Differential D30900 Diff 91389 sys/amd64/linux/linux_sysvec.c

Changeset View

Standalone View

sys/amd64/linux/linux_sysvec.c

Show First 20 Lines • Show All 44 Lines • ▼ Show 20 Lines
#include <sys/kernel.h>		#include <sys/kernel.h>
#include <sys/ktr.h>		#include <sys/ktr.h>
#include <sys/lock.h>		#include <sys/lock.h>
#include <sys/malloc.h>		#include <sys/malloc.h>
#include <sys/module.h>		#include <sys/module.h>
#include <sys/mutex.h>		#include <sys/mutex.h>
#include <sys/proc.h>		#include <sys/proc.h>
#include <sys/resourcevar.h>		#include <sys/resourcevar.h>
		#include <sys/stddef.h>
#include <sys/signalvar.h>		#include <sys/signalvar.h>
#include <sys/syscallsubr.h>		#include <sys/syscallsubr.h>
#include <sys/sysctl.h>		#include <sys/sysctl.h>
#include <sys/sysent.h>		#include <sys/sysent.h>
#include <sys/sysproto.h>		#include <sys/sysproto.h>
#include <sys/vnode.h>		#include <sys/vnode.h>
#include <sys/eventhandler.h>		#include <sys/eventhandler.h>

#include <vm/vm.h>		#include <vm/vm.h>
#include <vm/pmap.h>		#include <vm/pmap.h>
#include <vm/vm_extern.h>		#include <vm/vm_extern.h>
#include <vm/vm_map.h>		#include <vm/vm_map.h>
#include <vm/vm_object.h>		#include <vm/vm_object.h>
#include <vm/vm_page.h>		#include <vm/vm_page.h>
#include <vm/vm_param.h>		#include <vm/vm_param.h>

#include <machine/cpu.h>		#include <machine/cpu.h>
#include <machine/md_var.h>		#include <machine/md_var.h>
#include <machine/pcb.h>		#include <machine/pcb.h>
#include <machine/specialreg.h>		#include <machine/specialreg.h>
#include <machine/trap.h>		#include <machine/trap.h>

		#include <x86/linux/linux_x86.h>
#include <amd64/linux/linux.h>		#include <amd64/linux/linux.h>
#include <amd64/linux/linux_proto.h>		#include <amd64/linux/linux_proto.h>
#include <compat/linux/linux_emul.h>		#include <compat/linux/linux_emul.h>
#include <compat/linux/linux_ioctl.h>		#include <compat/linux/linux_ioctl.h>
#include <compat/linux/linux_mib.h>		#include <compat/linux/linux_mib.h>
#include <compat/linux/linux_misc.h>		#include <compat/linux/linux_misc.h>
#include <compat/linux/linux_signal.h>		#include <compat/linux/linux_signal.h>
#include <compat/linux/linux_sysproto.h>		#include <compat/linux/linux_sysproto.h>
#include <compat/linux/linux_util.h>		#include <compat/linux/linux_util.h>
#include <compat/linux/linux_vdso.h>		#include <compat/linux/linux_vdso.h>

MODULE_VERSION(linux64, 1);		MODULE_VERSION(linux64, 1);

		#define LINUX_VDSOPAGE_SIZE PAGE_SIZE * 2
		#define LINUX_VDSOPAGE_LA48 (VM_MAXUSER_ADDRESS_LA48 - \
		LINUX_VDSOPAGE_SIZE)
		#define LINUX_SHAREDPAGE_LA48 (LINUX_VDSOPAGE_LA48 - PAGE_SIZE)
		/*
		* PAGE_SIZE - the size
		* of the native SHAREDPAGE
		*/
		#define LINUX_USRSTACK_LA48 LINUX_SHAREDPAGE_LA48
		#define LINUX_PS_STRINGS_LA48 (LINUX_USRSTACK_LA48 - \
		sizeof(struct ps_strings))

static int linux_szsigcode;		static int linux_szsigcode;
static vm_object_t linux_shared_page_obj;		static vm_object_t linux_vdso_obj;
static char *linux_shared_page_mapping;		static char *linux_vdso_mapping;
extern char _binary_linux_locore_o_start;		extern char _binary_linux_vdso_so_o_start;
extern char _binary_linux_locore_o_end;		extern char _binary_linux_vdso_so_o_end;
		static vm_offset_t linux_vdso_base;

extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL];		extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL];

SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler);		SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler);

static int linux_copyout_strings(struct image_params *imgp,		static int linux_copyout_strings(struct image_params *imgp,
uintptr_t *stack_base);		uintptr_t *stack_base);
static int linux_fixup_elf(uintptr_t *stack_base,		static int linux_fixup_elf(uintptr_t *stack_base,
struct image_params *iparams);		struct image_params *iparams);
static bool linux_trans_osrel(const Elf_Note note, int32_t osrel);		static bool linux_trans_osrel(const Elf_Note note, int32_t osrel);
static void linux_vdso_install(void *param);		static void linux_vdso_install(void *param);
static void linux_vdso_deinstall(void *param);		static void linux_vdso_deinstall(void *param);
		static void linux_vdso_reloc(char *mapping, Elf_Addr offset);
static void linux_set_syscall_retval(struct thread *td, int error);		static void linux_set_syscall_retval(struct thread *td, int error);
static int linux_fetch_syscall_args(struct thread *td);		static int linux_fetch_syscall_args(struct thread *td);
static void linux_exec_setregs(struct thread td, struct image_params imgp,		static void linux_exec_setregs(struct thread td, struct image_params imgp,
uintptr_t stack);		uintptr_t stack);
		static void linux_exec_sysvec_init(void *param);
static int linux_on_exec_vmspace(struct proc *p,		static int linux_on_exec_vmspace(struct proc *p,
struct image_params *imgp);		struct image_params *imgp);
static int linux_vsyscall(struct thread *td);		static int linux_vsyscall(struct thread *td);

#define LINUX_T_UNKNOWN 255		#define LINUX_T_UNKNOWN 255
static int _bsd_to_linux_trapcode[] = {		static int _bsd_to_linux_trapcode[] = {
LINUX_T_UNKNOWN, /* 0 */		LINUX_T_UNKNOWN, /* 0 */
6, /* 1 T_PRIVINFLT */		6, /* 1 T_PRIVINFLT */
Show All 29 Lines
};		};
#define bsd_to_linux_trapcode(code) \		#define bsd_to_linux_trapcode(code) \
((code)<nitems(_bsd_to_linux_trapcode)? \		((code)<nitems(_bsd_to_linux_trapcode)? \
_bsd_to_linux_trapcode[(code)]: \		_bsd_to_linux_trapcode[(code)]: \
LINUX_T_UNKNOWN)		LINUX_T_UNKNOWN)

LINUX_VDSO_SYM_INTPTR(linux_rt_sigcode);		LINUX_VDSO_SYM_INTPTR(linux_rt_sigcode);
LINUX_VDSO_SYM_CHAR(linux_platform);		LINUX_VDSO_SYM_CHAR(linux_platform);
		LINUX_VDSO_SYM_INTPTR(kern_timekeep_base);
		LINUX_VDSO_SYM_INTPTR(kern_tsc_selector);

/*		/*
* If FreeBSD & Linux have a difference of opinion about what a trap		* If FreeBSD & Linux have a difference of opinion about what a trap
* means, deal with it here.		* means, deal with it here.
*		*
* MPSAFE		* MPSAFE
*/		*/
static int		static int
▲ Show 20 Lines • Show All 96 Lines • ▼ Show 20 Lines	linux_copyout_auxargs(struct image_params *imgp, uintptr_t base)
int error, issetugid;		int error, issetugid;

p = imgp->proc;		p = imgp->proc;
args = (Elf64_Auxargs *)imgp->auxargs;		args = (Elf64_Auxargs *)imgp->auxargs;
argarray = pos = malloc(LINUX_AT_COUNT * sizeof(*pos), M_TEMP,		argarray = pos = malloc(LINUX_AT_COUNT * sizeof(*pos), M_TEMP,
M_WAITOK \| M_ZERO);		M_WAITOK \| M_ZERO);

issetugid = p->p_flag & P_SUGID ? 1 : 0;		issetugid = p->p_flag & P_SUGID ? 1 : 0;
AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR,		AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR, linux_vdso_base);
imgp->proc->p_sysent->sv_shared_page_base);
AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature);		AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature);
AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);		AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz);		AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz);
AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);		AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
AUXARGS_ENTRY(pos, AT_PHENT, args->phent);		AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);		AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
AUXARGS_ENTRY(pos, AT_BASE, args->base);		AUXARGS_ENTRY(pos, AT_BASE, args->base);
AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);		AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
▲ Show 20 Lines • Show All 457 Lines • ▼ Show 20 Lines
}		}

