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_futex.h>
79 #include <compat/linux/linux_ioctl.h>
80 #include <compat/linux/linux_mib.h>
81 #include <compat/linux/linux_misc.h>
82 #include <compat/linux/linux_signal.h>
83 #include <compat/linux/linux_sysproto.h>
84 #include <compat/linux/linux_util.h>
85 #include <compat/linux/linux_vdso.h>
87 MODULE_VERSION(linux64, 1);
90 SYSCTL_PROC(_compat_linux, OID_AUTO, debug,
91 CTLTYPE_STRING | CTLFLAG_RW,
92 0, 0, linux_sysctl_debug, "A",
93 "Linux 64 debugging control");
97 * Allow the sendsig functions to use the ldebug() facility even though they
98 * are not syscalls themselves. Map them to syscall 0. This is slightly less
99 * bogus than using ldebug(sigreturn).
101 #define LINUX_SYS_linux_rt_sendsig 0
103 const char *linux_kplatform;
104 static int linux_szsigcode;
105 static vm_object_t linux_shared_page_obj;
106 static char *linux_shared_page_mapping;
107 extern char _binary_linux_locore_o_start;
108 extern char _binary_linux_locore_o_end;
110 extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL];
112 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler);
114 static register_t * linux_copyout_strings(struct image_params *imgp);
115 static int linux_fixup_elf(register_t **stack_base,
116 struct image_params *iparams);
117 static bool linux_trans_osrel(const Elf_Note *note, int32_t *osrel);
118 static void linux_vdso_install(void *param);
119 static void linux_vdso_deinstall(void *param);
120 static void linux_set_syscall_retval(struct thread *td, int error);
121 static int linux_fetch_syscall_args(struct thread *td);
122 static void linux_exec_setregs(struct thread *td, struct image_params *imgp,
124 static int linux_vsyscall(struct thread *td);
126 #define LINUX_T_UNKNOWN 255
127 static int _bsd_to_linux_trapcode[] = {
128 LINUX_T_UNKNOWN, /* 0 */
129 6, /* 1 T_PRIVINFLT */
130 LINUX_T_UNKNOWN, /* 2 */
132 LINUX_T_UNKNOWN, /* 4 */
133 LINUX_T_UNKNOWN, /* 5 */
134 16, /* 6 T_ARITHTRAP */
135 254, /* 7 T_ASTFLT */
136 LINUX_T_UNKNOWN, /* 8 */
137 13, /* 9 T_PROTFLT */
138 1, /* 10 T_TRCTRAP */
139 LINUX_T_UNKNOWN, /* 11 */
140 14, /* 12 T_PAGEFLT */
141 LINUX_T_UNKNOWN, /* 13 */
142 17, /* 14 T_ALIGNFLT */
143 LINUX_T_UNKNOWN, /* 15 */
144 LINUX_T_UNKNOWN, /* 16 */
145 LINUX_T_UNKNOWN, /* 17 */
151 8, /* 23 T_DOUBLEFLT */
152 9, /* 24 T_FPOPFLT */
153 10, /* 25 T_TSSFLT */
154 11, /* 26 T_SEGNPFLT */
155 12, /* 27 T_STKFLT */
157 19, /* 29 T_XMMFLT */
158 15 /* 30 T_RESERVED */
160 #define bsd_to_linux_trapcode(code) \
161 ((code)<nitems(_bsd_to_linux_trapcode)? \
162 _bsd_to_linux_trapcode[(code)]: \
165 LINUX_VDSO_SYM_INTPTR(linux_rt_sigcode);
166 LINUX_VDSO_SYM_CHAR(linux_platform);
169 * If FreeBSD & Linux have a difference of opinion about what a trap
170 * means, deal with it here.
