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);
90 #if BYTE_ORDER == LITTLE_ENDIAN
91 #define SHELLMAGIC 0x2123 /* #! */
93 #define SHELLMAGIC 0x2321
97 SYSCTL_PROC(_compat_linux, OID_AUTO, debug,
98 CTLTYPE_STRING | CTLFLAG_RW,
99 0, 0, linux_sysctl_debug, "A",
100 "Linux 64 debugging control");
104 * Allow the sendsig functions to use the ldebug() facility even though they
105 * are not syscalls themselves. Map them to syscall 0. This is slightly less
106 * bogus than using ldebug(sigreturn).
108 #define LINUX_SYS_linux_rt_sendsig 0
110 const char *linux_kplatform;
111 static int linux_szsigcode;
112 static vm_object_t linux_shared_page_obj;
113 static char *linux_shared_page_mapping;
114 extern char _binary_linux_locore_o_start;
115 extern char _binary_linux_locore_o_end;
117 extern struct sysent linux_sysent[LINUX_SYS_MAXSYSCALL];
119 SET_DECLARE(linux_ioctl_handler_set, struct linux_ioctl_handler);
121 static register_t * linux_copyout_strings(struct image_params *imgp);
122 static int elf_linux_fixup(register_t **stack_base,
123 struct image_params *iparams);
124 static bool linux_trans_osrel(const Elf_Note *note, int32_t *osrel);
125 static void linux_vdso_install(void *param);
126 static void linux_vdso_deinstall(void *param);
127 static void linux_set_syscall_retval(struct thread *td, int error);
128 static int linux_fetch_syscall_args(struct thread *td);
129 static void linux_exec_setregs(struct thread *td, struct image_params *imgp,
131 static int linux_vsyscall(struct thread *td);
133 #define LINUX_T_UNKNOWN 255
134 static int _bsd_to_linux_trapcode[] = {
135 LINUX_T_UNKNOWN, /* 0 */
136 6, /* 1 T_PRIVINFLT */
137 LINUX_T_UNKNOWN, /* 2 */
139 LINUX_T_UNKNOWN, /* 4 */
140 LINUX_T_UNKNOWN, /* 5 */
141 16, /* 6 T_ARITHTRAP */
142 254, /* 7 T_ASTFLT */
143 LINUX_T_UNKNOWN, /* 8 */
144 13, /* 9 T_PROTFLT */
145 1, /* 10 T_TRCTRAP */
146 LINUX_T_UNKNOWN, /* 11 */
147 14, /* 12 T_PAGEFLT */
148 LINUX_T_UNKNOWN, /* 13 */
149 17, /* 14 T_ALIGNFLT */
150 LINUX_T_UNKNOWN, /* 15 */
151 LINUX_T_UNKNOWN, /* 16 */
152 LINUX_T_UNKNOWN, /* 17 */
158 8, /* 23 T_DOUBLEFLT */
159 9, /* 24 T_FPOPFLT */
160 10, /* 25 T_TSSFLT */
161 11, /* 26 T_SEGNPFLT */
162 12, /* 27 T_STKFLT */
164 19, /* 29 T_XMMFLT */
165 15 /* 30 T_RESERVED */
167 #define bsd_to_linux_trapcode(code) \
168 ((code)<nitems(_bsd_to_linux_trapcode)? \
169 _bsd_to_linux_trapcode[(code)]: \
172 LINUX_VDSO_SYM_INTPTR(linux_rt_sigcode);
173 LINUX_VDSO_SYM_CHAR(linux_platform);
176 * If FreeBSD & Linux have a difference of opinion about what a trap
177 * means, deal with it here.
182 translate_traps(int signal, int trap_code)
185 if (signal != SIGBUS)
199 linux_fetch_syscall_args(struct thread *td)
202 struct trapframe *frame;
203 struct syscall_args *sa;
206 frame = td->td_frame;
209 sa->args[0] = frame->tf_rdi;
210 sa->args[1] = frame->tf_rsi;
211 sa->args[2] = frame->tf_rdx;
212 sa->args[3] = frame->tf_rcx;
213 sa->args[4] = frame->tf_r8;
214 sa->args[5] = frame->tf_r9;
215 sa->code = frame->tf_rax;
217 if (sa->code >= p->p_sysent->sv_size)
219 sa->callp = &p->p_sysent->sv_table[p->p_sysent->sv_size - 1];
221 sa->callp = &p->p_sysent->sv_table[sa->code];
222 sa->narg = sa->callp->sy_narg;
224 td->td_retval[0] = 0;
229 linux_set_syscall_retval(struct thread *td, int error)
231 struct trapframe *frame = td->td_frame;
234 * On Linux only %rcx and %r11 values are not preserved across
235 * the syscall. So, do not clobber %rdx and %r10.
