2 * Copyright (c) 1993, David Greenman
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD$");
30 #include "opt_capsicum.h"
31 #include "opt_hwpmc_hooks.h"
32 #include "opt_kdtrace.h"
33 #include "opt_ktrace.h"
36 #include <sys/param.h>
37 #include <sys/capability.h>
38 #include <sys/systm.h>
39 #include <sys/capability.h>
40 #include <sys/eventhandler.h>
42 #include <sys/mutex.h>
43 #include <sys/sysproto.h>
44 #include <sys/signalvar.h>
45 #include <sys/kernel.h>
46 #include <sys/mount.h>
47 #include <sys/filedesc.h>
48 #include <sys/fcntl.h>
51 #include <sys/imgact.h>
52 #include <sys/imgact_elf.h>
54 #include <sys/malloc.h>
57 #include <sys/pioctl.h>
58 #include <sys/namei.h>
59 #include <sys/resourcevar.h>
60 #include <sys/sched.h>
62 #include <sys/sf_buf.h>
63 #include <sys/syscallsubr.h>
64 #include <sys/sysent.h>
66 #include <sys/sysctl.h>
67 #include <sys/vnode.h>
70 #include <sys/ktrace.h>
74 #include <vm/vm_param.h>
76 #include <vm/vm_page.h>
77 #include <vm/vm_map.h>
78 #include <vm/vm_kern.h>
79 #include <vm/vm_extern.h>
80 #include <vm/vm_object.h>
81 #include <vm/vm_pager.h>
84 #include <sys/pmckern.h>
87 #include <machine/reg.h>
89 #include <security/audit/audit.h>
90 #include <security/mac/mac_framework.h>
93 #include <sys/dtrace_bsd.h>
94 dtrace_execexit_func_t dtrace_fasttrap_exec;
97 SDT_PROVIDER_DECLARE(proc);
98 SDT_PROBE_DEFINE1(proc, kernel, , exec, "char *");
99 SDT_PROBE_DEFINE1(proc, kernel, , exec__failure, "int");
100 SDT_PROBE_DEFINE1(proc, kernel, , exec__success, "char *");
102 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
104 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
105 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
106 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
107 static int do_execve(struct thread *td, struct image_args *args,
110 /* XXX This should be vm_size_t. */
111 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD,
112 NULL, 0, sysctl_kern_ps_strings, "LU", "");
114 /* XXX This should be vm_size_t. */
115 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD|
116 CTLFLAG_CAPRD, NULL, 0, sysctl_kern_usrstack, "LU", "");
118 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD,
119 NULL, 0, sysctl_kern_stackprot, "I", "");
121 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
122 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
123 &ps_arg_cache_limit, 0, "");
125 static int disallow_high_osrel;
126 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW,
127 &disallow_high_osrel, 0,
128 "Disallow execution of binaries built for higher version of the world");
130 static int map_at_zero = 0;
131 TUNABLE_INT("security.bsd.map_at_zero", &map_at_zero);
132 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RW, &map_at_zero, 0,
133 "Permit processes to map an object at virtual address 0.");
136 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
143 if (req->flags & SCTL_MASK32) {
145 val = (unsigned int)p->p_sysent->sv_psstrings;
146 error = SYSCTL_OUT(req, &val, sizeof(val));
149 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
150 sizeof(p->p_sysent->sv_psstrings));
155 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
162 if (req->flags & SCTL_MASK32) {
164 val = (unsigned int)p->p_sysent->sv_usrstack;
165 error = SYSCTL_OUT(req, &val, sizeof(val));
168 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
169 sizeof(p->p_sysent->sv_usrstack));
174 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
179 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
180 sizeof(p->p_sysent->sv_stackprot)));
184 * Each of the items is a pointer to a `const struct execsw', hence the
185 * double pointer here.