175 linux_translate_traps(int signal, int trap_code)
178 if (signal != SIGBUS)
192 linux_fetch_syscall_args(struct thread *td)
195 struct trapframe *frame;
196 struct syscall_args *sa;
199 frame = td->td_frame;
202 sa->args[0] = frame->tf_rdi;
203 sa->args[1] = frame->tf_rsi;
204 sa->args[2] = frame->tf_rdx;
205 sa->args[3] = frame->tf_rcx;
206 sa->args[4] = frame->tf_r8;
207 sa->args[5] = frame->tf_r9;
208 sa->code = frame->tf_rax;
210 if (sa->code >= p->p_sysent->sv_size)
212 sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1];
214 sa->callp = &p->p_sysent->sv_table[sa->code];
215 sa->narg = sa->callp->sy_narg;
217 td->td_retval[0] = 0;
222 linux_set_syscall_retval(struct thread *td, int error)
224 struct trapframe *frame = td->td_frame;
227 * On Linux only %rcx and %r11 values are not preserved across
228 * the syscall. So, do not clobber %rdx and %r10.
230 td->td_retval[1] = frame->tf_rdx;
231 if (error != EJUSTRETURN)
232 frame->tf_r10 = frame->tf_rcx;
234 cpu_set_syscall_retval(td, error);
236 /* Restore all registers. */
237 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
241 linux_fixup_elf(register_t **stack_base, struct image_params *imgp)
244 Elf_Auxinfo *argarray, *pos;
245 Elf_Addr *auxbase, *base;
246 struct ps_strings *arginfo;
248 int error, issetugid;
251 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
253 KASSERT(curthread->td_proc == imgp->proc,
254 ("unsafe linux_fixup_elf(), should be curproc"));
255 base = (Elf64_Addr *)*stack_base;
256 args = (Elf64_Auxargs *)imgp->auxargs;
257 auxbase = base + imgp->args->argc + 1 + imgp->args->envc + 1;
258 argarray = pos = malloc(LINUX_AT_COUNT * sizeof(*pos), M_TEMP,
261 issetugid = p->p_flag & P_SUGID ? 1 : 0;
262 AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR,
263 imgp->proc->p_sysent->sv_shared_page_base);
264 AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature);
265 AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz);
266 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
267 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
268 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
269 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
270 AUXARGS_ENTRY(pos, AT_BASE, args->base);
271 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
272 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
273 AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid);
274 AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid);
275 AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid);
276 AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid);
277 AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid);
278 AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform));
279 AUXARGS_ENTRY(pos, LINUX_AT_RANDOM, imgp->canary);
280 if (imgp->execpathp != 0)
281 AUXARGS_ENTRY(pos, LINUX_AT_EXECFN, imgp->execpathp);
282 if (args->execfd != -1)
283 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
284 AUXARGS_ENTRY(pos, AT_NULL, 0);
285 free(imgp->auxargs, M_TEMP);
286 imgp->auxargs = NULL;
287 KASSERT(pos - argarray <= LINUX_AT_COUNT, ("Too many auxargs"));
289 error = copyout(argarray, auxbase, sizeof(*argarray) * LINUX_AT_COUNT);
290 free(argarray, M_TEMP);
295 if (suword(base, (uint64_t)imgp->args->argc) == -1)
298 *stack_base = (register_t *)base;
303 * Copy strings out to the new process address space, constructing new arg
304 * and env vector tables. Return a pointer to the base so that it can be used
305 * as the initial stack pointer.