237 td->td_retval[1] = frame->tf_rdx;
238 if (error != EJUSTRETURN)
239 frame->tf_r10 = frame->tf_rcx;
241 cpu_set_syscall_retval(td, error);
243 /* Restore all registers. */
244 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
248 elf_linux_fixup(register_t **stack_base, struct image_params *imgp)
253 struct ps_strings *arginfo;
258 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
260 KASSERT(curthread->td_proc == imgp->proc,
261 ("unsafe elf_linux_fixup(), should be curproc"));
262 base = (Elf64_Addr *)*stack_base;
263 args = (Elf64_Auxargs *)imgp->auxargs;
264 pos = base + (imgp->args->argc + imgp->args->envc + 2);
266 issetugid = p->p_flag & P_SUGID ? 1 : 0;
267 AUXARGS_ENTRY(pos, LINUX_AT_SYSINFO_EHDR,
268 imgp->proc->p_sysent->sv_shared_page_base);
269 AUXARGS_ENTRY(pos, LINUX_AT_HWCAP, cpu_feature);
270 AUXARGS_ENTRY(pos, LINUX_AT_CLKTCK, stclohz);
271 AUXARGS_ENTRY(pos, AT_PHDR, args->phdr);
272 AUXARGS_ENTRY(pos, AT_PHENT, args->phent);
273 AUXARGS_ENTRY(pos, AT_PHNUM, args->phnum);
274 AUXARGS_ENTRY(pos, AT_PAGESZ, args->pagesz);
275 AUXARGS_ENTRY(pos, AT_BASE, args->base);
276 AUXARGS_ENTRY(pos, AT_FLAGS, args->flags);
277 AUXARGS_ENTRY(pos, AT_ENTRY, args->entry);
278 AUXARGS_ENTRY(pos, AT_UID, imgp->proc->p_ucred->cr_ruid);
279 AUXARGS_ENTRY(pos, AT_EUID, imgp->proc->p_ucred->cr_svuid);
280 AUXARGS_ENTRY(pos, AT_GID, imgp->proc->p_ucred->cr_rgid);
281 AUXARGS_ENTRY(pos, AT_EGID, imgp->proc->p_ucred->cr_svgid);
282 AUXARGS_ENTRY(pos, LINUX_AT_SECURE, issetugid);
283 AUXARGS_ENTRY(pos, LINUX_AT_PLATFORM, PTROUT(linux_platform));
284 AUXARGS_ENTRY(pos, LINUX_AT_RANDOM, imgp->canary);
285 if (imgp->execpathp != 0)
286 AUXARGS_ENTRY(pos, LINUX_AT_EXECFN, imgp->execpathp);
287 if (args->execfd != -1)
288 AUXARGS_ENTRY(pos, AT_EXECFD, args->execfd);
289 AUXARGS_ENTRY(pos, AT_NULL, 0);
290 free(imgp->auxargs, M_TEMP);
291 imgp->auxargs = NULL;
294 suword(base, (uint64_t)imgp->args->argc);
296 *stack_base = (register_t *)base;
301 * Copy strings out to the new process address space, constructing new arg
302 * and env vector tables. Return a pointer to the base so that it can be used
303 * as the initial stack pointer.
306 linux_copyout_strings(struct image_params *imgp)
310 char *stringp, *destp;
311 register_t *stack_base;
312 struct ps_strings *arginfo;
313 char canary[LINUX_AT_RANDOM_LEN];
317 /* Calculate string base and vector table pointers. */
318 if (imgp->execpath != NULL && imgp->auxargs != NULL)
319 execpath_len = strlen(imgp->execpath) + 1;
324 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
325 destp = (caddr_t)arginfo - SPARE_USRSPACE -
326 roundup(sizeof(canary), sizeof(char *)) -
327 roundup(execpath_len, sizeof(char *)) -
328 roundup(ARG_MAX - imgp->args->stringspace, sizeof(char *));
330 if (execpath_len != 0) {
331 imgp->execpathp = (uintptr_t)arginfo - execpath_len;
332 copyout(imgp->execpath, (void *)imgp->execpathp, execpath_len);
335 /* Prepare the canary for SSP. */
336 arc4rand(canary, sizeof(canary), 0);
337 imgp->canary = (uintptr_t)arginfo -
338 roundup(execpath_len, sizeof(char *)) -
339 roundup(sizeof(canary), sizeof(char *));
340 copyout(canary, (void *)imgp->canary, sizeof(canary));
342 vectp = (char **)destp;
345 * Allocate room on the stack for the ELF auxargs
346 * array. It has LINUX_AT_COUNT entries.
