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 #include "opt_compat.h"
39 #define __ELF_WORD_SIZE 64
41 #include <sys/param.h>
42 #include <sys/systm.h>
44 #include <sys/fcntl.h>
45 #include <sys/imgact.h>
46 #include <sys/imgact_elf.h>
47 #include <sys/kernel.h>
50 #include <sys/malloc.h>
51 #include <sys/module.h>
52 #include <sys/mutex.h>
54 #include <sys/resourcevar.h>
55 #include <sys/signalvar.h>
56 #include <sys/syscallsubr.h>
57 #include <sys/sysctl.h>
58 #include <sys/sysent.h>
59 #include <sys/sysproto.h>
60 #include <sys/vnode.h>
61 #include <sys/eventhandler.h>
65 #include <vm/vm_extern.h>
66 #include <vm/vm_map.h>
67 #include <vm/vm_object.h>
68 #include <vm/vm_page.h>
69 #include <vm/vm_param.h>
71 #include <machine/cpu.h>
72 #include <machine/md_var.h>
73 #include <machine/pcb.h>
74 #include <machine/specialreg.h>
76 #include <amd64/linux/linux.h>
77 #include <amd64/linux/linux_proto.h>
78 #include <compat/linux/linux_emul.h>
79 #include <compat/linux/linux_futex.h>
80 #include <compat/linux/linux_ioctl.h>
81 #include <compat/linux/linux_mib.h>
82 #include <compat/linux/linux_misc.h>
83 #include <compat/linux/linux_signal.h>
84 #include <compat/linux/linux_sysproto.h>
85 #include <compat/linux/linux_util.h>
86 #include <compat/linux/linux_vdso.h>
88 MODULE_VERSION(linux64, 1);
91 SYSCTL_PROC(_compat_linux, OID_AUTO, debug,
92 CTLTYPE_STRING | CTLFLAG_RW,
93 0, 0, linux_sysctl_debug, "A",
94 "Linux 64 debugging control");
98 * Allow the sendsig functions to use the ldebug() facility even though they
99 * are not syscalls themselves. Map them to syscall 0. This is slightly less
100 * bogus than using ldebug(sigreturn).
102 #define LINUX_SYS_linux_rt_sendsig 0
104 const char *linux_kplatform;
105 static int linux_szsigcode;
106 static vm_object_t linux_shared_page_obj;
107 static char *linux_shared_page_mapping;
108 extern char _binary_linux_locore_o_start;
109 extern char _binary_linux_locore_o_end;
111 extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL];
113 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler);
115 static register_t * linux_copyout_strings(struct image_params *imgp);
116 static int linux_fixup_elf(register_t **stack_base,
117 struct image_params *iparams);
118 static bool linux_trans_osrel(const Elf_Note *note, int32_t *osrel);
119 static void linux_vdso_install(void *param);
120 static void linux_vdso_deinstall(void *param);
121 static void linux_set_syscall_retval(struct thread *td, int error);
122 static int linux_fetch_syscall_args(struct thread *td);
123 static void linux_exec_setregs(struct thread *td, struct image_params *imgp,
125 static int linux_vsyscall(struct thread *td);
127 #define LINUX_T_UNKNOWN 255
128 static int _bsd_to_linux_trapcode[] = {
129 LINUX_T_UNKNOWN, /* 0 */
130 6, /* 1 T_PRIVINFLT */
131 LINUX_T_UNKNOWN, /* 2 */
133 LINUX_T_UNKNOWN, /* 4 */
134 LINUX_T_UNKNOWN, /* 5 */
135 16, /* 6 T_ARITHTRAP */
136 254, /* 7 T_ASTFLT */
137 LINUX_T_UNKNOWN, /* 8 */
138 13, /* 9 T_PROTFLT */
139 1, /* 10 T_TRCTRAP */
140 LINUX_T_UNKNOWN, /* 11 */
141 14, /* 12 T_PAGEFLT */
142 LINUX_T_UNKNOWN, /* 13 */
143 17, /* 14 T_ALIGNFLT */
144 LINUX_T_UNKNOWN, /* 15 */
145 LINUX_T_UNKNOWN, /* 16 */
146 LINUX_T_UNKNOWN, /* 17 */
152 8, /* 23 T_DOUBLEFLT */
153 9, /* 24 T_FPOPFLT */
154 10, /* 25 T_TSSFLT */
155 11, /* 26 T_SEGNPFLT */
156 12, /* 27 T_STKFLT */
158 19, /* 29 T_XMMFLT */
159 15 /* 30 T_RESERVED */
161 #define bsd_to_linux_trapcode(code) \
162 ((code)<nitems(_bsd_to_linux_trapcode)? \
163 _bsd_to_linux_trapcode[(code)]: \
166 LINUX_VDSO_SYM_INTPTR(linux_rt_sigcode);
167 LINUX_VDSO_SYM_CHAR(linux_platform);
170 * If FreeBSD & Linux have a difference of opinion about what a trap
171 * means, deal with it here.