308 linux_copyout_strings(struct image_params *imgp)
312 char *stringp, *destp;
313 register_t *stack_base;
314 struct ps_strings *arginfo;
315 char canary[LINUX_AT_RANDOM_LEN];
319 /* Calculate string base and vector table pointers. */
320 if (imgp->execpath != NULL && imgp->auxargs != NULL)
321 execpath_len = strlen(imgp->execpath) + 1;
326 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
327 destp = (caddr_t)arginfo - SPARE_USRSPACE -
328 roundup(sizeof(canary), sizeof(char *)) -
329 roundup(execpath_len, sizeof(char *)) -
330 roundup(ARG_MAX - imgp->args->stringspace, sizeof(char *));
332 if (execpath_len != 0) {
333 imgp->execpathp = (uintptr_t)arginfo - execpath_len;
334 copyout(imgp->execpath, (void *)imgp->execpathp, execpath_len);
337 /* Prepare the canary for SSP. */
338 arc4rand(canary, sizeof(canary), 0);
339 imgp->canary = (uintptr_t)arginfo -
340 roundup(execpath_len, sizeof(char *)) -
341 roundup(sizeof(canary), sizeof(char *));
342 copyout(canary, (void *)imgp->canary, sizeof(canary));
344 vectp = (char **)destp;
347 * Allocate room on the stack for the ELF auxargs
348 * array. It has LINUX_AT_COUNT entries.
350 vectp -= howmany(LINUX_AT_COUNT * sizeof(Elf64_Auxinfo),
355 * Allocate room for the argv[] and env vectors including the
356 * terminating NULL pointers.
358 vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
360 /* vectp also becomes our initial stack base. */
361 stack_base = (register_t *)vectp;
363 stringp = imgp->args->begin_argv;
364 argc = imgp->args->argc;
365 envc = imgp->args->envc;
367 /* Copy out strings - arguments and environment. */
368 copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
370 /* Fill in "ps_strings" struct for ps, w, etc. */
371 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
372 suword(&arginfo->ps_nargvstr, argc);
374 /* Fill in argument portion of vector table. */
375 for (; argc > 0; --argc) {
376 suword(vectp++, (long)(intptr_t)destp);
377 while (*stringp++ != 0)
382 /* A null vector table pointer separates the argp's from the envp's. */
385 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
386 suword(&arginfo->ps_nenvstr, envc);
388 /* Fill in environment portion of vector table. */
389 for (; envc > 0; --envc) {
390 suword(vectp++, (long)(intptr_t)destp);
391 while (*stringp++ != 0)
396 /* The end of the vector table is a null pointer. */
402 * Reset registers to default values on exec.
405 linux_exec_setregs(struct thread *td, struct image_params *imgp, u_long stack)
407 struct trapframe *regs;
409 register_t saved_rflags;
414 if (td->td_proc->p_md.md_ldt != NULL)
419 clear_pcb_flags(pcb, PCB_32BIT);
420 pcb->pcb_initial_fpucw = __LINUX_NPXCW__;
421 set_pcb_flags(pcb, PCB_FULL_IRET);
423 saved_rflags = regs->tf_rflags & PSL_T;
424 bzero((char *)regs, sizeof(struct trapframe));
425 regs->tf_rip = imgp->entry_addr;
426 regs->tf_rsp = stack;
427 regs->tf_rflags = PSL_USER | saved_rflags;
428 regs->tf_ss = _udatasel;
429 regs->tf_cs = _ucodesel;
430 regs->tf_ds = _udatasel;
431 regs->tf_es = _udatasel;
432 regs->tf_fs = _ufssel;
433 regs->tf_gs = _ugssel;
434 regs->tf_flags = TF_HASSEGS;
437 * Reset the hardware debug registers if they were in use.
438 * They won't have any meaning for the newly exec'd process.
440 if (pcb->pcb_flags & PCB_DBREGS) {
449 * Clear the debug registers on the running
450 * CPU, otherwise they will end up affecting
451 * the next process we switch to.
455 clear_pcb_flags(pcb, PCB_DBREGS);
459 * Drop the FP state if we hold it, so that the process gets a
460 * clean FP state if it uses the FPU again.
466 * Copied from amd64/amd64/machdep.c
468 * XXX fpu state need? don't think so
471 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
474 struct l_ucontext uc;
475 struct l_sigcontext *context;
476 struct trapframe *regs;
477 unsigned long rflags;
482 error = copyin((void *)regs->tf_rbx, &uc, sizeof(uc));
487 context = &uc.uc_mcontext;
488 rflags = context->sc_rflags;
491 * Don't allow users to change privileged or reserved flags.
