2 * Copyright (c) 2013 Dmitry Chagin
3 * Copyright (c) 2004 Tim J. Robbins
4 * Copyright (c) 2003 Peter Wemm
5 * Copyright (c) 2002 Doug Rabson
6 * Copyright (c) 1998-1999 Andrew Gallatin
7 * Copyright (c) 1994-1996 Søren Schmidt
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer
15 * in this position and unchanged.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 3. The name of the author may not be used to endorse or promote products
20 * derived from this software without specific prior written permission
22 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
23 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
24 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
25 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
27 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
31 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
37 #define __ELF_WORD_SIZE 64
39 #include <sys/param.h>
40 #include <sys/systm.h>
42 #include <sys/fcntl.h>
43 #include <sys/imgact.h>
44 #include <sys/imgact_elf.h>
45 #include <sys/kernel.h>
48 #include <sys/malloc.h>
49 #include <sys/module.h>
50 #include <sys/mutex.h>
52 #include <sys/resourcevar.h>
53 #include <sys/signalvar.h>
54 #include <sys/syscallsubr.h>
55 #include <sys/sysctl.h>
56 #include <sys/sysent.h>
57 #include <sys/sysproto.h>
58 #include <sys/vnode.h>
59 #include <sys/eventhandler.h>
63 #include <vm/vm_extern.h>
64 #include <vm/vm_map.h>
65 #include <vm/vm_object.h>
66 #include <vm/vm_page.h>
67 #include <vm/vm_param.h>
69 #include <machine/cpu.h>
70 #include <machine/md_var.h>
71 #include <machine/pcb.h>
72 #include <machine/specialreg.h>
73 #include <machine/trap.h>
75 #include <amd64/linux/linux.h>
76 #include <amd64/linux/linux_proto.h>
77 #include <compat/linux/linux_emul.h>
78 #include <compat/linux/linux_ioctl.h>
79 #include <compat/linux/linux_mib.h>
80 #include <compat/linux/linux_misc.h>
81 #include <compat/linux/linux_signal.h>
82 #include <compat/linux/linux_sysproto.h>
83 #include <compat/linux/linux_util.h>
84 #include <compat/linux/linux_vdso.h>
86 MODULE_VERSION(linux64, 1);
88 static int linux_szsigcode;
89 static vm_object_t linux_shared_page_obj;
90 static char *linux_shared_page_mapping;
91 extern char _binary_linux_locore_o_start;
92 extern char _binary_linux_locore_o_end;
94 extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL];
96 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler);
98 static int linux_copyout_strings(struct image_params *imgp,
99 uintptr_t *stack_base);
100 static int linux_fixup_elf(uintptr_t *stack_base,
101 struct image_params *iparams);
102 static bool linux_trans_osrel(const Elf_Note *note, int32_t *osrel);
103 static void linux_vdso_install(void *param);
104 static void linux_vdso_deinstall(void *param);
105 static void linux_set_syscall_retval(struct thread *td, int error);
106 static int linux_fetch_syscall_args(struct thread *td);
107 static void linux_exec_setregs(struct thread *td, struct image_params *imgp,
109 static int linux_vsyscall(struct thread *td);
111 #define LINUX_T_UNKNOWN 255
112 static int _bsd_to_linux_trapcode[] = {
113 LINUX_T_UNKNOWN, /* 0 */
114 6, /* 1 T_PRIVINFLT */
115 LINUX_T_UNKNOWN, /* 2 */
117 LINUX_T_UNKNOWN, /* 4 */
118 LINUX_T_UNKNOWN, /* 5 */
119 16, /* 6 T_ARITHTRAP */
120 254, /* 7 T_ASTFLT */
121 LINUX_T_UNKNOWN, /* 8 */
122 13, /* 9 T_PROTFLT */
123 1, /* 10 T_TRCTRAP */
124 LINUX_T_UNKNOWN, /* 11 */
125 14, /* 12 T_PAGEFLT */
126 LINUX_T_UNKNOWN, /* 13 */
127 17, /* 14 T_ALIGNFLT */
128 LINUX_T_UNKNOWN, /* 15 */
129 LINUX_T_UNKNOWN, /* 16 */
130 LINUX_T_UNKNOWN, /* 17 */
136 8, /* 23 T_DOUBLEFLT */
137 9, /* 24 T_FPOPFLT */
138 10, /* 25 T_TSSFLT */
139 11, /* 26 T_SEGNPFLT */
140 12, /* 27 T_STKFLT */
142 19, /* 29 T_XMMFLT */
143 15 /* 30 T_RESERVED */
145 #define bsd_to_linux_trapcode(code) \
146 ((code)<nitems(_bsd_to_linux_trapcode)? \
147 _bsd_to_linux_trapcode[(code)]: \
150 LINUX_VDSO_SYM_INTPTR(linux_rt_sigcode);
151 LINUX_VDSO_SYM_CHAR(linux_platform);
154 * If FreeBSD & Linux have a difference of opinion about what a trap
155 * means, deal with it here.
