2 * Copyright (c) 2004 Tim J. Robbins
3 * Copyright (c) 2003 Peter Wemm
4 * Copyright (c) 2002 Doug Rabson
5 * Copyright (c) 1998-1999 Andrew Gallatin
6 * Copyright (c) 1994-1996 Søren Schmidt
8 * Copyright (c) 2013, 2021 Dmitry Chagin <dchagin@FreeBSD.org>
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/stddef.h>
54 #include <sys/signalvar.h>
55 #include <sys/syscallsubr.h>
56 #include <sys/sysctl.h>
57 #include <sys/sysent.h>
58 #include <sys/sysproto.h>
59 #include <sys/vnode.h>
60 #include <sys/eventhandler.h>
64 #include <vm/vm_extern.h>
65 #include <vm/vm_map.h>
66 #include <vm/vm_object.h>
67 #include <vm/vm_page.h>
68 #include <vm/vm_param.h>
70 #include <machine/cpu.h>
71 #include <machine/md_var.h>
72 #include <machine/pcb.h>
73 #include <machine/specialreg.h>
74 #include <machine/trap.h>
76 #include <x86/linux/linux_x86.h>
77 #include <amd64/linux/linux.h>
78 #include <amd64/linux/linux_proto.h>
79 #include <compat/linux/linux_elf.h>
80 #include <compat/linux/linux_emul.h>
81 #include <compat/linux/linux_fork.h>
82 #include <compat/linux/linux_ioctl.h>
83 #include <compat/linux/linux_mib.h>
84 #include <compat/linux/linux_misc.h>
85 #include <compat/linux/linux_signal.h>
86 #include <compat/linux/linux_sysproto.h>
87 #include <compat/linux/linux_util.h>
88 #include <compat/linux/linux_vdso.h>
90 #include <x86/linux/linux_x86_sigframe.h>
92 MODULE_VERSION(linux64, 1);
94 #define LINUX_VDSOPAGE_SIZE PAGE_SIZE * 2
95 #define LINUX_VDSOPAGE_LA48 (VM_MAXUSER_ADDRESS_LA48 - \
97 #define LINUX_SHAREDPAGE_LA48 (LINUX_VDSOPAGE_LA48 - PAGE_SIZE)
99 * PAGE_SIZE - the size
100 * of the native SHAREDPAGE
102 #define LINUX_USRSTACK_LA48 LINUX_SHAREDPAGE_LA48
103 #define LINUX_PS_STRINGS_LA48 (LINUX_USRSTACK_LA48 - \
104 sizeof(struct ps_strings))
106 static int linux_szsigcode;
107 static vm_object_t linux_vdso_obj;
108 static char *linux_vdso_mapping;
109 extern char _binary_linux_vdso_so_o_start;
110 extern char _binary_linux_vdso_so_o_end;
111 static vm_offset_t linux_vdso_base;
113 extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL];
114 extern const char *linux_syscallnames[];
116 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler);
118 static int linux_copyout_strings(struct image_params *imgp,
119 uintptr_t *stack_base);
120 static bool linux_trans_osrel(const Elf_Note *note, int32_t *osrel);
121 static void linux_vdso_install(const void *param);
122 static void linux_vdso_deinstall(const void *param);
123 static void linux_vdso_reloc(char *mapping, Elf_Addr offset);
124 static void linux_set_syscall_retval(struct thread *td, int error);
125 static int linux_fetch_syscall_args(struct thread *td);
126 static void linux_exec_setregs(struct thread *td, struct image_params *imgp,
128 static void linux_exec_sysvec_init(void *param);
129 static int linux_on_exec_vmspace(struct proc *p,
130 struct image_params *imgp);
131 static void linux_set_fork_retval(struct thread *td);
132 static int linux_vsyscall(struct thread *td);
134 LINUX_VDSO_SYM_INTPTR(linux_rt_sigcode);
135 LINUX_VDSO_SYM_CHAR(linux_platform);
136 LINUX_VDSO_SYM_INTPTR(kern_timekeep_base);
137 LINUX_VDSO_SYM_INTPTR(kern_tsc_selector);
138 LINUX_VDSO_SYM_INTPTR(kern_cpu_selector);
141 linux_fetch_syscall_args(struct thread *td)
144 struct trapframe *frame;
145 struct syscall_args *sa;
148 frame = td->td_frame;
151 sa->args[0] = frame->tf_rdi;
