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_ktrace.h"
35 #include <sys/param.h>
36 #include <sys/capsicum.h>
37 #include <sys/systm.h>
38 #include <sys/eventhandler.h>
40 #include <sys/mutex.h>
41 #include <sys/sysproto.h>
42 #include <sys/signalvar.h>
43 #include <sys/kernel.h>
44 #include <sys/mount.h>
45 #include <sys/filedesc.h>
46 #include <sys/fcntl.h>
49 #include <sys/imgact.h>
50 #include <sys/imgact_elf.h>
52 #include <sys/malloc.h>
55 #include <sys/pioctl.h>
56 #include <sys/namei.h>
57 #include <sys/resourcevar.h>
58 #include <sys/rwlock.h>
59 #include <sys/sched.h>
61 #include <sys/sf_buf.h>
62 #include <sys/syscallsubr.h>
63 #include <sys/sysent.h>
65 #include <sys/sysctl.h>
66 #include <sys/vnode.h>
69 #include <sys/ktrace.h>
73 #include <vm/vm_param.h>
75 #include <vm/vm_page.h>
76 #include <vm/vm_map.h>
77 #include <vm/vm_kern.h>
78 #include <vm/vm_extern.h>
79 #include <vm/vm_object.h>
80 #include <vm/vm_pager.h>
83 #include <sys/pmckern.h>
86 #include <machine/reg.h>
88 #include <security/audit/audit.h>
89 #include <security/mac/mac_framework.h>
92 #include <sys/dtrace_bsd.h>
93 dtrace_execexit_func_t dtrace_fasttrap_exec;
96 SDT_PROVIDER_DECLARE(proc);
97 SDT_PROBE_DEFINE1(proc, kernel, , exec, "char *");
98 SDT_PROBE_DEFINE1(proc, kernel, , exec__failure, "int");
99 SDT_PROBE_DEFINE1(proc, kernel, , exec__success, "char *");
101 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
103 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
104 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
105 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
106 static int do_execve(struct thread *td, struct image_args *args,
109 /* XXX This should be vm_size_t. */
110 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD,
111 NULL, 0, sysctl_kern_ps_strings, "LU", "");
113 /* XXX This should be vm_size_t. */
114 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD|
115 CTLFLAG_CAPRD, NULL, 0, sysctl_kern_usrstack, "LU", "");
117 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD,
118 NULL, 0, sysctl_kern_stackprot, "I", "");
120 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
121 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
122 &ps_arg_cache_limit, 0, "");
124 static int disallow_high_osrel;
125 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW,
126 &disallow_high_osrel, 0,
127 "Disallow execution of binaries built for higher version of the world");
129 static int map_at_zero = 0;
130 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RWTUN, &map_at_zero, 0,
131 "Permit processes to map an object at virtual address 0.");
134 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
141 if (req->flags & SCTL_MASK32) {
143 val = (unsigned int)p->p_sysent->sv_psstrings;
144 error = SYSCTL_OUT(req, &val, sizeof(val));
147 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
148 sizeof(p->p_sysent->sv_psstrings));
153 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
160 if (req->flags & SCTL_MASK32) {
162 val = (unsigned int)p->p_sysent->sv_usrstack;
163 error = SYSCTL_OUT(req, &val, sizeof(val));
166 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
167 sizeof(p->p_sysent->sv_usrstack));
172 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
177 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
178 sizeof(p->p_sysent->sv_stackprot)));
182 * Each of the items is a pointer to a `const struct execsw', hence the
183 * double pointer here.