176 linux_translate_traps(int signal, int trap_code)
179 if (signal != SIGBUS)
193 linux_fetch_syscall_args(struct thread *td)
196 struct trapframe *frame;
197 struct syscall_args *sa;
200 frame = td->td_frame;
203 sa->args[0] = frame->tf_rdi;
204 sa->args[1] = frame->tf_rsi;
205 sa->args[2] = frame->tf_rdx;
206 sa->args[3] = frame->tf_rcx;
207 sa->args[4] = frame->tf_r8;
208 sa->args[5] = frame->tf_r9;
209 sa->code = frame->tf_rax;
211 if (sa->code >= p->p_sysent->sv_size)
213 sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1];
215 sa->callp = &p->p_sysent->sv_table[sa->code];
216 sa->narg = sa->callp->sy_narg;
218 td->td_retval[0] = 0;
223 linux_set_syscall_retval(struct thread *td, int error)
225 struct trapframe *frame = td->td_frame;
228 * On Linux only %rcx and %r11 values are not preserved across
229 * the syscall. So, do not clobber %rdx and %r10.
231 td->td_retval[1] = frame->tf_rdx;
232 if (error != EJUSTRETURN)
233 frame->tf_r10 = frame->tf_rcx;
235 cpu_set_syscall_retval(td, error);
237 /* Restore all registers. */
238 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
242 linux_fixup_elf(register_t **stack_base, struct image_params *imgp)
247 struct ps_strings *arginfo;
252 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
254 KASSERT(curthread->td_proc == imgp->proc,
255 ("unsafe linux_fixup_elf(), should be curproc"));
256 base = (Elf64_Addr *)*stack_base;
257 args = (Elf64_Auxargs *)imgp->auxargs;
258 pos = base + (imgp->args->argc + imgp->args->envc + 2);
260 issetugid = p->p_flag & P_SUGID ? 1 : 0;
261 AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR,
262 imgp->proc->p_sysent->sv_shared_page_base);
263 AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature);
264 AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz);
265 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
266 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
267 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
268 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
269 AUXARGS_ENTRY(pos, AT_BASE, args->base);
270 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
271 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
272 AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid);
273 AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid);
274 AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid);
275 AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid);
276 AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid);
277 AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform));
278 AUXARGS_ENTRY(pos, LINUX_AT_RANDOM, imgp->canary);
279 if (imgp->execpathp != 0)
280 AUXARGS_ENTRY(pos, LINUX_AT_EXECFN, imgp->execpathp);
281 if (args->execfd != -1)
282 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
283 AUXARGS_ENTRY(pos, AT_NULL, 0);
284 free(imgp->auxargs, M_TEMP);
285 imgp->auxargs = NULL;
288 suword(base, (uint64_t)imgp->args->argc);
290 *stack_base = (register_t *)base;
295 * Copy strings out to the new process address space, constructing new arg
296 * and env vector tables. Return a pointer to the base so that it can be used
297 * as the initial stack pointer.