494 * XXX do allow users to change the privileged flag PSL_RF.
495 * The cpu sets PSL_RF in tf_rflags for faults. Debuggers
496 * should sometimes set it there too. tf_rflags is kept in
497 * the signal context during signal handling and there is no
498 * other place to remember it, so the PSL_RF bit may be
499 * corrupted by the signal handler without us knowing.
500 * Corruption of the PSL_RF bit at worst causes one more or
501 * one less debugger trap, so allowing it is fairly harmless.
504 #define RFLAG_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
505 if (!RFLAG_SECURE(rflags & ~PSL_RF, regs->tf_rflags & ~PSL_RF)) {
506 printf("linux_rt_sigreturn: rflags = 0x%lx\n", rflags);
511 * Don't allow users to load a valid privileged %cs. Let the
512 * hardware check for invalid selectors, excess privilege in
513 * other selectors, invalid %eip's and invalid %esp's.
515 #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL)
516 if (!CS_SECURE(context->sc_cs)) {
517 printf("linux_rt_sigreturn: cs = 0x%x\n", context->sc_cs);
518 ksiginfo_init_trap(&ksi);
519 ksi.ksi_signo = SIGBUS;
520 ksi.ksi_code = BUS_OBJERR;
521 ksi.ksi_trapno = T_PROTFLT;
522 ksi.ksi_addr = (void *)regs->tf_rip;
523 trapsignal(td, &ksi);
528 linux_to_bsd_sigset(&uc.uc_sigmask, &td->td_sigmask);
529 SIG_CANTMASK(td->td_sigmask);
533 regs->tf_rdi = context->sc_rdi;
534 regs->tf_rsi = context->sc_rsi;
535 regs->tf_rdx = context->sc_rdx;
536 regs->tf_rbp = context->sc_rbp;
537 regs->tf_rbx = context->sc_rbx;
538 regs->tf_rcx = context->sc_rcx;
539 regs->tf_rax = context->sc_rax;
540 regs->tf_rip = context->sc_rip;
541 regs->tf_rsp = context->sc_rsp;
542 regs->tf_r8 = context->sc_r8;
543 regs->tf_r9 = context->sc_r9;
544 regs->tf_r10 = context->sc_r10;
545 regs->tf_r11 = context->sc_r11;
546 regs->tf_r12 = context->sc_r12;
547 regs->tf_r13 = context->sc_r13;
548 regs->tf_r14 = context->sc_r14;
549 regs->tf_r15 = context->sc_r15;
550 regs->tf_cs = context->sc_cs;
551 regs->tf_err = context->sc_err;
552 regs->tf_rflags = rflags;
554 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
555 return (EJUSTRETURN);
559 * copied from amd64/amd64/machdep.c
561 * Send an interrupt to process.