160 linux_translate_traps(int signal, int trap_code)
163 if (signal != SIGBUS)
177 linux_fetch_syscall_args(struct thread *td)
180 struct trapframe *frame;
181 struct syscall_args *sa;
184 frame = td->td_frame;
187 sa->args[0] = frame->tf_rdi;
188 sa->args[1] = frame->tf_rsi;
189 sa->args[2] = frame->tf_rdx;
190 sa->args[3] = frame->tf_rcx;
191 sa->args[4] = frame->tf_r8;
192 sa->args[5] = frame->tf_r9;
193 sa->code = frame->tf_rax;
195 if (sa->code >= p->p_sysent->sv_size)
197 sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1];
199 sa->callp = &p->p_sysent->sv_table[sa->code];
201 td->td_retval[0] = 0;
206 linux_set_syscall_retval(struct thread *td, int error)
208 struct trapframe *frame;
210 frame = td->td_frame;
214 frame->tf_rax = td->td_retval[0];
215 frame->tf_r10 = frame->tf_rcx;
220 * Reconstruct pc, we know that 'syscall' is 2 bytes,
221 * lcall $X,y is 7 bytes, int 0x80 is 2 bytes.
222 * We saved this in tf_err.
225 frame->tf_rip -= frame->tf_err;
226 frame->tf_r10 = frame->tf_rcx;
233 frame->tf_rax = bsd_to_linux_errno(error);
234 frame->tf_r10 = frame->tf_rcx;
239 * Differently from FreeBSD native ABI, on Linux only %rcx
240 * and %r11 values are not preserved across the syscall.
241 * Require full context restore to get all registers except
242 * those two restored at return to usermode.
244 * XXX: Would be great to be able to avoid PCB_FULL_IRET
245 * for the error == 0 case.
247 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
251 linux_copyout_auxargs(struct image_params *imgp, uintptr_t base)
254 Elf_Auxinfo *argarray, *pos;
256 int error, issetugid;
259 args = (Elf64_Auxargs *)imgp->auxargs;
260 argarray = pos = malloc(LINUX_AT_COUNT * sizeof(*pos), M_TEMP,
263 issetugid = p->p_flag & P_SUGID ? 1 : 0;
264 AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR,
265 imgp->proc->p_sysent->sv_shared_page_base);
266 AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature);
267 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
268 AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz);
269 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
270 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
271 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
272 AUXARGS_ENTRY(pos, AT_BASE, args->base);
273 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
274 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
275 AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid);
276 AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid);
277 AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid);
278 AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid);
279 AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid);
280 AUXARGS_ENTRY_PTR(pos, LINUX_AT_RANDOM, imgp->canary);
281 AUXARGS_ENTRY(pos, LINUX_AT_HWCAP2, 0);
282 if (imgp->execpathp != 0)
283 AUXARGS_ENTRY_PTR(pos, LINUX_AT_EXECFN, imgp->execpathp);
284 if (args->execfd != -1)
285 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
286 AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform));
287 AUXARGS_ENTRY(pos, AT_NULL, 0);
289 free(imgp->auxargs, M_TEMP);
290 imgp->auxargs = NULL;
291 KASSERT(pos - argarray <= LINUX_AT_COUNT, ("Too many auxargs"));
293 error = copyout(argarray, (void *)base,
294 sizeof(*argarray) * LINUX_AT_COUNT);
295 free(argarray, M_TEMP);
300 linux_fixup_elf(uintptr_t *stack_base, struct image_params *imgp)
304 base = (Elf64_Addr *)*stack_base;
306 if (suword(base, (uint64_t)imgp->args->argc) == -1)
309 *stack_base = (uintptr_t)base;
314 * Copy strings out to the new process address space, constructing new arg
315 * and env vector tables. Return a pointer to the base so that it can be used
316 * as the initial stack pointer.