300 linux_copyout_strings(struct image_params *imgp)
304 char *stringp, *destp;
305 register_t *stack_base;
306 struct ps_strings *arginfo;
307 char canary[LINUX_AT_RANDOM_LEN];
311 /* Calculate string base and vector table pointers. */
312 if (imgp->execpath != NULL && imgp->auxargs != NULL)
313 execpath_len = strlen(imgp->execpath) + 1;
318 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
319 destp = (caddr_t)arginfo - SPARE_USRSPACE -
320 roundup(sizeof(canary), sizeof(char *)) -
321 roundup(execpath_len, sizeof(char *)) -
322 roundup(ARG_MAX - imgp->args->stringspace, sizeof(char *));
324 if (execpath_len != 0) {
325 imgp->execpathp = (uintptr_t)arginfo - execpath_len;
326 copyout(imgp->execpath, (void *)imgp->execpathp, execpath_len);
329 /* Prepare the canary for SSP. */
330 arc4rand(canary, sizeof(canary), 0);
331 imgp->canary = (uintptr_t)arginfo -
332 roundup(execpath_len, sizeof(char *)) -
333 roundup(sizeof(canary), sizeof(char *));
334 copyout(canary, (void *)imgp->canary, sizeof(canary));
336 vectp = (char **)destp;
339 * Allocate room on the stack for the ELF auxargs
340 * array. It has LINUX_AT_COUNT entries.
342 vectp -= howmany(LINUX_AT_COUNT * sizeof(Elf64_Auxinfo),
347 * Allocate room for the argv[] and env vectors including the
348 * terminating NULL pointers.
350 vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
352 /* vectp also becomes our initial stack base. */
353 stack_base = (register_t *)vectp;
355 stringp = imgp->args->begin_argv;
356 argc = imgp->args->argc;
357 envc = imgp->args->envc;
359 /* Copy out strings - arguments and environment. */
360 copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
362 /* Fill in "ps_strings" struct for ps, w, etc. */
363 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
364 suword(&arginfo->ps_nargvstr, argc);
366 /* Fill in argument portion of vector table. */
367 for (; argc > 0; --argc) {
368 suword(vectp++, (long)(intptr_t)destp);
369 while (*stringp++ != 0)
374 /* A null vector table pointer separates the argp's from the envp's. */
377 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
378 suword(&arginfo->ps_nenvstr, envc);
380 /* Fill in environment portion of vector table. */
381 for (; envc > 0; --envc) {
382 suword(vectp++, (long)(intptr_t)destp);
383 while (*stringp++ != 0)
388 /* The end of the vector table is a null pointer. */
394 * Reset registers to default values on exec.
397 linux_exec_setregs(struct thread *td, struct image_params *imgp, u_long stack)
399 struct trapframe *regs;
401 register_t saved_rflags;
407 if (td->td_proc->p_md.md_ldt != NULL)
410 mtx_unlock(&dt_lock);
414 clear_pcb_flags(pcb, PCB_32BIT);
415 pcb->pcb_initial_fpucw = __LINUX_NPXCW__;
416 set_pcb_flags(pcb, PCB_FULL_IRET);
418 saved_rflags = regs->tf_rflags & PSL_T;
419 bzero((char *)regs, sizeof(struct trapframe));
420 regs->tf_rip = imgp->entry_addr;
421 regs->tf_rsp = stack;
422 regs->tf_rflags = PSL_USER | saved_rflags;
423 regs->tf_ss = _udatasel;
424 regs->tf_cs = _ucodesel;
425 regs->tf_ds = _udatasel;
426 regs->tf_es = _udatasel;
427 regs->tf_fs = _ufssel;
428 regs->tf_gs = _ugssel;
429 regs->tf_flags = TF_HASSEGS;
432 * Reset the hardware debug registers if they were in use.
433 * They won't have any meaning for the newly exec'd process.
435 if (pcb->pcb_flags & PCB_DBREGS) {
444 * Clear the debug registers on the running
445 * CPU, otherwise they will end up affecting
446 * the next process we switch to.
450 clear_pcb_flags(pcb, PCB_DBREGS);
454 * Drop the FP state if we hold it, so that the process gets a
455 * clean FP state if it uses the FPU again.