564 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
566 struct l_rt_sigframe sf, *sfp;
571 struct trapframe *regs;
577 PROC_LOCK_ASSERT(p, MA_OWNED);
578 sig = ksi->ksi_signo;
580 code = ksi->ksi_code;
581 mtx_assert(&psp->ps_mtx, MA_OWNED);
583 oonstack = sigonstack(regs->tf_rsp);
585 LINUX_CTR4(rt_sendsig, "%p, %d, %p, %u",
586 catcher, sig, mask, code);
588 /* Allocate space for the signal handler context. */
589 if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack &&
590 SIGISMEMBER(psp->ps_sigonstack, sig)) {
591 sp = (caddr_t)td->td_sigstk.ss_sp + td->td_sigstk.ss_size -
592 sizeof(struct l_rt_sigframe);
594 sp = (caddr_t)regs->tf_rsp - sizeof(struct l_rt_sigframe) - 128;
595 /* Align to 16 bytes. */
596 sfp = (struct l_rt_sigframe *)((unsigned long)sp & ~0xFul);
597 mtx_unlock(&psp->ps_mtx);
599 /* Translate the signal. */
600 sig = bsd_to_linux_signal(sig);
602 /* Save user context. */
603 bzero(&sf, sizeof(sf));
604 bsd_to_linux_sigset(mask, &sf.sf_sc.uc_sigmask);
605 bsd_to_linux_sigset(mask, &sf.sf_sc.uc_mcontext.sc_mask);
607 sf.sf_sc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp);
608 sf.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size;
609 sf.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
610 ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE;
613 sf.sf_sc.uc_mcontext.sc_rdi = regs->tf_rdi;
614 sf.sf_sc.uc_mcontext.sc_rsi = regs->tf_rsi;
615 sf.sf_sc.uc_mcontext.sc_rdx = regs->tf_rdx;
616 sf.sf_sc.uc_mcontext.sc_rbp = regs->tf_rbp;
617 sf.sf_sc.uc_mcontext.sc_rbx = regs->tf_rbx;
618 sf.sf_sc.uc_mcontext.sc_rcx = regs->tf_rcx;
619 sf.sf_sc.uc_mcontext.sc_rax = regs->tf_rax;
620 sf.sf_sc.uc_mcontext.sc_rip = regs->tf_rip;
621 sf.sf_sc.uc_mcontext.sc_rsp = regs->tf_rsp;
622 sf.sf_sc.uc_mcontext.sc_r8 = regs->tf_r8;
623 sf.sf_sc.uc_mcontext.sc_r9 = regs->tf_r9;
624 sf.sf_sc.uc_mcontext.sc_r10 = regs->tf_r10;
625 sf.sf_sc.uc_mcontext.sc_r11 = regs->tf_r11;
626 sf.sf_sc.uc_mcontext.sc_r12 = regs->tf_r12;
627 sf.sf_sc.uc_mcontext.sc_r13 = regs->tf_r13;
628 sf.sf_sc.uc_mcontext.sc_r14 = regs->tf_r14;
629 sf.sf_sc.uc_mcontext.sc_r15 = regs->tf_r15;
630 sf.sf_sc.uc_mcontext.sc_cs = regs->tf_cs;
631 sf.sf_sc.uc_mcontext.sc_rflags = regs->tf_rflags;
632 sf.sf_sc.uc_mcontext.sc_err = regs->tf_err;
633 sf.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code);
634 sf.sf_sc.uc_mcontext.sc_cr2 = (register_t)ksi->ksi_addr;
636 /* Build the argument list for the signal handler. */
637 regs->tf_rdi = sig; /* arg 1 in %rdi */
639 regs->tf_rsi = (register_t)&sfp->sf_si; /* arg 2 in %rsi */
640 regs->tf_rdx = (register_t)&sfp->sf_sc; /* arg 3 in %rdx */
642 sf.sf_handler = catcher;
643 /* Fill in POSIX parts. */
644 ksiginfo_to_lsiginfo(ksi, &sf.sf_si, sig);
646 /* Copy the sigframe out to the user's stack. */
647 if (copyout(&sf, sfp, sizeof(*sfp)) != 0) {
649 printf("process %ld has trashed its stack\n", (long)p->p_pid);
655 regs->tf_rsp = (long)sfp;
656 regs->tf_rip = linux_rt_sigcode;
657 regs->tf_rflags &= ~(PSL_T | PSL_D);
658 regs->tf_cs = _ucodesel;
659 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
661 mtx_lock(&psp->ps_mtx);
664 #define LINUX_VSYSCALL_START (-10UL << 20)
665 #define LINUX_VSYSCALL_SZ 1024
667 const unsigned long linux_vsyscall_vector[] = {
668 LINUX_SYS_gettimeofday,
669 LINUX_SYS_linux_time,
670 /* getcpu not implemented */
674 linux_vsyscall(struct thread *td)
676 struct trapframe *frame;
681 frame = td->td_frame;
683 /* Check %rip for vsyscall area. */
684 if (__predict_true(frame->tf_rip < LINUX_VSYSCALL_START))
686 if ((frame->tf_rip & (LINUX_VSYSCALL_SZ - 1)) != 0)
688 code = (frame->tf_rip - LINUX_VSYSCALL_START) / LINUX_VSYSCALL_SZ;
689 if (code >= nitems(linux_vsyscall_vector))
693 * vsyscall called as callq *(%rax), so we must
694 * use return address from %rsp and also fixup %rsp.