348 vectp -= howmany(LINUX_AT_COUNT * sizeof(Elf64_Auxinfo),
353 * Allocate room for the argv[] and env vectors including the
354 * terminating NULL pointers.
356 vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
358 /* vectp also becomes our initial stack base. */
359 stack_base = (register_t *)vectp;
361 stringp = imgp->args->begin_argv;
362 argc = imgp->args->argc;
363 envc = imgp->args->envc;
365 /* Copy out strings - arguments and environment. */
366 copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
368 /* Fill in "ps_strings" struct for ps, w, etc. */
369 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
370 suword(&arginfo->ps_nargvstr, argc);
372 /* Fill in argument portion of vector table. */
373 for (; argc > 0; --argc) {
374 suword(vectp++, (long)(intptr_t)destp);
375 while (*stringp++ != 0)
380 /* A null vector table pointer separates the argp's from the envp's. */
383 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
384 suword(&arginfo->ps_nenvstr, envc);
386 /* Fill in environment portion of vector table. */
387 for (; envc > 0; --envc) {
388 suword(vectp++, (long)(intptr_t)destp);
389 while (*stringp++ != 0)
394 /* The end of the vector table is a null pointer. */
400 * Reset registers to default values on exec.
403 linux_exec_setregs(struct thread *td, struct image_params *imgp, u_long stack)
405 struct trapframe *regs;
407 register_t saved_rflags;
413 if (td->td_proc->p_md.md_ldt != NULL)
416 mtx_unlock(&dt_lock);
420 clear_pcb_flags(pcb, PCB_32BIT);
421 pcb->pcb_initial_fpucw = __LINUX_NPXCW__;
422 set_pcb_flags(pcb, PCB_FULL_IRET);
424 saved_rflags = regs->tf_rflags & PSL_T;
425 bzero((char *)regs, sizeof(struct trapframe));
426 regs->tf_rip = imgp->entry_addr;
427 regs->tf_rsp = stack;
428 regs->tf_rflags = PSL_USER | saved_rflags;
429 regs->tf_ss = _udatasel;
430 regs->tf_cs = _ucodesel;
431 regs->tf_ds = _udatasel;
432 regs->tf_es = _udatasel;
433 regs->tf_fs = _ufssel;
434 regs->tf_gs = _ugssel;
435 regs->tf_flags = TF_HASSEGS;
438 * Reset the hardware debug registers if they were in use.
439 * They won't have any meaning for the newly exec'd process.
441 if (pcb->pcb_flags & PCB_DBREGS) {
450 * Clear the debug registers on the running
451 * CPU, otherwise they will end up affecting
452 * the next process we switch to.
456 clear_pcb_flags(pcb, PCB_DBREGS);
460 * Drop the FP state if we hold it, so that the process gets a
461 * clean FP state if it uses the FPU again.
467 * Copied from amd64/amd64/machdep.c
469 * XXX fpu state need? don't think so
472 linux_rt_sigreturn(struct thread *td, struct linux_rt_sigreturn_args *args)
475 struct l_ucontext uc;
476 struct l_sigcontext *context;
477 struct trapframe *regs;
478 unsigned long rflags;
483 error = copyin((void *)regs->tf_rbx, &uc, sizeof(uc));
488 context = &uc.uc_mcontext;
489 rflags = context->sc_rflags;
492 * Don't allow users to change privileged or reserved flags.
495 * XXX do allow users to change the privileged flag PSL_RF.
496 * The cpu sets PSL_RF in tf_rflags for faults. Debuggers
497 * should sometimes set it there too. tf_rflags is kept in
498 * the signal context during signal handling and there is no
499 * other place to remember it, so the PSL_RF bit may be
500 * corrupted by the signal handler without us knowing.
501 * Corruption of the PSL_RF bit at worst causes one more or
502 * one less debugger trap, so allowing it is fairly harmless.