152 sa->args[1] = frame->tf_rsi;
153 sa->args[2] = frame->tf_rdx;
154 sa->args[3] = frame->tf_rcx;
155 sa->args[4] = frame->tf_r8;
156 sa->args[5] = frame->tf_r9;
157 sa->code = frame->tf_rax;
158 sa->original_code = sa->code;
160 if (sa->code >= p->p_sysent->sv_size)
162 sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1];
164 sa->callp = &p->p_sysent->sv_table[sa->code];
166 td->td_retval[0] = 0;
171 linux_set_syscall_retval(struct thread *td, int error)
173 struct trapframe *frame;
175 frame = td->td_frame;
179 frame->tf_rax = td->td_retval[0];
180 frame->tf_r10 = frame->tf_rcx;
185 * Reconstruct pc, we know that 'syscall' is 2 bytes,
186 * lcall $X,y is 7 bytes, int 0x80 is 2 bytes.
187 * We saved this in tf_err.
190 frame->tf_rip -= frame->tf_err;
191 frame->tf_r10 = frame->tf_rcx;
198 frame->tf_rax = bsd_to_linux_errno(error);
199 frame->tf_r10 = frame->tf_rcx;
204 * Differently from FreeBSD native ABI, on Linux only %rcx
205 * and %r11 values are not preserved across the syscall.
206 * Require full context restore to get all registers except
207 * those two restored at return to usermode.
209 * XXX: Would be great to be able to avoid PCB_FULL_IRET
210 * for the error == 0 case.
212 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
216 linux_set_fork_retval(struct thread *td)
218 struct trapframe *frame = td->td_frame;
224 linux_copyout_auxargs(struct image_params *imgp, uintptr_t base)
227 Elf_Auxinfo *argarray, *pos;
229 int error, issetugid;
232 args = (Elf64_Auxargs *)imgp->auxargs;
233 argarray = pos = malloc(LINUX_AT_COUNT * sizeof(*pos), M_TEMP,
236 issetugid = p->p_flag & P_SUGID ? 1 : 0;
237 AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR, linux_vdso_base);
238 AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature);
239 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
240 AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz);
241 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
242 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
243 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
244 AUXARGS_ENTRY(pos, AT_BASE, args->base);
245 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
246 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
247 AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid);
248 AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid);
249 AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid);
250 AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid);
251 AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid);
252 AUXARGS_ENTRY_PTR(pos, LINUX_AT_RANDOM, imgp->canary);
253 AUXARGS_ENTRY(pos, LINUX_AT_HWCAP2, 0);
254 if (imgp->execpathp != 0)
255 AUXARGS_ENTRY_PTR(pos, LINUX_AT_EXECFN, imgp->execpathp);
256 if (args->execfd != -1)
257 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
258 AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform));
259 AUXARGS_ENTRY(pos, AT_NULL, 0);
261 free(imgp->auxargs, M_TEMP);
262 imgp->auxargs = NULL;
263 KASSERT(pos - argarray <= LINUX_AT_COUNT, ("Too many auxargs"));
265 error = copyout(argarray, (void *)base,
266 sizeof(*argarray) * LINUX_AT_COUNT);
267 free(argarray, M_TEMP);
272 * Copy strings out to the new process address space, constructing new arg
273 * and env vector tables. Return a pointer to the base so that it can be used
274 * as the initial stack pointer.
