rtld_bind_lock, &lockstate); if (handle == NULL || handle == RTLD_SELF) { void *retaddr; retaddr = __builtin_return_address(0); /* __GNUC__ only */ if ((obj = obj_from_addr(retaddr)) == NULL) _rtld_error("Cannot determine caller's shared object"); } else obj = dlcheck(handle); if (obj == NULL) { lock_release(rtld_bind_lock, &lockstate); return (-1); } error = 0; switch (request) { case RTLD_DI_LINKMAP: *((struct link_map const **)p) = &obj->linkmap; break; case RTLD_DI_ORIGIN: error = rtld_dirname(obj->path, p); break; case RTLD_DI_SERINFOSIZE: case RTLD_DI_SERINFO: error = do_search_info(obj, request, (struct dl_serinfo *)p); break; default: _rtld_error("Invalid request %d passed to dlinfo()", request); error = -1; } lock_release(rtld_bind_lock, &lockstate); return (error); } static void rtld_fill_dl_phdr_info(const Obj_Entry *obj, struct dl_phdr_info *phdr_info) { phdr_info->dlpi_addr = (Elf_Addr)obj->relocbase; phdr_info->dlpi_name = obj->path; phdr_info->dlpi_phdr = obj->phdr; phdr_info->dlpi_phnum = obj->phsize / sizeof(obj->phdr[0]); phdr_info->dlpi_tls_modid = obj->tlsindex; phdr_info->dlpi_tls_data = obj->tlsinit; phdr_info->dlpi_adds = obj_loads; phdr_info->dlpi_subs = obj_loads - obj_count; } int dl_iterate_phdr(__dl_iterate_hdr_callback callback, void *param) { struct dl_phdr_info phdr_info; const Obj_Entry *obj; RtldLockState bind_lockstate, phdr_lockstate; int error; wlock_acquire(rtld_phdr_lock, &phdr_lockstate); rlock_acquire(rtld_bind_lock, &bind_lockstate); error = 0; for (obj = obj_list; obj != NULL; obj = obj->next) { rtld_fill_dl_phdr_info(obj, &phdr_info); if ((error = callback(&phdr_info, sizeof phdr_info, param)) != 0) break; } if (error == 0) { rtld_fill_dl_phdr_info(&obj_rtld, &phdr_info); error = callback(&phdr_info, sizeof(phdr_info), param); } lock_release(rtld_bind_lock, &bind_lockstate); lock_release(rtld_phdr_lock, &phdr_lockstate); return (error); } static void * fill_search_info(const char *dir, size_t dirlen, void *param) { struct fill_search_info_args *arg; arg = param; if (arg->request == RTLD_DI_SERINFOSIZE) { arg->serinfo->dls_cnt ++; arg->serinfo->dls_size += sizeof(struct dl_serpath) + dirlen + 1; } else { struct dl_serpath *s_entry; s_entry = arg->serpath; s_entry->dls_name = arg->strspace; s_entry->dls_flags = arg->flags; strncpy(arg->strspace, dir, dirlen); arg->strspace[dirlen] = '\0'; arg->strspace += dirlen + 1; arg->serpath++; } return (NULL); } static int do_search_info(const Obj_Entry *obj, int request, struct dl_serinfo *info) { struct dl_serinfo _info; struct fill_search_info_args args; args.request = RTLD_DI_SERINFOSIZE; args.serinfo = &_info; _info.dls_size = __offsetof(struct dl_serinfo, dls_serpath); _info.dls_cnt = 0; path_enumerate(obj->rpath, fill_search_info, &args); path_enumerate(ld_library_path, fill_search_info, &args); path_enumerate(obj->runpath, fill_search_info, &args); path_enumerate(gethints(obj->z_nodeflib), fill_search_info, &args); if (!obj->z_nodeflib) path_enumerate(STANDARD_LIBRARY_PATH, fill_search_info, &args); if (request == RTLD_DI_SERINFOSIZE) { info->dls_size = _info.dls_size; info->dls_cnt = _info.dls_cnt; return (0); } if (info->dls_cnt != _info.dls_cnt || info->dls_size != _info.dls_size) { _rtld_error("Uninitialized Dl_serinfo struct passed to dlinfo()"); return (-1); } args.request = RTLD_DI_SERINFO; args.serinfo = info; args.serpath = &info->dls_serpath[0]; args.strspace = (char *)&info->dls_serpath[_info.dls_cnt]; args.flags = LA_SER_RUNPATH; if (path_enumerate(obj->rpath, fill_search_info, &args) != NULL) return (-1); args.flags = LA_SER_LIBPATH; if (path_enumerate(ld_library_path, fill_search_info, &args) != NULL) return (-1); args.flags = LA_SER_RUNPATH; if (path_enumerate(obj->runpath, fill_search_info, &args) != NULL) return (-1); args.flags = LA_SER_CONFIG; if (path_enumerate(gethints(obj->z_nodeflib), fill_search_info, &args) != NULL) return (-1); args.flags = LA_SER_DEFAULT; if (!obj->z_nodeflib && path_enumerate(STANDARD_LIBRARY_PATH, fill_search_info, &args) != NULL) return (-1); return (0); } static int rtld_dirname(const char *path, char *bname) { const char *endp; /* Empty or NULL string gets treated as "." */ if (path == NULL || *path == '\0') { bname[0] = '.'; bname[1] = '\0'; return (0); } /* Strip trailing slashes */ endp = path + strlen(path) - 1; while (endp > path && *endp == '/') endp--; /* Find the start of the dir */ while (endp > path && *endp != '/') endp--; /* Either the dir is "/" or there are no slashes */ if (endp == path) { bname[0] = *endp == '/' ? '/' : '.'; bname[1] = '\0'; return (0); } else { do { endp--; } while (endp > path && *endp == '/'); } if (endp - path + 2 > PATH_MAX) { _rtld_error("Filename is too long: %s", path); return(-1); } strncpy(bname, path, endp - path + 1); bname[endp - path + 1] = '\0'; return (0); } static int rtld_dirname_abs(const char *path, char *base) { char *last; if (realpath(path, base) == NULL) return (-1); dbg("%s -> %s", path, base); last = strrchr(base, '/'); if (last == NULL) return (-1); if (last != base) *last = '\0'; return (0); } static void linkmap_add(Obj_Entry *obj) { struct link_map *l = &obj->linkmap; struct link_map *prev; obj->linkmap.l_name = obj->path; obj->linkmap.l_addr = obj->mapbase; obj->linkmap.l_ld = obj->dynamic; #ifdef __mips__ /* GDB needs load offset on MIPS to use the symbols */ obj->linkmap.l_offs = obj->relocbase; #endif if (r_debug.r_map == NULL) { r_debug.r_map = l; return; } /* * Scan to the end of the list, but not past the entry for the * dynamic linker, which we want to keep at the very end. */ for (prev = r_debug.r_map; prev->l_next != NULL && prev->l_next != &obj_rtld.linkmap; prev = prev->l_next) ; /* Link in the new entry. */ l->l_prev = prev; l->l_next = prev->l_next; if (l->l_next != NULL) l->l_next->l_prev = l; prev->l_next = l; } static void linkmap_delete(Obj_Entry *obj) { struct link_map *l = &obj->linkmap; if (l->l_prev == NULL) { if ((r_debug.r_map = l->l_next) != NULL) l->l_next->l_prev = NULL; return; } if ((l->l_prev->l_next = l->l_next) != NULL) l->l_next->l_prev = l->l_prev; } /* * Function for the debugger to set a breakpoint on to gain control. * * The two parameters allow the debugger to easily find and determine * what the runtime loader is doing and to whom it is doing it. * * When the loadhook trap is hit (r_debug_state, set at program * initialization), the arguments can be found on the stack: * * +8 struct link_map *m * +4 struct r_debug *rd * +0 RetAddr */ void r_debug_state(struct r_debug* rd, struct link_map *m) { /* * The following is a hack to force the compiler to emit calls to * this function, even when optimizing. If the function is empty, * the compiler is not obliged to emit any code for calls to it, * even when marked __noinline. However, gdb depends on those * calls being made. */ __compiler_membar(); } /* * A function called after init routines have completed. This can be used to * break before a program's entry routine is called, and can be used when * main is not available in the symbol table. */ void _r_debug_postinit(struct link_map *m) { /* See r_debug_state(). */ __compiler_membar(); } /* * Get address of the pointer variable in the main program. * Prefer non-weak symbol over the weak one. */ static const void ** get_program_var_addr(const char *name, RtldLockState *lockstate) { SymLook req; DoneList donelist; symlook_init(&req, name); req.lockstate = lockstate; donelist_init(&donelist); if (symlook_global(&req, &donelist) != 0) return (NULL); if (ELF_ST_TYPE(req.sym_out->st_info) == STT_FUNC) return ((const void **)make_function_pointer(req.sym_out, req.defobj_out)); else if (ELF_ST_TYPE(req.sym_out->st_info) == STT_GNU_IFUNC) return ((const void **)rtld_resolve_ifunc(req.defobj_out, req.sym_out)); else return ((const void **)(req.defobj_out->relocbase + req.sym_out->st_value)); } /* * Set a pointer variable in the main program to the given value. This * is used to set key variables such as "environ" before any of the * init functions are called. */ static void set_program_var(const char *name, const void *value) { const void **addr; if ((addr = get_program_var_addr(name, NULL)) != NULL) { dbg("\"%s\": *%p <-- %p", name, addr, value); *addr = value; } } /* * Search the global objects, including dependencies and main object, * for the given symbol. */ static int symlook_global(SymLook *req, DoneList *donelist) { SymLook req1; const Objlist_Entry *elm; int res; symlook_init_from_req(&req1, req); /* Search all objects loaded at program start up. */ if (req->defobj_out == NULL || ELF_ST_BIND(req->sym_out->st_info) == STB_WEAK) { res = symlook_list(&req1, &list_main, donelist); if (res == 0 && (req->defobj_out == NULL || ELF_ST_BIND(req1.sym_out->st_info) != STB_WEAK)) { req->sym_out = req1.sym_out; req->defobj_out = req1.defobj_out; assert(req->defobj_out != NULL); } } /* Search all DAGs whose roots are RTLD_GLOBAL objects. */ STAILQ_FOREACH(elm, &list_global, link) { if (req->defobj_out != NULL && ELF_ST_BIND(req->sym_out->st_info) != STB_WEAK) break; res = symlook_list(&req1, &elm->obj->dagmembers, donelist); if (res == 0 && (req->defobj_out == NULL || ELF_ST_BIND(req1.sym_out->st_info) != STB_WEAK)) { req->sym_out = req1.sym_out; req->defobj_out = req1.defobj_out; assert(req->defobj_out != NULL); } } return (req->sym_out != NULL ? 0 : ESRCH); } /* * Given a symbol name in a referencing object, find the corresponding * definition of the symbol. Returns a pointer to the symbol, or NULL if * no definition was found. Returns a pointer to the Obj_Entry of the * defining object via the reference parameter DEFOBJ_OUT. */ static int symlook_default(SymLook *req, const Obj_Entry *refobj) { DoneList donelist; const Objlist_Entry *elm; SymLook req1; int res; donelist_init(&donelist); symlook_init_from_req(&req1, req); /* Look first in the referencing object if linked symbolically. */ if (refobj->symbolic && !donelist_check(&donelist, refobj)) { res = symlook_obj(&req1, refobj); if (res == 0) { req->sym_out = req1.sym_out; req->defobj_out = req1.defobj_out; assert(req->defobj_out != NULL); } } symlook_global(req, &donelist); /* Search all dlopened DAGs containing the referencing object. */ STAILQ_FOREACH(elm, &refobj->dldags, link) { if (req->sym_out != NULL && ELF_ST_BIND(req->sym_out->st_info) != STB_WEAK) break; res = symlook_list(&req1, &elm->obj->dagmembers, &donelist); if (res == 0 && (req->sym_out == NULL || ELF_ST_BIND(req1.sym_out->st_info) != STB_WEAK)) { req->sym_out = req1.sym_out; req->defobj_out = req1.defobj_out; assert(req->defobj_out != NULL); } } /* * Search the dynamic linker itself, and possibly resolve the * symbol from there. This is how the application links to * dynamic linker services such as dlopen. */ if (req->sym_out == NULL || ELF_ST_BIND(req->sym_out->st_info) == STB_WEAK) { res = symlook_obj(&req1, &obj_rtld); if (res == 0) { req->sym_out = req1.sym_out; req->defobj_out = req1.defobj_out; assert(req->defobj_out != NULL); } } return (req->sym_out != NULL ? 0 : ESRCH); } static int symlook_list(SymLook *req, const Objlist *objlist, DoneList *dlp) { const Elf_Sym *def; const Obj_Entry *defobj; const Objlist_Entry *elm; SymLook req1; int res; def = NULL; defobj = NULL; STAILQ_FOREACH(elm, objlist, link) { if (donelist_check(dlp, elm->obj)) continue; symlook_init_from_req(&req1, req); if ((res = symlook_obj(&req1, elm->obj)) == 0) { if (def == NULL || ELF_ST_BIND(req1.sym_out->st_info) != STB_WEAK) { def = req1.sym_out; defobj = req1.defobj_out; if (ELF_ST_BIND(def->st_info) != STB_WEAK) break; } } } if (def != NULL) { req->sym_out = def; req->defobj_out = defobj; return (0); } return (ESRCH); } /* * Search the chain of DAGS cointed to by the given Needed_Entry * for a symbol of the given name. Each DAG is scanned completely * before advancing to the next one. Returns a pointer to the symbol, * or NULL if no definition was found. */ static int symlook_needed(SymLook *req, const Needed_Entry *needed, DoneList *dlp) { const Elf_Sym *def; const Needed_Entry *n; const Obj_Entry *defobj; SymLook req1; int res; def = NULL; defobj = NULL; symlook_init_from_req(&req1, req); for (n = needed; n != NULL; n = n->next) { if (n->obj == NULL || (res = symlook_list(&req1, &n->obj->dagmembers, dlp)) != 0) continue; if (def == NULL || ELF_ST_BIND(req1.sym_out->st_info) != STB_WEAK) { def = req1.sym_out; defobj = req1.defobj_out; if (ELF_ST_BIND(def->st_info) != STB_WEAK) break; } } if (def != NULL) { req->sym_out = def; req->defobj_out = defobj; return (0); } return (ESRCH); } /* * Search the symbol table of a single shared object for a symbol of * the given name and version, if requested. Returns a pointer to the * symbol, or NULL if no definition was found. If the object is * filter, return filtered symbol from filtee. * * The symbol's hash value is passed in for efficiency reasons; that * eliminates many recomputations of the hash value. */ int symlook_obj(SymLook *req, const Obj_Entry *obj) { DoneList donelist; SymLook req1; int flags, res, mres; /* * If there is at least one valid hash at this point, we prefer to * use the faster GNU version if available. */ if (obj->valid_hash_gnu) mres = symlook_obj1_gnu(req, obj); else if (obj->valid_hash_sysv) mres = symlook_obj1_sysv(req, obj); else return (EINVAL); if (mres == 0) { if (obj->needed_filtees != NULL) { flags = (req->flags & SYMLOOK_EARLY) ? RTLD_LO_EARLY : 0; load_filtees(__DECONST(Obj_Entry *, obj), flags, req->lockstate); donelist_init(&donelist); symlook_init_from_req(&req1, req); res = symlook_needed(&req1, obj->needed_filtees, &donelist); if (res == 0) { req->sym_out = req1.sym_out; req->defobj_out = req1.defobj_out; } return (res); } if (obj->needed_aux_filtees != NULL) { flags = (req->flags & SYMLOOK_EARLY) ? RTLD_LO_EARLY : 0; load_filtees(__DECONST(Obj_Entry *, obj), flags, req->lockstate); donelist_init(&donelist); symlook_init_from_req(&req1, req); res = symlook_needed(&req1, obj->needed_aux_filtees, &donelist); if (res == 0) { req->sym_out = req1.sym_out; req->defobj_out = req1.defobj_out; return (res); } } } return (mres); } /* Symbol match routine common to both hash functions */ static bool matched_symbol(SymLook *req, const Obj_Entry *obj, Sym_Match_Result *result, const unsigned long symnum) { Elf_Versym verndx; const Elf_Sym *symp; const char *strp; symp = obj->symtab + symnum; strp = obj->strtab + symp->st_name; switch (ELF_ST_TYPE(symp->st_info)) { case STT_FUNC: case STT_NOTYPE: case STT_OBJECT: case STT_COMMON: case STT_GNU_IFUNC: if (symp->st_value == 0) return (false); /* fallthrough */ case STT_TLS: if (symp->st_shndx != SHN_UNDEF) break; #ifndef __mips__ else if (((req->flags & SYMLOOK_IN_PLT) == 0) && (ELF_ST_TYPE(symp->st_info) == STT_FUNC)) break; /* fallthrough */ #endif default: return (false); } if (req->name[0] != strp[0] || strcmp(req->name, strp) != 0) return (false); if (req->ventry == NULL) { if (obj->versyms != NULL) { verndx = VER_NDX(obj->versyms[symnum]); if (verndx > obj->vernum) { _rtld_error( "%s: symbol %s references wrong version %d", obj->path, obj->strtab + symnum, verndx); return (false); } /* * If we are not called from dlsym (i.e. this * is a normal relocation from unversioned * binary), accept the symbol immediately if * it happens to have first version after this * shared object became versioned. Otherwise, * if symbol is versioned and not hidden, * remember it. If it is the only symbol with * this name exported by the shared object, it * will be returned as a match by the calling * function. If symbol is global (verndx < 2) * accept it unconditionally. */ if ((req->flags & SYMLOOK_DLSYM) == 0 && verndx == VER_NDX_GIVEN) { result->sym_out = symp; return (true); } else if (verndx >= VER_NDX_GIVEN) { if ((obj->versyms[symnum] & VER_NDX_HIDDEN) == 0) { if (result->vsymp == NULL) result->vsymp = symp; result->vcount++; } return (false); } } result->sym_out = symp; return (true); } if (obj->versyms == NULL) { if (object_match_name(obj, req->ventry->name)) { _rtld_error("%s: object %s should provide version %s " "for symbol %s", obj_rtld.path, obj->path, req->ventry->name, obj->strtab + symnum); return (false); } } else { verndx = VER_NDX(obj->versyms[symnum]); if (verndx > obj->vernum) { _rtld_error("%s: symbol %s references wrong version %d", obj->path, obj->strtab + symnum, verndx); return (false); } if (obj->vertab[verndx].hash != req->ventry->hash || strcmp(obj->vertab[verndx].name, req->ventry->name)) { /* * Version does not match. Look if this is a * global symbol and if it is not hidden. If * global symbol (verndx < 2) is available, * use it. Do not return symbol if we are * called by dlvsym, because dlvsym looks for * a specific version and default one is not * what dlvsym wants. */ if ((req->flags & SYMLOOK_DLSYM) || (verndx >= VER_NDX_GIVEN) || (obj->versyms[symnum] & VER_NDX_HIDDEN)) return (false); } } result->sym_out = symp; return (true); } /* * Search for symbol using SysV hash function. * obj->buckets is known not to be NULL at this point; the test for this was * performed with the obj->valid_hash_sysv assignment. */ static int symlook_obj1_sysv(SymLook *req, const Obj_Entry *obj) { unsigned long symnum; Sym_Match_Result matchres; matchres.sym_out = NULL; matchres.vsymp = NULL; matchres.vcount = 0; for (symnum = obj->buckets[req->hash % obj->nbuckets]; symnum != STN_UNDEF; symnum = obj->chains[symnum]) { if (symnum >= obj->nchains) return (ESRCH); /* Bad object */ if (matched_symbol(req, obj, &matchres, symnum)) { req->sym_out = matchres.sym_out; req->defobj_out = obj; return (0); } } if (matchres.vcount == 1) { req->sym_out = matchres.vsymp; req->defobj_out = obj; return (0); } return (ESRCH); } /* Search for symbol using GNU hash function */ static int symlook_obj1_gnu(SymLook *req, const Obj_Entry *obj) { Elf_Addr bloom_word; const Elf32_Word *hashval; Elf32_Word bucket; Sym_Match_Result matchres; unsigned int h1, h2; unsigned long symnum; matchres.sym_out = NULL; matchres.vsymp = NULL; matchres.vcount = 0; /* Pick right bitmask word from Bloom filter array */ bloom_word = obj->bloom_gnu[(req->hash_gnu / __ELF_WORD_SIZE) & obj->maskwords_bm_gnu]; /* Calculate modulus word size of gnu hash and its derivative */ h1 = req->hash_gnu & (__ELF_WORD_SIZE - 1); h2 = ((req->hash_gnu >> obj->shift2_gnu) & (__ELF_WORD_SIZE - 1)); /* Filter out the "definitely not in set" queries */ if (((bloom_word >> h1) & (bloom_word >> h2) & 1) == 0) return (ESRCH); /* Locate hash chain and corresponding value element*/ bucket = obj->buckets_gnu[req->hash_gnu % obj->nbuckets_gnu]; if (bucket == 0) return (ESRCH); hashval = &obj->chain_zero_gnu[bucket]; do { if (((*hashval ^ req->hash_gnu) >> 1) == 0) { symnum = hashval - obj->chain_zero_gnu; if (matched_symbol(req, obj, &matchres, symnum)) { req->sym_out = matchres.sym_out; req->defobj_out = obj; return (0); } } } while ((*hashval++ & 1) == 0); if (matchres.vcount == 1) { req->sym_out = matchres.vsymp; req->defobj_out = obj; return (0); } return (ESRCH); } static void trace_loaded_objects(Obj_Entry *obj) { char *fmt1, *fmt2, *fmt, *main_local, *list_containers; int c; if ((main_local = getenv(LD_ "TRACE_LOADED_OBJECTS_PROGNAME")) == NULL) main_local = ""; if ((fmt1 = getenv(LD_ "TRACE_LOADED_OBJECTS_FMT1")) == NULL) fmt1 = "\t%o => %p (%x)\n"; if ((fmt2 = getenv(LD_ "TRACE_LOADED_OBJECTS_FMT2")) == NULL) fmt2 = "\t%o (%x)\n"; list_containers = getenv(LD_ "TRACE_LOADED_OBJECTS_ALL"); for (; obj; obj = obj->next) { Needed_Entry *needed; char *name, *path; bool is_lib; if (list_containers && obj->needed != NULL) rtld_printf("%s:\n", obj->path); for (needed = obj->needed; needed; needed = needed->next) { if (needed->obj != NULL) { if (needed->obj->traced && !list_containers) continue; needed->obj->traced = true; path = needed->obj->path; } else path = "not found"; name = (char *)obj->strtab + needed->name; is_lib = strncmp(name, "lib", 3) == 0; /* XXX - bogus */ fmt = is_lib ? fmt1 : fmt2; while ((c = *fmt++) != '\0') { switch (c) { default: rtld_putchar(c); continue; case '\\': switch (c = *fmt) { case '\0': continue; case 'n': rtld_putchar('\n'); break; case 't': rtld_putchar('\t'); break; } break; case '%': switch (c = *fmt) { case '\0': continue; case '%': default: rtld_putchar(c); break; case 'A': rtld_putstr(main_local); break; case 'a': rtld_putstr(obj_main->path); break; case 'o': rtld_putstr(name); break; #if 0 case 'm': rtld_printf("%d", sodp->sod_major); break; case 'n': rtld_printf("%d", sodp->sod_minor); break; #endif case 'p': rtld_putstr(path); break; case 'x': rtld_printf("%p", needed->obj ? needed->obj->mapbase : 0); break; } break; } ++fmt; } } } } /* * Unload a dlopened object and its dependencies from memory and from * our data structures. It is assumed that the DAG rooted in the * object has already been unreferenced, and that the object has a * reference count of 0. */ static void unload_object(Obj_Entry *root) { Obj_Entry *obj; Obj_Entry **linkp; assert(root->refcount == 0); /* * Pass over the DAG removing unreferenced objects from * appropriate lists. */ unlink_object(root); /* Unmap all objects that are no longer referenced. */ linkp = &obj_list->next; while ((obj = *linkp) != NULL) { if (obj->refcount == 0) { LD_UTRACE(UTRACE_UNLOAD_OBJECT, obj, obj->mapbase, obj->mapsize, 0, obj->path); dbg("unloading \"%s\"", obj->path); unload_filtees(root); munmap(obj->mapbase, obj->mapsize); linkmap_delete(obj); *linkp = obj->next; obj_count--; obj_free(obj); } else linkp = &obj->next; } obj_tail = linkp; } static void unlink_object(Obj_Entry *root) { Objlist_Entry *elm; if (root->refcount == 0) { /* Remove the object from the RTLD_GLOBAL list. */ objlist_remove(&list_global, root); /* Remove the object from all objects' DAG lists. */ STAILQ_FOREACH(elm, &root->dagmembers, link) { objlist_remove(&elm->obj->dldags, root); if (elm->obj != root) unlink_object(elm->obj); } } } static void ref_dag(Obj_Entry *root) { Objlist_Entry *elm; assert(root->dag_inited); STAILQ_FOREACH(elm, &root->dagmembers, link) elm->obj->refcount++; } static void unref_dag(Obj_Entry *root) { Objlist_Entry *elm; assert(root->dag_inited); STAILQ_FOREACH(elm, &root->dagmembers, link) elm->obj->refcount--; } /* * Common code for MD __tls_get_addr(). */ static void *tls_get_addr_slow(Elf_Addr **, int, size_t) __noinline; static void * tls_get_addr_slow(Elf_Addr **dtvp, int index, size_t offset) { Elf_Addr *newdtv, *dtv; RtldLockState lockstate; int to_copy; dtv = *dtvp; /* Check dtv generation in case new modules have arrived */ if (dtv[0] != tls_dtv_generation) { wlock_acquire(rtld_bind_lock, &lockstate); newdtv = xcalloc(tls_max_index + 2, sizeof(Elf_Addr)); to_copy = dtv[1]; if (to_copy > tls_max_index) to_copy = tls_max_index; memcpy(&newdtv[2], &dtv[2], to_copy * sizeof(Elf_Addr)); newdtv[0] = tls_dtv_generation; newdtv[1] = tls_max_index; free(dtv); lock_release(rtld_bind_lock, &lockstate); dtv = *dtvp = newdtv; } /* Dynamically allocate module TLS if necessary */ if (dtv[index + 1] == 0) { /* Signal safe, wlock will block out signals. */ wlock_acquire(rtld_bind_lock, &lockstate); if (!dtv[index + 1]) dtv[index + 1] = (Elf_Addr)allocate_module_tls(index); lock_release(rtld_bind_lock, &lockstate); } return ((void *)(dtv[index + 1] + offset)); } void * tls_get_addr_common(Elf_Addr **dtvp, int index, size_t offset) { Elf_Addr *dtv; dtv = *dtvp; /* Check dtv generation in case new modules have arrived */ if (__predict_true(dtv[0] == tls_dtv_generation && dtv[index + 1] != 0)) return ((void *)(dtv[index + 1] + offset)); return (tls_get_addr_slow(dtvp, index, offset)); } #if defined(__aarch64__) || defined(__arm__) || defined(__mips__) || \ defined(__powerpc__) /* * Allocate Static TLS using the Variant I method. */ void * allocate_tls(Obj_Entry *objs, void *oldtcb, size_t tcbsize, size_t tcbalign) { Obj_Entry *obj; char *tcb; Elf_Addr **tls; Elf_Addr *dtv; Elf_Addr addr; int i; if (oldtcb != NULL && tcbsize == TLS_TCB_SIZE) return (oldtcb); assert(tcbsize >= TLS_TCB_SIZE); tcb = xcalloc(1, tls_static_space - TLS_TCB_SIZE + tcbsize); tls = (Elf_Addr **)(tcb + tcbsize - TLS_TCB_SIZE); if (oldtcb != NULL) { memcpy(tls, oldtcb, tls_static_space); free(oldtcb); /* Adjust the DTV. */ dtv = tls[0]; for (i = 0; i < dtv[1]; i++) { if (dtv[i+2] >= (Elf_Addr)oldtcb && dtv[i+2] < (Elf_Addr)oldtcb + tls_static_space) { dtv[i+2] = dtv[i+2] - (Elf_Addr)oldtcb + (Elf_Addr)tls; } } } else { dtv = xcalloc(tls_max_index + 2, sizeof(Elf_Addr)); tls[0] = dtv; dtv[0] = tls_dtv_generation; dtv[1] = tls_max_index; for (obj = objs; obj; obj = obj->next) { if (obj->tlsoffset > 0) { addr = (Elf_Addr)tls + obj->tlsoffset; if (obj->tlsinitsize > 0) memcpy((void*) addr, obj->tlsinit, obj->tlsinitsize); if (obj->tlssize > obj->tlsinitsize) memset((void*) (addr + obj->tlsinitsize), 0, obj->tlssize - obj->tlsinitsize); dtv[obj->tlsindex + 1] = addr; } } } return (tcb); } void free_tls(void *tcb, size_t tcbsize, size_t tcbalign) { Elf_Addr *dtv; Elf_Addr tlsstart, tlsend; int dtvsize, i; assert(tcbsize >= TLS_TCB_SIZE); tlsstart = (Elf_Addr)tcb + tcbsize - TLS_TCB_SIZE; tlsend = tlsstart + tls_static_space; dtv = *(Elf_Addr **)tlsstart; dtvsize = dtv[1]; for (i = 0; i < dtvsize; i++) { if (dtv[i+2] && (dtv[i+2] < tlsstart || dtv[i+2] >= tlsend)) { free((void*)dtv[i+2]); } } free(dtv); free(tcb); } #endif #if defined(__i386__) || defined(__amd64__) || defined(__sparc64__) /* * Allocate Static TLS using the Variant II method. */ void * allocate_tls(Obj_Entry *objs, void *oldtls, size_t tcbsize, size_t tcbalign) { Obj_Entry *obj; size_t size, ralign; char *tls; Elf_Addr *dtv, *olddtv; Elf_Addr segbase, oldsegbase, addr; int i; ralign = tcbalign; if (tls_static_max_align > ralign) ralign = tls_static_max_align; size = round(tls_static_space, ralign) + round(tcbsize, ralign); assert(tcbsize >= 2*sizeof(Elf_Addr)); tls = malloc_aligned(size, ralign); dtv = xcalloc(tls_max_index + 2, sizeof(Elf_Addr)); segbase = (Elf_Addr)(tls + round(tls_static_space, ralign)); ((Elf_Addr*)segbase)[0] = segbase; ((Elf_Addr*)segbase)[1] = (Elf_Addr) dtv; dtv[0] = tls_dtv_generation; dtv[1] = tls_max_index; if (oldtls) { /* * Copy the static TLS block over whole. */ oldsegbase = (Elf_Addr) oldtls; memcpy((void *)(segbase - tls_static_space), (const void *)(oldsegbase - tls_static_space), tls_static_space); /* * If any dynamic TLS blocks have been created tls_get_addr(), * move them over. */ olddtv = ((Elf_Addr**)oldsegbase)[1]; for (i = 0; i < olddtv[1]; i++) { if (olddtv[i+2] < oldsegbase - size || olddtv[i+2] > oldsegbase) { dtv[i+2] = olddtv[i+2]; olddtv[i+2] = 0; } } /* * We assume that this block was the one we created with * allocate_initial_tls(). */ free_tls(oldtls, 2*sizeof(Elf_Addr), sizeof(Elf_Addr)); } else { for (obj = objs; obj; obj = obj->next) { if (obj->tlsoffset) { addr = segbase - obj->tlsoffset; memset((void*) (addr + obj->tlsinitsize), 0, obj->tlssize - obj->tlsinitsize); if (obj->tlsinit) memcpy((void*) addr, obj->tlsinit, obj->tlsinitsize); dtv[obj->tlsindex + 1] = addr; } } } return (void*) segbase; } void free_tls(void *tls, size_t tcbsize, size_t tcbalign) { Elf_Addr* dtv; size_t size, ralign; int dtvsize, i; Elf_Addr tlsstart, tlsend; /* * Figure out the size of the initial TLS block so that we can * find stuff which ___tls_get_addr() allocated dynamically. */ ralign = tcbalign; if (tls_static_max_align > ralign) ralign = tls_static_max_align; size = round(tls_static_space, ralign); dtv = ((Elf_Addr**)tls)[1]; dtvsize = dtv[1]; tlsend = (Elf_Addr) tls; tlsstart = tlsend - size; for (i = 0; i < dtvsize; i++) { if (dtv[i + 2] != 0 && (dtv[i + 2] < tlsstart || dtv[i + 2] > tlsend)) { free_aligned((void *)dtv[i + 2]); } } free_aligned((void *)tlsstart); free((void*) dtv); } #endif /* * Allocate TLS block for module with given index. */ void * allocate_module_tls(int index) { Obj_Entry* obj; char* p; for (obj = obj_list; obj; obj = obj->next) { if (obj->tlsindex == index) break; } if (!obj) { _rtld_error("Can't find module with TLS index %d", index); rtld_die(); } p = malloc_aligned(obj->tlssize, obj->tlsalign); memcpy(p, obj->tlsinit, obj->tlsinitsize); memset(p + obj->tlsinitsize, 0, obj->tlssize - obj->tlsinitsize); return p; } bool allocate_tls_offset(Obj_Entry *obj) { size_t off; if (obj->tls_done) return true; if (obj->tlssize == 0) { obj->tls_done = true; return true; } if (obj->tlsindex == 1) off = calculate_first_tls_offset(obj->tlssize, obj->tlsalign); else off = calculate_tls_offset(tls_last_offset, tls_last_size, obj->tlssize, obj->tlsalign); /* * If we have already fixed the size of the static TLS block, we * must stay within that size. When allocating the static TLS, we * leave a small amount of space spare to be used for dynamically * loading modules which use static TLS. */ if (tls_static_space != 0) { if (calculate_tls_end(off, obj->tlssize) > tls_static_space) return false; } else if (obj->tlsalign > tls_static_max_align) { tls_static_max_align = obj->tlsalign; } tls_last_offset = obj->tlsoffset = off; tls_last_size = obj->tlssize; obj->tls_done = true; return true; } void free_tls_offset(Obj_Entry *obj) { /* * If we were the last thing to allocate out of the static TLS * block, we give our space back to the 'allocator'. This is a * simplistic workaround to allow libGL.so.1 to be loaded and * unloaded multiple times. */ if (calculate_tls_end(obj->tlsoffset, obj->tlssize) == calculate_tls_end(tls_last_offset, tls_last_size)) { tls_last_offset -= obj->tlssize; tls_last_size = 0; } } void * _rtld_allocate_tls(void *oldtls, size_t tcbsize, size_t tcbalign) { void *ret; RtldLockState lockstate; wlock_acquire(rtld_bind_lock, &lockstate); ret = allocate_tls(obj_list, oldtls, tcbsize, tcbalign); lock_release(rtld_bind_lock, &lockstate); return (ret); } void _rtld_free_tls(void *tcb, size_t tcbsize, size_t tcbalign) { RtldLockState lockstate; wlock_acquire(rtld_bind_lock, &lockstate); free_tls(tcb, tcbsize, tcbalign); lock_release(rtld_bind_lock, &lockstate); } static void object_add_name(Obj_Entry *obj, const char *name) { Name_Entry *entry; size_t len; len = strlen(name); entry = malloc(sizeof(Name_Entry) + len); if (entry != NULL) { strcpy(entry->name, name); STAILQ_INSERT_TAIL(&obj->names, entry, link); } } static int object_match_name(const Obj_Entry *obj, const char *name) { Name_Entry *entry; STAILQ_FOREACH(entry, &obj->names, link) { if (strcmp(name, entry->name) == 0) return (1); } return (0); } static Obj_Entry * locate_dependency(const Obj_Entry *obj, const char *name) { const Objlist_Entry *entry; const Needed_Entry *needed; STAILQ_FOREACH(entry, &list_main, link) { if (object_match_name(entry->obj, name)) return entry->obj; } for (needed = obj->needed; needed != NULL; needed = needed->next) { if (strcmp(obj->strtab + needed->name, name) == 0 || (needed->obj != NULL && object_match_name(needed->obj, name))) { /* * If there is DT_NEEDED for the name we are looking for, * we are all set. Note that object might not be found if * dependency was not loaded yet, so the function can * return NULL here. This is expected and handled * properly by the caller. */ return (needed->obj); } } _rtld_error("%s: Unexpected inconsistency: dependency %s not found", obj->path, name); rtld_die(); } static int check_object_provided_version(Obj_Entry *refobj, const Obj_Entry *depobj, const Elf_Vernaux *vna) { const Elf_Verdef *vd; const char *vername; vername = refobj->strtab + vna->vna_name; vd = depobj->verdef; if (vd == NULL) { _rtld_error("%s: version %s required by %s not defined", depobj->path, vername, refobj->path); return (-1); } for (;;) { if (vd->vd_version != VER_DEF_CURRENT) { _rtld_error("%s: Unsupported version %d of Elf_Verdef entry", depobj->path, vd->vd_version); return (-1); } if (vna->vna_hash == vd->vd_hash) { const Elf_Verdaux *aux = (const Elf_Verdaux *) ((char *)vd + vd->vd_aux); if (strcmp(vername, depobj->strtab + aux->vda_name) == 0) return (0); } if (vd->vd_next == 0) break; vd = (const Elf_Verdef *) ((char *)vd + vd->vd_next); } if (vna->vna_flags & VER_FLG_WEAK) return (0); _rtld_error("%s: version %s required by %s not found", depobj->path, vername, refobj->path); return (-1); } static int rtld_verify_object_versions(Obj_Entry *obj) { const Elf_Verneed *vn; const Elf_Verdef *vd; const Elf_Verdaux *vda; const Elf_Vernaux *vna; const Obj_Entry *depobj; int maxvernum, vernum; if (obj->ver_checked) return (0); obj->ver_checked = true; maxvernum = 0; /* * Walk over defined and required version records and figure out * max index used by any of them. Do very basic sanity checking * while there. */ vn = obj->verneed; while (vn != NULL) { if (vn->vn_version != VER_NEED_CURRENT) { _rtld_error("%s: Unsupported version %d of Elf_Verneed entry", obj->path, vn->vn_version); return (-1); } vna = (const Elf_Vernaux *) ((char *)vn + vn->vn_aux); for (;;) { vernum = VER_NEED_IDX(vna->vna_other); if (vernum > maxvernum) maxvernum = vernum; if (vna->vna_next == 0) break; vna = (const Elf_Vernaux *) ((char *)vna + vna->vna_next); } if (vn->vn_next == 0) break; vn = (const Elf_Verneed *) ((char *)vn + vn->vn_next); } vd = obj->verdef; while (vd != NULL) { if (vd->vd_version != VER_DEF_CURRENT) { _rtld_error("%s: Unsupported version %d of Elf_Verdef entry", obj->path, vd->vd_version); return (-1); } vernum = VER_DEF_IDX(vd->vd_ndx); if (vernum > maxvernum) maxvernum = vernum; if (vd->vd_next == 0) break; vd = (const Elf_Verdef *) ((char *)vd + vd->vd_next); } if (maxvernum == 0) return (0); /* * Store version information in array indexable by version index. * Verify that object version requirements are satisfied along the * way. */ obj->vernum = maxvernum + 1; obj->vertab = xcalloc(obj->vernum, sizeof(Ver_Entry)); vd = obj->verdef; while (vd != NULL) { if ((vd->vd_flags & VER_FLG_BASE) == 0) { vernum = VER_DEF_IDX(vd->vd_ndx); assert(vernum <= maxvernum); vda = (const Elf_Verdaux *)((char *)vd + vd->vd_aux); obj->vertab[vernum].hash = vd->vd_hash; obj->vertab[vernum].name = obj->strtab + vda->vda_name; obj->vertab[vernum].file = NULL; obj->vertab[vernum].flags = 0; } if (vd->vd_next == 0) break; vd = (const Elf_Verdef *) ((char *)vd + vd->vd_next); } vn = obj->verneed; while (vn != NULL) { depobj = locate_dependency(obj, obj->strtab + vn->vn_file); if (depobj == NULL) return (-1); vna = (const Elf_Vernaux *) ((char *)vn + vn->vn_aux); for (;;) { if (check_object_provided_version(obj, depobj, vna)) return (-1); vernum = VER_NEED_IDX(vna->vna_other); assert(vernum <= maxvernum); obj->vertab[vernum].hash = vna->vna_hash; obj->vertab[vernum].name = obj->strtab + vna->vna_name; obj->vertab[vernum].file = obj->strtab + vn->vn_file; obj->vertab[vernum].flags = (vna->vna_other & VER_NEED_HIDDEN) ? VER_INFO_HIDDEN : 0; if (vna->vna_next == 0) break; vna = (const Elf_Vernaux *) ((char *)vna + vna->vna_next); } if (vn->vn_next == 0) break; vn = (const Elf_Verneed *) ((char *)vn + vn->vn_next); } return 0; } static int rtld_verify_versions(const Objlist *objlist) { Objlist_Entry *entry; int rc; rc = 0; STAILQ_FOREACH(entry, objlist, link) { /* * Skip dummy objects or objects that have their version requirements * already checked. */ if (entry->obj->strtab == NULL || entry->obj->vertab != NULL) continue; if (rtld_verify_object_versions(entry->obj) == -1) { rc = -1; if (ld_tracing == NULL) break; } } if (rc == 0 || ld_tracing != NULL) rc = rtld_verify_object_versions(&obj_rtld); return rc; } const Ver_Entry * fetch_ventry(const Obj_Entry *obj, unsigned long symnum) { Elf_Versym vernum; if (obj->vertab) { vernum = VER_NDX(obj->versyms[symnum]); if (vernum >= obj->vernum) { _rtld_error("%s: symbol %s has wrong verneed value %d", obj->path, obj->strtab + symnum, vernum); } else if (obj->vertab[vernum].hash != 0) { return &obj->vertab[vernum]; } } return NULL; } int _rtld_get_stack_prot(void) { return (stack_prot); } int _rtld_is_dlopened(void *arg) { Obj_Entry *obj; RtldLockState lockstate; int res; rlock_acquire(rtld_bind_lock, &lockstate); obj = dlcheck(arg); if (obj == NULL) obj = obj_from_addr(arg); if (obj == NULL) { _rtld_error("No shared object contains address"); lock_release(rtld_bind_lock, &lockstate); return (-1); } res = obj->dlopened ? 1 : 0; lock_release(rtld_bind_lock, &lockstate); return (res); } static void map_stacks_exec(RtldLockState *lockstate) { void (*thr_map_stacks_exec)(void); if ((max_stack_flags & PF_X) == 0 || (stack_prot & PROT_EXEC) != 0) return; thr_map_stacks_exec = (void (*)(void))(uintptr_t) get_program_var_addr("__pthread_map_stacks_exec", lockstate); if (thr_map_stacks_exec != NULL) { stack_prot |= PROT_EXEC; thr_map_stacks_exec(); } } void symlook_init(SymLook *dst, const char *name) { bzero(dst, sizeof(*dst)); dst->name = name; dst->hash = elf_hash(name); dst->hash_gnu = gnu_hash(name); } static void symlook_init_from_req(SymLook *dst, const SymLook *src) { dst->name = src->name; dst->hash = src->hash; dst->hash_gnu = src->hash_gnu; dst->ventry = src->ventry; dst->flags = src->flags; dst->defobj_out = NULL; dst->sym_out = NULL; dst->lockstate = src->lockstate; } /* * Parse a file descriptor number without pulling in more of libc (e.g. atoi). */ static int parse_libdir(const char *str) { static const int RADIX = 10; /* XXXJA: possibly support hex? */ const char *orig; int fd; char c; orig = str; fd = 0; for (c = *str; c != '\0'; c = *++str) { if (c < '0' || c > '9') return (-1); fd *= RADIX; fd += c - '0'; } /* Make sure we actually parsed something. */ if (str == orig) { _rtld_error("failed to parse directory FD from '%s'", str); return (-1); } return (fd); } /* * Overrides for libc_pic-provided functions. */ int __getosreldate(void) { size_t len; int oid[2]; int error, osrel; if (osreldate != 0) return (osreldate); oid[0] = CTL_KERN; oid[1] = KERN_OSRELDATE; osrel = 0; len = sizeof(osrel); error = sysctl(oid, 2, &osrel, &len, NULL, 0); if (error == 0 && osrel > 0 && len == sizeof(osrel)) osreldate = osrel; return (osreldate); } void exit(int status) { _exit(status); } void (*__cleanup)(void); int __isthreaded = 0; int _thread_autoinit_dummy_decl = 1; /* * No unresolved symbols for rtld. */ void __pthread_cxa_finalize(struct dl_phdr_info *a) { } void __stack_chk_fail(void) { _rtld_error("stack overflow detected; terminated"); rtld_die(); } __weak_reference(__stack_chk_fail, __stack_chk_fail_local); void __chk_fail(void) { _rtld_error("buffer overflow detected; terminated"); rtld_die(); } const char * rtld_strerror(int errnum) { if (errnum < 0 || errnum >= sys_nerr) return ("Unknown error"); return (sys_errlist[errnum]); } Index: head/sys/kern/vfs_vnops.c =================================================================== --- head/sys/kern/vfs_vnops.c (revision 281844) +++ head/sys/kern/vfs_vnops.c (revision 281845) @@ -1,2354 +1,2360 @@ /*- * Copyright (c) 1982, 1986, 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. * * Copyright (c) 2012 Konstantin Belousov * Copyright (c) 2013, 2014 The FreeBSD Foundation * * Portions of this software were developed by Konstantin Belousov * under sponsorship from the FreeBSD Foundation. * * 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. * 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. * * @(#)vfs_vnops.c 8.2 (Berkeley) 1/21/94 */ #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 #include #include #include #include #include #include #include #include static fo_rdwr_t vn_read; static fo_rdwr_t vn_write; static fo_rdwr_t vn_io_fault; static fo_truncate_t vn_truncate; static fo_ioctl_t vn_ioctl; static fo_poll_t vn_poll; static fo_kqfilter_t vn_kqfilter; static fo_stat_t vn_statfile; static fo_close_t vn_closefile; struct fileops vnops = { .fo_read = vn_io_fault, .fo_write = vn_io_fault, .fo_truncate = vn_truncate, .fo_ioctl = vn_ioctl, .fo_poll = vn_poll, .fo_kqfilter = vn_kqfilter, .fo_stat = vn_statfile, .fo_close = vn_closefile, .fo_chmod = vn_chmod, .fo_chown = vn_chown, .fo_sendfile = vn_sendfile, .fo_seek = vn_seek, .fo_fill_kinfo = vn_fill_kinfo, .fo_flags = DFLAG_PASSABLE | DFLAG_SEEKABLE }; static const int io_hold_cnt = 16; static int vn_io_fault_enable = 1; SYSCTL_INT(_debug, OID_AUTO, vn_io_fault_enable, CTLFLAG_RW, &vn_io_fault_enable, 0, "Enable vn_io_fault lock avoidance"); static u_long vn_io_faults_cnt; SYSCTL_ULONG(_debug, OID_AUTO, vn_io_faults, CTLFLAG_RD, &vn_io_faults_cnt, 0, "Count of vn_io_fault lock avoidance triggers"); /* * Returns true if vn_io_fault mode of handling the i/o request should * be used. */ static bool do_vn_io_fault(struct vnode *vp, struct uio *uio) { struct mount *mp; return (uio->uio_segflg == UIO_USERSPACE && vp->v_type == VREG && (mp = vp->v_mount) != NULL && (mp->mnt_kern_flag & MNTK_NO_IOPF) != 0 && vn_io_fault_enable); } /* * Structure used to pass arguments to vn_io_fault1(), to do either * file- or vnode-based I/O calls. */ struct vn_io_fault_args { enum { VN_IO_FAULT_FOP, VN_IO_FAULT_VOP } kind; struct ucred *cred; int flags; union { struct fop_args_tag { struct file *fp; fo_rdwr_t *doio; } fop_args; struct vop_args_tag { struct vnode *vp; } vop_args; } args; }; static int vn_io_fault1(struct vnode *vp, struct uio *uio, struct vn_io_fault_args *args, struct thread *td); int vn_open(ndp, flagp, cmode, fp) struct nameidata *ndp; int *flagp, cmode; struct file *fp; { struct thread *td = ndp->ni_cnd.cn_thread; return (vn_open_cred(ndp, flagp, cmode, 0, td->td_ucred, fp)); } /* * Common code for vnode open operations via a name lookup. * Lookup the vnode and invoke VOP_CREATE if needed. * Check permissions, and call the VOP_OPEN or VOP_CREATE routine. * * Note that this does NOT free nameidata for the successful case, * due to the NDINIT being done elsewhere. */ int vn_open_cred(struct nameidata *ndp, int *flagp, int cmode, u_int vn_open_flags, struct ucred *cred, struct file *fp) { struct vnode *vp; struct mount *mp; struct thread *td = ndp->ni_cnd.cn_thread; struct vattr vat; struct vattr *vap = &vat; int fmode, error; restart: fmode = *flagp; if (fmode & O_CREAT) { ndp->ni_cnd.cn_nameiop = CREATE; /* * Set NOCACHE to avoid flushing the cache when * rolling in many files at once. */ ndp->ni_cnd.cn_flags = ISOPEN | LOCKPARENT | LOCKLEAF | NOCACHE; if ((fmode & O_EXCL) == 0 && (fmode & O_NOFOLLOW) == 0) ndp->ni_cnd.cn_flags |= FOLLOW; if (!(vn_open_flags & VN_OPEN_NOAUDIT)) ndp->ni_cnd.cn_flags |= AUDITVNODE1; if (vn_open_flags & VN_OPEN_NOCAPCHECK) ndp->ni_cnd.cn_flags |= NOCAPCHECK; bwillwrite(); if ((error = namei(ndp)) != 0) return (error); if (ndp->ni_vp == NULL) { VATTR_NULL(vap); vap->va_type = VREG; vap->va_mode = cmode; if (fmode & O_EXCL) vap->va_vaflags |= VA_EXCLUSIVE; if (vn_start_write(ndp->ni_dvp, &mp, V_NOWAIT) != 0) { NDFREE(ndp, NDF_ONLY_PNBUF); vput(ndp->ni_dvp); if ((error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH)) != 0) return (error); goto restart; } if ((vn_open_flags & VN_OPEN_NAMECACHE) != 0) ndp->ni_cnd.cn_flags |= MAKEENTRY; #ifdef MAC error = mac_vnode_check_create(cred, ndp->ni_dvp, &ndp->ni_cnd, vap); if (error == 0) #endif error = VOP_CREATE(ndp->ni_dvp, &ndp->ni_vp, &ndp->ni_cnd, vap); vput(ndp->ni_dvp); vn_finished_write(mp); if (error) { NDFREE(ndp, NDF_ONLY_PNBUF); return (error); } fmode &= ~O_TRUNC; vp = ndp->ni_vp; } else { if (ndp->ni_dvp == ndp->ni_vp) vrele(ndp->ni_dvp); else vput(ndp->ni_dvp); ndp->ni_dvp = NULL; vp = ndp->ni_vp; if (fmode & O_EXCL) { error = EEXIST; goto bad; } fmode &= ~O_CREAT; } } else { ndp->ni_cnd.cn_nameiop = LOOKUP; ndp->ni_cnd.cn_flags = ISOPEN | ((fmode & O_NOFOLLOW) ? NOFOLLOW : FOLLOW) | LOCKLEAF; if (!(fmode & FWRITE)) ndp->ni_cnd.cn_flags |= LOCKSHARED; if (!(vn_open_flags & VN_OPEN_NOAUDIT)) ndp->ni_cnd.cn_flags |= AUDITVNODE1; if (vn_open_flags & VN_OPEN_NOCAPCHECK) ndp->ni_cnd.cn_flags |= NOCAPCHECK; if ((error = namei(ndp)) != 0) return (error); vp = ndp->ni_vp; } error = vn_open_vnode(vp, fmode, cred, td, fp); if (error) goto bad; *flagp = fmode; return (0); bad: NDFREE(ndp, NDF_ONLY_PNBUF); vput(vp); *flagp = fmode; ndp->ni_vp = NULL; return (error); } /* * Common code for vnode open operations once a vnode is located. * Check permissions, and call the VOP_OPEN routine. */ int vn_open_vnode(struct vnode *vp, int fmode, struct ucred *cred, struct thread *td, struct file *fp) { struct mount *mp; accmode_t accmode; struct flock lf; int error, have_flock, lock_flags, type; if (vp->v_type == VLNK) return (EMLINK); if (vp->v_type == VSOCK) return (EOPNOTSUPP); if (vp->v_type != VDIR && fmode & O_DIRECTORY) return (ENOTDIR); accmode = 0; if (fmode & (FWRITE | O_TRUNC)) { if (vp->v_type == VDIR) return (EISDIR); accmode |= VWRITE; } if (fmode & FREAD) accmode |= VREAD; if (fmode & FEXEC) accmode |= VEXEC; if ((fmode & O_APPEND) && (fmode & FWRITE)) accmode |= VAPPEND; #ifdef MAC + if (fmode & O_CREAT) + accmode |= VCREAT; + if (fmode & O_VERIFY) + accmode |= VVERIFY; error = mac_vnode_check_open(cred, vp, accmode); if (error) return (error); + + accmode &= ~(VCREAT | VVERIFY); #endif if ((fmode & O_CREAT) == 0) { if (accmode & VWRITE) { error = vn_writechk(vp); if (error) return (error); } if (accmode) { error = VOP_ACCESS(vp, accmode, cred, td); if (error) return (error); } } if (vp->v_type == VFIFO && VOP_ISLOCKED(vp) != LK_EXCLUSIVE) vn_lock(vp, LK_UPGRADE | LK_RETRY); if ((error = VOP_OPEN(vp, fmode, cred, td, fp)) != 0) return (error); if (fmode & (O_EXLOCK | O_SHLOCK)) { KASSERT(fp != NULL, ("open with flock requires fp")); lock_flags = VOP_ISLOCKED(vp); VOP_UNLOCK(vp, 0); lf.l_whence = SEEK_SET; lf.l_start = 0; lf.l_len = 0; if (fmode & O_EXLOCK) lf.l_type = F_WRLCK; else lf.l_type = F_RDLCK; type = F_FLOCK; if ((fmode & FNONBLOCK) == 0) type |= F_WAIT; error = VOP_ADVLOCK(vp, (caddr_t)fp, F_SETLK, &lf, type); have_flock = (error == 0); vn_lock(vp, lock_flags | LK_RETRY); if (error == 0 && vp->v_iflag & VI_DOOMED) error = ENOENT; /* * Another thread might have used this vnode as an * executable while the vnode lock was dropped. * Ensure the vnode is still able to be opened for * writing after the lock has been obtained. */ if (error == 0 && accmode & VWRITE) error = vn_writechk(vp); if (error) { VOP_UNLOCK(vp, 0); if (have_flock) { lf.l_whence = SEEK_SET; lf.l_start = 0; lf.l_len = 0; lf.l_type = F_UNLCK; (void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK); } vn_start_write(vp, &mp, V_WAIT); vn_lock(vp, lock_flags | LK_RETRY); (void)VOP_CLOSE(vp, fmode, cred, td); vn_finished_write(mp); /* Prevent second close from fdrop()->vn_close(). */ if (fp != NULL) fp->f_ops= &badfileops; return (error); } fp->f_flag |= FHASLOCK; } if (fmode & FWRITE) { VOP_ADD_WRITECOUNT(vp, 1); CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d", __func__, vp, vp->v_writecount); } ASSERT_VOP_LOCKED(vp, "vn_open_vnode"); return (0); } /* * Check for write permissions on the specified vnode. * Prototype text segments cannot be written. */ int vn_writechk(vp) register struct vnode *vp; { ASSERT_VOP_LOCKED(vp, "vn_writechk"); /* * If there's shared text associated with * the vnode, try to free it up once. If * we fail, we can't allow writing. */ if (VOP_IS_TEXT(vp)) return (ETXTBSY); return (0); } /* * Vnode close call */ int vn_close(vp, flags, file_cred, td) register struct vnode *vp; int flags; struct ucred *file_cred; struct thread *td; { struct mount *mp; int error, lock_flags; if (vp->v_type != VFIFO && (flags & FWRITE) == 0 && MNT_EXTENDED_SHARED(vp->v_mount)) lock_flags = LK_SHARED; else lock_flags = LK_EXCLUSIVE; vn_start_write(vp, &mp, V_WAIT); vn_lock(vp, lock_flags | LK_RETRY); if (flags & FWRITE) { VNASSERT(vp->v_writecount > 0, vp, ("vn_close: negative writecount")); VOP_ADD_WRITECOUNT(vp, -1); CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d", __func__, vp, vp->v_writecount); } error = VOP_CLOSE(vp, flags, file_cred, td); vput(vp); vn_finished_write(mp); return (error); } /* * Heuristic to detect sequential operation. */ static int sequential_heuristic(struct uio *uio, struct file *fp) { ASSERT_VOP_LOCKED(fp->f_vnode, __func__); if (fp->f_flag & FRDAHEAD) return (fp->f_seqcount << IO_SEQSHIFT); /* * Offset 0 is handled specially. open() sets f_seqcount to 1 so * that the first I/O is normally considered to be slightly * sequential. Seeking to offset 0 doesn't change sequentiality * unless previous seeks have reduced f_seqcount to 0, in which * case offset 0 is not special. */ if ((uio->uio_offset == 0 && fp->f_seqcount > 0) || uio->uio_offset == fp->f_nextoff) { /* * f_seqcount is in units of fixed-size blocks so that it * depends mainly on the amount of sequential I/O and not * much on the number of sequential I/O's. The fixed size * of 16384 is hard-coded here since it is (not quite) just * a magic size that works well here. This size is more * closely related to the best I/O size for real disks than * to any block size used by software. */ fp->f_seqcount += howmany(uio->uio_resid, 16384); if (fp->f_seqcount > IO_SEQMAX) fp->f_seqcount = IO_SEQMAX; return (fp->f_seqcount << IO_SEQSHIFT); } /* Not sequential. Quickly draw-down sequentiality. */ if (fp->f_seqcount > 1) fp->f_seqcount = 1; else fp->f_seqcount = 0; return (0); } /* * Package up an I/O request on a vnode into a uio and do it. */ int vn_rdwr(enum uio_rw rw, struct vnode *vp, void *base, int len, off_t offset, enum uio_seg segflg, int ioflg, struct ucred *active_cred, struct