319 linux_copyout_strings(struct image_params *imgp, uintptr_t *stack_base)
321 int argc, envc, error;
324 uintptr_t destp, ustringp;
325 struct ps_strings *arginfo;
326 char canary[LINUX_AT_RANDOM_LEN];
330 /* Calculate string base and vector table pointers. */
331 if (imgp->execpath != NULL && imgp->auxargs != NULL)
332 execpath_len = strlen(imgp->execpath) + 1;
337 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
338 destp = (uintptr_t)arginfo;
340 if (execpath_len != 0) {
341 destp -= execpath_len;
342 destp = rounddown2(destp, sizeof(void *));
343 imgp->execpathp = (void *)destp;
344 error = copyout(imgp->execpath, imgp->execpathp, execpath_len);
349 /* Prepare the canary for SSP. */
350 arc4rand(canary, sizeof(canary), 0);
351 destp -= roundup(sizeof(canary), sizeof(void *));
352 imgp->canary = (void *)destp;
353 error = copyout(canary, imgp->canary, sizeof(canary));
357 /* Allocate room for the argument and environment strings. */
358 destp -= ARG_MAX - imgp->args->stringspace;
359 destp = rounddown2(destp, sizeof(void *));
364 * Allocate room on the stack for the ELF auxargs
365 * array. It has LINUX_AT_COUNT entries.
367 destp -= LINUX_AT_COUNT * sizeof(Elf64_Auxinfo);
368 destp = rounddown2(destp, sizeof(void *));
371 vectp = (char **)destp;
374 * Allocate room for the argv[] and env vectors including the
375 * terminating NULL pointers.
377 vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
380 * Starting with 2.24, glibc depends on a 16-byte stack alignment.
381 * One "long argc" will be prepended later.
383 vectp = (char **)((((uintptr_t)vectp + 8) & ~0xF) - 8);
385 /* vectp also becomes our initial stack base. */
386 *stack_base = (uintptr_t)vectp;
388 stringp = imgp->args->begin_argv;
389 argc = imgp->args->argc;
390 envc = imgp->args->envc;
392 /* Copy out strings - arguments and environment. */
393 error = copyout(stringp, (void *)ustringp,
394 ARG_MAX - imgp->args->stringspace);
398 /* Fill in "ps_strings" struct for ps, w, etc. */
399 if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 ||
400 suword(&arginfo->ps_nargvstr, argc) != 0)
403 /* Fill in argument portion of vector table. */
404 for (; argc > 0; --argc) {
405 if (suword(vectp++, ustringp) != 0)
407 while (*stringp++ != 0)
412 /* A null vector table pointer separates the argp's from the envp's. */
413 if (suword(vectp++, 0) != 0)
416 if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 ||
417 suword(&arginfo->ps_nenvstr, envc) != 0)
420 /* Fill in environment portion of vector table. */
421 for (; envc > 0; --envc) {
422 if (suword(vectp++, ustringp) != 0)
424 while (*stringp++ != 0)
429 /* The end of the vector table is a null pointer. */
430 if (suword(vectp, 0) != 0)
435 error = imgp->sysent->sv_copyout_auxargs(imgp,
445 * Reset registers to default values on exec.