struct sysentvec elf_linux_sysvec = {		struct sysentvec elf_linux_sysvec = {
.sv_size = LINUX_SYS_MAXSYSCALL,		.sv_size = LINUX_SYS_MAXSYSCALL,
.sv_table = linux_sysent,		.sv_table = linux_sysent,
.sv_transtrap = linux_translate_traps,		.sv_transtrap = linux_translate_traps,
.sv_fixup = linux_fixup_elf,		.sv_fixup = linux_fixup_elf,
.sv_sendsig = linux_rt_sendsig,		.sv_sendsig = linux_rt_sendsig,
.sv_sigcode = &_binary_linux_locore_o_start,		.sv_sigcode = &_binary_linux_vdso_so_o_start,
.sv_szsigcode = &linux_szsigcode,		.sv_szsigcode = &linux_szsigcode,
.sv_name = "Linux ELF64",		.sv_name = "Linux ELF64",
.sv_coredump = elf64_coredump,		.sv_coredump = elf64_coredump,
.sv_imgact_try = linux_exec_imgact_try,		.sv_imgact_try = linux_exec_imgact_try,
.sv_minsigstksz = LINUX_MINSIGSTKSZ,		.sv_minsigstksz = LINUX_MINSIGSTKSZ,
.sv_minuser = VM_MIN_ADDRESS,		.sv_minuser = VM_MIN_ADDRESS,
.sv_maxuser = VM_MAXUSER_ADDRESS_LA48,		.sv_maxuser = VM_MAXUSER_ADDRESS_LA48,
.sv_usrstack = USRSTACK_LA48,		.sv_usrstack = LINUX_USRSTACK_LA48,
.sv_psstrings = PS_STRINGS_LA48,		.sv_psstrings = LINUX_PS_STRINGS_LA48,
.sv_stackprot = VM_PROT_ALL,		.sv_stackprot = VM_PROT_ALL,
.sv_copyout_auxargs = linux_copyout_auxargs,		.sv_copyout_auxargs = linux_copyout_auxargs,
.sv_copyout_strings = linux_copyout_strings,		.sv_copyout_strings = linux_copyout_strings,
.sv_setregs = linux_exec_setregs,		.sv_setregs = linux_exec_setregs,
.sv_fixlimit = NULL,		.sv_fixlimit = NULL,
.sv_maxssiz = NULL,		.sv_maxssiz = NULL,
.sv_flags = SV_ABI_LINUX \| SV_LP64 \| SV_SHP \| SV_SIG_DISCIGN \|		.sv_flags = SV_ABI_LINUX \| SV_LP64 \| SV_SHP \| SV_SIG_DISCIGN \|
SV_SIG_WAITNDQ,		SV_SIG_WAITNDQ \| SV_TIMEKEEP,
.sv_set_syscall_retval = linux_set_syscall_retval,		.sv_set_syscall_retval = linux_set_syscall_retval,
.sv_fetch_syscall_args = linux_fetch_syscall_args,		.sv_fetch_syscall_args = linux_fetch_syscall_args,
.sv_syscallnames = NULL,		.sv_syscallnames = NULL,
.sv_shared_page_base = SHAREDPAGE_LA48,		.sv_shared_page_base = LINUX_SHAREDPAGE_LA48,
.sv_shared_page_len = PAGE_SIZE,		.sv_shared_page_len = PAGE_SIZE,
.sv_schedtail = linux_schedtail,		.sv_schedtail = linux_schedtail,
.sv_thread_detach = linux_thread_detach,		.sv_thread_detach = linux_thread_detach,
.sv_trap = linux_vsyscall,		.sv_trap = linux_vsyscall,
.sv_onexec = linux_on_exec_vmspace,		.sv_onexec = linux_on_exec_vmspace,
.sv_onexit = linux_on_exit,		.sv_onexit = linux_on_exit,
.sv_ontdexit = linux_thread_dtor,		.sv_ontdexit = linux_thread_dtor,
.sv_setid_allowed = &linux_setid_allowed_query,		.sv_setid_allowed = &linux_setid_allowed_query,
};		};

static int		static int
linux_on_exec_vmspace(struct proc p, struct image_params imgp)		linux_on_exec_vmspace(struct proc p, struct image_params imgp)
{		{
		int error;

		error = linux_map_vdso(p, linux_vdso_obj, linux_vdso_base,
		LINUX_VDSOPAGE_SIZE, imgp);
		if (error == 0)
linux_on_exec(p, imgp);		linux_on_exec(p, imgp);
return (0);		return (error);
}		}

static void		static void
linux_vdso_install(void *param)		linux_exec_sysvec_init(void *param)
{		{
		l_uintptr_t ktimekeep_base, ktsc_selector;
		struct sysentvec *sv;
		ptrdiff_t tkoff;

amd64_lower_shared_page(&elf_linux_sysvec);		sv = param;
		amd64_lower_shared_page(sv);
		/* Fill timekeep_base */
		exec_sysvec_init(sv);

linux_szsigcode = (&_binary_linux_locore_o_end -		tkoff = kern_timekeep_base - linux_vdso_base;
&_binary_linux_locore_o_start);		ktimekeep_base = (l_uintptr_t *)(linux_vdso_mapping + tkoff);
		*ktimekeep_base = sv->sv_timekeep_base;

if (linux_szsigcode > elf_linux_sysvec.sv_shared_page_len)		tkoff = kern_tsc_selector - linux_vdso_base;
panic("Linux invalid vdso size\n");		ktsc_selector = (l_uintptr_t *)(linux_vdso_mapping + tkoff);
		*ktsc_selector = linux_vdso_tsc_selector_idx();
		if (bootverbose)
		printf("Linux x86-64 vDSO tsc_selector: %lu\n", *ktsc_selector);
		}
		SYSINIT(elf_linux_exec_sysvec_init, SI_SUB_EXEC, SI_ORDER_ANY,
		linux_exec_sysvec_init, &elf_linux_sysvec);