ucred *file_cred, ssize_t *aresid, struct thread *td) { struct uio auio; struct iovec aiov; struct mount *mp; struct ucred *cred; void *rl_cookie; struct vn_io_fault_args args; int error, lock_flags; auio.uio_iov = &aiov; auio.uio_iovcnt = 1; aiov.iov_base = base; aiov.iov_len = len; auio.uio_resid = len; auio.uio_offset = offset; auio.uio_segflg = segflg; auio.uio_rw = rw; auio.uio_td = td; error = 0; if ((ioflg & IO_NODELOCKED) == 0) { if ((ioflg & IO_RANGELOCKED) == 0) { if (rw == UIO_READ) { rl_cookie = vn_rangelock_rlock(vp, offset, offset + len); } else { rl_cookie = vn_rangelock_wlock(vp, offset, offset + len); } } else rl_cookie = NULL; mp = NULL; if (rw == UIO_WRITE) { if (vp->v_type != VCHR && (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0) goto out; if (MNT_SHARED_WRITES(mp) || ((mp == NULL) && MNT_SHARED_WRITES(vp->v_mount))) lock_flags = LK_SHARED; else lock_flags = LK_EXCLUSIVE; } else lock_flags = LK_SHARED; vn_lock(vp, lock_flags | LK_RETRY); } else rl_cookie = NULL; ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); #ifdef MAC if ((ioflg & IO_NOMACCHECK) == 0) { if (rw == UIO_READ) error = mac_vnode_check_read(active_cred, file_cred, vp); else error = mac_vnode_check_write(active_cred, file_cred, vp); } #endif if (error == 0) { if (file_cred != NULL) cred = file_cred; else cred = active_cred; if (do_vn_io_fault(vp, &auio)) { args.kind = VN_IO_FAULT_VOP; args.cred = cred; args.flags = ioflg; args.args.vop_args.vp = vp; error = vn_io_fault1(vp, &auio, &args, td); } else if (rw == UIO_READ) { error = VOP_READ(vp, &auio, ioflg, cred); } else /* if (rw == UIO_WRITE) */ { error = VOP_WRITE(vp, &auio, ioflg, cred); } } if (aresid) *aresid = auio.uio_resid; else if (auio.uio_resid && error == 0) error = EIO; if ((ioflg & IO_NODELOCKED) == 0) { VOP_UNLOCK(vp, 0); if (mp != NULL) vn_finished_write(mp); } out: if (rl_cookie != NULL) vn_rangelock_unlock(vp, rl_cookie); return (error); } /* * Package up an I/O request on a vnode into a uio and do it. The I/O * request is split up into smaller chunks and we try to avoid saturating * the buffer cache while potentially holding a vnode locked, so we * check bwillwrite() before calling vn_rdwr(). We also call kern_yield() * to give other processes a chance to lock the vnode (either other processes * core'ing the same binary, or unrelated processes scanning the directory). */ int vn_rdwr_inchunks(rw, vp, base, len, offset, segflg, ioflg, active_cred, file_cred, aresid, td) enum uio_rw rw; struct vnode *vp; void *base; size_t len; off_t offset; enum uio_seg segflg; int ioflg; struct ucred *active_cred; struct ucred *file_cred; size_t *aresid; struct thread *td; { int error = 0; ssize_t iaresid; do { int chunk; /* * Force `offset' to a multiple of MAXBSIZE except possibly * for the first chunk, so that filesystems only need to * write full blocks except possibly for the first and last * chunks. */ chunk = MAXBSIZE - (uoff_t)offset % MAXBSIZE; if (chunk > len) chunk = len; if (rw != UIO_READ && vp->v_type == VREG) bwillwrite(); iaresid = 0; error = vn_rdwr(rw, vp, base, chunk, offset, segflg, ioflg, active_cred, file_cred, &iaresid, td); len -= chunk; /* aresid calc already includes length */ if (error) break; offset += chunk; base = (char *)base + chunk; kern_yield(PRI_USER); } while (len); if (aresid) *aresid = len + iaresid; return (error); } off_t foffset_lock(struct file *fp, int flags) { struct mtx *mtxp; off_t res; KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed")); #if OFF_MAX <= LONG_MAX /* * Caller only wants the current f_offset value. Assume that * the long and shorter integer types reads are atomic. */ if ((flags & FOF_NOLOCK) != 0) return (fp->f_offset); #endif /* * According to McKusick the vn lock was protecting f_offset here. * It is now protected by the FOFFSET_LOCKED flag. */ mtxp = mtx_pool_find(mtxpool_sleep, fp); mtx_lock(mtxp); if ((flags & FOF_NOLOCK) == 0) { while (fp->f_vnread_flags & FOFFSET_LOCKED) { fp->f_vnread_flags |= FOFFSET_LOCK_WAITING; msleep(&fp->f_vnread_flags, mtxp, PUSER -1, "vofflock", 0); } fp->f_vnread_flags |= FOFFSET_LOCKED; } res = fp->f_offset; mtx_unlock(mtxp); return (res); } void foffset_unlock(struct file *fp, off_t val, int flags) { struct mtx *mtxp; KASSERT((flags & FOF_OFFSET) == 0, ("FOF_OFFSET passed")); #if OFF_MAX <= LONG_MAX if ((flags & FOF_NOLOCK) != 0) { if ((flags & FOF_NOUPDATE) == 0) fp->f_offset = val; if ((flags & FOF_NEXTOFF) != 0) fp->f_nextoff = val; return; } #endif mtxp = mtx_pool_find(mtxpool_sleep, fp); mtx_lock(mtxp); if ((flags & FOF_NOUPDATE) == 0) fp->f_offset = val; if ((flags & FOF_NEXTOFF) != 0) fp->f_nextoff = val; if ((flags & FOF_NOLOCK) == 0) { KASSERT((fp->f_vnread_flags & FOFFSET_LOCKED) != 0, ("Lost FOFFSET_LOCKED")); if (fp->f_vnread_flags & FOFFSET_LOCK_WAITING) wakeup(&fp->f_vnread_flags); fp->f_vnread_flags = 0; } mtx_unlock(mtxp); } void foffset_lock_uio(struct file *fp, struct uio *uio, int flags) { if ((flags & FOF_OFFSET) == 0) uio->uio_offset = foffset_lock(fp, flags); } void foffset_unlock_uio(struct file *fp, struct uio *uio, int flags) { if ((flags & FOF_OFFSET) == 0) foffset_unlock(fp, uio->uio_offset, flags); } static int get_advice(struct file *fp, struct uio *uio) { struct mtx *mtxp; int ret; ret = POSIX_FADV_NORMAL; if (fp->f_advice == NULL) return (ret); mtxp = mtx_pool_find(mtxpool_sleep, fp); mtx_lock(mtxp); if (uio->uio_offset >= fp->f_advice->fa_start && uio->uio_offset + uio->uio_resid <= fp->f_advice->fa_end) ret = fp->f_advice->fa_advice; mtx_unlock(mtxp); return (ret); } /* * File table vnode read routine. */ static int vn_read(fp, uio, active_cred, flags, td) struct file *fp; struct uio *uio; struct ucred *active_cred; int flags; struct thread *td; { struct vnode *vp; struct mtx *mtxp; int error, ioflag; int advice; off_t offset, start, end; KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", uio->uio_td, td)); KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET")); vp = fp->f_vnode; ioflag = 0; if (fp->f_flag & FNONBLOCK) ioflag |= IO_NDELAY; if (fp->f_flag & O_DIRECT) ioflag |= IO_DIRECT; advice = get_advice(fp, uio); vn_lock(vp, LK_SHARED | LK_RETRY); switch (advice) { case POSIX_FADV_NORMAL: case POSIX_FADV_SEQUENTIAL: case POSIX_FADV_NOREUSE: ioflag |= sequential_heuristic(uio, fp); break; case POSIX_FADV_RANDOM: /* Disable read-ahead for random I/O. */ break; } offset = uio->uio_offset; #ifdef MAC error = mac_vnode_check_read(active_cred, fp->f_cred, vp); if (error == 0) #endif error = VOP_READ(vp, uio, ioflag, fp->f_cred); fp->f_nextoff = uio->uio_offset; VOP_UNLOCK(vp, 0); if (error == 0 && advice == POSIX_FADV_NOREUSE && offset != uio->uio_offset) { /* * Use POSIX_FADV_DONTNEED to flush clean pages and * buffers for the backing file after a * POSIX_FADV_NOREUSE read(2). To optimize the common * case of using POSIX_FADV_NOREUSE with sequential * access, track the previous implicit DONTNEED * request and grow this request to include the * current read(2) in addition to the previous * DONTNEED. With purely sequential access this will * cause the DONTNEED requests to continously grow to * cover all of the previously read regions of the * file. This allows filesystem blocks that are * accessed by multiple calls to read(2) to be flushed * once the last read(2) finishes. */ start = offset; end = uio->uio_offset - 1; mtxp = mtx_pool_find(mtxpool_sleep, fp); mtx_lock(mtxp); if (fp->f_advice != NULL && fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) { if (start != 0 && fp->f_advice->fa_prevend + 1 == start) start = fp->f_advice->fa_prevstart; else if (fp->f_advice->fa_prevstart != 0 && fp->f_advice->fa_prevstart == end + 1) end = fp->f_advice->fa_prevend; fp->f_advice->fa_prevstart = start; fp->f_advice->fa_prevend = end; } mtx_unlock(mtxp); error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED); } return (error); } /* * File table vnode write routine. */ static int vn_write(fp, uio, active_cred, flags, td) struct file *fp; struct uio *uio; struct ucred *active_cred; int flags; struct thread *td; { struct vnode *vp; struct mount *mp; struct mtx *mtxp; int error, ioflag, lock_flags; int advice; off_t offset, start, end; KASSERT(uio->uio_td == td, ("uio_td %p is not td %p", uio->uio_td, td)); KASSERT(flags & FOF_OFFSET, ("No FOF_OFFSET")); vp = fp->f_vnode; if (vp->v_type == VREG) bwillwrite(); ioflag = IO_UNIT; if (vp->v_type == VREG && (fp->f_flag & O_APPEND)) ioflag |= IO_APPEND; if (fp->f_flag & FNONBLOCK) ioflag |= IO_NDELAY; if (fp->f_flag & O_DIRECT) ioflag |= IO_DIRECT; if ((fp->f_flag & O_FSYNC) || (vp->v_mount && (vp->v_mount->mnt_flag & MNT_SYNCHRONOUS))) ioflag |= IO_SYNC; mp = NULL; if (vp->v_type != VCHR && (error = vn_start_write(vp, &mp, V_WAIT | PCATCH)) != 0) goto unlock; advice = get_advice(fp, uio); if (MNT_SHARED_WRITES(mp) || (mp == NULL && MNT_SHARED_WRITES(vp->v_mount))) { lock_flags = LK_SHARED; } else { lock_flags = LK_EXCLUSIVE; } vn_lock(vp, lock_flags | LK_RETRY); switch (advice) { case POSIX_FADV_NORMAL: case POSIX_FADV_SEQUENTIAL: case POSIX_FADV_NOREUSE: ioflag |= sequential_heuristic(uio, fp); break; case POSIX_FADV_RANDOM: /* XXX: Is this correct? */ break; } offset = uio->uio_offset; #ifdef MAC error = mac_vnode_check_write(active_cred, fp->f_cred, vp); if (error == 0) #endif error = VOP_WRITE(vp, uio, ioflag, fp->f_cred); fp->f_nextoff = uio->uio_offset; VOP_UNLOCK(vp, 0); if (vp->v_type != VCHR) vn_finished_write(mp); if (error == 0 && advice == POSIX_FADV_NOREUSE && offset != uio->uio_offset) { /* * Use POSIX_FADV_DONTNEED to flush clean pages and * buffers for the backing file after a * POSIX_FADV_NOREUSE write(2). To optimize the * common case of using POSIX_FADV_NOREUSE with * sequential access, track the previous implicit * DONTNEED request and grow this request to include * the current write(2) in addition to the previous * DONTNEED. With purely sequential access this will * cause the DONTNEED requests to continously grow to * cover all of the previously written regions of the * file. * * Note that the blocks just written are almost * certainly still dirty, so this only works when * VOP_ADVISE() calls from subsequent writes push out * the data written by this write(2) once the backing * buffers are clean. However, as compared to forcing * IO_DIRECT, this gives much saner behavior. Write * clustering is still allowed, and clean pages are * merely moved to the cache page queue rather than * outright thrown away. This means a subsequent * read(2) can still avoid hitting the disk if the * pages have not been reclaimed. * * This does make POSIX_FADV_NOREUSE largely useless * with non-sequential access. However, sequential * access is the more common use case and the flag is * merely advisory. */ start = offset; end = uio->uio_offset - 1; mtxp = mtx_pool_find(mtxpool_sleep, fp); mtx_lock(mtxp); if (fp->f_advice != NULL && fp->f_advice->fa_advice == POSIX_FADV_NOREUSE) { if (start != 0 && fp->f_advice->fa_prevend + 1 == start) start = fp->f_advice->fa_prevstart; else if (fp->f_advice->fa_prevstart != 0 && fp->f_advice->fa_prevstart == end + 1) end = fp->f_advice->fa_prevend; fp->f_advice->fa_prevstart = start; fp->f_advice->fa_prevend = end; } mtx_unlock(mtxp); error = VOP_ADVISE(vp, start, end, POSIX_FADV_DONTNEED); } unlock: return (error); } /* * The vn_io_fault() is a wrapper around vn_read() and vn_write() to * prevent the following deadlock: * * Assume that the thread A reads from the vnode vp1 into userspace * buffer buf1 backed by the pages of vnode vp2. If a page in buf1 is * currently not resident, then system ends up with the call chain * vn_read() -> VOP_READ(vp1) -> uiomove() -> [Page Fault] -> * vm_fault(buf1) -> vnode_pager_getpages(vp2) -> VOP_GETPAGES(vp2) * which establishes lock order vp1->vn_lock, then vp2->vn_lock. * If, at the same time, thread B reads from vnode vp2 into buffer buf2 * backed by the pages of vnode vp1, and some page in buf2 is not * resident, we get a reversed order vp2->vn_lock, then vp1->vn_lock. * * To prevent the lock order reversal and deadlock, vn_io_fault() does * not allow page faults to happen during VOP_READ() or VOP_WRITE(). * Instead, it first tries to do the whole range i/o with pagefaults * disabled. If all pages in the i/o buffer are resident and mapped, * VOP will succeed (ignoring the genuine filesystem errors). * Otherwise, we get back EFAULT, and vn_io_fault() falls back to do * i/o in chunks, with all pages in the chunk prefaulted and held * using vm_fault_quick_hold_pages(). * * Filesystems using this deadlock avoidance scheme should use the * array of the held pages from uio, saved in the curthread->td_ma, * instead of doing uiomove(). A helper function * vn_io_fault_uiomove() converts uiomove request into * uiomove_fromphys() over td_ma array. * * Since vnode locks do not cover the whole i/o anymore, rangelocks * make the current i/o request atomic with respect to other i/os and * truncations. */ /* * Decode vn_io_fault_args and perform the corresponding i/o. */ static int vn_io_fault_doio(struct vn_io_fault_args *args, struct uio *uio, struct thread *td) { switch (args->kind) { case VN_IO_FAULT_FOP: return ((args->args.fop_args.doio)(args->args.fop_args.fp, uio, args->cred, args->flags, td)); case VN_IO_FAULT_VOP: if (uio->uio_rw == UIO_READ) { return (VOP_READ(args->args.vop_args.vp, uio, args->flags, args->cred)); } else if (uio->uio_rw == UIO_WRITE) { return (VOP_WRITE(args->args.vop_args.vp, uio, args->flags, args->cred)); } break; } panic("vn_io_fault_doio: unknown kind of io %d %d", args->kind, uio->uio_rw); } /* * Common code for vn_io_fault(), agnostic to the kind of i/o request. * Uses vn_io_fault_doio() to make the call to an actual i/o function. * Used from vn_rdwr() and vn_io_fault(), which encode the i/o request * into args and call vn_io_fault1() to handle faults during the user * mode buffer accesses. */ static int vn_io_fault1(struct vnode *vp, struct uio *uio, struct vn_io_fault_args *args, struct thread *td) { vm_page_t ma[io_hold_cnt + 2]; struct uio *uio_clone, short_uio; struct iovec short_iovec[1]; vm_page_t *prev_td_ma; vm_prot_t prot; vm_offset_t addr, end; size_t len, resid; ssize_t adv; int error, cnt, save, saveheld, prev_td_ma_cnt; prot = uio->uio_rw == UIO_READ ? VM_PROT_WRITE : VM_PROT_READ; /* * The UFS follows IO_UNIT directive and replays back both * uio_offset and uio_resid if an error is encountered during the * operation. But, since the iovec may be already advanced, * uio is still in an inconsistent state. * * Cache a copy of the original uio, which is advanced to the redo * point using UIO_NOCOPY below. */ uio_clone = cloneuio(uio); resid = uio->uio_resid; short_uio.uio_segflg = UIO_USERSPACE; short_uio.uio_rw = uio->uio_rw; short_uio.uio_td = uio->uio_td; save = vm_fault_disable_pagefaults(); error = vn_io_fault_doio(args, uio, td); if (error != EFAULT) goto out; atomic_add_long(&vn_io_faults_cnt, 1); uio_clone->uio_segflg = UIO_NOCOPY; uiomove(NULL, resid - uio->uio_resid, uio_clone); uio_clone->uio_segflg = uio->uio_segflg; saveheld = curthread_pflags_set(TDP_UIOHELD); prev_td_ma = td->td_ma; prev_td_ma_cnt = td->td_ma_cnt; while (uio_clone->uio_resid != 0) { len = uio_clone->uio_iov->iov_len; if (len == 0) { KASSERT(uio_clone->uio_iovcnt >= 1, ("iovcnt underflow")); uio_clone->uio_iov++; uio_clone->uio_iovcnt--; continue; } if (len > io_hold_cnt * PAGE_SIZE) len = io_hold_cnt * PAGE_SIZE; addr = (uintptr_t)uio_clone->uio_iov->iov_base; end = round_page(addr + len); if (end < addr) { error = EFAULT; break; } cnt = atop(end - trunc_page(addr)); /* * A perfectly misaligned address and length could cause * both the start and the end of the chunk to use partial * page. +2 accounts for such a situation. */ cnt = vm_fault_quick_hold_pages(&td->td_proc->p_vmspace->vm_map, addr, len, prot, ma, io_hold_cnt + 2); if (cnt == -1) { error = EFAULT; break; } short_uio.uio_iov = &short_iovec[0]; short_iovec[0].iov_base = (void *)addr; short_uio.uio_iovcnt = 1; short_uio.uio_resid = short_iovec[0].iov_len = len; short_uio.uio_offset = uio_clone->uio_offset; td->td_ma = ma; td->td_ma_cnt = cnt; error = vn_io_fault_doio(args, &short_uio, td); vm_page_unhold_pages(ma, cnt); adv = len - short_uio.uio_resid; uio_clone->uio_iov->iov_base = (char *)uio_clone->uio_iov->iov_base + adv; uio_clone->uio_iov->iov_len -= adv; uio_clone->uio_resid -= adv; uio_clone->uio_offset += adv; uio->uio_resid -= adv; uio->uio_offset += adv; if (error != 0 || adv == 0) break; } td->td_ma = prev_td_ma; td->td_ma_cnt = prev_td_ma_cnt; curthread_pflags_restore(saveheld); out: vm_fault_enable_pagefaults(save); free(uio_clone, M_IOV); return (error); } static int vn_io_fault(struct file *fp, struct uio *uio, struct ucred *active_cred, int flags, struct thread *td) { fo_rdwr_t *doio; struct vnode *vp; void *rl_cookie; struct vn_io_fault_args args; int error; doio = uio->uio_rw == UIO_READ ? vn_read : vn_write; vp = fp->f_vnode; foffset_lock_uio(fp, uio, flags); if (do_vn_io_fault(vp, uio)) { args.kind = VN_IO_FAULT_FOP; args.args.fop_args.fp = fp; args.args.fop_args.doio = doio; args.cred = active_cred; args.flags = flags | FOF_OFFSET; if (uio->uio_rw == UIO_READ) { rl_cookie = vn_rangelock_rlock(vp, uio->uio_offset, uio->uio_offset + uio->uio_resid); } else if ((fp->f_flag & O_APPEND) != 0 || (flags & FOF_OFFSET) == 0) { /* For appenders, punt and lock the whole range. */ rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX); } else { rl_cookie = vn_rangelock_wlock(vp, uio->uio_offset, uio->uio_offset + uio->uio_resid); } error = vn_io_fault1(vp, uio, &args, td); vn_rangelock_unlock(vp, rl_cookie); } else { error = doio(fp, uio, active_cred, flags | FOF_OFFSET, td); } foffset_unlock_uio(fp, uio, flags); return (error); } /* * Helper function to perform the requested uiomove operation using * the held pages for io->uio_iov[0].iov_base buffer instead of * copyin/copyout. Access to the pages with uiomove_fromphys() * instead of iov_base prevents page faults that could occur due to * pmap_collect() invalidating the mapping created by * vm_fault_quick_hold_pages(), or pageout daemon, page laundry or * object cleanup revoking the write access from page mappings. * * Filesystems specified MNTK_NO_IOPF shall use vn_io_fault_uiomove() * instead of plain uiomove(). */ int vn_io_fault_uiomove(char *data, int xfersize, struct uio *uio) { struct uio transp_uio; struct iovec transp_iov[1]; struct thread *td; size_t adv; int error, pgadv; td = curthread; if ((td->td_pflags & TDP_UIOHELD) == 0 || uio->uio_segflg != UIO_USERSPACE) return (uiomove(data, xfersize, uio)); KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt)); transp_iov[0].iov_base = data; transp_uio.uio_iov = &transp_iov[0]; transp_uio.uio_iovcnt = 1; if (xfersize > uio->uio_resid) xfersize = uio->uio_resid; transp_uio.uio_resid = transp_iov[0].iov_len = xfersize; transp_uio.uio_offset = 0; transp_uio.uio_segflg = UIO_SYSSPACE; /* * Since transp_iov points to data, and td_ma page array * corresponds to original uio->uio_iov, we need to invert the * direction of the i/o operation as passed to * uiomove_fromphys(). */ switch (uio->uio_rw) { case UIO_WRITE: transp_uio.uio_rw = UIO_READ; break; case UIO_READ: transp_uio.uio_rw = UIO_WRITE; break; } transp_uio.uio_td = uio->uio_td; error = uiomove_fromphys(td->td_ma, ((vm_offset_t)uio->uio_iov->iov_base) & PAGE_MASK, xfersize, &transp_uio); adv = xfersize - transp_uio.uio_resid; pgadv = (((vm_offset_t)uio->uio_iov->iov_base + adv) >> PAGE_SHIFT) - (((vm_offset_t)uio->uio_iov->iov_base) >> PAGE_SHIFT); td->td_ma += pgadv; KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt, pgadv)); td->td_ma_cnt -= pgadv; uio->uio_iov->iov_base = (char *)uio->uio_iov->iov_base + adv; uio->uio_iov->iov_len -= adv; uio->uio_resid -= adv; uio->uio_offset += adv; return (error); } int vn_io_fault_pgmove(vm_page_t ma[], vm_offset_t offset, int xfersize, struct uio *uio) { struct thread *td; vm_offset_t iov_base; int cnt, pgadv; td = curthread; if ((td->td_pflags & TDP_UIOHELD) == 0 || uio->uio_segflg != UIO_USERSPACE) return (uiomove_fromphys(ma, offset, xfersize, uio)); KASSERT(uio->uio_iovcnt == 1, ("uio_iovcnt %d", uio->uio_iovcnt)); cnt = xfersize > uio->uio_resid ? uio->uio_resid : xfersize; iov_base = (vm_offset_t)uio->uio_iov->iov_base; switch (uio->uio_rw) { case UIO_WRITE: pmap_copy_pages(td->td_ma, iov_base & PAGE_MASK, ma, offset, cnt); break; case UIO_READ: pmap_copy_pages(ma, offset, td->td_ma, iov_base & PAGE_MASK, cnt); break; } pgadv = ((iov_base + cnt) >> PAGE_SHIFT) - (iov_base >> PAGE_SHIFT); td->td_ma += pgadv; KASSERT(td->td_ma_cnt >= pgadv, ("consumed pages %d %d", td->td_ma_cnt, pgadv)); td->td_ma_cnt -= pgadv; uio->uio_iov->iov_base = (char *)(iov_base + cnt); uio->uio_iov->iov_len -= cnt; uio->uio_resid -= cnt; uio->uio_offset += cnt; return (0); } /* * File table truncate routine. */ static int vn_truncate(struct file *fp, off_t length, struct ucred *active_cred, struct thread *td) { struct vattr vattr; struct mount *mp; struct vnode *vp; void *rl_cookie; int error; vp = fp->f_vnode; /* * Lock the whole range for truncation. Otherwise split i/o * might happen partly before and partly after the truncation. */ rl_cookie = vn_rangelock_wlock(vp, 0, OFF_MAX); error = vn_start_write(vp, &mp, V_WAIT | PCATCH); if (error) goto out1; vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); if (vp->v_type == VDIR) { error = EISDIR; goto out; } #ifdef MAC error = mac_vnode_check_write(active_cred, fp->f_cred, vp); if (error) goto out; #endif error = vn_writechk(vp); if (error == 0) { VATTR_NULL(&vattr); vattr.va_size = length; error = VOP_SETATTR(vp, &vattr, fp->f_cred); } out: VOP_UNLOCK(vp, 0); vn_finished_write(mp); out1: vn_rangelock_unlock(vp, rl_cookie); return (error); } /* * File table vnode stat routine. */ static int vn_statfile(fp, sb, active_cred, td) struct file *fp; struct stat *sb; struct ucred *active_cred; struct thread *td; { struct vnode *vp = fp->f_vnode; int error; vn_lock(vp, LK_SHARED | LK_RETRY); error = vn_stat(vp, sb, active_cred, fp->f_cred, td); VOP_UNLOCK(vp, 0); return (error); } /* * Stat a vnode; implementation for the stat syscall */ int vn_stat(vp, sb, active_cred, file_cred, td) struct vnode *vp; register struct stat *sb; struct ucred *active_cred; struct ucred *file_cred; struct thread *td; { struct vattr vattr; register struct vattr *vap; int error; u_short mode; #ifdef MAC error = mac_vnode_check_stat(active_cred, file_cred, vp); if (error) return (error); #endif vap = &vattr; /* * Initialize defaults for new and unusual fields, so that file * systems which don't support these fields don't need to know * about them. */ vap->va_birthtime.tv_sec = -1; vap->va_birthtime.tv_nsec = 0; vap->va_fsid = VNOVAL; vap->va_rdev = NODEV; error = VOP_GETATTR(vp, vap, active_cred); if (error) return (error); /* * Zero the spare stat fields */ bzero(sb, sizeof *sb); /* * Copy from vattr table */ if (vap->va_fsid != VNOVAL) sb->st_dev = vap->va_fsid; else sb->st_dev = vp->v_mount->mnt_stat.f_fsid.val[0]; sb->st_ino = vap->va_fileid; mode = vap->va_mode; switch (vap->va_type) { case VREG: mode |= S_IFREG; break; case VDIR: mode |= S_IFDIR; break; case VBLK: mode |= S_IFBLK; break; case VCHR: mode |= S_IFCHR; break; case VLNK: mode |= S_IFLNK; break; case VSOCK: mode |= S_IFSOCK; break; case VFIFO: mode |= S_IFIFO; break; default: return (EBADF); }; sb->st_mode = mode; sb->st_nlink = vap->va_nlink; sb->st_uid = vap->va_uid; sb->st_gid = vap->va_gid; sb->st_rdev = vap->va_rdev; if (vap->va_size > OFF_MAX) return (EOVERFLOW); sb->st_size = vap->va_size; sb->st_atim = vap->va_atime; sb->st_mtim = vap->va_mtime; sb->st_ctim = vap->va_ctime; sb->st_birthtim = vap->va_birthtime; /* * According to www.opengroup.org, the meaning of st_blksize is * "a filesystem-specific preferred I/O block size for this * object. In some filesystem types, this may vary from file * to file" * Use miminum/default of PAGE_SIZE (e.g. for VCHR). */ sb->st_blksize = max(PAGE_SIZE, vap->va_blocksize); sb->st_flags = vap->va_flags; if (priv_check(td, PRIV_VFS_GENERATION)) sb->st_gen = 0; else sb->st_gen = vap->va_gen; sb->st_blocks = vap->va_bytes / S_BLKSIZE; return (0); } /* * File table vnode ioctl routine. */ static int vn_ioctl(fp, com, data, active_cred, td) struct file *fp; u_long com; void *data; struct ucred *active_cred; struct thread *td; { struct vattr vattr; struct vnode *vp; int error; vp = fp->f_vnode; switch (vp->v_type) { case VDIR: case VREG: switch (com) { case FIONREAD: vn_lock(vp, LK_SHARED | LK_RETRY); error = VOP_GETATTR(vp, &vattr, active_cred); VOP_UNLOCK(vp, 0); if (error == 0) *(int *)data = vattr.va_size - fp->f_offset; return (error); case FIONBIO: case FIOASYNC: return (0); default: return (VOP_IOCTL(vp, com, data, fp->f_flag, active_cred, td)); } default: return (ENOTTY); } } /* * File table vnode poll routine. */ static int vn_poll(fp, events, active_cred, td) struct file *fp; int events; struct ucred *active_cred; struct thread *td; { struct vnode *vp; int error; vp = fp->f_vnode; #ifdef MAC vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); error = mac_vnode_check_poll(active_cred, fp->f_cred, vp); VOP_UNLOCK(vp, 0); if (!error) #endif error = VOP_POLL(vp, events, fp->f_cred, td); return (error); } /* * Acquire the requested lock and then check for validity. LK_RETRY * permits vn_lock to return doomed vnodes. */ int _vn_lock(struct vnode *vp, int flags, char *file, int line) { int error; VNASSERT((flags & LK_TYPE_MASK) != 0, vp, ("vn_lock called with no locktype.")); do { #ifdef DEBUG_VFS_LOCKS KASSERT(vp->v_holdcnt != 0, ("vn_lock %p: zero hold count", vp)); #endif error = VOP_LOCK1(vp, flags, file, line); flags &= ~LK_INTERLOCK; /* Interlock is always dropped. */ KASSERT((flags & LK_RETRY) == 0 || error == 0, ("LK_RETRY set with incompatible flags (0x%x) or an error occured (%d)", flags, error)); /* * Callers specify LK_RETRY if they wish to get dead vnodes. * If RETRY is not set, we return ENOENT instead. */ if (error == 0 && vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0) { VOP_UNLOCK(vp, 0); error = ENOENT; break; } } while (flags & LK_RETRY && error != 0); return (error); } /* * File table vnode close routine. */ static int vn_closefile(fp, td) struct file *fp; struct thread *td; { struct vnode *vp; struct flock lf; int error; vp = fp->f_vnode; fp->f_ops = &badfileops; if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK) vref(vp); error = vn_close(vp, fp->f_flag, fp->f_cred, td); if (fp->f_type == DTYPE_VNODE && fp->f_flag & FHASLOCK) { lf.l_whence = SEEK_SET; lf.l_start = 0; lf.l_len = 0; lf.l_type = F_UNLCK; (void) VOP_ADVLOCK(vp, fp, F_UNLCK, &lf, F_FLOCK); vrele(vp); } return (error); } static bool vn_suspendable_mp(struct mount *mp) { return ((mp->mnt_kern_flag & MNTK_SUSPENDABLE) != 0); } static bool vn_suspendable(struct vnode *vp, struct mount **mpp) { if (vp != NULL) *mpp = vp->v_mount; if (*mpp == NULL) return (false); return (vn_suspendable_mp(*mpp)); } /* * Preparing to start a filesystem write operation. If the operation is * permitted, then we bump the count of operations in progress and * proceed. If a suspend request is in progress, we wait until the * suspension is over, and then proceed. */ static int vn_start_write_locked(struct mount *mp, int flags) { int error, mflags; mtx_assert(MNT_MTX(mp), MA_OWNED); error = 0; /* * Check on status of suspension. */ if ((curthread->td_pflags & TDP_IGNSUSP) == 0 || mp->mnt_susp_owner != curthread) { mflags = ((mp->mnt_vfc->vfc_flags & VFCF_SBDRY) != 0 ? (flags & PCATCH) : 0) | (PUSER - 1); while ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) { if (flags & V_NOWAIT) { error = EWOULDBLOCK; goto unlock; } error = msleep(&mp->mnt_flag, MNT_MTX(mp), mflags, "suspfs", 0); if (error) goto unlock; } } if (flags & V_XSLEEP) goto unlock; mp->mnt_writeopcount++; unlock: if (error != 0 || (flags & V_XSLEEP) != 0) MNT_REL(mp); MNT_IUNLOCK(mp); return (error); } int vn_start_write(vp, mpp, flags) struct vnode *vp; struct mount **mpp; int flags; { struct mount *mp; int error; if (!vn_suspendable(vp, mpp)) return (0); error = 0; /* * If a vnode is provided, get and return the mount point that * to which it will write. */ if (vp != NULL) { if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) { *mpp = NULL; if (error != EOPNOTSUPP) return (error); return (0); } } if ((mp = *mpp) == NULL) return (0); /* * VOP_GETWRITEMOUNT() returns with the mp refcount held through * a vfs_ref(). * As long as a vnode is not provided we need to acquire a * refcount for the provided mountpoint too, in order to * emulate a vfs_ref(). */ MNT_ILOCK(mp); if (vp == NULL) MNT_REF(mp); return (vn_start_write_locked(mp, flags)); } /* * Secondary suspension. Used by operations such as vop_inactive * routines that are needed by the higher level functions. These * are allowed to proceed until all the higher level functions have * completed (indicated by mnt_writeopcount dropping to zero). At that * time, these operations are halted until the suspension is over. */ int vn_start_secondary_write(vp, mpp, flags) struct vnode *vp; struct mount **mpp; int flags; { struct mount *mp; int error; if (!vn_suspendable(vp, mpp)) return (0); retry: if (vp != NULL) { if ((error = VOP_GETWRITEMOUNT(vp, mpp)) != 0) { *mpp = NULL; if (error != EOPNOTSUPP) return (error); return (0); } } /* * If we are not suspended or have not yet reached suspended * mode, then let the operation proceed. */ if ((mp = *mpp) == NULL) return (0); /* * VOP_GETWRITEMOUNT() returns with the mp refcount held through * a vfs_ref(). * As long as a vnode is not provided we need to acquire a * refcount for the provided mountpoint too, in order to * emulate a vfs_ref(). */ MNT_ILOCK(mp); if (vp == NULL) MNT_REF(mp); if ((mp->mnt_kern_flag & (MNTK_SUSPENDED | MNTK_SUSPEND2)) == 0) { mp->mnt_secondary_writes++; mp->mnt_secondary_accwrites++; MNT_IUNLOCK(mp); return (0); } if (flags & V_NOWAIT) { MNT_REL(mp); MNT_IUNLOCK(mp); return (EWOULDBLOCK); } /* * Wait for the suspension to finish. */ error = msleep(&mp->mnt_flag, MNT_MTX(mp), (PUSER - 1) | PDROP | ((mp->mnt_vfc->vfc_flags & VFCF_SBDRY) != 0 ? (flags & PCATCH) : 0), "suspfs", 0); vfs_rel(mp); if (error == 0) goto retry; return (error); } /* * Filesystem write operation has completed. If we are suspending and this * operation is the last one, notify the suspender that the suspension is * now in effect. */ void vn_finished_write(mp) struct mount *mp; { if (mp == NULL || !vn_suspendable_mp(mp)) return; MNT_ILOCK(mp); MNT_REL(mp); mp->mnt_writeopcount--; if (mp->mnt_writeopcount < 0) panic("vn_finished_write: neg cnt"); if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 && mp->mnt_writeopcount <= 0) wakeup(&mp->mnt_writeopcount); MNT_IUNLOCK(mp); } /* * Filesystem secondary write operation has completed. If we are * suspending and this operation is the last one, notify the suspender * that the suspension is now in effect. */ void vn_finished_secondary_write(mp) struct mount *mp; { if (mp == NULL || !vn_suspendable_mp(mp)) return; MNT_ILOCK(mp); MNT_REL(mp); mp->mnt_secondary_writes--; if (mp->mnt_secondary_writes < 0) panic("vn_finished_secondary_write: neg cnt"); if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0 && mp->mnt_secondary_writes <= 0) wakeup(&mp->mnt_secondary_writes); MNT_IUNLOCK(mp); } /* * Request a filesystem to suspend write operations. */ int vfs_write_suspend(struct mount *mp, int flags) { int error; MPASS(vn_suspendable_mp(mp)); MNT_ILOCK(mp); if (mp->mnt_susp_owner == curthread) { MNT_IUNLOCK(mp); return (EALREADY); } while (mp->mnt_kern_flag & MNTK_SUSPEND) msleep(&mp->mnt_flag, MNT_MTX(mp), PUSER - 1, "wsuspfs", 0); /* * Unmount holds a write reference on the mount point. If we * own busy reference and drain for writers, we deadlock with * the reference draining in the unmount path. Callers of * vfs_write_suspend() must specify VS_SKIP_UNMOUNT if * vfs_busy() reference is owned and caller is not in the * unmount context. */ if ((flags & VS_SKIP_UNMOUNT) != 0 && (mp->mnt_kern_flag & MNTK_UNMOUNT) != 0) { MNT_IUNLOCK(mp); return (EBUSY); } mp->mnt_kern_flag |= MNTK_SUSPEND; mp->mnt_susp_owner = curthread; if (mp->mnt_writeopcount > 0) (void) msleep(&mp->mnt_writeopcount, MNT_MTX(mp), (PUSER - 1)|PDROP, "suspwt", 0); else MNT_IUNLOCK(mp); if ((error = VFS_SYNC(mp, MNT_SUSPEND)) != 0) vfs_write_resume(mp, 0); return (error); } /* * Request a filesystem to resume write operations. */ void vfs_write_resume(struct mount *mp, int flags) { MPASS(vn_suspendable_mp(mp)); MNT_ILOCK(mp); if ((mp->mnt_kern_flag & MNTK_SUSPEND) != 0) { KASSERT(mp->mnt_susp_owner == curthread, ("mnt_susp_owner")); mp->mnt_kern_flag &= ~(MNTK_SUSPEND | MNTK_SUSPEND2 | MNTK_SUSPENDED); mp->mnt_susp_owner = NULL; wakeup(&mp->mnt_writeopcount); wakeup(&mp->mnt_flag); curthread->td_pflags &= ~TDP_IGNSUSP; if ((flags & VR_START_WRITE) != 0) { MNT_REF(mp); mp->mnt_writeopcount++; } MNT_IUNLOCK(mp); if ((flags & VR_NO_SUSPCLR) == 0) VFS_SUSP_CLEAN(mp); } else if ((flags & VR_START_WRITE) != 0) { MNT_REF(mp); vn_start_write_locked(mp, 0); } else { MNT_IUNLOCK(mp); } } /* * Helper loop around vfs_write_suspend() for filesystem unmount VFS * methods. */ int vfs_write_suspend_umnt(struct mount *mp) { int error; MPASS(vn_suspendable_mp(mp)); KASSERT((curthread->td_pflags & TDP_IGNSUSP) == 0, ("vfs_write_suspend_umnt: recursed")); /* dounmount() already called vn_start_write(). */ for (;;) { vn_finished_write(mp); error = vfs_write_suspend(mp, 0); if (error != 0) { vn_start_write(NULL, &mp, V_WAIT); return (error); } MNT_ILOCK(mp); if ((mp->mnt_kern_flag & MNTK_SUSPENDED) != 0) break; MNT_IUNLOCK(mp); vn_start_write(NULL, &mp, V_WAIT); } mp->mnt_kern_flag &= ~(MNTK_SUSPENDED | MNTK_SUSPEND2); wakeup(&mp->mnt_flag); MNT_IUNLOCK(mp); curthread->td_pflags |= TDP_IGNSUSP; return (0); } /* * Implement kqueues for files by translating it to vnode operation. */ static int vn_kqfilter(struct file *fp, struct knote *kn) { return (VOP_KQFILTER(fp->f_vnode, kn)); } /* * Simplified in-kernel wrapper calls for extended attribute access. * Both calls pass in a NULL credential, authorizing as "kernel" access. * Set IO_NODELOCKED in ioflg if the vnode is already locked. */ int vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace, const char *attrname, int *buflen, char *buf, struct thread *td) { struct uio auio; struct iovec iov; int error; iov.iov_len = *buflen; iov.iov_base = buf; auio.uio_iov = &iov; auio.uio_iovcnt = 1; auio.uio_rw = UIO_READ; auio.uio_segflg = UIO_SYSSPACE; auio.uio_td = td; auio.uio_offset = 0; auio.uio_resid = *buflen; if ((ioflg & IO_NODELOCKED) == 0) vn_lock(vp, LK_SHARED | LK_RETRY); ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); /* authorize attribute retrieval as kernel */ error = VOP_GETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, NULL, td); if ((ioflg & IO_NODELOCKED) == 0) VOP_UNLOCK(vp, 0); if (error == 0) { *buflen = *buflen - auio.uio_resid; } return (error); } /* * XXX failure mode if partially written? */ int vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace, const char *attrname, int buflen, char *buf, struct thread *td) { struct uio auio; struct iovec iov; struct mount *mp; int error; iov.iov_len = buflen; iov.iov_base = buf; auio.uio_iov = &iov; auio.uio_iovcnt = 1; auio.uio_rw = UIO_WRITE; auio.uio_segflg = UIO_SYSSPACE; auio.uio_td = td; auio.uio_offset = 0; auio.uio_resid = buflen; if ((ioflg & IO_NODELOCKED) == 0) { if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0) return (error); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); } ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); /* authorize attribute setting as kernel */ error = VOP_SETEXTATTR(vp, attrnamespace, attrname, &auio, NULL, td); if ((ioflg & IO_NODELOCKED) == 0) { vn_finished_write(mp); VOP_UNLOCK(vp, 0); } return (error); } int vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace, const char *attrname, struct thread *td) { struct mount *mp; int error; if ((ioflg & IO_NODELOCKED) == 0) { if ((error = vn_start_write(vp, &mp, V_WAIT)) != 0) return (error); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); } ASSERT_VOP_LOCKED(vp, "IO_NODELOCKED with no vp lock held"); /* authorize attribute removal as kernel */ error = VOP_DELETEEXTATTR(vp, attrnamespace, attrname, NULL, td); if (error == EOPNOTSUPP) error = VOP_SETEXTATTR(vp, attrnamespace, attrname, NULL, NULL, td); if ((ioflg & IO_NODELOCKED) == 0) { vn_finished_write(mp); VOP_UNLOCK(vp, 0); } return (error); } static int vn_get_ino_alloc_vget(struct mount *mp, void *arg, int lkflags, struct vnode **rvp) { return (VFS_VGET(mp, *(ino_t *)arg, lkflags, rvp)); } int vn_vget_ino(struct vnode *vp, ino_t ino, int lkflags, struct vnode **rvp) { return (vn_vget_ino_gen(vp, vn_get_ino_alloc_vget, &ino, lkflags, rvp)); } int vn_vget_ino_gen(struct vnode *vp, vn_get_ino_t alloc, void *alloc_arg, int lkflags, struct vnode **rvp) { struct mount *mp; int ltype, error; ASSERT_VOP_LOCKED(vp, "vn_vget_ino_get"); mp = vp->v_mount; ltype = VOP_ISLOCKED(vp); KASSERT(ltype == LK_EXCLUSIVE || ltype == LK_SHARED, ("vn_vget_ino: vp not locked")); error = vfs_busy(mp, MBF_NOWAIT); if (error != 0) { vfs_ref(mp); VOP_UNLOCK(vp, 0); error = vfs_busy(mp, 0); vn_lock(vp, ltype | LK_RETRY); vfs_rel(mp); if (error != 0) return (ENOENT); if (vp->v_iflag & VI_DOOMED) { vfs_unbusy(mp); return (ENOENT); } } VOP_UNLOCK(vp, 0); error = alloc(mp, alloc_arg, lkflags, rvp); vfs_unbusy(mp); if (*rvp != vp) vn_lock(vp, ltype | LK_RETRY); if (vp->v_iflag & VI_DOOMED) { if (error == 0) { if (*rvp == vp) vunref(vp); else vput(*rvp); } error = ENOENT; } return (error); } int vn_rlimit_fsize(const struct vnode *vp, const struct uio *uio, const struct thread *td) { if (vp->v_type != VREG || td == NULL) return (0); PROC_LOCK(td->td_proc); if ((uoff_t)uio->uio_offset + uio->uio_resid > lim_cur(td->td_proc, RLIMIT_FSIZE)) { kern_psignal(td->td_proc, SIGXFSZ); PROC_UNLOCK(td->td_proc); return (EFBIG); } PROC_UNLOCK(td->td_proc); return (0); } int vn_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, struct thread *td) { struct vnode *vp; vp = fp->f_vnode; #ifdef AUDIT vn_lock(vp, LK_SHARED | LK_RETRY); AUDIT_ARG_VNODE1(vp); VOP_UNLOCK(vp, 0); #endif return (setfmode(td, active_cred, vp, mode)); } int vn_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, struct thread *td) { struct vnode *vp; vp = fp->f_vnode; #ifdef AUDIT vn_lock(vp, LK_SHARED | LK_RETRY); AUDIT_ARG_VNODE1(vp); VOP_UNLOCK(vp, 0); #endif return (setfown(td, active_cred, vp, uid, gid)); } void vn_pages_remove(struct vnode *vp, vm_pindex_t start, vm_pindex_t end) { vm_object_t object; if ((object = vp->v_object) == NULL) return; VM_OBJECT_WLOCK(object); vm_object_page_remove(object, start, end, 0); VM_OBJECT_WUNLOCK(object); } int vn_bmap_seekhole(struct vnode *vp, u_long cmd, off_t *off, struct ucred *cred) { struct vattr va; daddr_t bn, bnp; uint64_t bsize; off_t noff; int error; KASSERT(cmd == FIOSEEKHOLE || cmd == FIOSEEKDATA, ("Wrong command %lu", cmd)); if (vn_lock(vp, LK_SHARED) != 0) return (EBADF); if (vp->v_type != VREG) { error = ENOTTY; goto unlock; } error = VOP_GETATTR(vp, &va, cred); if (error != 0) goto unlock; noff = *off; if (noff >= va.va_size) { error = ENXIO; goto unlock; } bsize = vp->v_mount->mnt_stat.f_iosize; for (bn = noff / bsize; noff < va.va_size; bn++, noff += bsize) { error = VOP_BMAP(vp, bn, NULL, &bnp, NULL, NULL); if (error == EOPNOTSUPP) { error = ENOTTY; goto unlock; } if ((bnp == -1 && cmd == FIOSEEKHOLE) || (bnp != -1 && cmd == FIOSEEKDATA)) { noff = bn * bsize; if (noff < *off) noff = *off; goto unlock; } } if (noff > va.va_size) noff = va.va_size; /* noff == va.va_size. There is an implicit hole at the end of file. */ if (cmd == FIOSEEKDATA) error = ENXIO; unlock: VOP_UNLOCK(vp, 0); if (error == 0) *off = noff; return (error); } int vn_seek(struct file *fp, off_t offset, int whence, struct thread *td) { struct ucred *cred; struct vnode *vp; struct vattr vattr; off_t foffset, size; int error, noneg; cred = td->td_ucred; vp = fp->f_vnode; foffset = foffset_lock(fp, 0); noneg = (vp->v_type != VCHR); error = 0; switch (whence) { case L_INCR: if (noneg && (foffset < 0 || (offset > 0 && foffset > OFF_MAX - offset))) { error = EOVERFLOW; break; } offset += foffset; break; case L_XTND: vn_lock(vp, LK_SHARED | LK_RETRY); error = VOP_GETATTR(vp, &vattr, cred); VOP_UNLOCK(vp, 0); if (error) break; /* * If the file references a disk device, then fetch * the media size and use that to determine the ending * offset. */ if (vattr.va_size == 0 && vp->v_type == VCHR && fo_ioctl(fp, DIOCGMEDIASIZE, &size, cred, td) == 0) vattr.va_size = size; if (noneg && (vattr.va_size > OFF_MAX || (offset > 0 && vattr.va_size > OFF_MAX - offset))) { error = EOVERFLOW; break; } offset += vattr.va_size; break; case L_SET: break; case SEEK_DATA: error = fo_ioctl(fp, FIOSEEKDATA, &offset, cred, td); break; case SEEK_HOLE: error = fo_ioctl(fp, FIOSEEKHOLE, &offset, cred, td); break; default: error = EINVAL; } if (error == 0 && noneg && offset < 0) error = EINVAL; if (error != 0) goto drop; VFS_KNOTE_UNLOCKED(vp, 0); td->td_uretoff.tdu_off = offset; drop: foffset_unlock(fp, offset, error != 0 ? FOF_NOUPDATE : 0); return (error); } int vn_utimes_perm(struct vnode *vp, struct vattr *vap, struct ucred *cred, struct thread *td) { int error; /* * Grant permission if the caller is the owner of the file, or * the super-user, or has ACL_WRITE_ATTRIBUTES permission on * on the file. If the time pointer is null, then write * permission on the file is also sufficient. * * From NFSv4.1, draft 21, 6.2.1.3.1, Discussion of Mask Attributes: * A user having ACL_WRITE_DATA or ACL_WRITE_ATTRIBUTES * will be allowed to set the times [..] to the current * server time. */ error = VOP_ACCESSX(vp, VWRITE_ATTRIBUTES, cred, td); if (error != 0 && (vap->va_vaflags & VA_UTIMES_NULL) != 0) error = VOP_ACCESS(vp, VWRITE, cred, td); return (error); } int vn_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp) { struct vnode *vp; int error; if (fp->f_type == DTYPE_FIFO) kif->kf_type = KF_TYPE_FIFO; else kif->kf_type = KF_TYPE_VNODE; vp = fp->f_vnode; vref(vp); FILEDESC_SUNLOCK(fdp); error = vn_fill_kinfo_vnode(vp, kif); vrele(vp); FILEDESC_SLOCK(fdp); return (error); } int vn_fill_kinfo_vnode(struct vnode *vp, struct kinfo_file *kif) { struct vattr va; char *fullpath, *freepath; int error; kif->kf_vnode_type = vntype_to_kinfo(vp->v_type); freepath = NULL; fullpath = "-"; error = vn_fullpath(curthread, vp, &fullpath, &freepath); if (error == 0) { strlcpy(kif->kf_path, fullpath, sizeof(kif->kf_path)); } if (freepath != NULL) free(freepath, M_TEMP); /* * Retrieve vnode attributes. */ va.va_fsid = VNOVAL; va.va_rdev = NODEV; vn_lock(vp, LK_SHARED | LK_RETRY); error = VOP_GETATTR(vp, &va, curthread->td_ucred); VOP_UNLOCK(vp, 0); if (error != 0) return (error); if (va.va_fsid != VNOVAL) kif->kf_un.kf_file.kf_file_fsid = va.va_fsid; else kif->kf_un.kf_file.kf_file_fsid = vp->v_mount->mnt_stat.f_fsid.val[0]; kif->kf_un.kf_file.kf_file_fileid = va.va_fileid; kif->kf_un.kf_file.kf_file_mode = MAKEIMODE(va.va_type, va.va_mode); kif->kf_un.kf_file.kf_file_size = va.va_size; kif->kf_un.kf_file.kf_file_rdev = va.va_rdev; return (0); } Index: head/sys/sys/fcntl.h =================================================================== --- head/sys/sys/fcntl.h (revision 281844) +++ head/sys/sys/fcntl.h (revision 281845) @@ -1,321 +1,325 @@ /*- * Copyright (c) 1983, 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. * 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. * * @(#)fcntl.h 8.3 (Berkeley) 1/21/94 * $FreeBSD$ */ #ifndef _SYS_FCNTL_H_ #define _SYS_FCNTL_H_ /* * This file includes the definitions for open and fcntl * described by POSIX for ; it also includes * related kernel definitions. */ #include #include #ifndef _MODE_T_DECLARED typedef __mode_t mode_t; #define _MODE_T_DECLARED #endif #ifndef _OFF_T_DECLARED typedef __off_t off_t; #define _OFF_T_DECLARED #endif #ifndef _PID_T_DECLARED typedef __pid_t pid_t; #define _PID_T_DECLARED #endif /* * File status flags: these are used by open(2), fcntl(2). * They are also used (indirectly) in the kernel file structure f_flags, * which is a superset of the open/fcntl flags. Open flags and f_flags * are inter-convertible using OFLAGS(fflags) and FFLAGS(oflags). * Open/fcntl flags begin with O_; kernel-internal flags begin with F. */ /* open-only flags */ #define O_RDONLY 0x0000 /* open for reading only */ #define O_WRONLY 0x0001 /* open for writing only */ #define O_RDWR 0x0002 /* open for reading and writing */ #define O_ACCMODE 0x0003 /* mask for above modes */ /* * Kernel encoding of open mode; separate read and write bits that are * independently testable: 1 greater than the above. * * XXX * FREAD and FWRITE are excluded from the #ifdef _KERNEL so that TIOCFLUSH, * which was documented to use FREAD/FWRITE, continues to work. */ #if __BSD_VISIBLE #define FREAD 0x0001 #define FWRITE 0x0002 #endif #define O_NONBLOCK 0x0004 /* no delay */ #define O_APPEND 0x0008 /* set append mode */ #if __BSD_VISIBLE #define O_SHLOCK 0x0010 /* open with shared file lock */ #define O_EXLOCK 0x0020 /* open with exclusive file lock */ #define O_ASYNC 0x0040 /* signal pgrp when data ready */ #define O_FSYNC 0x0080 /* synchronous writes */ #endif #define O_SYNC 0x0080 /* POSIX synonym for O_FSYNC */ #if __POSIX_VISIBLE >= 200809 #define O_NOFOLLOW 0x0100 /* don't follow symlinks */ #endif #define O_CREAT 0x0200 /* create if nonexistent */ #define O_TRUNC 0x0400 /* truncate to zero length */ #define O_EXCL 0x0800 /* error if already exists */ #ifdef _KERNEL #define FHASLOCK 0x4000 /* descriptor holds advisory lock */ #endif /* Defined by POSIX 1003.1; BSD default, but must be distinct from O_RDONLY. */ #define O_NOCTTY 0x8000 /* don't assign controlling terminal */ #if __BSD_VISIBLE /* Attempt to bypass buffer cache */ #define O_DIRECT 0x00010000 #endif #if __POSIX_VISIBLE >= 200809 #define O_DIRECTORY 0x00020000 /* Fail if not directory */ #define O_EXEC 0x00040000 /* Open for execute only */ #endif #ifdef _KERNEL #define FEXEC O_EXEC #endif #if __POSIX_VISIBLE >= 200809 /* Defined by POSIX 1003.1-2008; BSD default, but reserve for future use. */ #define O_TTY_INIT 0x00080000 /* Restore default termios attributes */ #define O_CLOEXEC 0x00100000 #endif +#if __BSD_VISIBLE +#define O_VERIFY 0x00200000 /* open only after verification */ +#endif + /* * XXX missing O_DSYNC, O_RSYNC. */ #ifdef _KERNEL /* convert from open() flags to/from fflags; convert O_RD/WR to FREAD/FWRITE */ #define FFLAGS(oflags) ((oflags) & O_EXEC ? (oflags) : (oflags) + 1) #define OFLAGS(fflags) ((fflags) & O_EXEC ? (fflags) : (fflags) - 1) /* bits to save after open */ #define FMASK (FREAD|FWRITE|FAPPEND|FASYNC|FFSYNC|FNONBLOCK|O_DIRECT|FEXEC) /* bits settable by fcntl(F_SETFL, ...) */ #define FCNTLFLAGS (FAPPEND|FASYNC|FFSYNC|FNONBLOCK|FRDAHEAD|O_DIRECT) #if defined(COMPAT_FREEBSD7) || defined(COMPAT_FREEBSD6) || \ defined(COMPAT_FREEBSD5) || defined(COMPAT_FREEBSD4) /* * Set by shm_open(3) in older libc's to get automatic MAP_ASYNC * behavior for POSIX shared memory objects (which are otherwise * implemented as plain files). */ #define FPOSIXSHM O_NOFOLLOW #undef FCNTLFLAGS #define FCNTLFLAGS (FAPPEND|FASYNC|FFSYNC|FNONBLOCK|FPOSIXSHM|FRDAHEAD| \ O_DIRECT) #endif #endif /* * The O_* flags used to have only F* names, which were used in the kernel * and by fcntl. We retain the F* names for the kernel f_flag field * and for backward compatibility for fcntl. These flags are deprecated. */ #if __BSD_VISIBLE #define FAPPEND O_APPEND /* kernel/compat */ #define FASYNC O_ASYNC /* kernel/compat */ #define FFSYNC O_FSYNC /* kernel */ #define FNONBLOCK O_NONBLOCK /* kernel */ #define FNDELAY O_NONBLOCK /* compat */ #define O_NDELAY O_NONBLOCK /* compat */ #endif /* * We are out of bits in f_flag (which is a short). However, * the flag bits not set in FMASK are only meaningful in the * initial open syscall. Those bits can thus be given a * different meaning for fcntl(2). */ #if __BSD_VISIBLE /* Read ahead */ #define FRDAHEAD O_CREAT #endif #if __POSIX_VISIBLE >= 200809 /* * Magic value that specify the use of the current working directory * to determine the target of relative file paths in the openat() and * similar syscalls. */ #define AT_FDCWD -100 /* * Miscellaneous flags for the *at() syscalls. */ #define AT_EACCESS 0x100 /* Check access using effective user and group ID */ #define AT_SYMLINK_NOFOLLOW 0x200 /* Do not follow symbolic links */ #define AT_SYMLINK_FOLLOW 0x400 /* Follow symbolic link */ #define AT_REMOVEDIR 0x800 /* Remove directory instead of file */ #endif /* * Constants used for fcntl(2) */ /* command values */ #define F_DUPFD 0 /* duplicate file descriptor */ #define F_GETFD 1 /* get file descriptor flags */ #define F_SETFD 2 /* set file descriptor flags */ #define F_GETFL 3 /* get file status flags */ #define F_SETFL 4 /* set file status flags */ #if __XSI_VISIBLE || __POSIX_VISIBLE >= 200112 #define F_GETOWN 5 /* get SIGIO/SIGURG proc/pgrp */ #define F_SETOWN 6 /* set SIGIO/SIGURG proc/pgrp */ #endif #if __BSD_VISIBLE #define F_OGETLK 7 /* get