277 linux_copyout_strings(struct image_params *imgp, uintptr_t *stack_base)
279 int argc, envc, error;
282 uintptr_t destp, ustringp;
283 struct ps_strings *arginfo;
284 char canary[LINUX_AT_RANDOM_LEN];
289 arginfo = (struct ps_strings *)PROC_PS_STRINGS(p);
290 destp = (uintptr_t)arginfo;
292 if (imgp->execpath != NULL && imgp->auxargs != NULL) {
293 execpath_len = strlen(imgp->execpath) + 1;
294 destp -= execpath_len;
295 destp = rounddown2(destp, sizeof(void *));
296 imgp->execpathp = (void *)destp;
297 error = copyout(imgp->execpath, imgp->execpathp, execpath_len);
302 /* Prepare the canary for SSP. */
303 arc4rand(canary, sizeof(canary), 0);
304 destp -= roundup(sizeof(canary), sizeof(void *));
305 imgp->canary = (void *)destp;
306 error = copyout(canary, imgp->canary, sizeof(canary));
310 /* Allocate room for the argument and environment strings. */
311 destp -= ARG_MAX - imgp->args->stringspace;
312 destp = rounddown2(destp, sizeof(void *));
317 * Allocate room on the stack for the ELF auxargs
318 * array. It has LINUX_AT_COUNT entries.
320 destp -= LINUX_AT_COUNT * sizeof(Elf64_Auxinfo);
321 destp = rounddown2(destp, sizeof(void *));
324 vectp = (char **)destp;
327 * Allocate room for the argv[] and env vectors including the
328 * terminating NULL pointers.
330 vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
333 * Starting with 2.24, glibc depends on a 16-byte stack alignment.
334 * One "long argc" will be prepended later.
336 vectp = (char **)((((uintptr_t)vectp + 8) & ~0xF) - 8);
338 /* vectp also becomes our initial stack base. */
339 *stack_base = (uintptr_t)vectp;
341 stringp = imgp->args->begin_argv;
342 argc = imgp->args->argc;
343 envc = imgp->args->envc;
345 /* Copy out strings - arguments and environment. */
346 error = copyout(stringp, (void *)ustringp,
347 ARG_MAX - imgp->args->stringspace);
351 /* Fill in "ps_strings" struct for ps, w, etc. */
352 if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 ||
353 suword(&arginfo->ps_nargvstr, argc) != 0)
356 /* Fill in argument portion of vector table. */
357 for (; argc > 0; --argc) {
358 if (suword(vectp++, ustringp) != 0)
360 while (*stringp++ != 0)
365 /* A null vector table pointer separates the argp's from the envp's. */
366 if (suword(vectp++, 0) != 0)
369 if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 ||
370 suword(&arginfo->ps_nenvstr, envc) != 0)
373 /* Fill in environment portion of vector table. */
374 for (; envc > 0; --envc) {
375 if (suword(vectp++, ustringp) != 0)
377 while (*stringp++ != 0)
382 /* The end of the vector table is a null pointer. */
383 if (suword(vectp, 0) != 0)
388 error = imgp->sysent->sv_copyout_auxargs(imgp,
398 * Reset registers to default values on exec.
401 linux_exec_setregs(struct thread *td, struct image_params *imgp,
404 struct trapframe *regs;
406 register_t saved_rflags;
411 if (td->td_proc->p_md.md_ldt != NULL)
416 clear_pcb_flags(pcb, PCB_32BIT);
417 pcb->pcb_initial_fpucw = __LINUX_NPXCW__;
418 set_pcb_flags(pcb, PCB_FULL_IRET);
420 saved_rflags = regs->tf_rflags & PSL_T;
421 bzero((char *)regs, sizeof(struct trapframe));
422 regs->tf_rip = imgp->entry_addr;
423 regs->tf_rsp = stack;
424 regs->tf_rflags = PSL_USER | saved_rflags;
425 regs->tf_ss = _udatasel;
426 regs->tf_cs = _ucodesel;
427 regs->tf_ds = _udatasel;
428 regs->tf_es = _udatasel;
429 regs->tf_fs = _ufssel;
430 regs->tf_gs = _ugssel;
431 regs->tf_flags = TF_HASSEGS;
433 x86_clear_dbregs(pcb);
436 * Drop the FP state if we hold it, so that the process gets a
437 * clean FP state if it uses the FPU again.