187 static const struct execsw **execsw;
189 #ifndef _SYS_SYSPROTO_H_
200 struct execve_args /* {
207 struct image_args args;
209 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
210 uap->argv, uap->envv);
212 error = kern_execve(td, &args, NULL);
216 #ifndef _SYS_SYSPROTO_H_
217 struct fexecve_args {
224 sys_fexecve(struct thread *td, struct fexecve_args *uap)
227 struct image_args args;
229 error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
230 uap->argv, uap->envv);
233 error = kern_execve(td, &args, NULL);
238 #ifndef _SYS_SYSPROTO_H_
239 struct __mac_execve_args {
248 sys___mac_execve(td, uap)
250 struct __mac_execve_args /* {
259 struct image_args args;
261 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
262 uap->argv, uap->envv);
264 error = kern_execve(td, &args, uap->mac_p);
272 * XXX: kern_execve has the astonishing property of not always returning to
273 * the caller. If sufficiently bad things happen during the call to
274 * do_execve(), it can end up calling exit1(); as a result, callers must
275 * avoid doing anything which they might need to undo (e.g., allocating
279 kern_execve(td, args, mac_p)
281 struct image_args *args;
284 struct proc *p = td->td_proc;
285 struct vmspace *oldvmspace;
288 AUDIT_ARG_ARGV(args->begin_argv, args->argc,
289 args->begin_envv - args->begin_argv);
290 AUDIT_ARG_ENVV(args->begin_envv, args->envc,
291 args->endp - args->begin_envv);
292 if (p->p_flag & P_HADTHREADS) {
294 if (thread_single(SINGLE_BOUNDARY)) {
296 exec_free_args(args);
297 return (ERESTART); /* Try again later. */
302 KASSERT((td->td_pflags & TDP_EXECVMSPC) == 0, ("nested execve"));
303 oldvmspace = td->td_proc->p_vmspace;
304 error = do_execve(td, args, mac_p);
306 if (p->p_flag & P_HADTHREADS) {
309 * If success, we upgrade to SINGLE_EXIT state to
310 * force other threads to suicide.
313 thread_single(SINGLE_EXIT);
318 if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
319 KASSERT(td->td_proc->p_vmspace != oldvmspace,
320 ("oldvmspace still used"));
321 vmspace_free(oldvmspace);
322 td->td_pflags &= ~TDP_EXECVMSPC;
329 * In-kernel implementation of execve(). All arguments are assumed to be
330 * userspace pointers from the passed thread.
333 do_execve(td, args, mac_p)
335 struct image_args *args;
338 struct proc *p = td->td_proc;
340 struct ucred *newcred = NULL, *oldcred;
341 struct uidinfo *euip;
342 register_t *stack_base;
344 struct image_params image_params, *imgp;
346 int (*img_first)(struct image_params *);
347 struct pargs *oldargs = NULL, *newargs = NULL;
348 struct sigacts *oldsigacts, *newsigacts;
350 struct vnode *tracevp = NULL;
351 struct ucred *tracecred = NULL;
353 struct vnode *textvp = NULL, *binvp = NULL;
354 int credential_changing;
358 struct label *interpvplabel = NULL;
362 struct pmckern_procexec pe;
364 static const char fexecv_proc_title[] = "(fexecv)";
367 imgp = &image_params;
370 * Lock the process and set the P_INEXEC flag to indicate that
371 * it should be left alone until we're done here. This is
372 * necessary to avoid race conditions - e.g. in ptrace() -
373 * that might allow a local user to illicitly obtain elevated
377 KASSERT((p->p_flag & P_INEXEC) == 0,
378 ("%s(): process already has P_INEXEC flag", __func__));
379 p->p_flag |= P_INEXEC;
383 * Initialize part of the common data
386 imgp->execlabel = NULL;
388 imgp->entry_addr = 0;
389 imgp->reloc_base = 0;
390 imgp->vmspace_destroyed = 0;
391 imgp->interpreted = 0;
393 imgp->interpreter_name = NULL;
394 imgp->auxargs = NULL;
397 imgp->firstpage = NULL;
398 imgp->ps_strings = 0;
399 imgp->auxarg_size = 0;
401 imgp->execpath = imgp->freepath = NULL;
406 imgp->pagesizeslen = 0;
407 imgp->stack_prot = 0;
410 error = mac_execve_enter(imgp, mac_p);
415 imgp->image_header = NULL;
418 * Translate the file name. namei() returns a vnode pointer
419 * in ni_vp amoung other things.
421 * XXXAUDIT: It would be desirable to also audit the name of the
422 * interpreter if this is an interpreted binary.