448 linux_exec_setregs(struct thread *td, struct image_params *imgp,
451 struct trapframe *regs;
453 register_t saved_rflags;
458 if (td->td_proc->p_md.md_ldt != NULL)
463 clear_pcb_flags(pcb, PCB_32BIT);
464 pcb->pcb_initial_fpucw = __LINUX_NPXCW__;
465 set_pcb_flags(pcb, PCB_FULL_IRET);
467 saved_rflags = regs->tf_rflags & PSL_T;
468 bzero((char *)regs, sizeof(struct trapframe));
469 regs->tf_rip = imgp->entry_addr;
470 regs->tf_rsp = stack;
471 regs->tf_rflags = PSL_USER | saved_rflags;
472 regs->tf_ss = _udatasel;
473 regs->tf_cs = _ucodesel;
474 regs->tf_ds = _udatasel;
475 regs->tf_es = _udatasel;
476 regs->tf_fs = _ufssel;
477 regs->tf_gs = _ugssel;
478 regs->tf_flags = TF_HASSEGS;
480 x86_clear_dbregs(pcb);
483 * Drop the FP state if we hold it, so that the process gets a
484 * clean FP state if it uses the FPU again.
490 * Copied from amd64/amd64/machdep.c
492 * XXX fpu state need? don't think so
495 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
498 struct l_ucontext uc;
499 struct l_sigcontext *context;
500 struct trapframe *regs;
501 unsigned long rflags;
506 error = copyin((void *)regs->tf_rbx, &uc, sizeof(uc));
511 context = &uc.uc_mcontext;
512 rflags = context->sc_rflags;
515 * Don't allow users to change privileged or reserved flags.
518 * XXX do allow users to change the privileged flag PSL_RF.
519 * The cpu sets PSL_RF in tf_rflags for faults. Debuggers
520 * should sometimes set it there too. tf_rflags is kept in
521 * the signal context during signal handling and there is no
522 * other place to remember it, so the PSL_RF bit may be
523 * corrupted by the signal handler without us knowing.
524 * Corruption of the PSL_RF bit at worst causes one more or
525 * one less debugger trap, so allowing it is fairly harmless.
528 #define RFLAG_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
529 if (!RFLAG_SECURE(rflags & ~PSL_RF, regs->tf_rflags & ~PSL_RF)) {
530 printf("linux_rt_sigreturn: rflags = 0x%lx\n", rflags);
535 * Don't allow users to load a valid privileged %cs. Let the
536 * hardware check for invalid selectors, excess privilege in
537 * other selectors, invalid %eip's and invalid %esp's.
539 #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL)
540 if (!CS_SECURE(context->sc_cs)) {
541 printf("linux_rt_sigreturn: cs = 0x%x\n", context->sc_cs);
542 ksiginfo_init_trap(&ksi);
543 ksi.ksi_signo = SIGBUS;
544 ksi.ksi_code = BUS_OBJERR;
545 ksi.ksi_trapno = T_PROTFLT;
546 ksi.ksi_addr = (void *)regs->tf_rip;
547 trapsignal(td, &ksi);
552 linux_to_bsd_sigset(&uc.uc_sigmask, &td->td_sigmask);
553 SIG_CANTMASK(td->td_sigmask);
557 regs->tf_rdi = context->sc_rdi;
558 regs->tf_rsi = context->sc_rsi;
559 regs->tf_rdx = context->sc_rdx;
560 regs->tf_rbp = context->sc_rbp;
561 regs->tf_rbx = context->sc_rbx;
562 regs->tf_rcx = context->sc_rcx;
563 regs->tf_rax = context->sc_rax;
564 regs->tf_rip = context->sc_rip;
565 regs->tf_rsp = context->sc_rsp;
566 regs->tf_r8 = context->sc_r8;
567 regs->tf_r9 = context->sc_r9;
568 regs->tf_r10 = context->sc_r10;
569 regs->tf_r11 = context->sc_r11;
570 regs->tf_r12 = context->sc_r12;
571 regs->tf_r13 = context->sc_r13;
572 regs->tf_r14 = context->sc_r14;
573 regs->tf_r15 = context->sc_r15;
574 regs->tf_cs = context->sc_cs;
575 regs->tf_err = context->sc_err;
576 regs->tf_rflags = rflags;
578 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
579 return (EJUSTRETURN);
583 * copied from amd64/amd64/machdep.c
585 * Send an interrupt to process.