505 #define RFLAG_SECURE(ef, oef) ((((ef) ^ (oef)) & ~PSL_USERCHANGE) == 0)
506 if (!RFLAG_SECURE(rflags & ~PSL_RF, regs->tf_rflags & ~PSL_RF)) {
507 printf("linux_rt_sigreturn: rflags = 0x%lx\n", rflags);
512 * Don't allow users to load a valid privileged %cs. Let the
513 * hardware check for invalid selectors, excess privilege in
514 * other selectors, invalid %eip's and invalid %esp's.
516 #define CS_SECURE(cs) (ISPL(cs) == SEL_UPL)
517 if (!CS_SECURE(context->sc_cs)) {
518 printf("linux_rt_sigreturn: cs = 0x%x\n", context->sc_cs);
519 ksiginfo_init_trap(&ksi);
520 ksi.ksi_signo = SIGBUS;
521 ksi.ksi_code = BUS_OBJERR;
522 ksi.ksi_trapno = T_PROTFLT;
523 ksi.ksi_addr = (void *)regs->tf_rip;
524 trapsignal(td, &ksi);
529 linux_to_bsd_sigset(&uc.uc_sigmask, &td->td_sigmask);
530 SIG_CANTMASK(td->td_sigmask);
534 regs->tf_rdi = context->sc_rdi;
535 regs->tf_rsi = context->sc_rsi;
536 regs->tf_rdx = context->sc_rdx;
537 regs->tf_rbp = context->sc_rbp;
538 regs->tf_rbx = context->sc_rbx;
539 regs->tf_rcx = context->sc_rcx;
540 regs->tf_rax = context->sc_rax;
541 regs->tf_rip = context->sc_rip;
542 regs->tf_rsp = context->sc_rsp;
543 regs->tf_r8 = context->sc_r8;
544 regs->tf_r9 = context->sc_r9;
545 regs->tf_r10 = context->sc_r10;
546 regs->tf_r11 = context->sc_r11;
547 regs->tf_r12 = context->sc_r12;
548 regs->tf_r13 = context->sc_r13;
549 regs->tf_r14 = context->sc_r14;
550 regs->tf_r15 = context->sc_r15;
551 regs->tf_cs = context->sc_cs;
552 regs->tf_err = context->sc_err;
553 regs->tf_rflags = rflags;
555 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
556 return (EJUSTRETURN);
560 * copied from amd64/amd64/machdep.c
562 * Send an interrupt to process.
565 linux_rt_sendsig(sig_t catcher, ksiginfo_t *ksi, sigset_t *mask)
567 struct l_rt_sigframe sf, *sfp;
572 struct trapframe *regs;
578 PROC_LOCK_ASSERT(p, MA_OWNED);
579 sig = ksi->ksi_signo;
581 code = ksi->ksi_code;
582 mtx_assert(&psp->ps_mtx, MA_OWNED);
584 oonstack = sigonstack(regs->tf_rsp);
586 LINUX_CTR4(rt_sendsig, "%p, %d, %p, %u",
587 catcher, sig, mask, code);
589 /* Allocate space for the signal handler context. */
590 if ((td->td_pflags & TDP_ALTSTACK) != 0 && !oonstack &&
591 SIGISMEMBER(psp->ps_sigonstack, sig)) {
592 sp = (caddr_t)td->td_sigstk.ss_sp + td->td_sigstk.ss_size -
593 sizeof(struct l_rt_sigframe);
595 sp = (caddr_t)regs->tf_rsp - sizeof(struct l_rt_sigframe) - 128;
596 /* Align to 16 bytes. */
597 sfp = (struct l_rt_sigframe *)((unsigned long)sp & ~0xFul);
598 mtx_unlock(&psp->ps_mtx);
600 /* Translate the signal. */
601 sig = bsd_to_linux_signal(sig);
603 /* Save user context. */
604 bzero(&sf, sizeof(sf));
605 bsd_to_linux_sigset(mask, &sf.sf_sc.uc_sigmask);
606 bsd_to_linux_sigset(mask, &sf.sf_sc.uc_mcontext.sc_mask);
608 sf.sf_sc.uc_stack.ss_sp = PTROUT(td->td_sigstk.ss_sp);