443 * Copied from amd64/amd64/machdep.c
445 * XXX fpu state need? don't think so
448 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
451 struct l_ucontext uc;
452 struct l_sigcontext *context;
453 struct trapframe *regs;
454 unsigned long rflags;
460 error = copyin((void *)regs->tf_rbx, &uc, sizeof(uc));
465 context = &uc.uc_mcontext;
466 rflags = context->sc_rflags;
469 * Don't allow users to change privileged or reserved flags.
472 * XXX do allow users to change the privileged flag PSL_RF.
473 * The cpu sets PSL_RF in tf_rflags for faults. Debuggers
474 * should sometimes set it there too. tf_rflags is kept in
475 * the signal context during signal handling and there is no
476 * other place to remember it, so the PSL_RF bit may be
477 * corrupted by the signal handler without us knowing.
478 * Corruption of the PSL_RF bit at worst causes one more or
479 * one less debugger trap, so allowing it is fairly harmless.
481 if (!EFL_SECURE(rflags & ~PSL_RF, regs->tf_rflags & ~PSL_RF)) {
482 uprintf("pid %d comm %s linux mangled rflags %#lx\n",
483 p->p_pid, p->p_comm, rflags);
488 * Don't allow users to load a valid privileged %cs. Let the
489 * hardware check for invalid selectors, excess privilege in
490 * other selectors, invalid %eip's and invalid %esp's.
492 if (!CS_SECURE(context->sc_cs)) {
493 uprintf("pid %d comm %s linux mangled cs %#x\n",
494 p->p_pid, p->p_comm, context->sc_cs);
495 ksiginfo_init_trap(&ksi);
496 ksi.ksi_signo = SIGBUS;
497 ksi.ksi_code = BUS_OBJERR;
498 ksi.ksi_trapno = T_PROTFLT;
499 ksi.ksi_addr = (void *)regs->tf_rip;
500 trapsignal(td, &ksi);
504 linux_to_bsd_sigset(&uc.uc_sigmask, &bmask);
505 kern_sigprocmask(td, SIG_SETMASK, &bmask, NULL, 0);
507 regs->tf_rdi = context->sc_rdi;
508 regs->tf_rsi = context->sc_rsi;
509 regs->tf_rdx = context->sc_rdx;
510 regs->tf_rbp = context->sc_rbp;
511 regs->tf_rbx = context->sc_rbx;
512 regs->tf_rcx = context->sc_rcx;
513 regs->tf_rax = context->sc_rax;
514 regs->tf_rip = context->sc_rip;
515 regs->tf_rsp = context->sc_rsp;
516 regs->tf_r8 = context->sc_r8;
517 regs->tf_r9 = context->sc_r9;
518 regs->tf_r10 = context->sc_r10;
519 regs->tf_r11 = context->sc_r11;
520 regs->tf_r12 = context->sc_r12;
521 regs->tf_r13 = context->sc_r13;
522 regs->tf_r14 = context->sc_r14;
523 regs->tf_r15 = context->sc_r15;
524 regs->tf_cs = context->sc_cs;
525 regs->tf_err = context->sc_err;
526 regs->tf_rflags = rflags;
528 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
529 return (EJUSTRETURN);
533 * copied from amd64/amd64/machdep.c
535 * Send an interrupt to process.