461 * Copied from amd64/amd64/machdep.c
463 * XXX fpu state need? don't think so
466 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
469 struct l_ucontext uc;
470 struct l_sigcontext *context;
471 struct trapframe *regs;
472 unsigned long rflags;
477 error = copyin((void *)regs->tf_rbx, &uc, sizeof(uc));
482 context = &uc.uc_mcontext;
483 rflags = context->sc_rflags;
486 * Don't allow users to change privileged or reserved flags.
489 * XXX do allow users to change the privileged flag PSL_RF.
490 * The cpu sets PSL_RF in tf_rflags for faults. Debuggers
491 * should sometimes set it there too. tf_rflags is kept in
492 * the signal context during signal handling and there is no
493 * other place to remember it, so the PSL_RF bit may be
494 * corrupted by the signal handler without us knowing.
495 * Corruption of the PSL_RF bit at worst causes one more or
496 * one less debugger trap, so allowing it is fairly harmless.
499 #define RFLAG_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
500 if (!RFLAG_SECURE(rflags & ~PSL_RF, regs->tf_rflags & ~PSL_RF)) {
501 printf("linux_rt_sigreturn: rflags = 0x%lx\n", rflags);
506 * Don't allow users to load a valid privileged %cs. Let the
507 * hardware check for invalid selectors, excess privilege in
508 * other selectors, invalid %eip's and invalid %esp's.
510 #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL)
511 if (!CS_SECURE(context->sc_cs)) {
512 printf("linux_rt_sigreturn: cs = 0x%x\n", context->sc_cs);
513 ksiginfo_init_trap(&ksi);
514 ksi.ksi_signo = SIGBUS;
515 ksi.ksi_code = BUS_OBJERR;
516 ksi.ksi_trapno = T_PROTFLT;
517 ksi.ksi_addr = (void *)regs->tf_rip;
518 trapsignal(td, &ksi);
523 linux_to_bsd_sigset(&uc.uc_sigmask, &td->td_sigmask);
524 SIG_CANTMASK(td->td_sigmask);
528 regs->tf_rdi = context->sc_rdi;
529 regs->tf_rsi = context->sc_rsi;
530 regs->tf_rdx = context->sc_rdx;
531 regs->tf_rbp = context->sc_rbp;
532 regs->tf_rbx = context->sc_rbx;
533 regs->tf_rcx = context->sc_rcx;
534 regs->tf_rax = context->sc_rax;
535 regs->tf_rip = context->sc_rip;
536 regs->tf_rsp = context->sc_rsp;
537 regs->tf_r8 = context->sc_r8;
538 regs->tf_r9 = context->sc_r9;
539 regs->tf_r10 = context->sc_r10;
540 regs->tf_r11 = context->sc_r11;
541 regs->tf_r12 = context->sc_r12;
542 regs->tf_r13 = context->sc_r13;
543 regs->tf_r14 = context->sc_r14;
544 regs->tf_r15 = context->sc_r15;
545 regs->tf_cs = context->sc_cs;
546 regs->tf_err = context->sc_err;
547 regs->tf_rflags = rflags;
549 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
550 return (EJUSTRETURN);
554 * copied from amd64/amd64/machdep.c
556 * Send an interrupt to process.