424 if (args->fname != NULL) {
425 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME
426 | MPSAFE | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
429 SDT_PROBE(proc, kernel, , exec, args->fname, 0, 0, 0, 0 );
432 if (args->fname != NULL) {
433 #ifdef CAPABILITY_MODE
435 * While capability mode can't reach this point via direct
436 * path arguments to execve(), we also don't allow
437 * interpreters to be used in capability mode (for now).
438 * Catch indirect lookups and return a permissions error.
440 if (IN_CAPABILITY_MODE(td)) {
449 vfslocked = NDHASGIANT(&nd);
453 AUDIT_ARG_FD(args->fd);
455 * Some might argue that CAP_READ and/or CAP_MMAP should also
456 * be required here; such arguments will be entertained.
458 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
460 error = fgetvp_exec(td, args->fd, CAP_FEXECVE, &binvp);
463 vfslocked = VFS_LOCK_GIANT(binvp->v_mount);
464 vn_lock(binvp, LK_EXCLUSIVE | LK_RETRY);
465 AUDIT_ARG_VNODE1(binvp);
470 * Check file permissions (also 'opens' file)
472 error = exec_check_permissions(imgp);
474 goto exec_fail_dealloc;
476 imgp->object = imgp->vp->v_object;
477 if (imgp->object != NULL)
478 vm_object_reference(imgp->object);
481 * Set VV_TEXT now so no one can write to the executable while we're
484 * Remember if this was set before and unset it in case this is not
485 * actually an executable image.
487 textset = VOP_IS_TEXT(imgp->vp);
488 VOP_SET_TEXT(imgp->vp);
490 error = exec_map_first_page(imgp);
492 goto exec_fail_dealloc;
494 imgp->proc->p_osrel = 0;
496 * If the current process has a special image activator it
497 * wants to try first, call it. For example, emulating shell
498 * scripts differently.
501 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
502 error = img_first(imgp);
505 * Loop through the list of image activators, calling each one.
506 * An activator returns -1 if there is no match, 0 on success,
507 * and an error otherwise.
509 for (i = 0; error == -1 && execsw[i]; ++i) {
510 if (execsw[i]->ex_imgact == NULL ||
511 execsw[i]->ex_imgact == img_first) {
514 error = (*execsw[i]->ex_imgact)(imgp);
520 VOP_UNSET_TEXT(imgp->vp);
523 goto exec_fail_dealloc;
527 * Special interpreter operation, cleanup and loop up to try to
528 * activate the interpreter.
530 if (imgp->interpreted) {
531 exec_unmap_first_page(imgp);
533 * VV_TEXT needs to be unset for scripts. There is a short
534 * period before we determine that something is a script where
535 * VV_TEXT will be set. The vnode lock is held over this
536 * entire period so nothing should illegitimately be blocked.
538 VOP_UNSET_TEXT(imgp->vp);
539 /* free name buffer and old vnode */
540 if (args->fname != NULL)
541 NDFREE(&nd, NDF_ONLY_PNBUF);
543 mac_execve_interpreter_enter(binvp, &interpvplabel);
546 VOP_CLOSE(binvp, FREAD, td->td_ucred, td);
550 vm_object_deallocate(imgp->object);
552 VFS_UNLOCK_GIANT(vfslocked);
554 /* set new name to that of the interpreter */
555 NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME | MPSAFE,
556 UIO_SYSSPACE, imgp->interpreter_name, td);
557 args->fname = imgp->interpreter_name;
562 * NB: We unlock the vnode here because it is believed that none
563 * of the sv_copyout_strings/sv_fixup operations require the vnode.
565 VOP_UNLOCK(imgp->vp, 0);
568 * Do the best to calculate the full path to the image file.
570 if (imgp->auxargs != NULL &&
571 ((args->fname != NULL && args->fname[0] == '/') ||
572 vn_fullpath(td, imgp->vp, &imgp->execpath, &imgp->freepath) != 0))
573 imgp->execpath = args->fname;
575 if (disallow_high_osrel &&
576 P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
578 uprintf("Osrel %d for image %s too high\n", p->p_osrel,
579 imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
580 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
581 goto exec_fail_dealloc;
585 * Copy out strings (args and env) and initialize stack base
587 if (p->p_sysent->sv_copyout_strings)
588 stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
590 stack_base = exec_copyout_strings(imgp);
593 * If custom stack fixup routine present for this process
594 * let it do the stack setup.