609 sf.sf_sc.uc_stack.ss_size = td->td_sigstk.ss_size;
610 sf.sf_sc.uc_stack.ss_flags = (td->td_pflags & TDP_ALTSTACK)
611 ? ((oonstack) ? LINUX_SS_ONSTACK : 0) : LINUX_SS_DISABLE;
614 sf.sf_sc.uc_mcontext.sc_rdi = regs->tf_rdi;
615 sf.sf_sc.uc_mcontext.sc_rsi = regs->tf_rsi;
616 sf.sf_sc.uc_mcontext.sc_rdx = regs->tf_rdx;
617 sf.sf_sc.uc_mcontext.sc_rbp = regs->tf_rbp;
618 sf.sf_sc.uc_mcontext.sc_rbx = regs->tf_rbx;
619 sf.sf_sc.uc_mcontext.sc_rcx = regs->tf_rcx;
620 sf.sf_sc.uc_mcontext.sc_rax = regs->tf_rax;
621 sf.sf_sc.uc_mcontext.sc_rip = regs->tf_rip;
622 sf.sf_sc.uc_mcontext.sc_rsp = regs->tf_rsp;
623 sf.sf_sc.uc_mcontext.sc_r8 = regs->tf_r8;
624 sf.sf_sc.uc_mcontext.sc_r9 = regs->tf_r9;
625 sf.sf_sc.uc_mcontext.sc_r10 = regs->tf_r10;
626 sf.sf_sc.uc_mcontext.sc_r11 = regs->tf_r11;
627 sf.sf_sc.uc_mcontext.sc_r12 = regs->tf_r12;
628 sf.sf_sc.uc_mcontext.sc_r13 = regs->tf_r13;
629 sf.sf_sc.uc_mcontext.sc_r14 = regs->tf_r14;
630 sf.sf_sc.uc_mcontext.sc_r15 = regs->tf_r15;
631 sf.sf_sc.uc_mcontext.sc_cs = regs->tf_cs;
632 sf.sf_sc.uc_mcontext.sc_rflags = regs->tf_rflags;
633 sf.sf_sc.uc_mcontext.sc_err = regs->tf_err;
634 sf.sf_sc.uc_mcontext.sc_trapno = bsd_to_linux_trapcode(code);
635 sf.sf_sc.uc_mcontext.sc_cr2 = (register_t)ksi->ksi_addr;
637 /* Build the argument list for the signal handler. */
638 regs->tf_rdi = sig; /* arg 1 in %rdi */
640 regs->tf_rsi = (register_t)&sfp->sf_si; /* arg 2 in %rsi */
641 regs->tf_rdx = (register_t)&sfp->sf_sc; /* arg 3 in %rdx */
643 sf.sf_handler = catcher;
644 /* Fill in POSIX parts. */
645 ksiginfo_to_lsiginfo(ksi, &sf.sf_si, sig);
647 /* Copy the sigframe out to the user's stack. */
648 if (copyout(&sf, sfp, sizeof(*sfp)) != 0) {
650 printf("process %ld has trashed its stack\n", (long)p->p_pid);
656 regs->tf_rsp = (long)sfp;
657 regs->tf_rip = linux_rt_sigcode;
658 regs->tf_rflags &= ~(PSL_T | PSL_D);
659 regs->tf_cs = _ucodesel;
660 set_pcb_flags(td->td_pcb, PCB_FULL_IRET);
662 mtx_lock(&psp->ps_mtx);
666 * If a Linux binary is exec'ing something, try this image activator
667 * first. We override standard shell script execution in order to
668 * be able to modify the interpreter path. We only do this if a Linux
669 * binary is doing the exec, so we do not create an EXEC module for it.
671 static int exec_linux_imgact_try(struct image_params *iparams);
674 exec_linux_imgact_try(struct image_params *imgp)
676 const char *head = (const char *)imgp->image_header;
681 * The interpreter for shell scripts run from a Linux binary needs
682 * to be located in /compat/linux if possible in order to recursively
683 * maintain Linux path emulation.
685 if (((const short *)head)[0] == SHELLMAGIC) {
687 * Run our normal shell image activator. If it succeeds
688 * attempt to use the alternate path for the interpreter.