538 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
540 struct l_rt_sigframe sf, *sfp;
545 struct trapframe *regs;
547 int oonstack, issiginfo;
551 PROC_LOCK_ASSERT(p, MA_OWNED);
552 sig = linux_translate_traps(ksi->ksi_signo, ksi->ksi_trapno);
554 issiginfo = SIGISMEMBER(psp->ps_siginfo, sig);
555 code = ksi->ksi_code;
556 mtx_assert(&psp->ps_mtx, MA_OWNED);
558 oonstack = sigonstack(regs->tf_rsp);
560 LINUX_CTR4(rt_sendsig, "%p, %d, %p, %u",
561 catcher, sig, mask, code);
563 /* Save user context. */
564 bzero(&sf, sizeof(sf));
565 bsd_to_linux_sigset(mask, &sf.sf_uc.uc_sigmask);
566 sf.sf_uc.uc_mcontext.sc_mask = sf.sf_uc.uc_sigmask;
568 sf.sf_uc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp);
569 sf.sf_uc.uc_stack.ss_size = td->td_sigstk.ss_size;
570 sf.sf_uc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
571 ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE;
573 sf.sf_uc.uc_mcontext.sc_rdi = regs->tf_rdi;
574 sf.sf_uc.uc_mcontext.sc_rsi = regs->tf_rsi;
575 sf.sf_uc.uc_mcontext.sc_rdx = regs->tf_rdx;
576 sf.sf_uc.uc_mcontext.sc_rbp = regs->tf_rbp;
577 sf.sf_uc.uc_mcontext.sc_rbx = regs->tf_rbx;
578 sf.sf_uc.uc_mcontext.sc_rcx = regs->tf_rcx;
579 sf.sf_uc.uc_mcontext.sc_rax = regs->tf_rax;
580 sf.sf_uc.uc_mcontext.sc_rip = regs->tf_rip;
581 sf.sf_uc.uc_mcontext.sc_rsp = regs->tf_rsp;
582 sf.sf_uc.uc_mcontext.sc_r8 = regs->tf_r8;
583 sf.sf_uc.uc_mcontext.sc_r9 = regs->tf_r9;
584 sf.sf_uc.uc_mcontext.sc_r10 = regs->tf_r10;
585 sf.sf_uc.uc_mcontext.sc_r11 = regs->tf_r11;
586 sf.sf_uc.uc_mcontext.sc_r12 = regs->tf_r12;
587 sf.sf_uc.uc_mcontext.sc_r13 = regs->tf_r13;
588 sf.sf_uc.uc_mcontext.sc_r14 = regs->tf_r14;
589 sf.sf_uc.uc_mcontext.sc_r15 = regs->tf_r15;
590 sf.sf_uc.uc_mcontext.sc_cs = regs->tf_cs;
591 sf.sf_uc.uc_mcontext.sc_rflags = regs->tf_rflags;
592 sf.sf_uc.uc_mcontext.sc_err = regs->tf_err;
593 sf.sf_uc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code);
594 sf.sf_uc.uc_mcontext.sc_cr2 = (register_t)ksi->ksi_addr;
596 /* Allocate space for the signal handler context. */
597 if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack &&
598 SIGISMEMBER(psp->ps_sigonstack, sig)) {
599 sp = (caddr_t)td->td_sigstk.ss_sp + td->td_sigstk.ss_size;
601 sp = (caddr_t)regs->tf_rsp - 128;
602 sp -= sizeof(struct l_rt_sigframe);
603 /* Align to 16 bytes. */
604 sfp = (struct l_rt_sigframe *)((unsigned long)sp & ~0xFul);
606 mtx_unlock(&psp->ps_mtx);
609 /* Translate the signal. */
610 sig = bsd_to_linux_signal(sig);
611 /* Fill in POSIX parts. */
612 siginfo_to_lsiginfo(&ksi->ksi_info, &sf.sf_si, sig);
614 /* Copy the sigframe out to the user's stack. */
615 if (copyout(&sf, sfp, sizeof(*sfp)) != 0) {
616 uprintf("pid %d comm %s has trashed its stack, killing\n",
617 p->p_pid, p->p_comm);
622 /* Build the argument list for the signal handler. */
623 regs->tf_rdi = sig; /* arg 1 in %rdi */
626 regs->tf_rsi = (register_t)&sfp->sf_si; /* arg 2 in %rsi */
627 regs->tf_rdx = (register_t)&sfp->sf_uc; /* arg 3 in %rdx */
632 regs->tf_rcx = (register_t)catcher;
633 regs->tf_rsp = (long)sfp;
634 regs->tf_rip = linux_rt_sigcode;
635 regs->tf_rflags &= ~(PSL_T | PSL_D);
636 regs->tf_cs = _ucodesel;
637 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
639 mtx_lock(&psp->ps_mtx);
642 #define LINUX_VSYSCALL_START (-10UL << 20)
643 #define LINUX_VSYSCALL_SZ 1024
645 const unsigned long linux_vsyscall_vector[] = {
646 LINUX_SYS_gettimeofday,
647 LINUX_SYS_linux_time,
648 LINUX_SYS_linux_getcpu,
652 linux_vsyscall(struct thread *td)
654 struct trapframe *frame;
659 frame = td->td_frame;
661 /* Check %rip for vsyscall area. */
662 if (__predict_true(frame->tf_rip < LINUX_VSYSCALL_START))
664 if ((frame->tf_rip & (LINUX_VSYSCALL_SZ - 1)) != 0)
666 code = (frame->tf_rip - LINUX_VSYSCALL_START) / LINUX_VSYSCALL_SZ;
667 if (code >= nitems(linux_vsyscall_vector))
671 * vsyscall called as callq *(%rax), so we must
672 * use return address from %rsp and also fixup %rsp.