595 * Else stuff argument count as first item on stack
597 if (p->p_sysent->sv_fixup != NULL)
598 (*p->p_sysent->sv_fixup)(&stack_base, imgp);
600 suword(--stack_base, imgp->args->argc);
603 * For security and other reasons, the file descriptor table cannot
604 * be shared after an exec.
609 * Malloc things before we need locks.
612 euip = uifind(attr.va_uid);
613 i = imgp->args->begin_envv - imgp->args->begin_argv;
614 /* Cache arguments if they fit inside our allowance */
615 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
616 newargs = pargs_alloc(i);
617 bcopy(imgp->args->begin_argv, newargs->ar_args, i);
620 /* close files on exec */
622 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
624 /* Get a reference to the vnode prior to locking the proc */
628 * For security and other reasons, signal handlers cannot
629 * be shared after an exec. The new process gets a copy of the old
630 * handlers. In execsigs(), the new process will have its signals
634 oldcred = crcopysafe(p, newcred);
635 if (sigacts_shared(p->p_sigacts)) {
636 oldsigacts = p->p_sigacts;
638 newsigacts = sigacts_alloc();
639 sigacts_copy(newsigacts, oldsigacts);
641 p->p_sigacts = newsigacts;
648 /* reset caught signals */
651 /* name this process - nameiexec(p, ndp) */
652 bzero(p->p_comm, sizeof(p->p_comm));
654 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
655 min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
656 else if (vn_commname(binvp, p->p_comm, sizeof(p->p_comm)) != 0)
657 bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
658 bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
660 sched_clear_tdname(td);
664 * mark as execed, wakeup the process that vforked (if any) and tell
665 * it that it now has its own resources back
668 if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
669 p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
670 cv_broadcast(&p->p_pwait);
674 * Implement image setuid/setgid.
676 * Don't honor setuid/setgid if the filesystem prohibits it or if
677 * the process is being traced.
679 * We disable setuid/setgid/etc in compatibility mode on the basis
680 * that most setugid applications are not written with that
681 * environment in mind, and will therefore almost certainly operate
682 * incorrectly. In principle there's no reason that setugid
683 * applications might not be useful in capability mode, so we may want
684 * to reconsider this conservative design choice in the future.
686 * XXXMAC: For the time being, use NOSUID to also prohibit
687 * transitions on the file system.
689 credential_changing = 0;
690 credential_changing |= (attr.va_mode & S_ISUID) && oldcred->cr_uid !=
692 credential_changing |= (attr.va_mode & S_ISGID) && oldcred->cr_gid !=
695 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
696 interpvplabel, imgp);
697 credential_changing |= will_transition;
700 if (credential_changing &&
701 #ifdef CAPABILITY_MODE
702 ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
704 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
705 (p->p_flag & P_TRACED) == 0) {
707 * Turn off syscall tracing for set-id programs, except for
708 * root. Record any set-id flags first to make sure that
709 * we do not regain any tracing during a possible block.
714 if (p->p_tracecred != NULL &&
715 priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED, 0))
716 ktrprocexec(p, &tracecred, &tracevp);
719 * Close any file descriptors 0..2 that reference procfs,
720 * then make sure file descriptors 0..2 are in use.
722 * setugidsafety() may call closef() and then pfind()
723 * which may grab the process lock.
724 * fdcheckstd() may call falloc() which may block to
725 * allocate memory, so temporarily drop the process lock.
728 VOP_UNLOCK(imgp->vp, 0);
730 error = fdcheckstd(td);
731 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
736 * Set the new credentials.
738 if (attr.va_mode & S_ISUID)
739 change_euid(newcred, euip);
740 if (attr.va_mode & S_ISGID)
741 change_egid(newcred, attr.va_gid);
743 if (will_transition) {
744 mac_vnode_execve_transition(oldcred, newcred, imgp->vp,
745 interpvplabel, imgp);
749 * Implement correct POSIX saved-id behavior.
751 * XXXMAC: Note that the current logic will save the
752 * uid and gid if a MAC domain transition occurs, even
753 * though maybe it shouldn't.
755 change_svuid(newcred, newcred->cr_uid);
756 change_svgid(newcred, newcred->cr_gid);
757 p->p_ucred = newcred;
760 if (oldcred->cr_uid == oldcred->cr_ruid &&
761 oldcred->cr_gid == oldcred->cr_rgid)
762 p->p_flag &= ~P_SUGID;
764 * Implement correct POSIX saved-id behavior.