689 * If an alternate path is found, use our stringspace
692 if ((error = exec_shell_imgact(imgp)) == 0) {
693 linux_emul_convpath(FIRST_THREAD_IN_PROC(imgp->proc),
694 imgp->interpreter_name, UIO_SYSSPACE,
695 &rpath, 0, AT_FDCWD);
697 imgp->args->fname_buf =
698 imgp->interpreter_name = rpath;
704 #define LINUX_VSYSCALL_START (-10UL << 20)
705 #define LINUX_VSYSCALL_SZ 1024
707 const unsigned long linux_vsyscall_vector[] = {
708 LINUX_SYS_gettimeofday,
709 LINUX_SYS_linux_time,
710 /* getcpu not implemented */
714 linux_vsyscall(struct thread *td)
716 struct trapframe *frame;
721 frame = td->td_frame;
723 /* Check %rip for vsyscall area. */
724 if (__predict_true(frame->tf_rip < LINUX_VSYSCALL_START))
726 if ((frame->tf_rip & (LINUX_VSYSCALL_SZ - 1)) != 0)
728 code = (frame->tf_rip - LINUX_VSYSCALL_START) / LINUX_VSYSCALL_SZ;
729 if (code >= nitems(linux_vsyscall_vector))
733 * vsyscall called as callq *(%rax), so we must
734 * use return address from %rsp and also fixup %rsp.
736 error = copyin((void *)frame->tf_rsp, &retqaddr, sizeof(retqaddr));
740 frame->tf_rip = retqaddr;
741 frame->tf_rax = linux_vsyscall_vector[code];
744 traced = (frame->tf_flags & PSL_T);
746 amd64_syscall(td, traced);
751 struct sysentvec elf_linux_sysvec = {
752 .sv_size = LINUX_SYS_MAXSYSCALL,
753 .sv_table = linux_sysent,
755 .sv_errsize = ELAST + 1,
756 .sv_errtbl = bsd_to_linux_errno_generic,
757 .sv_transtrap = translate_traps,
758 .sv_fixup = elf_linux_fixup,
759 .sv_sendsig = linux_rt_sendsig,
760 .sv_sigcode = &_binary_linux_locore_o_start,
761 .sv_szsigcode = &linux_szsigcode,
762 .sv_name = "Linux ELF64",
763 .sv_coredump = elf64_coredump,
764 .sv_imgact_try = exec_linux_imgact_try,
765 .sv_minsigstksz = LINUX_MINSIGSTKSZ,
766 .sv_pagesize = PAGE_SIZE,
767 .sv_minuser = VM_MIN_ADDRESS,
768 .sv_maxuser = VM_MAXUSER_ADDRESS,
769 .sv_usrstack = USRSTACK,
770 .sv_psstrings = PS_STRINGS,
771 .sv_stackprot = VM_PROT_ALL,
772 .sv_copyout_strings = linux_copyout_strings,
773 .sv_setregs = linux_exec_setregs,
776 .sv_flags = SV_ABI_LINUX | SV_LP64 | SV_SHP,
777 .sv_set_syscall_retval = linux_set_syscall_retval,
778 .sv_fetch_syscall_args = linux_fetch_syscall_args,
779 .sv_syscallnames = NULL,
780 .sv_shared_page_base = SHAREDPAGE,
781 .sv_shared_page_len = PAGE_SIZE,
782 .sv_schedtail = linux_schedtail,
783 .sv_thread_detach = linux_thread_detach,
784 .sv_trap = linux_vsyscall,
788 linux_vdso_install(void *param)
791 amd64_lower_shared_page(&elf_linux_sysvec);
793 linux_szsigcode = (&_binary_linux_locore_o_end -
794 &_binary_linux_locore_o_start);
796 if (linux_szsigcode > elf_linux_sysvec.sv_shared_page_len)
797 panic("Linux invalid vdso size\n");
799 __elfN(linux_vdso_fixup)(&elf_linux_sysvec);
801 linux_shared_page_obj = __elfN(linux_shared_page_init)
802 (&linux_shared_page_mapping);
804 __elfN(linux_vdso_reloc)(&elf_linux_sysvec);
806 bcopy(elf_linux_sysvec.sv_sigcode, linux_shared_page_mapping,
808 elf_linux_sysvec.sv_shared_page_obj = linux_shared_page_obj;
810 linux_kplatform = linux_shared_page_mapping +
811 (linux_platform - (caddr_t)elf_linux_sysvec.sv_shared_page_base);
813 SYSINIT(elf_linux_vdso_init, SI_SUB_EXEC, SI_ORDER_ANY,
814 (sysinit_cfunc_t)linux_vdso_install, NULL);
817 linux_vdso_deinstall(void *param)
820 __elfN(linux_shared_page_fini)(linux_shared_page_obj);
822 SYSUNINIT(elf_linux_vdso_uninit, SI_SUB_EXEC, SI_ORDER_FIRST,
823 (sysinit_cfunc_t)linux_vdso_deinstall, NULL);
825 static char GNULINUX_ABI_VENDOR[] = "GNU";
826 static int GNULINUX_ABI_DESC = 0;
829 linux_trans_osrel(const Elf_Note *note, int32_t *osrel)
831 const Elf32_Word *desc;
834 p = (uintptr_t)(note + 1);
835 p += roundup2(note->n_namesz, sizeof(Elf32_Addr));
837 desc = (const Elf32_Word *)p;
838 if (desc[0] != GNULINUX_ABI_DESC)
842 * For Linux we encode osrel as follows (see linux_mib.c):
843 * VVVMMMIII (version, major, minor), see linux_mib.c.