559 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
561 struct l_rt_sigframe sf, *sfp;
566 struct trapframe *regs;
572 PROC_LOCK_ASSERT(p, MA_OWNED);
573 sig = ksi->ksi_signo;
575 code = ksi->ksi_code;
576 mtx_assert(&psp->ps_mtx, MA_OWNED);
578 oonstack = sigonstack(regs->tf_rsp);
580 LINUX_CTR4(rt_sendsig, "%p, %d, %p, %u",
581 catcher, sig, mask, code);
583 /* Allocate space for the signal handler context. */
584 if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack &&
585 SIGISMEMBER(psp->ps_sigonstack, sig)) {
586 sp = (caddr_t)td->td_sigstk.ss_sp + td->td_sigstk.ss_size -
587 sizeof(struct l_rt_sigframe);
589 sp = (caddr_t)regs->tf_rsp - sizeof(struct l_rt_sigframe) - 128;
590 /* Align to 16 bytes. */
591 sfp = (struct l_rt_sigframe *)((unsigned long)sp & ~0xFul);
592 mtx_unlock(&psp->ps_mtx);
594 /* Translate the signal. */
595 sig = bsd_to_linux_signal(sig);
597 /* Save user context. */
598 bzero(&sf, sizeof(sf));
599 bsd_to_linux_sigset(mask, &sf.sf_sc.uc_sigmask);
600 bsd_to_linux_sigset(mask, &sf.sf_sc.uc_mcontext.sc_mask);
602 sf.sf_sc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp);
603 sf.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size;
604 sf.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
605 ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE;
608 sf.sf_sc.uc_mcontext.sc_rdi = regs->tf_rdi;
609 sf.sf_sc.uc_mcontext.sc_rsi = regs->tf_rsi;
610 sf.sf_sc.uc_mcontext.sc_rdx = regs->tf_rdx;
611 sf.sf_sc.uc_mcontext.sc_rbp = regs->tf_rbp;
612 sf.sf_sc.uc_mcontext.sc_rbx = regs->tf_rbx;
613 sf.sf_sc.uc_mcontext.sc_rcx = regs->tf_rcx;
614 sf.sf_sc.uc_mcontext.sc_rax = regs->tf_rax;
615 sf.sf_sc.uc_mcontext.sc_rip = regs->tf_rip;
616 sf.sf_sc.uc_mcontext.sc_rsp = regs->tf_rsp;
617 sf.sf_sc.uc_mcontext.sc_r8 = regs->tf_r8;
618 sf.sf_sc.uc_mcontext.sc_r9 = regs->tf_r9;
619 sf.sf_sc.uc_mcontext.sc_r10 = regs->tf_r10;
620 sf.sf_sc.uc_mcontext.sc_r11 = regs->tf_r11;
621 sf.sf_sc.uc_mcontext.sc_r12 = regs->tf_r12;
622 sf.sf_sc.uc_mcontext.sc_r13 = regs->tf_r13;
623 sf.sf_sc.uc_mcontext.sc_r14 = regs->tf_r14;
624 sf.sf_sc.uc_mcontext.sc_r15 = regs->tf_r15;
625 sf.sf_sc.uc_mcontext.sc_cs = regs->tf_cs;
626 sf.sf_sc.uc_mcontext.sc_rflags = regs->tf_rflags;
627 sf.sf_sc.uc_mcontext.sc_err = regs->tf_err;
628 sf.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code);
629 sf.sf_sc.uc_mcontext.sc_cr2 = (register_t)ksi->ksi_addr;
631 /* Build the argument list for the signal handler. */
632 regs->tf_rdi = sig; /* arg 1 in %rdi */
634 regs->tf_rsi = (register_t)&sfp->sf_si; /* arg 2 in %rsi */
635 regs->tf_rdx = (register_t)&sfp->sf_sc; /* arg 3 in %rdx */
637 sf.sf_handler = catcher;
638 /* Fill in POSIX parts. */
639 ksiginfo_to_lsiginfo(ksi, &sf.sf_si, sig);
641 /* Copy the sigframe out to the user's stack. */
642 if (copyout(&sf, sfp, sizeof(*sfp)) != 0) {
644 printf("process %ld has trashed its stack\n", (long)p->p_pid);
650 regs->tf_rsp = (long)sfp;
651 regs->tf_rip = linux_rt_sigcode;
652 regs->tf_rflags &= ~(PSL_T | PSL_D);
653 regs->tf_cs = _ucodesel;
654 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
656 mtx_lock(&psp->ps_mtx);
659 #define LINUX_VSYSCALL_START (-10UL << 20)
660 #define LINUX_VSYSCALL_SZ 1024
662 const unsigned long linux_vsyscall_vector[] = {
663 LINUX_SYS_gettimeofday,
664 LINUX_SYS_linux_time,
665 /* getcpu not implemented */
669 linux_vsyscall(struct thread *td)
671 struct trapframe *frame;
676 frame = td->td_frame;
678 /* Check %rip for vsyscall area. */
679 if (__predict_true(frame->tf_rip < LINUX_VSYSCALL_START))
681 if ((frame->tf_rip & (LINUX_VSYSCALL_SZ - 1)) != 0)
683 code = (frame->tf_rip - LINUX_VSYSCALL_START) / LINUX_VSYSCALL_SZ;
684 if (code >= nitems(linux_vsyscall_vector))
688 * vsyscall called as callq *(%rax), so we must
689 * use return address from %rsp and also fixup %rsp.