696 error = copyin((void *)frame->tf_rsp, &retqaddr, sizeof(retqaddr));
700 frame->tf_rip = retqaddr;
701 frame->tf_rax = linux_vsyscall_vector[code];
704 traced = (frame->tf_flags & PSL_T);
706 amd64_syscall(td, traced);
711 struct sysentvec elf_linux_sysvec = {
712 .sv_size = LINUX_SYS_MAXSYSCALL,
713 .sv_table = linux_sysent,
714 .sv_errsize = ELAST + 1,
715 .sv_errtbl = linux_errtbl,
716 .sv_transtrap = linux_translate_traps,
717 .sv_fixup = linux_fixup_elf,
718 .sv_sendsig = linux_rt_sendsig,
719 .sv_sigcode = &_binary_linux_locore_o_start,
720 .sv_szsigcode = &linux_szsigcode,
721 .sv_name = "Linux ELF64",
722 .sv_coredump = elf64_coredump,
723 .sv_imgact_try = linux_exec_imgact_try,
724 .sv_minsigstksz = LINUX_MINSIGSTKSZ,
725 .sv_pagesize = PAGE_SIZE,
726 .sv_minuser = VM_MIN_ADDRESS,
727 .sv_maxuser = VM_MAXUSER_ADDRESS,
728 .sv_usrstack = USRSTACK,
729 .sv_psstrings = PS_STRINGS,
730 .sv_stackprot = VM_PROT_ALL,
731 .sv_copyout_strings = linux_copyout_strings,
732 .sv_setregs = linux_exec_setregs,
735 .sv_flags = SV_ABI_LINUX | SV_LP64 | SV_SHP,
736 .sv_set_syscall_retval = linux_set_syscall_retval,
737 .sv_fetch_syscall_args = linux_fetch_syscall_args,
738 .sv_syscallnames = NULL,
739 .sv_shared_page_base = SHAREDPAGE,
740 .sv_shared_page_len = PAGE_SIZE,
741 .sv_schedtail = linux_schedtail,
742 .sv_thread_detach = linux_thread_detach,
743 .sv_trap = linux_vsyscall,
747 linux_vdso_install(void *param)
750 amd64_lower_shared_page(&elf_linux_sysvec);
752 linux_szsigcode = (&_binary_linux_locore_o_end -
753 &_binary_linux_locore_o_start);
755 if (linux_szsigcode > elf_linux_sysvec.sv_shared_page_len)
756 panic("Linux invalid vdso size\n");
758 __elfN(linux_vdso_fixup)(&elf_linux_sysvec);
760 linux_shared_page_obj = __elfN(linux_shared_page_init)
761 (&linux_shared_page_mapping);
763 __elfN(linux_vdso_reloc)(&elf_linux_sysvec);
765 bcopy(elf_linux_sysvec.sv_sigcode, linux_shared_page_mapping,
767 elf_linux_sysvec.sv_shared_page_obj = linux_shared_page_obj;
769 linux_kplatform = linux_shared_page_mapping +
770 (linux_platform - (caddr_t)elf_linux_sysvec.sv_shared_page_base);
772 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC, SI_ORDER_ANY,
773 linux_vdso_install, NULL);
776 linux_vdso_deinstall(void *param)
779 __elfN(linux_shared_page_fini)(linux_shared_page_obj);
781 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
782 linux_vdso_deinstall, NULL);
784 static char GNULINUX_ABI_VENDOR[] = "GNU";
785 static int GNULINUX_ABI_DESC = 0;
788 linux_trans_osrel(const Elf_Note *note, int32_t *osrel)
790 const Elf32_Word *desc;
793 p = (uintptr_t)(note + 1);
794 p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
796 desc = (const Elf32_Word *)p;
797 if (desc[0] != GNULINUX_ABI_DESC)
801 * For Linux we encode osrel using the Linux convention of