845 *osrel = desc[1] * 1000000 + desc[2] * 1000 + desc[3];
850 static Elf_Brandnote linux64_brandnote = {
851 .hdr.n_namesz = sizeof(GNULINUX_ABI_VENDOR),
854 .vendor = GNULINUX_ABI_VENDOR,
855 .flags = BN_TRANSLATE_OSREL,
856 .trans_osrel = linux_trans_osrel
859 static Elf64_Brandinfo linux_glibc2brand = {
860 .brand = ELFOSABI_LINUX,
861 .machine = EM_X86_64,
862 .compat_3_brand = "Linux",
863 .emul_path = "/compat/linux",
864 .interp_path = "/lib64/ld-linux-x86-64.so.2",
865 .sysvec = &elf_linux_sysvec,
866 .interp_newpath = NULL,
867 .brand_note = &linux64_brandnote,
868 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
871 static Elf64_Brandinfo linux_glibc2brandshort = {
872 .brand = ELFOSABI_LINUX,
873 .machine = EM_X86_64,
874 .compat_3_brand = "Linux",
875 .emul_path = "/compat/linux",
876 .interp_path = "/lib64/ld-linux.so.2",
877 .sysvec = &elf_linux_sysvec,
878 .interp_newpath = NULL,
879 .brand_note = &linux64_brandnote,
880 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
883 static Elf64_Brandinfo linux_muslbrand = {
884 .brand = ELFOSABI_LINUX,
885 .machine = EM_X86_64,
886 .compat_3_brand = "Linux",
887 .emul_path = "/compat/linux",
888 .interp_path = "/lib/ld-musl-x86_64.so.1",
889 .sysvec = &elf_linux_sysvec,
890 .interp_newpath = NULL,
891 .brand_note = &linux64_brandnote,
892 .flags = BI_CAN_EXEC_DYN | BI_BRAND_NOTE
895 Elf64_Brandinfo *linux_brandlist[] = {
897 &linux_glibc2brandshort,
903 linux64_elf_modevent(module_t mod, int type, void *data)
905 Elf64_Brandinfo **brandinfo;
907 struct linux_ioctl_handler **lihp;
913 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
915 if (elf64_insert_brand_entry(*brandinfo) < 0)
918 SET_FOREACH(lihp, linux_ioctl_handler_set)
919 linux_ioctl_register_handler(*lihp);
920 LIST_INIT(&futex_list);
921 mtx_init(&futex_mtx, "ftllk64", NULL, MTX_DEF);
922 stclohz = (stathz ? stathz : hz);
924 printf("Linux x86-64 ELF exec handler installed\n");
926 printf("cannot insert Linux x86-64 ELF brand handler\n");
929 for (brandinfo = &linux_brandlist[0]; *brandinfo != NULL;
931 if (elf64_brand_inuse(*brandinfo))
934 for (brandinfo = &linux_brandlist[0];
935 *brandinfo != NULL; ++brandinfo)
936 if (elf64_remove_brand_entry(*brandinfo) < 0)
940 SET_FOREACH(lihp, linux_ioctl_handler_set)
941 linux_ioctl_unregister_handler(*lihp);
942 mtx_destroy(&futex_mtx);
944 printf("Linux ELF exec handler removed\n");
946 printf("Could not deinstall ELF interpreter entry\n");
954 static moduledata_t linux64_elf_mod = {
956 linux64_elf_modevent,
960 DECLARE_MODULE_TIED(linux64elf, linux64_elf_mod, SI_SUB_EXEC, SI_ORDER_ANY);
961 MODULE_DEPEND(linux64elf, linux_common, 1, 1, 1);
962 FEATURE(linux64, "Linux 64bit support");