766 * XXX: It's not clear that the existing behavior is
767 * POSIX-compliant. A number of sources indicate that the
768 * saved uid/gid should only be updated if the new ruid is
769 * not equal to the old ruid, or the new euid is not equal
770 * to the old euid and the new euid is not equal to the old
771 * ruid. The FreeBSD code always updates the saved uid/gid.
772 * Also, this code uses the new (replaced) euid and egid as
773 * the source, which may or may not be the right ones to use.
775 if (oldcred->cr_svuid != oldcred->cr_uid ||
776 oldcred->cr_svgid != oldcred->cr_gid) {
777 change_svuid(newcred, newcred->cr_uid);
778 change_svgid(newcred, newcred->cr_gid);
779 p->p_ucred = newcred;
785 * Store the vp for use in procfs. This vnode was referenced prior
786 * to locking the proc lock.
788 textvp = p->p_textvp;
793 * Tell the DTrace fasttrap provider about the exec if it
794 * has declared an interest.
796 if (dtrace_fasttrap_exec)
797 dtrace_fasttrap_exec(p);
801 * Notify others that we exec'd, and clear the P_INEXEC flag
802 * as we're now a bona fide freshly-execed process.
804 KNOTE_LOCKED(&p->p_klist, NOTE_EXEC);
805 p->p_flag &= ~P_INEXEC;
807 /* clear "fork but no exec" flag, as we _are_ execing */
808 p->p_acflag &= ~AFORK;
811 * Free any previous argument cache and replace it with
812 * the new argument cache, if any.
820 * Check if system-wide sampling is in effect or if the
821 * current process is using PMCs. If so, do exec() time
822 * processing. This processing needs to happen AFTER the
823 * P_INEXEC flag is cleared.
825 * The proc lock needs to be released before taking the PMC
828 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
830 VOP_UNLOCK(imgp->vp, 0);
831 pe.pm_credentialschanged = credential_changing;
832 pe.pm_entryaddr = imgp->entry_addr;
834 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
835 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
838 #else /* !HWPMC_HOOKS */
842 /* Set values passed into the program in registers. */
843 if (p->p_sysent->sv_setregs)
844 (*p->p_sysent->sv_setregs)(td, imgp,
845 (u_long)(uintptr_t)stack_base);
847 exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base);
849 vfs_mark_atime(imgp->vp, td->td_ucred);
851 SDT_PROBE(proc, kernel, , exec__success, args->fname, 0, 0, 0, 0);
855 * Free any resources malloc'd earlier that we didn't use.
862 VOP_UNLOCK(imgp->vp, 0);
865 * Handle deferred decrement of ref counts.
867 if (textvp != NULL) {
870 tvfslocked = VFS_LOCK_GIANT(textvp->v_mount);
872 VFS_UNLOCK_GIANT(tvfslocked);
874 if (binvp && error != 0)
877 if (tracevp != NULL) {
880 tvfslocked = VFS_LOCK_GIANT(tracevp->v_mount);
882 VFS_UNLOCK_GIANT(tvfslocked);
884 if (tracecred != NULL)
887 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
890 if (oldsigacts != NULL)
891 sigacts_free(oldsigacts);
896 * free various allocated resources
898 if (imgp->firstpage != NULL)
899 exec_unmap_first_page(imgp);
901 if (imgp->vp != NULL) {
903 NDFREE(&nd, NDF_ONLY_PNBUF);
905 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
909 if (imgp->object != NULL)
910 vm_object_deallocate(imgp->object);
912 free(imgp->freepath, M_TEMP);
916 td->td_dbgflags |= TDB_EXEC;
920 * Stop the process here if its stop event mask has
921 * the S_EXEC bit set.