691 error = copyin((void *)frame->tf_rsp, &retqaddr, sizeof(retqaddr));
695 frame->tf_rip = retqaddr;
696 frame->tf_rax = linux_vsyscall_vector[code];
699 traced = (frame->tf_flags & PSL_T);
701 amd64_syscall(td, traced);
706 struct sysentvec elf_linux_sysvec = {
707 .sv_size = LINUX_SYS_MAXSYSCALL,
708 .sv_table = linux_sysent,
710 .sv_errsize = ELAST + 1,
711 .sv_errtbl = linux_errtbl,
712 .sv_transtrap = linux_translate_traps,
713 .sv_fixup = linux_fixup_elf,
714 .sv_sendsig = linux_rt_sendsig,
715 .sv_sigcode = &_binary_linux_locore_o_start,
716 .sv_szsigcode = &linux_szsigcode,
717 .sv_name = "Linux ELF64",
718 .sv_coredump = elf64_coredump,
719 .sv_imgact_try = linux_exec_imgact_try,
720 .sv_minsigstksz = LINUX_MINSIGSTKSZ,
721 .sv_pagesize = PAGE_SIZE,
722 .sv_minuser = VM_MIN_ADDRESS,
723 .sv_maxuser = VM_MAXUSER_ADDRESS,
724 .sv_usrstack = USRSTACK,
725 .sv_psstrings = PS_STRINGS,
726 .sv_stackprot = VM_PROT_ALL,
727 .sv_copyout_strings = linux_copyout_strings,
728 .sv_setregs = linux_exec_setregs,
731 .sv_flags = SV_ABI_LINUX | SV_LP64 | SV_SHP,
732 .sv_set_syscall_retval = linux_set_syscall_retval,
733 .sv_fetch_syscall_args = linux_fetch_syscall_args,
734 .sv_syscallnames = NULL,
735 .sv_shared_page_base = SHAREDPAGE,
736 .sv_shared_page_len = PAGE_SIZE,
737 .sv_schedtail = linux_schedtail,
738 .sv_thread_detach = linux_thread_detach,
739 .sv_trap = linux_vsyscall,
743 linux_vdso_install(void *param)
746 amd64_lower_shared_page(&elf_linux_sysvec);
748 linux_szsigcode = (&_binary_linux_locore_o_end -
749 &_binary_linux_locore_o_start);
751 if (linux_szsigcode > elf_linux_sysvec.sv_shared_page_len)
752 panic("Linux invalid vdso size\n");
754 __elfN(linux_vdso_fixup)(&elf_linux_sysvec);
756 linux_shared_page_obj = __elfN(linux_shared_page_init)
757 (&linux_shared_page_mapping);
759 __elfN(linux_vdso_reloc)(&elf_linux_sysvec);
761 bcopy(elf_linux_sysvec.sv_sigcode, linux_shared_page_mapping,
763 elf_linux_sysvec.sv_shared_page_obj = linux_shared_page_obj;
765 linux_kplatform = linux_shared_page_mapping +
766 (linux_platform - (caddr_t)elf_linux_sysvec.sv_shared_page_base);
768 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC, SI_ORDER_ANY,
769 linux_vdso_install, NULL);
772 linux_vdso_deinstall(void *param)
775 __elfN(linux_shared_page_fini)(linux_shared_page_obj);
777 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
778 linux_vdso_deinstall, NULL);
780 static char GNULINUX_ABI_VENDOR[] = "GNU";
781 static int GNULINUX_ABI_DESC = 0;
784 linux_trans_osrel(const Elf_Note *note, int32_t *osrel)
786 const Elf32_Word *desc;
789 p = (uintptr_t)(note + 1);
790 p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
792 desc = (const Elf32_Word *)p;
793 if (desc[0] != GNULINUX_ABI_DESC)
797 * For Linux we encode osrel using the Linux convention of