__elfN(linux_vdso_fixup)(&elf_linux_sysvec);		static void
		linux_vdso_install(void *param)
		{
		char *vdso_start = &_binary_linux_vdso_so_o_start;
		char *vdso_end = &_binary_linux_vdso_so_o_end;

linux_shared_page_obj = __elfN(linux_shared_page_init)		linux_szsigcode = vdso_end - vdso_start;
(&linux_shared_page_mapping);		MPASS(linux_szsigcode <= LINUX_VDSOPAGE_SIZE);

__elfN(linux_vdso_reloc)(&elf_linux_sysvec);		linux_vdso_base = LINUX_VDSOPAGE_LA48;
		if (hw_lower_amd64_sharedpage != 0)
		linux_vdso_base -= PAGE_SIZE;

bcopy(elf_linux_sysvec.sv_sigcode, linux_shared_page_mapping,		__elfN(linux_vdso_fixup)(vdso_start, linux_vdso_base);
linux_szsigcode);
elf_linux_sysvec.sv_shared_page_obj = linux_shared_page_obj;		linux_vdso_obj = __elfN(linux_shared_page_init)
		(&linux_vdso_mapping, LINUX_VDSOPAGE_SIZE);
		bcopy(vdso_start, linux_vdso_mapping, linux_szsigcode);

		linux_vdso_reloc(linux_vdso_mapping, linux_vdso_base);
}		}
SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC, SI_ORDER_ANY,		SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC, SI_ORDER_FIRST,
linux_vdso_install, NULL);		linux_vdso_install, NULL);

static void		static void
linux_vdso_deinstall(void *param)		linux_vdso_deinstall(void *param)
{		{

__elfN(linux_shared_page_fini)(linux_shared_page_obj,		__elfN(linux_shared_page_fini)(linux_vdso_obj,
linux_shared_page_mapping);		linux_vdso_mapping, LINUX_VDSOPAGE_SIZE);
}		}
SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,		SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
linux_vdso_deinstall, NULL);		linux_vdso_deinstall, NULL);

		static void
		linux_vdso_reloc(char *mapping, Elf_Addr offset)
		{
		const Elf_Ehdr *ehdr;
		const Elf_Shdr *shdr;
		Elf64_Addr *where, val;
		Elf_Size rtype, symidx;
		const Elf_Rela *rela;
		Elf_Addr addr, addend;
		int relacnt;
		int i, j;

		MPASS(offset != 0);

		relacnt = 0;
		ehdr = (const Elf_Ehdr *)mapping;
		shdr = (const Elf_Shdr *)(mapping + ehdr->e_shoff);
		for (i = 0; i < ehdr->e_shnum; i++)
		{
		switch (shdr[i].sh_type) {
		case SHT_REL:
		printf("Linux x86_64 vDSO: unexpected Rel section\n");
		break;
		case SHT_RELA:
		rela = (const Elf_Rela *)(mapping + shdr[i].sh_offset);
		relacnt = shdr[i].sh_size / sizeof(*rela);
		}
		}

		for (j = 0; j < relacnt; j++, rela++) {
		where = (Elf_Addr *)(mapping + rela->r_offset);
		addend = rela->r_addend;
		rtype = ELF_R_TYPE(rela->r_info);
		symidx = ELF_R_SYM(rela->r_info);

		switch (rtype) {
		case R_X86_64_NONE: /* none */
		break;

		case R_X86_64_RELATIVE: /* B + A */
		addr = (Elf_Addr)(offset + addend);
		val = addr;
		if (*where != val)
		*where = val;
		break;
		case R_X86_64_IRELATIVE:
		printf("Linux x86_64 vDSO: unexpected ifunc relocation, "
		"symbol index %ld\n", symidx);
		break;
		default:
		printf("Linux x86_64 vDSO: unexpected relocation type %ld, "
		"symbol index %ld\n", rtype, symidx);
		}
		}
		}

static char GNULINUX_ABI_VENDOR[] = "GNU";		static char GNULINUX_ABI_VENDOR[] = "GNU";
static int GNULINUX_ABI_DESC = 0;		static int GNULINUX_ABI_DESC = 0;

static bool		static bool
linux_trans_osrel(const Elf_Note note, int32_t osrel)		linux_trans_osrel(const Elf_Note note, int32_t osrel)
{		{
const Elf32_Word *desc;		const Elf32_Word *desc;
▲ Show 20 Lines • Show All 129 Lines • Show Last 20 Lines