674 error = copyin((void *)frame->tf_rsp, &retqaddr, sizeof(retqaddr));
678 frame->tf_rip = retqaddr;
679 frame->tf_rax = linux_vsyscall_vector[code];
682 traced = (frame->tf_flags & PSL_T);
684 amd64_syscall(td, traced);
689 struct sysentvec elf_linux_sysvec = {
690 .sv_size = LINUX_SYS_MAXSYSCALL,
691 .sv_table = linux_sysent,
692 .sv_fixup = __elfN(freebsd_fixup),
693 .sv_sendsig = linux_rt_sendsig,
694 .sv_sigcode = &_binary_linux_vdso_so_o_start,
695 .sv_szsigcode = &linux_szsigcode,
696 .sv_name = "Linux ELF64",
697 .sv_coredump = elf64_coredump,
698 .sv_elf_core_osabi = ELFOSABI_NONE,
699 .sv_elf_core_abi_vendor = LINUX_ABI_VENDOR,
700 .sv_elf_core_prepare_notes = linux64_prepare_notes,
701 .sv_imgact_try = linux_exec_imgact_try,
702 .sv_minsigstksz = LINUX_MINSIGSTKSZ,
703 .sv_minuser = VM_MIN_ADDRESS,
704 .sv_maxuser = VM_MAXUSER_ADDRESS_LA48,
705 .sv_usrstack = LINUX_USRSTACK_LA48,
706 .sv_psstrings = LINUX_PS_STRINGS_LA48,
707 .sv_psstringssz = sizeof(struct ps_strings),
708 .sv_stackprot = VM_PROT_ALL,
709 .sv_copyout_auxargs = linux_copyout_auxargs,
710 .sv_copyout_strings = linux_copyout_strings,
711 .sv_setregs = linux_exec_setregs,
714 .sv_flags = SV_ABI_LINUX | SV_LP64 | SV_SHP | SV_SIG_DISCIGN |
715 SV_SIG_WAITNDQ | SV_TIMEKEEP,
716 .sv_set_syscall_retval = linux_set_syscall_retval,
717 .sv_fetch_syscall_args = linux_fetch_syscall_args,
718 .sv_syscallnames = linux_syscallnames,
719 .sv_shared_page_base = LINUX_SHAREDPAGE_LA48,
720 .sv_shared_page_len = PAGE_SIZE,
721 .sv_schedtail = linux_schedtail,
722 .sv_thread_detach = linux_thread_detach,
723 .sv_trap = linux_vsyscall,
724 .sv_onexec = linux_on_exec_vmspace,
725 .sv_onexit = linux_on_exit,
726 .sv_ontdexit = linux_thread_dtor,
727 .sv_setid_allowed = &linux_setid_allowed_query,
728 .sv_set_fork_retval = linux_set_fork_retval,
732 linux_on_exec_vmspace(struct proc *p, struct image_params *imgp)
736 error = linux_map_vdso(p, linux_vdso_obj, linux_vdso_base,
737 LINUX_VDSOPAGE_SIZE, imgp);
739 linux_on_exec(p, imgp);
744 * linux_vdso_install() and linux_exec_sysvec_init() must be called
745 * after exec_sysvec_init() which is SI_SUB_EXEC (SI_ORDER_ANY).