923 STOPEVENT(p, S_EXEC, 0);
928 /* we're done here, clear P_INEXEC */
930 p->p_flag &= ~P_INEXEC;
933 SDT_PROBE(proc, kernel, , exec__failure, error, 0, 0, 0, 0);
937 mac_execve_exit(imgp);
938 mac_execve_interpreter_exit(interpvplabel);
940 VFS_UNLOCK_GIANT(vfslocked);
941 exec_free_args(args);
943 if (error && imgp->vmspace_destroyed) {
944 /* sorry, no more process anymore. exit gracefully */
945 exit1(td, W_EXITCODE(0, SIGABRT));
958 exec_map_first_page(imgp)
959 struct image_params *imgp;
963 vm_page_t ma[VM_INITIAL_PAGEIN];
966 if (imgp->firstpage != NULL)
967 exec_unmap_first_page(imgp);
969 object = imgp->vp->v_object;
972 VM_OBJECT_LOCK(object);
973 #if VM_NRESERVLEVEL > 0
974 if ((object->flags & OBJ_COLORED) == 0) {
975 object->flags |= OBJ_COLORED;
976 object->pg_color = 0;
979 ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
980 if (ma[0]->valid != VM_PAGE_BITS_ALL) {
981 initial_pagein = VM_INITIAL_PAGEIN;
982 if (initial_pagein > object->size)
983 initial_pagein = object->size;
984 for (i = 1; i < initial_pagein; i++) {
985 if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) {
988 if ((ma[i]->oflags & VPO_BUSY) || ma[i]->busy)
992 ma[i] = vm_page_alloc(object, i,
993 VM_ALLOC_NORMAL | VM_ALLOC_IFNOTCACHED);
999 rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
1000 ma[0] = vm_page_lookup(object, 0);
1001 if ((rv != VM_PAGER_OK) || (ma[0] == NULL)) {
1002 if (ma[0] != NULL) {
1003 vm_page_lock(ma[0]);
1004 vm_page_free(ma[0]);
1005 vm_page_unlock(ma[0]);
1007 VM_OBJECT_UNLOCK(object);
1011 vm_page_lock(ma[0]);
1012 vm_page_hold(ma[0]);
1013 vm_page_unlock(ma[0]);
1014 vm_page_wakeup(ma[0]);
1015 VM_OBJECT_UNLOCK(object);
1017 imgp->firstpage = sf_buf_alloc(ma[0], 0);
1018 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
1024 exec_unmap_first_page(imgp)
1025 struct image_params *imgp;
1029 if (imgp->firstpage != NULL) {
1030 m = sf_buf_page(imgp->firstpage);
1031 sf_buf_free(imgp->firstpage);
1032 imgp->firstpage = NULL;
1040 * Destroy old address space, and allocate a new stack
1041 * The new stack is only SGROWSIZ large because it is grown
1042 * automatically in trap.c.
1045 exec_new_vmspace(imgp, sv)
1046 struct image_params *imgp;
1047 struct sysentvec *sv;
1050 struct proc *p = imgp->proc;
1051 struct vmspace *vmspace = p->p_vmspace;
1053 vm_offset_t sv_minuser, stack_addr;
1057 imgp->vmspace_destroyed = 1;
1060 /* May be called with Giant held */
1061 EVENTHANDLER_INVOKE(process_exec, p, imgp);
1064 * Blow away entire process VM, if address space not shared,
1065 * otherwise, create a new VM space so that other threads are
1068 map = &vmspace->vm_map;
1070 sv_minuser = sv->sv_minuser;
1072 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1073 if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser &&
1074 vm_map_max(map) == sv->sv_maxuser) {
1076 pmap_remove_pages(vmspace_pmap(vmspace));
1077 vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1079 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1082 vmspace = p->p_vmspace;
1083 map = &vmspace->vm_map;
1086 /* Map a shared page */
1087 obj = sv->sv_shared_page_obj;
1089 vm_object_reference(obj);
1090 error = vm_map_fixed(map, obj, 0,
1091 sv->sv_shared_page_base, sv->sv_shared_page_len,
1092 VM_PROT_READ | VM_PROT_EXECUTE,
1093 VM_PROT_READ | VM_PROT_EXECUTE,
1094 MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1096 vm_object_deallocate(obj);
1101 /* Allocate a new stack */
1102 if (sv->sv_maxssiz != NULL)
1103 ssiz = *sv->sv_maxssiz;
1106 stack_addr = sv->sv_usrstack - ssiz;
1107 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1108 obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1110 VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1115 /* Allocate a new register stack */
1116 stack_addr = IA64_BACKINGSTORE;
1117 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1118 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_UP);
1123 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
1124 * VM_STACK case, but they are still used to monitor the size of the
1125 * process stack so we can check the stack rlimit.