record locking information */ #define F_OSETLK 8 /* set record locking information */ #define F_OSETLKW 9 /* F_SETLK; wait if blocked */ #define F_DUP2FD 10 /* duplicate file descriptor to arg */ #endif #define F_GETLK 11 /* get record locking information */ #define F_SETLK 12 /* set record locking information */ #define F_SETLKW 13 /* F_SETLK; wait if blocked */ #if __BSD_VISIBLE #define F_SETLK_REMOTE 14 /* debugging support for remote locks */ #define F_READAHEAD 15 /* read ahead */ #define F_RDAHEAD 16 /* Darwin compatible read ahead */ #endif #if __POSIX_VISIBLE >= 200809 #define F_DUPFD_CLOEXEC 17 /* Like F_DUPFD, but FD_CLOEXEC is set */ #endif #if __BSD_VISIBLE #define F_DUP2FD_CLOEXEC 18 /* Like F_DUP2FD, but FD_CLOEXEC is set */ #endif /* file descriptor flags (F_GETFD, F_SETFD) */ #define FD_CLOEXEC 1 /* close-on-exec flag */ /* record locking flags (F_GETLK, F_SETLK, F_SETLKW) */ #define F_RDLCK 1 /* shared or read lock */ #define F_UNLCK 2 /* unlock */ #define F_WRLCK 3 /* exclusive or write lock */ #if __BSD_VISIBLE #define F_UNLCKSYS 4 /* purge locks for a given system ID */ #define F_CANCEL 5 /* cancel an async lock request */ #endif #ifdef _KERNEL #define F_WAIT 0x010 /* Wait until lock is granted */ #define F_FLOCK 0x020 /* Use flock(2) semantics for lock */ #define F_POSIX 0x040 /* Use POSIX semantics for lock */ #define F_REMOTE 0x080 /* Lock owner is remote NFS client */ #define F_NOINTR 0x100 /* Ignore signals when waiting */ #endif /* * Advisory file segment locking data type - * information passed to system by user */ struct flock { off_t l_start; /* starting offset */ off_t l_len; /* len = 0 means until end of file */ pid_t l_pid; /* lock owner */ short l_type; /* lock type: read/write, etc. */ short l_whence; /* type of l_start */ int l_sysid; /* remote system id or zero for local */ }; #if __BSD_VISIBLE /* * Old advisory file segment locking data type, * before adding l_sysid. */ struct __oflock { off_t l_start; /* starting offset */ off_t l_len; /* len = 0 means until end of file */ pid_t l_pid; /* lock owner */ short l_type; /* lock type: read/write, etc. */ short l_whence; /* type of l_start */ }; #endif #if __BSD_VISIBLE /* lock operations for flock(2) */ #define LOCK_SH 0x01 /* shared file lock */ #define LOCK_EX 0x02 /* exclusive file lock */ #define LOCK_NB 0x04 /* don't block when locking */ #define LOCK_UN 0x08 /* unlock file */ #endif #if __POSIX_VISIBLE >= 200112 /* * Advice to posix_fadvise */ #define POSIX_FADV_NORMAL 0 /* no special treatment */ #define POSIX_FADV_RANDOM 1 /* expect random page references */ #define POSIX_FADV_SEQUENTIAL 2 /* expect sequential page references */ #define POSIX_FADV_WILLNEED 3 /* will need these pages */ #define POSIX_FADV_DONTNEED 4 /* dont need these pages */ #define POSIX_FADV_NOREUSE 5 /* access data only once */ #endif #ifndef _KERNEL __BEGIN_DECLS int open(const char *, int, ...); int creat(const char *, mode_t); int fcntl(int, int, ...); #if __BSD_VISIBLE int flock(int, int); #endif #if __POSIX_VISIBLE >= 200809 int openat(int, const char *, int, ...); #endif #if __POSIX_VISIBLE >= 200112 int posix_fadvise(int, off_t, off_t, int); int posix_fallocate(int, off_t, off_t); #endif __END_DECLS #endif #endif /* !_SYS_FCNTL_H_ */ Index: head/sys/sys/vnode.h =================================================================== --- head/sys/sys/vnode.h (revision 281844) +++ head/sys/sys/vnode.h (revision 281845) @@ -1,866 +1,868 @@ /*- * Copyright (c) 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. * 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. * * @(#)vnode.h 8.7 (Berkeley) 2/4/94 * $FreeBSD$ */ #ifndef _SYS_VNODE_H_ #define _SYS_VNODE_H_ #include #include #include #include #include #include #include #include #include #include /* * The vnode is the focus of all file activity in UNIX. There is a * unique vnode allocated for each active file, each current directory, * each mounted-on file, text file, and the root. */ /* * Vnode types. VNON means no type. */ enum vtype { VNON, VREG, VDIR, VBLK, VCHR, VLNK, VSOCK, VFIFO, VBAD, VMARKER }; /* * Each underlying filesystem allocates its own private area and hangs * it from v_data. If non-null, this area is freed in getnewvnode(). */ struct namecache; struct vpollinfo { struct mtx vpi_lock; /* lock to protect below */ struct selinfo vpi_selinfo; /* identity of poller(s) */ short vpi_events; /* what they are looking for */ short vpi_revents; /* what has happened */ }; /* * Reading or writing any of these items requires holding the appropriate lock. * * Lock reference: * c - namecache mutex * f - freelist mutex * i - interlock * m - mount point interlock * p - pollinfo lock * u - Only a reference to the vnode is needed to read. * v - vnode lock * * Vnodes may be found on many lists. The general way to deal with operating * on a vnode that is on a list is: * 1) Lock the list and find the vnode. * 2) Lock interlock so that the vnode does not go away. * 3) Unlock the list to avoid lock order reversals. * 4) vget with LK_INTERLOCK and check for ENOENT, or * 5) Check for DOOMED if the vnode lock is not required. * 6) Perform your operation, then vput(). */ #if defined(_KERNEL) || defined(_KVM_VNODE) struct vnode { /* * Fields which define the identity of the vnode. These fields are * owned by the filesystem (XXX: and vgone() ?) */ const char *v_tag; /* u type of underlying data */ struct vop_vector *v_op; /* u vnode operations vector */ void *v_data; /* u private data for fs */ /* * Filesystem instance stuff */ struct mount *v_mount; /* u ptr to vfs we are in */ TAILQ_ENTRY(vnode) v_nmntvnodes; /* m vnodes for mount point */ /* * Type specific fields, only one applies to any given vnode. * See #defines below for renaming to v_* namespace. */ union { struct mount *vu_mount; /* v ptr to mountpoint (VDIR) */ struct socket *vu_socket; /* v unix domain net (VSOCK) */ struct cdev *vu_cdev; /* v device (VCHR, VBLK) */ struct fifoinfo *vu_fifoinfo; /* v fifo (VFIFO) */ } v_un; /* * vfs_hash: (mount + inode) -> vnode hash. The hash value * itself is grouped with other int fields, to avoid padding. */ LIST_ENTRY(vnode) v_hashlist; /* * VFS_namecache stuff */ LIST_HEAD(, namecache) v_cache_src; /* c Cache entries from us */ TAILQ_HEAD(, namecache) v_cache_dst; /* c Cache entries to us */ struct namecache *v_cache_dd; /* c Cache entry for .. vnode */ /* * Locking */ struct lock v_lock; /* u (if fs don't have one) */ struct mtx v_interlock; /* lock for "i" things */ struct lock *v_vnlock; /* u pointer to vnode lock */ /* * The machinery of being a vnode */ TAILQ_ENTRY(vnode) v_actfreelist; /* f vnode active/free lists */ struct bufobj v_bufobj; /* * Buffer cache object */ /* * Hooks for various subsystems and features. */ struct vpollinfo *v_pollinfo; /* i Poll events, p for *v_pi */ struct label *v_label; /* MAC label for vnode */ struct lockf *v_lockf; /* Byte-level advisory lock list */ struct rangelock v_rl; /* Byte-range lock */ /* * clustering stuff */ daddr_t v_cstart; /* v start block of cluster */ daddr_t v_lasta; /* v last allocation */ daddr_t v_lastw; /* v last write */ int v_clen; /* v length of cur. cluster */ int v_holdcnt; /* i prevents recycling. */ int v_usecount; /* i ref count of users */ u_int v_iflag; /* i vnode flags (see below) */ u_int v_vflag; /* v vnode flags */ int v_writecount; /* v ref count of writers */ u_int v_hash; enum vtype v_type; /* u vnode type */ }; #endif /* defined(_KERNEL) || defined(_KVM_VNODE) */ #define v_mountedhere v_un.vu_mount #define v_socket v_un.vu_socket #define v_rdev v_un.vu_cdev #define v_fifoinfo v_un.vu_fifoinfo /* XXX: These are temporary to avoid a source sweep at this time */ #define v_object v_bufobj.bo_object /* * Userland version of struct vnode, for sysctl. */ struct xvnode { size_t xv_size; /* sizeof(struct xvnode) */ void *xv_vnode; /* address of real vnode */ u_long xv_flag; /* vnode vflags */ int xv_usecount; /* reference count of users */ int xv_writecount; /* reference count of writers */ int xv_holdcnt; /* page & buffer references */ u_long xv_id; /* capability identifier */ void *xv_mount; /* address of parent mount */ long xv_numoutput; /* num of writes in progress */ enum vtype xv_type; /* vnode type */ union { void *xvu_socket; /* socket, if VSOCK */ void *xvu_fifo; /* fifo, if VFIFO */ dev_t xvu_rdev; /* maj/min, if VBLK/VCHR */ struct { dev_t xvu_dev; /* device, if VDIR/VREG/VLNK */ ino_t xvu_ino; /* id, if VDIR/VREG/VLNK */ } xv_uns; } xv_un; }; #define xv_socket xv_un.xvu_socket #define xv_fifo xv_un.xvu_fifo #define xv_rdev xv_un.xvu_rdev #define xv_dev xv_un.xv_uns.xvu_dev #define xv_ino xv_un.xv_uns.xvu_ino /* We don't need to lock the knlist */ #define VN_KNLIST_EMPTY(vp) ((vp)->v_pollinfo == NULL || \ KNLIST_EMPTY(&(vp)->v_pollinfo->vpi_selinfo.si_note)) #define VN_KNOTE(vp, b, a) \ do { \ if (!VN_KNLIST_EMPTY(vp)) \ KNOTE(&vp->v_pollinfo->vpi_selinfo.si_note, (b), \ (a) | KNF_NOKQLOCK); \ } while (0) #define VN_KNOTE_LOCKED(vp, b) VN_KNOTE(vp, b, KNF_LISTLOCKED) #define VN_KNOTE_UNLOCKED(vp, b) VN_KNOTE(vp, b, 0) /* * Vnode flags. * VI flags are protected by interlock and live in v_iflag * VV flags are protected by the vnode lock and live in v_vflag * * VI_DOOMED is doubly protected by the interlock and vnode lock. Both * are required for writing but the status may be checked with either. */ #define VI_MOUNT 0x0020 /* Mount in progress */ #define VI_AGE 0x0040 /* Insert vnode at head of free list */ #define VI_DOOMED 0x0080 /* This vnode is being recycled */ #define VI_FREE 0x0100 /* This vnode is on the freelist */ #define VI_ACTIVE 0x0200 /* This vnode is on the active list */ #define VI_DOINGINACT 0x0800 /* VOP_INACTIVE is in progress */ #define VI_OWEINACT 0x1000 /* Need to call inactive */ #define VV_ROOT 0x0001 /* root of its filesystem */ #define VV_ISTTY 0x0002 /* vnode represents a tty */ #define VV_NOSYNC 0x0004 /* unlinked, stop syncing */ #define VV_ETERNALDEV 0x0008 /* device that is never destroyed */ #define VV_CACHEDLABEL 0x0010 /* Vnode has valid cached MAC label */ #define VV_TEXT 0x0020 /* vnode is a pure text prototype */ #define VV_COPYONWRITE 0x0040 /* vnode is doing copy-on-write */ #define VV_SYSTEM 0x0080 /* vnode being used by kernel */ #define VV_PROCDEP 0x0100 /* vnode is process dependent */ #define VV_NOKNOTE 0x0200 /* don't activate knotes on this vnode */ #define VV_DELETED 0x0400 /* should be removed */ #define VV_MD 0x0800 /* vnode backs the md device */ #define VV_FORCEINSMQ 0x1000 /* force the insmntque to succeed */ /* * Vnode attributes. A field value of VNOVAL represents a field whose value * is unavailable (getattr) or which is not to be changed (setattr). */ struct vattr { enum vtype va_type; /* vnode type (for create) */ u_short va_mode; /* files access mode and type */ short va_nlink; /* number of references to file */ uid_t va_uid; /* owner user id */ gid_t va_gid; /* owner group id */ dev_t va_fsid; /* filesystem id */ long va_fileid; /* file id */ u_quad_t va_size; /* file size in bytes */ long va_blocksize; /* blocksize preferred for i/o */ struct timespec va_atime; /* time of last access */ struct timespec va_mtime; /* time of last modification */ struct timespec va_ctime; /* time file changed */ struct timespec va_birthtime; /* time file created */ u_long va_gen; /* generation number of file */ u_long va_flags; /* flags defined for file */ dev_t va_rdev; /* device the special file represents */ u_quad_t va_bytes; /* bytes of disk space held by file */ u_quad_t va_filerev; /* file modification number */ u_int va_vaflags; /* operations flags, see below */ long va_spare; /* remain quad aligned */ }; /* * Flags for va_vaflags. */ #define VA_UTIMES_NULL 0x01 /* utimes argument was NULL */ #define VA_EXCLUSIVE 0x02 /* exclusive create request */ /* * Flags for ioflag. (high 16 bits used to ask for read-ahead and * help with write clustering) * NB: IO_NDELAY and IO_DIRECT are linked to fcntl.h */ #define IO_UNIT 0x0001 /* do I/O as atomic unit */ #define IO_APPEND 0x0002 /* append write to end */ #define IO_NDELAY 0x0004 /* FNDELAY flag set in file table */ #define IO_NODELOCKED 0x0008 /* underlying node already locked */ #define IO_ASYNC 0x0010 /* bawrite rather then bdwrite */ #define IO_VMIO 0x0020 /* data already in VMIO space */ #define IO_INVAL 0x0040 /* invalidate after I/O */ #define IO_SYNC 0x0080 /* do I/O synchronously */ #define IO_DIRECT 0x0100 /* attempt to bypass buffer cache */ #define IO_EXT 0x0400 /* operate on external attributes */ #define IO_NORMAL 0x0800 /* operate on regular data */ #define IO_NOMACCHECK 0x1000 /* MAC checks unnecessary */ #define IO_BUFLOCKED 0x2000 /* ffs flag; indir buf is locked */ #define IO_RANGELOCKED 0x4000 /* range locked */ #define IO_SEQMAX 0x7F /* seq heuristic max value */ #define IO_SEQSHIFT 16 /* seq heuristic in upper 16 bits */ /* * Flags for accmode_t. */ #define VEXEC 000000000100 /* execute/search permission */ #define VWRITE 000000000200 /* write permission */ #define VREAD 000000000400 /* read permission */ #define VADMIN 000000010000 /* being the file owner */ #define VAPPEND 000000040000 /* permission to write/append */ /* * VEXPLICIT_DENY makes VOP_ACCESSX(9) return EPERM or EACCES only * if permission was denied explicitly, by a "deny" rule in NFSv4 ACL, * and 0 otherwise. This never happens with ordinary unix access rights * or POSIX.1e ACLs. Obviously, VEXPLICIT_DENY must be OR-ed with * some other V* constant. */ #define VEXPLICIT_DENY 000000100000 #define VREAD_NAMED_ATTRS 000000200000 /* not used */ #define VWRITE_NAMED_ATTRS 000000400000 /* not used */ #define VDELETE_CHILD 000001000000 #define VREAD_ATTRIBUTES 000002000000 /* permission to stat(2) */ #define VWRITE_ATTRIBUTES 000004000000 /* change {m,c,a}time */ #define VDELETE 000010000000 #define VREAD_ACL 000020000000 /* read ACL and file mode */ #define VWRITE_ACL 000040000000 /* change ACL and/or file mode */ #define VWRITE_OWNER 000100000000 /* change file owner */ #define VSYNCHRONIZE 000200000000 /* not used */ +#define VCREAT 000400000000 /* creating new file */ +#define VVERIFY 001000000000 /* verification required */ /* * Permissions that were traditionally granted only to the file owner. */ #define VADMIN_PERMS (VADMIN | VWRITE_ATTRIBUTES | VWRITE_ACL | \ VWRITE_OWNER) /* * Permissions that were traditionally granted to everyone. */ #define VSTAT_PERMS (VREAD_ATTRIBUTES | VREAD_ACL) /* * Permissions that allow to change the state of the file in any way. */ #define VMODIFY_PERMS (VWRITE | VAPPEND | VADMIN_PERMS | VDELETE_CHILD | \ VDELETE) /* * Token indicating no attribute value yet assigned. */ #define VNOVAL (-1) /* * LK_TIMELOCK timeout for vnode locks (used mainly by the pageout daemon) */ #define VLKTIMEOUT (hz / 20 + 1) #ifdef _KERNEL #ifdef MALLOC_DECLARE MALLOC_DECLARE(M_VNODE); #endif /* * Convert between vnode types and inode formats (since POSIX.1 * defines mode word of stat structure in terms of inode formats). */ extern enum vtype iftovt_tab[]; extern int vttoif_tab[]; #define IFTOVT(mode) (iftovt_tab[((mode) & S_IFMT) >> 12]) #define VTTOIF(indx) (vttoif_tab[(int)(indx)]) #define MAKEIMODE(indx, mode) (int)(VTTOIF(indx) | (mode)) /* * Flags to various vnode functions. */ #define SKIPSYSTEM 0x0001 /* vflush: skip vnodes marked VSYSTEM */ #define FORCECLOSE 0x0002 /* vflush: force file closure */ #define WRITECLOSE 0x0004 /* vflush: only close writable files */ #define EARLYFLUSH 0x0008 /* vflush: early call for ffs_flushfiles */ #define V_SAVE 0x0001 /* vinvalbuf: sync file first */ #define V_ALT 0x0002 /* vinvalbuf: invalidate only alternate bufs */ #define V_NORMAL 0x0004 /* vinvalbuf: invalidate only regular bufs */ #define V_CLEANONLY 0x0008 /* vinvalbuf: invalidate only clean bufs */ #define REVOKEALL 0x0001 /* vop_revoke: revoke all aliases */ #define V_WAIT 0x0001 /* vn_start_write: sleep for suspend */ #define V_NOWAIT 0x0002 /* vn_start_write: don't sleep for suspend */ #define V_XSLEEP 0x0004 /* vn_start_write: just return after sleep */ #define VR_START_WRITE 0x0001 /* vfs_write_resume: start write atomically */ #define VR_NO_SUSPCLR 0x0002 /* vfs_write_resume: do not clear suspension */ #define VS_SKIP_UNMOUNT 0x0001 /* vfs_write_suspend: fail if the filesystem is being unmounted */ #define VREF(vp) vref(vp) #ifdef DIAGNOSTIC #define VATTR_NULL(vap) vattr_null(vap) #else #define VATTR_NULL(vap) (*(vap) = va_null) /* initialize