748 linux_exec_sysvec_init(void *param)
750 l_uintptr_t *ktimekeep_base, *ktsc_selector;
751 struct sysentvec *sv;
755 amd64_lower_shared_page(sv);
756 /* Fill timekeep_base */
757 exec_sysvec_init(sv);
759 tkoff = kern_timekeep_base - linux_vdso_base;
760 ktimekeep_base = (l_uintptr_t *)(linux_vdso_mapping + tkoff);
761 *ktimekeep_base = sv->sv_shared_page_base + sv->sv_timekeep_offset;
763 tkoff = kern_tsc_selector - linux_vdso_base;
764 ktsc_selector = (l_uintptr_t *)(linux_vdso_mapping + tkoff);
765 *ktsc_selector = linux_vdso_tsc_selector_idx();
767 printf("Linux x86-64 vDSO tsc_selector: %lu\n", *ktsc_selector);
769 tkoff = kern_cpu_selector - linux_vdso_base;
770 ktsc_selector = (l_uintptr_t *)(linux_vdso_mapping + tkoff);
771 *ktsc_selector = linux_vdso_cpu_selector_idx();
773 printf("Linux x86-64 vDSO cpu_selector: %lu\n", *ktsc_selector);
775 SYSINIT(elf_linux_exec_sysvec_init, SI_SUB_EXEC + 1, SI_ORDER_ANY,
776 linux_exec_sysvec_init, &elf_linux_sysvec);
779 linux_vdso_install(const void *param)
781 char *vdso_start = &_binary_linux_vdso_so_o_start;
782 char *vdso_end = &_binary_linux_vdso_so_o_end;
784 linux_szsigcode = vdso_end - vdso_start;
785 MPASS(linux_szsigcode <= LINUX_VDSOPAGE_SIZE);
787 linux_vdso_base = LINUX_VDSOPAGE_LA48;
788 if (hw_lower_amd64_sharedpage != 0)
789 linux_vdso_base -= PAGE_SIZE;
791 __elfN(linux_vdso_fixup)(vdso_start, linux_vdso_base);
793 linux_vdso_obj = __elfN(linux_shared_page_init)
794 (&linux_vdso_mapping, LINUX_VDSOPAGE_SIZE);
795 bcopy(vdso_start, linux_vdso_mapping, linux_szsigcode);
797 linux_vdso_reloc(linux_vdso_mapping, linux_vdso_base);
799 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC + 1, SI_ORDER_FIRST,
800 linux_vdso_install, NULL);
803 linux_vdso_deinstall(const void *param)
806 __elfN(linux_shared_page_fini)(linux_vdso_obj,
807 linux_vdso_mapping, LINUX_VDSOPAGE_SIZE);
809 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
810 linux_vdso_deinstall, NULL);
813 linux_vdso_reloc(char *mapping, Elf_Addr offset)
815 const Elf_Ehdr *ehdr;
816 const Elf_Shdr *shdr;
817 Elf64_Addr *where, val;
818 Elf_Size rtype, symidx;
819 const Elf_Rela *rela;
820 Elf_Addr addr, addend;
827 ehdr = (const Elf_Ehdr *)mapping;
828 shdr = (const Elf_Shdr *)(mapping + ehdr->e_shoff);
829 for (i = 0; i < ehdr->e_shnum; i++)
831 switch (shdr[i].sh_type) {
833 printf("Linux x86_64 vDSO: unexpected Rel section\n");
836 rela = (const Elf_Rela *)(mapping + shdr[i].sh_offset);
837 relacnt = shdr[i].sh_size / sizeof(*rela);
841 for (j = 0; j < relacnt; j++, rela++) {
842 where = (Elf_Addr *)(mapping + rela->r_offset);
843 addend = rela->r_addend;
844 rtype = ELF_R_TYPE(rela->r_info);
845 symidx = ELF_R_SYM(rela->r_info);
848 case R_X86_64_NONE: /* none */
851 case R_X86_64_RELATIVE: /* B + A */
852 addr = (Elf_Addr)(offset + addend);
857 case R_X86_64_IRELATIVE:
858 printf("Linux x86_64 vDSO: unexpected ifunc relocation, "
859 "symbol index %ld\n", symidx);
862 printf("Linux x86_64 vDSO: unexpected relocation type %ld, "
863 "symbol index %ld\n", rtype, symidx);
868 static char GNULINUX_ABI_VENDOR[] = "GNU";