588 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
590 struct l_rt_sigframe sf, *sfp;
595 struct trapframe *regs;
601 PROC_LOCK_ASSERT(p, MA_OWNED);
602 sig = ksi->ksi_signo;
604 code = ksi->ksi_code;
605 mtx_assert(&psp->ps_mtx, MA_OWNED);
607 oonstack = sigonstack(regs->tf_rsp);
609 LINUX_CTR4(rt_sendsig, "%p, %d, %p, %u",
610 catcher, sig, mask, code);
612 /* Translate the signal. */
613 sig = bsd_to_linux_signal(sig);
615 /* Save user context. */
616 bzero(&sf, sizeof(sf));
617 bsd_to_linux_sigset(mask, &sf.sf_sc.uc_sigmask);
618 bsd_to_linux_sigset(mask, &sf.sf_sc.uc_mcontext.sc_mask);
620 sf.sf_sc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp);
621 sf.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size;
622 sf.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
623 ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE;
625 sf.sf_sc.uc_mcontext.sc_rdi = regs->tf_rdi;
626 sf.sf_sc.uc_mcontext.sc_rsi = regs->tf_rsi;
627 sf.sf_sc.uc_mcontext.sc_rdx = regs->tf_rdx;
628 sf.sf_sc.uc_mcontext.sc_rbp = regs->tf_rbp;
629 sf.sf_sc.uc_mcontext.sc_rbx = regs->tf_rbx;
630 sf.sf_sc.uc_mcontext.sc_rcx = regs->tf_rcx;
631 sf.sf_sc.uc_mcontext.sc_rax = regs->tf_rax;
632 sf.sf_sc.uc_mcontext.sc_rip = regs->tf_rip;
633 sf.sf_sc.uc_mcontext.sc_rsp = regs->tf_rsp;
634 sf.sf_sc.uc_mcontext.sc_r8 = regs->tf_r8;
635 sf.sf_sc.uc_mcontext.sc_r9 = regs->tf_r9;
636 sf.sf_sc.uc_mcontext.sc_r10 = regs->tf_r10;
637 sf.sf_sc.uc_mcontext.sc_r11 = regs->tf_r11;
638 sf.sf_sc.uc_mcontext.sc_r12 = regs->tf_r12;
639 sf.sf_sc.uc_mcontext.sc_r13 = regs->tf_r13;
640 sf.sf_sc.uc_mcontext.sc_r14 = regs->tf_r14;
641 sf.sf_sc.uc_mcontext.sc_r15 = regs->tf_r15;
642 sf.sf_sc.uc_mcontext.sc_cs = regs->tf_cs;
643 sf.sf_sc.uc_mcontext.sc_rflags = regs->tf_rflags;
644 sf.sf_sc.uc_mcontext.sc_err = regs->tf_err;
645 sf.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code);
646 sf.sf_sc.uc_mcontext.sc_cr2 = (register_t)ksi->ksi_addr;
648 /* Allocate space for the signal handler context. */
649 if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack &&
650 SIGISMEMBER(psp->ps_sigonstack, sig)) {
651 sp = (caddr_t)td->td_sigstk.ss_sp + td->td_sigstk.ss_size -
652 sizeof(struct l_rt_sigframe);
654 sp = (caddr_t)regs->tf_rsp - sizeof(struct l_rt_sigframe) - 128;
655 /* Align to 16 bytes. */
656 sfp = (struct l_rt_sigframe *)((unsigned long)sp & ~0xFul);
658 /* Build the argument list for the signal handler. */
659 regs->tf_rdi = sig; /* arg 1 in %rdi */
661 regs->tf_rsi = (register_t)&sfp->sf_si; /* arg 2 in %rsi */
662 regs->tf_rdx = (register_t)&sfp->sf_sc; /* arg 3 in %rdx */
664 /* Fill in POSIX parts. */
665 siginfo_to_lsiginfo(&ksi->ksi_info, &sf.sf_si, sig);
666 sf.sf_handler = catcher;
668 mtx_unlock(&psp->ps_mtx);
671 /* Copy the sigframe out to the user's stack. */
672 if (copyout(&sf, sfp, sizeof(*sfp)) != 0) {
677 regs->tf_rsp = (long)sfp;
678 regs->tf_rip = linux_rt_sigcode;
679 regs->tf_rflags &= ~(PSL_T | PSL_D);
680 regs->tf_cs = _ucodesel;
681 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
683 mtx_lock(&psp->ps_mtx);
686 #define LINUX_VSYSCALL_START (-10UL << 20)
687 #define LINUX_VSYSCALL_SZ 1024
689 const unsigned long linux_vsyscall_vector[] = {
690 LINUX_SYS_gettimeofday,
691 LINUX_SYS_linux_time,
692 LINUX_SYS_linux_getcpu,
696 linux_vsyscall(struct thread *td)
698 struct trapframe *frame;
703 frame = td->td_frame;
705 /* Check %rip for vsyscall area. */
706 if (__predict_true(frame->tf_rip < LINUX_VSYSCALL_START))
708 if ((frame->tf_rip & (LINUX_VSYSCALL_SZ - 1)) != 0)
710 code = (frame->tf_rip - LINUX_VSYSCALL_START) / LINUX_VSYSCALL_SZ;
711 if (code >= nitems(linux_vsyscall_vector))
715 * vsyscall called as callq *(%rax), so we must
716 * use return address from %rsp and also fixup %rsp.