802 * (version << 16) | (major << 8) | (minor)
803 * See macro in linux_mib.h
805 *osrel = LINUX_KERNVER(desc[1], desc[2], desc[3]);
810 static Elf_Brandnote linux64_brandnote = {
811 .hdr.n_namesz = sizeof(GNULINUX_ABI_VENDOR),
814 .vendor = GNULINUX_ABI_VENDOR,
815 .flags = BN_TRANSLATE_OSREL,
816 .trans_osrel = linux_trans_osrel
819 static Elf64_Brandinfo linux_glibc2brand = {
820 .brand = ELFOSABI_LINUX,
821 .machine = EM_X86_64,
822 .compat_3_brand = "Linux",
823 .emul_path = "/compat/linux",
824 .interp_path = "/lib64/ld-linux-x86-64.so.2",
825 .sysvec = &elf_linux_sysvec,
826 .interp_newpath = NULL,
827 .brand_note = &linux64_brandnote,
828 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
831 static Elf64_Brandinfo linux_glibc2brandshort = {
832 .brand = ELFOSABI_LINUX,
833 .machine = EM_X86_64,
834 .compat_3_brand = "Linux",
835 .emul_path = "/compat/linux",
836 .interp_path = "/lib64/ld-linux.so.2",
837 .sysvec = &elf_linux_sysvec,
838 .interp_newpath = NULL,
839 .brand_note = &linux64_brandnote,
840 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
843 static Elf64_Brandinfo linux_muslbrand = {
844 .brand = ELFOSABI_LINUX,
845 .machine = EM_X86_64,
846 .compat_3_brand = "Linux",
847 .emul_path = "/compat/linux",
848 .interp_path = "/lib/ld-musl-x86_64.so.1",
849 .sysvec = &elf_linux_sysvec,
850 .interp_newpath = NULL,
851 .brand_note = &linux64_brandnote,
852 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
855 Elf64_Brandinfo *linux_brandlist[] = {
857 &linux_glibc2brandshort,
863 linux64_elf_modevent(module_t mod, int type, void *data)
865 Elf64_Brandinfo **brandinfo;
867 struct linux_ioctl_handler **lihp;
873 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
875 if (elf64_insert_brand_entry(*brandinfo) < 0)
878 SET_FOREACH(lihp, linux_ioctl_handler_set)
879 linux_ioctl_register_handler(*lihp);
880 LIST_INIT(&futex_list);
881 mtx_init(&futex_mtx, "ftllk64", NULL, MTX_DEF);
882 stclohz = (stathz ? stathz : hz);
884 printf("Linux x86-64 ELF exec handler installed\n");
886 printf("cannot insert Linux x86-64 ELF brand handler\n");
889 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
891 if (elf64_brand_inuse(*brandinfo))
894 for (brandinfo = &linux_brandlist[0];
895 *brandinfo != NULL; ++brandinfo)
896 if (elf64_remove_brand_entry(*brandinfo) < 0)
900 SET_FOREACH(lihp, linux_ioctl_handler_set)
901 linux_ioctl_unregister_handler(*lihp);
902 mtx_destroy(&futex_mtx);
904 printf("Linux ELF exec handler removed\n");
906 printf("Could not deinstall ELF interpreter entry\n");
914 static moduledata_t linux64_elf_mod = {
916 linux64_elf_modevent,
920 DECLARE_MODULE_TIED(linux64elf, linux64_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);
921 MODULE_DEPEND(linux64elf, linux_common, 1, 1, 1);
922 FEATURE(linux64, "Linux 64bit support");