869 static int GNULINUX_ABI_DESC = 0;
872 linux_trans_osrel(const Elf_Note *note, int32_t *osrel)
874 const Elf32_Word *desc;
877 p = (uintptr_t)(note + 1);
878 p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
880 desc = (const Elf32_Word *)p;
881 if (desc[0] != GNULINUX_ABI_DESC)
885 * For Linux we encode osrel using the Linux convention of
886 * (version << 16) | (major << 8) | (minor)
887 * See macro in linux_mib.h
889 *osrel = LINUX_KERNVER(desc[1], desc[2], desc[3]);
894 static Elf_Brandnote linux64_brandnote = {
895 .hdr.n_namesz = sizeof(GNULINUX_ABI_VENDOR),
898 .vendor = GNULINUX_ABI_VENDOR,
899 .flags = BN_TRANSLATE_OSREL,
900 .trans_osrel = linux_trans_osrel
903 static Elf64_Brandinfo linux_glibc2brand = {
904 .brand = ELFOSABI_LINUX,
905 .machine = EM_X86_64,
906 .compat_3_brand = "Linux",
907 .emul_path = linux_emul_path,
908 .interp_path = "/lib64/ld-linux-x86-64.so.2",
909 .sysvec = &elf_linux_sysvec,
910 .interp_newpath = NULL,
911 .brand_note = &linux64_brandnote,
912 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
915 static Elf64_Brandinfo linux_glibc2brandshort = {
916 .brand = ELFOSABI_LINUX,
917 .machine = EM_X86_64,
918 .compat_3_brand = "Linux",
919 .emul_path = linux_emul_path,
920 .interp_path = "/lib64/ld-linux.so.2",
921 .sysvec = &elf_linux_sysvec,
922 .interp_newpath = NULL,
923 .brand_note = &linux64_brandnote,
924 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
927 static Elf64_Brandinfo linux_muslbrand = {
928 .brand = ELFOSABI_LINUX,
929 .machine = EM_X86_64,
930 .compat_3_brand = "Linux",
931 .emul_path = linux_emul_path,
932 .interp_path = "/lib/ld-musl-x86_64.so.1",
933 .sysvec = &elf_linux_sysvec,
934 .interp_newpath = NULL,
935 .brand_note = &linux64_brandnote,
936 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE |
937 LINUX_BI_FUTEX_REQUEUE
940 Elf64_Brandinfo *linux_brandlist[] = {
942 &linux_glibc2brandshort,
948 linux64_elf_modevent(module_t mod, int type, void *data)
950 Elf64_Brandinfo **brandinfo;
952 struct linux_ioctl_handler **lihp;
958 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
960 if (elf64_insert_brand_entry(*brandinfo) < 0)
963 SET_FOREACH(lihp, linux_ioctl_handler_set)
964 linux_ioctl_register_handler(*lihp);
965 stclohz = (stathz ? stathz : hz);
967 printf("Linux x86-64 ELF exec handler installed\n");
969 printf("cannot insert Linux x86-64 ELF brand handler\n");
972 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
974 if (elf64_brand_inuse(*brandinfo))
977 for (brandinfo = &linux_brandlist[0];
978 *brandinfo != NULL; ++brandinfo)
979 if (elf64_remove_brand_entry(*brandinfo) < 0)
983 SET_FOREACH(lihp, linux_ioctl_handler_set)
984 linux_ioctl_unregister_handler(*lihp);
986 printf("Linux x86_64 ELF exec handler removed\n");
988 printf("Could not deinstall Linux x86_64 ELF interpreter entry\n");
996 static moduledata_t linux64_elf_mod = {
998 linux64_elf_modevent,
1002 DECLARE_MODULE_TIED(linux64elf, linux64_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);
1003 MODULE_DEPEND(linux64elf, linux_common, 1, 1, 1);
1004 FEATURE(linux64, "Linux 64bit support");