a vattr */ #endif /* DIAGNOSTIC */ #define NULLVP ((struct vnode *)NULL) /* * Global vnode data. */ extern struct vnode *rootvnode; /* root (i.e. "/") vnode */ extern int async_io_version; /* 0 or POSIX version of AIO i'face */ extern int desiredvnodes; /* number of vnodes desired */ extern struct uma_zone *namei_zone; extern struct vattr va_null; /* predefined null vattr structure */ #define VI_LOCK(vp) mtx_lock(&(vp)->v_interlock) #define VI_LOCK_FLAGS(vp, flags) mtx_lock_flags(&(vp)->v_interlock, (flags)) #define VI_TRYLOCK(vp) mtx_trylock(&(vp)->v_interlock) #define VI_UNLOCK(vp) mtx_unlock(&(vp)->v_interlock) #define VI_MTX(vp) (&(vp)->v_interlock) #define VN_LOCK_AREC(vp) lockallowrecurse((vp)->v_vnlock) #define VN_LOCK_ASHARE(vp) lockallowshare((vp)->v_vnlock) #define VN_LOCK_DSHARE(vp) lockdisableshare((vp)->v_vnlock) #endif /* _KERNEL */ /* * Mods for extensibility. */ /* * Flags for vdesc_flags: */ #define VDESC_MAX_VPS 16 /* Low order 16 flag bits are reserved for willrele flags for vp arguments. */ #define VDESC_VP0_WILLRELE 0x0001 #define VDESC_VP1_WILLRELE 0x0002 #define VDESC_VP2_WILLRELE 0x0004 #define VDESC_VP3_WILLRELE 0x0008 #define VDESC_NOMAP_VPP 0x0100 #define VDESC_VPP_WILLRELE 0x0200 /* * A generic structure. * This can be used by bypass routines to identify generic arguments. */ struct vop_generic_args { struct vnodeop_desc *a_desc; /* other random data follows, presumably */ }; typedef int vop_bypass_t(struct vop_generic_args *); /* * VDESC_NO_OFFSET is used to identify the end of the offset list * and in places where no such field exists. */ #define VDESC_NO_OFFSET -1 /* * This structure describes the vnode operation taking place. */ struct vnodeop_desc { char *vdesc_name; /* a readable name for debugging */ int vdesc_flags; /* VDESC_* flags */ vop_bypass_t *vdesc_call; /* Function to call */ /* * These ops are used by bypass routines to map and locate arguments. * Creds and procs are not needed in bypass routines, but sometimes * they are useful to (for example) transport layers. * Nameidata is useful because it has a cred in it. */ int *vdesc_vp_offsets; /* list ended by VDESC_NO_OFFSET */ int vdesc_vpp_offset; /* return vpp location */ int vdesc_cred_offset; /* cred location, if any */ int vdesc_thread_offset; /* thread location, if any */ int vdesc_componentname_offset; /* if any */ }; #ifdef _KERNEL /* * A list of all the operation descs. */ extern struct vnodeop_desc *vnodeop_descs[]; #define VOPARG_OFFSETOF(s_type, field) __offsetof(s_type, field) #define VOPARG_OFFSETTO(s_type, s_offset, struct_p) \ ((s_type)(((char*)(struct_p)) + (s_offset))) #ifdef DEBUG_VFS_LOCKS /* * Support code to aid in debugging VFS locking problems. Not totally * reliable since if the thread sleeps between changing the lock * state and checking it with the assert, some other thread could * change the state. They are good enough for debugging a single * filesystem using a single-threaded test. Note that the unreliability is * limited to false negatives; efforts were made to ensure that false * positives cannot occur. */ void assert_vi_locked(struct vnode *vp, const char *str); void assert_vi_unlocked(struct vnode *vp, const char *str); void assert_vop_elocked(struct vnode *vp, const char *str); #if 0 void assert_vop_elocked_other(struct vnode *vp, const char *str); #endif void assert_vop_locked(struct vnode *vp, const char *str); #if 0 voi0 assert_vop_slocked(struct vnode *vp, const char *str); #endif void assert_vop_unlocked(struct vnode *vp, const char *str); #define ASSERT_VI_LOCKED(vp, str) assert_vi_locked((vp), (str)) #define ASSERT_VI_UNLOCKED(vp, str) assert_vi_unlocked((vp), (str)) #define ASSERT_VOP_ELOCKED(vp, str) assert_vop_elocked((vp), (str)) #if 0 #define ASSERT_VOP_ELOCKED_OTHER(vp, str) assert_vop_locked_other((vp), (str)) #endif #define ASSERT_VOP_LOCKED(vp, str) assert_vop_locked((vp), (str)) #if 0 #define ASSERT_VOP_SLOCKED(vp, str) assert_vop_slocked((vp), (str)) #endif #define ASSERT_VOP_UNLOCKED(vp, str) assert_vop_unlocked((vp), (str)) #else /* !DEBUG_VFS_LOCKS */ #define ASSERT_VI_LOCKED(vp, str) ((void)0) #define ASSERT_VI_UNLOCKED(vp, str) ((void)0) #define ASSERT_VOP_ELOCKED(vp, str) ((void)0) #if 0 #define ASSERT_VOP_ELOCKED_OTHER(vp, str) #endif #define ASSERT_VOP_LOCKED(vp, str) ((void)0) #if 0 #define ASSERT_VOP_SLOCKED(vp, str) #endif #define ASSERT_VOP_UNLOCKED(vp, str) ((void)0) #endif /* DEBUG_VFS_LOCKS */ /* * This call works for vnodes in the kernel. */ #define VCALL(c) ((c)->a_desc->vdesc_call(c)) #define DOINGASYNC(vp) \ (((vp)->v_mount->mnt_kern_flag & MNTK_ASYNC) != 0 && \ ((curthread->td_pflags & TDP_SYNCIO) == 0)) /* * VMIO support inline */ extern int vmiodirenable; static __inline int vn_canvmio(struct vnode *vp) { if (vp && (vp->v_type == VREG || (vmiodirenable && vp->v_type == VDIR))) return(TRUE); return(FALSE); } /* * Finally, include the default set of vnode operations. */ typedef void vop_getpages_iodone_t(void *, vm_page_t *, int, int); #include "vnode_if.h" /* vn_open_flags */ #define VN_OPEN_NOAUDIT 0x00000001 #define VN_OPEN_NOCAPCHECK 0x00000002 #define VN_OPEN_NAMECACHE 0x00000004 /* * Public vnode manipulation functions. */ struct componentname; struct file; struct mount; struct nameidata; struct ostat; struct thread; struct proc; struct stat; struct nstat; struct ucred; struct uio; struct vattr; struct vnode; typedef int (*vn_get_ino_t)(struct mount *, void *, int, struct vnode **); /* cache_* may belong in namei.h. */ #define cache_enter(dvp, vp, cnp) \ cache_enter_time(dvp, vp, cnp, NULL, NULL) void cache_enter_time(struct vnode *dvp, struct vnode *vp, struct componentname *cnp, struct timespec *tsp, struct timespec *dtsp); int cache_lookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp, struct timespec *tsp, int *ticksp); void cache_purge(struct vnode *vp); void cache_purge_negative(struct vnode *vp); void cache_purgevfs(struct mount *mp); int change_dir(struct vnode *vp, struct thread *td); int change_root(struct vnode *vp, struct thread *td); void cvtstat(struct stat *st, struct ostat *ost); void cvtnstat(struct stat *sb, struct nstat *nsb); int getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops, struct vnode **vpp); void getnewvnode_reserve(u_int count); void getnewvnode_drop_reserve(void); int insmntque1(struct vnode *vp, struct mount *mp, void (*dtr)(struct vnode *, void *), void *dtr_arg); int insmntque(struct vnode *vp, struct mount *mp); u_quad_t init_va_filerev(void); int speedup_syncer(void); int vn_vptocnp(struct vnode **vp, struct ucred *cred, char *buf, u_int *buflen); #define textvp_fullpath(p, rb, rfb) \ vn_fullpath(FIRST_THREAD_IN_PROC(p), (p)->p_textvp, rb, rfb) int vn_fullpath(struct thread *td, struct vnode *vn, char **retbuf, char **freebuf); int vn_fullpath_global(struct thread *td, struct vnode *vn, char **retbuf, char **freebuf); struct vnode * vn_dir_dd_ino(struct vnode *vp); int vn_commname(struct vnode *vn, char *buf, u_int buflen); int vn_path_to_global_path(struct thread *td, struct vnode *vp, char *path, u_int pathlen); 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); int vaccess_acl_nfs4(enum vtype type, uid_t file_uid, gid_t file_gid, struct acl *aclp, accmode_t accmode, struct ucred *cred, int *privused); int vaccess_acl_posix1e(enum vtype type, uid_t file_uid, gid_t file_gid, struct acl *acl, accmode_t accmode, struct ucred *cred, int *privused); void vattr_null(struct vattr *vap); int vcount(struct vnode *vp); void vdrop(struct vnode *); void vdropl(struct vnode *); int vflush(struct mount *mp, int rootrefs, int flags, struct thread *td); int vget(struct vnode *vp, int lockflag, struct thread *td); void vgone(struct vnode *vp); void vhold(struct vnode *); void vholdl(struct vnode *); void vinactive(struct vnode *, struct thread *); int vinvalbuf(struct vnode *vp, int save, int slpflag, int slptimeo); int vtruncbuf(struct vnode *vp, struct ucred *cred, off_t length, int blksize); void vunref(struct vnode *); void vn_printf(struct vnode *vp, const char *fmt, ...) __printflike(2,3); #define vprint(label, vp) vn_printf((vp), "%s\n", (label)) int vrecycle(struct vnode *vp); int vn_bmap_seekhole(struct vnode *vp, u_long cmd, off_t *off, struct ucred *cred); int vn_close(struct vnode *vp, int flags, struct ucred *file_cred, struct thread *td); void vn_finished_write(struct mount *mp); void vn_finished_secondary_write(struct mount *mp); int vn_isdisk(struct vnode *vp, int *errp); int _vn_lock(struct vnode *vp, int flags, char *file, int line); #define vn_lock(vp, flags) _vn_lock(vp, flags, __FILE__, __LINE__) int vn_open(struct nameidata *ndp, int *flagp, int cmode, struct file *fp); int vn_open_cred(struct nameidata *ndp, int *flagp, int cmode, u_int vn_open_flags, struct ucred *cred, struct file *fp); int vn_open_vnode(struct vnode *vp, int fmode, struct ucred *cred, struct thread *td, struct file *fp); void vn_pages_remove(struct vnode *vp, vm_pindex_t start, vm_pindex_t end); int vn_pollrecord(struct vnode *vp, struct thread *p, int events); int vn_rdwr(enum uio_rw rw, struct vnode *vp, void *base, int len, off_t offset, enum uio_seg segflg, int ioflg, struct ucred *active_cred, struct ucred *file_cred, ssize_t *aresid, struct thread *td); int vn_rdwr_inchunks(enum uio_rw rw, struct vnode *vp, void *base, size_t len, off_t offset, enum uio_seg segflg, int ioflg, struct ucred *active_cred, struct ucred *file_cred, size_t *aresid, struct thread *td); int vn_rlimit_fsize(const struct vnode *vn, const struct uio *uio, const struct thread *td); int vn_stat(struct vnode *vp, struct stat *sb, struct ucred *active_cred, struct ucred *file_cred, struct thread *td); int vn_start_write(struct vnode *vp, struct mount **mpp, int flags); int vn_start_secondary_write(struct vnode *vp, struct mount **mpp, int flags); int vn_writechk(struct vnode *vp); int vn_extattr_get(struct vnode *vp, int ioflg, int attrnamespace, const char *attrname, int *buflen, char *buf, struct thread *td); int vn_extattr_set(struct vnode *vp, int ioflg, int attrnamespace, const char *attrname, int buflen, char *buf, struct thread *td); int vn_extattr_rm(struct vnode *vp, int ioflg, int attrnamespace, const char *attrname, struct thread *td); int vn_vget_ino(struct vnode *vp, ino_t ino, int lkflags, struct vnode **rvp); int vn_vget_ino_gen(struct vnode *vp, vn_get_ino_t alloc, void *alloc_arg, int lkflags, struct vnode **rvp); int vn_utimes_perm(struct vnode *vp, struct vattr *vap, struct ucred *cred, struct thread *td); int vn_io_fault_uiomove(char *data, int xfersize, struct uio *uio); int vn_io_fault_pgmove(vm_page_t ma[], vm_offset_t offset, int xfersize, struct uio *uio); #define vn_rangelock_unlock(vp, cookie) \ rangelock_unlock(&(vp)->v_rl, (cookie), VI_MTX(vp)) #define vn_rangelock_unlock_range(vp, cookie, start, end) \ rangelock_unlock_range(&(vp)->v_rl, (cookie), (start), (end), \ VI_MTX(vp)) #define vn_rangelock_rlock(vp, start, end) \ rangelock_rlock(&(vp)->v_rl, (start), (end), VI_MTX(vp)) #define vn_rangelock_wlock(vp, start, end) \ rangelock_wlock(&(vp)->v_rl, (start), (end), VI_MTX(vp)) int vfs_cache_lookup(struct vop_lookup_args *ap); void vfs_timestamp(struct timespec *); void vfs_write_resume(struct mount *mp, int flags); int vfs_write_suspend(struct mount *mp, int flags); int vfs_write_suspend_umnt(struct mount *mp); int vop_stdbmap(struct vop_bmap_args *); int vop_stdfsync(struct vop_fsync_args *); int vop_stdgetwritemount(struct vop_getwritemount_args *); int vop_stdgetpages(struct vop_getpages_args *); int vop_stdinactive(struct vop_inactive_args *); int vop_stdislocked(struct vop_islocked_args *); int vop_stdkqfilter(struct vop_kqfilter_args *); int vop_stdlock(struct vop_lock1_args *); int vop_stdputpages(struct vop_putpages_args *); int vop_stdunlock(struct vop_unlock_args *); int vop_nopoll(struct vop_poll_args *); int vop_stdaccess(struct vop_access_args *ap); int vop_stdaccessx(struct vop_accessx_args *ap); int vop_stdadvise(struct vop_advise_args *ap); int vop_stdadvlock(struct vop_advlock_args *ap); int vop_stdadvlockasync(struct vop_advlockasync_args *ap); int vop_stdadvlockpurge(struct vop_advlockpurge_args *ap); int vop_stdallocate(struct vop_allocate_args *ap); int vop_stdpathconf(struct vop_pathconf_args *); int vop_stdpoll(struct vop_poll_args *); int vop_stdvptocnp(struct vop_vptocnp_args *ap); int vop_stdvptofh(struct vop_vptofh_args *ap); int vop_stdunp_bind(struct vop_unp_bind_args *ap); int vop_stdunp_connect(struct vop_unp_connect_args *ap); int vop_stdunp_detach(struct vop_unp_detach_args *ap); int vop_eopnotsupp(struct vop_generic_args *ap); int vop_ebadf(struct vop_generic_args *ap); int vop_einval(struct vop_generic_args *ap); int vop_enoent(struct vop_generic_args *ap); int vop_enotty(struct vop_generic_args *ap); int vop_null(struct vop_generic_args *ap); int vop_panic(struct vop_generic_args *ap); int dead_poll(struct vop_poll_args *ap); int dead_read(struct vop_read_args *ap); int dead_write(struct vop_write_args *ap); /* These are called from within the actual VOPS. */ void vop_create_post(void *a, int rc); void vop_deleteextattr_post(void *a, int rc); void vop_link_post(void *a, int rc); void vop_lock_pre(void *a); void vop_lock_post(void *a, int rc); void vop_lookup_post(void *a, int rc); void vop_lookup_pre(void *a); void vop_mkdir_post(void *a, int rc); void vop_mknod_post(void *a, int rc); void vop_remove_post(void *a, int rc); void vop_rename_post(void *a, int rc); void vop_rename_pre(void *a); void vop_rmdir_post(void *a, int rc); void vop_setattr_post(void *a, int rc); void vop_setextattr_post(void *a, int rc); void vop_strategy_pre(void *a); void vop_symlink_post(void *a, int rc); void vop_unlock_post(void *a, int rc); void vop_unlock_pre(void *a); void vop_rename_fail(struct vop_rename_args *ap); #define VOP_WRITE_PRE(ap) \ struct vattr va; \ int error, osize, ooffset, noffset; \ \ osize = ooffset = noffset = 0; \ if (!VN_KNLIST_EMPTY((ap)->a_vp)) { \ error = VOP_GETATTR((ap)->a_vp, &va, (ap)->a_cred); \ if (error) \ return (error); \ ooffset = (ap)->a_uio->uio_offset; \ osize = va.va_size; \ } #define VOP_WRITE_POST(ap, ret) \ noffset = (ap)->a_uio->uio_offset; \ if (noffset > ooffset && !VN_KNLIST_EMPTY((ap)->a_vp)) { \ VFS_KNOTE_LOCKED((ap)->a_vp, NOTE_WRITE \ | (noffset > osize ? NOTE_EXTEND : 0)); \ } #define VOP_LOCK(vp, flags) VOP_LOCK1(vp, flags, __FILE__, __LINE__) void vput(struct vnode *vp); void vrele(struct vnode *vp); void vref(struct vnode *vp); int vrefcnt(struct vnode *vp); void v_addpollinfo(struct vnode *vp); int vnode_create_vobject(struct vnode *vp, off_t size, struct thread *td); void vnode_destroy_vobject(struct vnode *vp); extern struct vop_vector fifo_specops; extern struct vop_vector dead_vnodeops; extern struct vop_vector default_vnodeops; extern struct vop_vector devfs_specops; #define VOP_PANIC ((void*)(uintptr_t)vop_panic) #define VOP_NULL ((void*)(uintptr_t)vop_null) #define VOP_EBADF ((void*)(uintptr_t)vop_ebadf) #define VOP_ENOTTY ((void*)(uintptr_t)vop_enotty) #define VOP_EINVAL ((void*)(uintptr_t)vop_einval) #define VOP_ENOENT ((void*)(uintptr_t)vop_enoent) #define VOP_EOPNOTSUPP ((void*)(uintptr_t)vop_eopnotsupp) /* fifo_vnops.c */ int fifo_printinfo(struct vnode *); /* vfs_hash.c */ typedef int vfs_hash_cmp_t(struct vnode *vp, void *arg); int vfs_hash_get(const struct mount *mp, u_int hash, int flags, struct thread *td, struct vnode **vpp, vfs_hash_cmp_t *fn, void *arg); u_int vfs_hash_index(struct vnode *vp); int vfs_hash_insert(struct vnode *vp, u_int hash, int flags, struct thread *td, struct vnode **vpp, vfs_hash_cmp_t *fn, void *arg); void vfs_hash_rehash(struct vnode *vp, u_int hash); void vfs_hash_remove(struct vnode *vp); int vfs_kqfilter(struct vop_kqfilter_args *); void vfs_mark_atime(struct vnode *vp, struct ucred *cred); struct dirent; int vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off); int vfs_unixify_accmode(accmode_t *accmode); void vfs_unp_reclaim(struct vnode *vp); int setfmode(struct thread *td, struct ucred *cred, struct vnode *vp, int mode); int setfown(struct thread *td, struct ucred *cred, struct vnode *vp, uid_t uid, gid_t gid); int vn_chmod(struct file *fp, mode_t mode, struct ucred *active_cred, struct thread *td); int vn_chown(struct file *fp, uid_t uid, gid_t gid, struct ucred *active_cred, struct thread *td); #endif /* _KERNEL */ #endif /* !_SYS_VNODE_H_ */