718 error = copyin((void *)frame->tf_rsp, &retqaddr, sizeof(retqaddr));
722 frame->tf_rip = retqaddr;
723 frame->tf_rax = linux_vsyscall_vector[code];
726 traced = (frame->tf_flags & PSL_T);
728 amd64_syscall(td, traced);
733 struct sysentvec elf_linux_sysvec = {
734 .sv_size = LINUX_SYS_MAXSYSCALL,
735 .sv_table = linux_sysent,
736 .sv_transtrap = linux_translate_traps,
737 .sv_fixup = linux_fixup_elf,
738 .sv_sendsig = linux_rt_sendsig,
739 .sv_sigcode = &_binary_linux_locore_o_start,
740 .sv_szsigcode = &linux_szsigcode,
741 .sv_name = "Linux ELF64",
742 .sv_coredump = elf64_coredump,
743 .sv_imgact_try = linux_exec_imgact_try,
744 .sv_minsigstksz = LINUX_MINSIGSTKSZ,
745 .sv_minuser = VM_MIN_ADDRESS,
746 .sv_maxuser = VM_MAXUSER_ADDRESS_LA48,
747 .sv_usrstack = USRSTACK_LA48,
748 .sv_psstrings = PS_STRINGS_LA48,
749 .sv_stackprot = VM_PROT_ALL,
750 .sv_copyout_auxargs = linux_copyout_auxargs,
751 .sv_copyout_strings = linux_copyout_strings,
752 .sv_setregs = linux_exec_setregs,
755 .sv_flags = SV_ABI_LINUX | SV_LP64 | SV_SHP | SV_SIG_DISCIGN |
757 .sv_set_syscall_retval = linux_set_syscall_retval,
758 .sv_fetch_syscall_args = linux_fetch_syscall_args,
759 .sv_syscallnames = NULL,
760 .sv_shared_page_base = SHAREDPAGE_LA48,
761 .sv_shared_page_len = PAGE_SIZE,
762 .sv_schedtail = linux_schedtail,
763 .sv_thread_detach = linux_thread_detach,
764 .sv_trap = linux_vsyscall,
765 .sv_onexec = linux_on_exec,
766 .sv_onexit = linux_on_exit,
767 .sv_ontdexit = linux_thread_dtor,
768 .sv_setid_allowed = &linux_setid_allowed_query,
772 linux_vdso_install(void *param)
775 amd64_lower_shared_page(&elf_linux_sysvec);
777 linux_szsigcode = (&_binary_linux_locore_o_end -
778 &_binary_linux_locore_o_start);
780 if (linux_szsigcode > elf_linux_sysvec.sv_shared_page_len)
781 panic("Linux invalid vdso size\n");
783 __elfN(linux_vdso_fixup)(&elf_linux_sysvec);
785 linux_shared_page_obj = __elfN(linux_shared_page_init)
786 (&linux_shared_page_mapping);
788 __elfN(linux_vdso_reloc)(&elf_linux_sysvec);
790 bcopy(elf_linux_sysvec.sv_sigcode, linux_shared_page_mapping,
792 elf_linux_sysvec.sv_shared_page_obj = linux_shared_page_obj;
794 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC, SI_ORDER_ANY,
795 linux_vdso_install, NULL);
798 linux_vdso_deinstall(void *param)
801 __elfN(linux_shared_page_fini)(linux_shared_page_obj,
802 linux_shared_page_mapping);
804 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
805 linux_vdso_deinstall, NULL);
807 static char GNULINUX_ABI_VENDOR[] = "GNU";
808 static int GNULINUX_ABI_DESC = 0;
811 linux_trans_osrel(const Elf_Note *note, int32_t *osrel)
813 const Elf32_Word *desc;
816 p = (uintptr_t)(note + 1);
817 p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
819 desc = (const Elf32_Word *)p;