798 * (version << 16) | (major << 8) | (minor)
799 * See macro in linux_mib.h
801 *osrel = LINUX_KERNVER(desc[1], desc[2], desc[3]);
806 static Elf_Brandnote linux64_brandnote = {
807 .hdr.n_namesz = sizeof(GNULINUX_ABI_VENDOR),
810 .vendor = GNULINUX_ABI_VENDOR,
811 .flags = BN_TRANSLATE_OSREL,
812 .trans_osrel = linux_trans_osrel
815 static Elf64_Brandinfo linux_glibc2brand = {
816 .brand = ELFOSABI_LINUX,
817 .machine = EM_X86_64,
818 .compat_3_brand = "Linux",
819 .emul_path = "/compat/linux",
820 .interp_path = "/lib64/ld-linux-x86-64.so.2",
821 .sysvec = &elf_linux_sysvec,
822 .interp_newpath = NULL,
823 .brand_note = &linux64_brandnote,
824 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
827 static Elf64_Brandinfo linux_glibc2brandshort = {
828 .brand = ELFOSABI_LINUX,
829 .machine = EM_X86_64,
830 .compat_3_brand = "Linux",
831 .emul_path = "/compat/linux",
832 .interp_path = "/lib64/ld-linux.so.2",
833 .sysvec = &elf_linux_sysvec,
834 .interp_newpath = NULL,
835 .brand_note = &linux64_brandnote,
836 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
839 static Elf64_Brandinfo linux_muslbrand = {
840 .brand = ELFOSABI_LINUX,
841 .machine = EM_X86_64,
842 .compat_3_brand = "Linux",
843 .emul_path = "/compat/linux",
844 .interp_path = "/lib/ld-musl-x86_64.so.1",
845 .sysvec = &elf_linux_sysvec,
846 .interp_newpath = NULL,
847 .brand_note = &linux64_brandnote,
848 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
851 Elf64_Brandinfo *linux_brandlist[] = {
853 &linux_glibc2brandshort,
859 linux64_elf_modevent(module_t mod, int type, void *data)
861 Elf64_Brandinfo **brandinfo;
863 struct linux_ioctl_handler **lihp;
869 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
871 if (elf64_insert_brand_entry(*brandinfo) < 0)
874 SET_FOREACH(lihp, linux_ioctl_handler_set)
875 linux_ioctl_register_handler(*lihp);
876 LIST_INIT(&futex_list);
877 mtx_init(&futex_mtx, "ftllk64", NULL, MTX_DEF);
878 stclohz = (stathz ? stathz : hz);
880 printf("Linux x86-64 ELF exec handler installed\n");
882 printf("cannot insert Linux x86-64 ELF brand handler\n");
885 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
887 if (elf64_brand_inuse(*brandinfo))
890 for (brandinfo = &linux_brandlist[0];
891 *brandinfo != NULL; ++brandinfo)
892 if (elf64_remove_brand_entry(*brandinfo) < 0)
896 SET_FOREACH(lihp, linux_ioctl_handler_set)
897 linux_ioctl_unregister_handler(*lihp);
898 mtx_destroy(&futex_mtx);
900 printf("Linux ELF exec handler removed\n");
902 printf("Could not deinstall ELF interpreter entry\n");
910 static moduledata_t linux64_elf_mod = {
912 linux64_elf_modevent,
916 DECLARE_MODULE_TIED(linux64elf, linux64_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);
917 MODULE_DEPEND(linux64elf, linux_common, 1, 1, 1);
918 FEATURE(linux64, "Linux 64bit support");