1127 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1128 vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - ssiz;
1134 * Copy out argument and environment strings from the old process address
1135 * space into the temporary string buffer.
1138 exec_copyin_args(struct image_args *args, char *fname,
1139 enum uio_seg segflg, char **argv, char **envv)
1145 bzero(args, sizeof(*args));
1150 * Allocate demand-paged memory for the file name, argument, and
1151 * environment strings.
1153 error = exec_alloc_args(args);
1158 * Copy the file name.
1160 if (fname != NULL) {
1161 args->fname = args->buf;
1162 error = (segflg == UIO_SYSSPACE) ?
1163 copystr(fname, args->fname, PATH_MAX, &length) :
1164 copyinstr(fname, args->fname, PATH_MAX, &length);
1170 args->begin_argv = args->buf + length;
1171 args->endp = args->begin_argv;
1172 args->stringspace = ARG_MAX;
1175 * extract arguments first
1177 while ((argp = (caddr_t) (intptr_t) fuword(argv++))) {
1178 if (argp == (caddr_t) -1) {
1182 if ((error = copyinstr(argp, args->endp,
1183 args->stringspace, &length))) {
1184 if (error == ENAMETOOLONG)
1188 args->stringspace -= length;
1189 args->endp += length;
1193 args->begin_envv = args->endp;
1196 * extract environment strings
1199 while ((envp = (caddr_t)(intptr_t)fuword(envv++))) {
1200 if (envp == (caddr_t)-1) {
1204 if ((error = copyinstr(envp, args->endp,
1205 args->stringspace, &length))) {
1206 if (error == ENAMETOOLONG)
1210 args->stringspace -= length;
1211 args->endp += length;
1219 exec_free_args(args);
1224 * Allocate temporary demand-paged, zero-filled memory for the file name,
1225 * argument, and environment strings. Returns zero if the allocation succeeds
1226 * and ENOMEM otherwise.
1229 exec_alloc_args(struct image_args *args)
1232 args->buf = (char *)kmem_alloc_wait(exec_map, PATH_MAX + ARG_MAX);
1233 return (args->buf != NULL ? 0 : ENOMEM);
1237 exec_free_args(struct image_args *args)
1240 if (args->buf != NULL) {
1241 kmem_free_wakeup(exec_map, (vm_offset_t)args->buf,
1242 PATH_MAX + ARG_MAX);
1245 if (args->fname_buf != NULL) {
1246 free(args->fname_buf, M_TEMP);
1247 args->fname_buf = NULL;
1252 * Copy strings out to the new process address space, constructing new arg
1253 * and env vector tables. Return a pointer to the base so that it can be used
1254 * as the initial stack pointer.
1257 exec_copyout_strings(imgp)
1258 struct image_params *imgp;
1264 register_t *stack_base;
1265 struct ps_strings *arginfo;
1267 size_t execpath_len;
1268 int szsigcode, szps;
1269 char canary[sizeof(long) * 8];
1271 szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1273 * Calculate string base and vector table pointers.
1274 * Also deal with signal trampoline code for this exec type.
1276 if (imgp->execpath != NULL && imgp->auxargs != NULL)
1277 execpath_len = strlen(imgp->execpath) + 1;
1282 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1283 if (p->p_sysent->sv_sigcode_base == 0) {
1284 if (p->p_sysent->sv_szsigcode != NULL)
1285 szsigcode = *(p->p_sysent->sv_szsigcode);
1287 destp = (uintptr_t)arginfo;
1292 if (szsigcode != 0) {
1294 destp = rounddown2(destp, sizeof(void *));
1295 copyout(p->p_sysent->sv_sigcode, (void *)destp, szsigcode);
1299 * Copy the image path for the rtld.
1301 if (execpath_len != 0) {
1302 destp -= execpath_len;
1303 imgp->execpathp = destp;
1304 copyout(imgp->execpath, (void *)destp, execpath_len);
1308 * Prepare the canary for SSP.
1310 arc4rand(canary, sizeof(canary), 0);
1311 destp -= sizeof(canary);
1312 imgp->canary = destp;
1313 copyout(canary, (void *)destp, sizeof(canary));
1314 imgp->canarylen = sizeof(canary);
1317 * Prepare the pagesizes array.