820 if (desc[0] != GNULINUX_ABI_DESC)
824 * For Linux we encode osrel using the Linux convention of
825 * (version << 16) | (major << 8) | (minor)
826 * See macro in linux_mib.h
828 *osrel = LINUX_KERNVER(desc[1], desc[2], desc[3]);
833 static Elf_Brandnote linux64_brandnote = {
834 .hdr.n_namesz = sizeof(GNULINUX_ABI_VENDOR),
837 .vendor = GNULINUX_ABI_VENDOR,
838 .flags = BN_TRANSLATE_OSREL,
839 .trans_osrel = linux_trans_osrel
842 static Elf64_Brandinfo linux_glibc2brand = {
843 .brand = ELFOSABI_LINUX,
844 .machine = EM_X86_64,
845 .compat_3_brand = "Linux",
846 .emul_path = linux_emul_path,
847 .interp_path = "/lib64/ld-linux-x86-64.so.2",
848 .sysvec = &elf_linux_sysvec,
849 .interp_newpath = NULL,
850 .brand_note = &linux64_brandnote,
851 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
854 static Elf64_Brandinfo linux_glibc2brandshort = {
855 .brand = ELFOSABI_LINUX,
856 .machine = EM_X86_64,
857 .compat_3_brand = "Linux",
858 .emul_path = linux_emul_path,
859 .interp_path = "/lib64/ld-linux.so.2",
860 .sysvec = &elf_linux_sysvec,
861 .interp_newpath = NULL,
862 .brand_note = &linux64_brandnote,
863 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
866 static Elf64_Brandinfo linux_muslbrand = {
867 .brand = ELFOSABI_LINUX,
868 .machine = EM_X86_64,
869 .compat_3_brand = "Linux",
870 .emul_path = linux_emul_path,
871 .interp_path = "/lib/ld-musl-x86_64.so.1",
872 .sysvec = &elf_linux_sysvec,
873 .interp_newpath = NULL,
874 .brand_note = &linux64_brandnote,
875 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
878 Elf64_Brandinfo *linux_brandlist[] = {
880 &linux_glibc2brandshort,
886 linux64_elf_modevent(module_t mod, int type, void *data)
888 Elf64_Brandinfo **brandinfo;
890 struct linux_ioctl_handler **lihp;
896 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
898 if (elf64_insert_brand_entry(*brandinfo) < 0)
901 SET_FOREACH(lihp, linux_ioctl_handler_set)
902 linux_ioctl_register_handler(*lihp);
903 stclohz = (stathz ? stathz : hz);
905 printf("Linux x86-64 ELF exec handler installed\n");
907 printf("cannot insert Linux x86-64 ELF brand handler\n");
910 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
912 if (elf64_brand_inuse(*brandinfo))
915 for (brandinfo = &linux_brandlist[0];
916 *brandinfo != NULL; ++brandinfo)
917 if (elf64_remove_brand_entry(*brandinfo) < 0)
921 SET_FOREACH(lihp, linux_ioctl_handler_set)
922 linux_ioctl_unregister_handler(*lihp);
924 printf("Linux ELF exec handler removed\n");
926 printf("Could not deinstall ELF interpreter entry\n");
934 static moduledata_t linux64_elf_mod = {
936 linux64_elf_modevent,
940 DECLARE_MODULE_TIED(linux64elf, linux64_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);
941 MODULE_DEPEND(linux64elf, linux_common, 1, 1, 1);
942 FEATURE(linux64, "Linux 64bit support");