1320 destp = rounddown2(destp, sizeof(void *));
1321 imgp->pagesizes = destp;
1322 copyout(pagesizes, (void *)destp, szps);
1323 imgp->pagesizeslen = szps;
1325 destp -= ARG_MAX - imgp->args->stringspace;
1326 destp = rounddown2(destp, sizeof(void *));
1329 * If we have a valid auxargs ptr, prepare some room
1332 if (imgp->auxargs) {
1334 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1335 * lower compatibility.
1337 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1340 * The '+ 2' is for the null pointers at the end of each of
1341 * the arg and env vector sets,and imgp->auxarg_size is room
1342 * for argument of Runtime loader.
1344 vectp = (char **)(destp - (imgp->args->argc +
1345 imgp->args->envc + 2 + imgp->auxarg_size)
1349 * The '+ 2' is for the null pointers at the end of each of
1350 * the arg and env vector sets
1352 vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc
1353 + 2) * sizeof(char *));
1357 * vectp also becomes our initial stack base
1359 stack_base = (register_t *)vectp;
1361 stringp = imgp->args->begin_argv;
1362 argc = imgp->args->argc;
1363 envc = imgp->args->envc;
1366 * Copy out strings - arguments and environment.
1368 copyout(stringp, (void *)destp, ARG_MAX - imgp->args->stringspace);
1371 * Fill in "ps_strings" struct for ps, w, etc.
1373 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1374 suword32(&arginfo->ps_nargvstr, argc);
1377 * Fill in argument portion of vector table.
1379 for (; argc > 0; --argc) {
1380 suword(vectp++, (long)(intptr_t)destp);
1381 while (*stringp++ != 0)
1386 /* a null vector table pointer separates the argp's from the envp's */
1389 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1390 suword32(&arginfo->ps_nenvstr, envc);
1393 * Fill in environment portion of vector table.
1395 for (; envc > 0; --envc) {
1396 suword(vectp++, (long)(intptr_t)destp);
1397 while (*stringp++ != 0)
1402 /* end of vector table is a null pointer */
1405 return (stack_base);
1409 * Check permissions of file to execute.
1410 * Called with imgp->vp locked.
1411 * Return 0 for success or error code on failure.
1414 exec_check_permissions(imgp)
1415 struct image_params *imgp;
1417 struct vnode *vp = imgp->vp;
1418 struct vattr *attr = imgp->attr;
1420 int error, writecount;
1424 /* Get file attributes */
1425 error = VOP_GETATTR(vp, attr, td->td_ucred);
1430 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1436 * 1) Check if file execution is disabled for the filesystem that
1437 * this file resides on.
1438 * 2) Ensure that at least one execute bit is on. Otherwise, a
1439 * privileged user will always succeed, and we don't want this
1440 * to happen unless the file really is executable.
1441 * 3) Ensure that the file is a regular file.
1443 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1444 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1445 (attr->va_type != VREG))
1449 * Zero length files can't be exec'd
1451 if (attr->va_size == 0)
1455 * Check for execute permission to file based on current credentials.
1457 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1462 * Check number of open-for-writes on the file and deny execution
1465 error = VOP_GET_WRITECOUNT(vp, &writecount);
1468 if (writecount != 0)
1472 * Call filesystem specific open routine (which does nothing in the
1475 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1482 * Exec handler registration
1485 exec_register(execsw_arg)
1486 const struct execsw *execsw_arg;
1488 const struct execsw **es, **xs, **newexecsw;
1489 int count = 2; /* New slot and trailing NULL */
1492 for (es = execsw; *es; es++)
1494 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1495 if (newexecsw == NULL)
1499 for (es = execsw; *es; es++)
1504 free(execsw, M_TEMP);
1510 exec_unregister(execsw_arg)
1511 const struct execsw *execsw_arg;
1513 const struct execsw **es, **xs, **newexecsw;
1517 panic("unregister with no handlers left?\n");
1519 for (es = execsw; *es; es++) {
1520 if (*es == execsw_arg)
1525 for (es = execsw; *es; es++)
1526 if (*es != execsw_arg)
1528 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1529 if (newexecsw == NULL)
1532 for (es = execsw; *es; es++)
1533 if (*es != execsw_arg)
1537 free(execsw, M_TEMP);