185 static const struct execsw **execsw;
187 #ifndef _SYS_SYSPROTO_H_
196 sys_execve(struct thread *td, struct execve_args *uap)
198 struct image_args args;
199 struct vmspace *oldvmspace;
202 error = pre_execve(td, &oldvmspace);
205 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
206 uap->argv, uap->envv);
208 error = kern_execve(td, &args, NULL);
209 post_execve(td, error, oldvmspace);
213 #ifndef _SYS_SYSPROTO_H_
214 struct fexecve_args {
221 sys_fexecve(struct thread *td, struct fexecve_args *uap)
223 struct image_args args;
224 struct vmspace *oldvmspace;
227 error = pre_execve(td, &oldvmspace);
230 error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
231 uap->argv, uap->envv);
234 error = kern_execve(td, &args, NULL);
236 post_execve(td, error, oldvmspace);
240 #ifndef _SYS_SYSPROTO_H_
241 struct __mac_execve_args {
250 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap)
253 struct image_args args;
254 struct vmspace *oldvmspace;
257 error = pre_execve(td, &oldvmspace);
260 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
261 uap->argv, uap->envv);
263 error = kern_execve(td, &args, uap->mac_p);
264 post_execve(td, error, oldvmspace);
272 pre_execve(struct thread *td, struct vmspace **oldvmspace)
277 KASSERT(td == curthread, ("non-current thread %p", td));
280 if ((p->p_flag & P_HADTHREADS) != 0) {
282 if (thread_single(p, SINGLE_BOUNDARY) != 0)
286 KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0,
288 *oldvmspace = p->p_vmspace;
293 post_execve(struct thread *td, int error, struct vmspace *oldvmspace)
297 KASSERT(td == curthread, ("non-current thread %p", td));
299 if ((p->p_flag & P_HADTHREADS) != 0) {
302 * If success, we upgrade to SINGLE_EXIT state to
303 * force other threads to suicide.
306 thread_single(p, SINGLE_EXIT);
308 thread_single_end(p, SINGLE_BOUNDARY);
311 if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
312 KASSERT(p->p_vmspace != oldvmspace,
313 ("oldvmspace still used"));
314 vmspace_free(oldvmspace);
315 td->td_pflags &= ~TDP_EXECVMSPC;
320 * XXX: kern_execve has the astonishing property of not always returning to
321 * the caller. If sufficiently bad things happen during the call to
322 * do_execve(), it can end up calling exit1(); as a result, callers must
323 * avoid doing anything which they might need to undo (e.g., allocating
327 kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p)
330 AUDIT_ARG_ARGV(args->begin_argv, args->argc,
331 args->begin_envv - args->begin_argv);
332 AUDIT_ARG_ENVV(args->begin_envv, args->envc,
333 args->endp - args->begin_envv);
334 return (do_execve(td, args, mac_p));
338 * In-kernel implementation of execve(). All arguments are assumed to be
339 * userspace pointers from the passed thread.
342 do_execve(td, args, mac_p)
344 struct image_args *args;
347 struct proc *p = td->td_proc;
349 struct ucred *newcred = NULL, *oldcred;
350 struct uidinfo *euip = NULL;
351 register_t *stack_base;
353 struct image_params image_params, *imgp;
355 int (*img_first)(struct image_params *);
356 struct pargs *oldargs = NULL, *newargs = NULL;
357 struct sigacts *oldsigacts, *newsigacts;
359 struct vnode *tracevp = NULL;
360 struct ucred *tracecred = NULL;
362 struct vnode *textvp = NULL, *binvp;
364 int credential_changing;
367 struct label *interpvplabel = NULL;
371 struct pmckern_procexec pe;
373 static const char fexecv_proc_title[] = "(fexecv)";
375 imgp = &image_params;
378 * Lock the process and set the P_INEXEC flag to indicate that
379 * it should be left alone until we're done here. This is
380 * necessary to avoid race conditions - e.g. in ptrace() -
381 * that might allow a local user to illicitly obtain elevated
385 KASSERT((p->p_flag & P_INEXEC) == 0,
386 ("%s(): process already has P_INEXEC flag", __func__));
387 p->p_flag |= P_INEXEC;
391 * Initialize part of the common data
393 bzero(imgp, sizeof(*imgp));
399 error = mac_execve_enter(imgp, mac_p);
405 * Translate the file name. namei() returns a vnode pointer
406 * in ni_vp amoung other things.
408 * XXXAUDIT: It would be desirable to also audit the name of the
409 * interpreter if this is an interpreted binary.
411 if (args->fname != NULL) {
412 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME
413 | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
416 SDT_PROBE(proc, kernel, , exec, args->fname, 0, 0, 0, 0 );
419 if (args->fname != NULL) {
420 #ifdef CAPABILITY_MODE
422 * While capability mode can't reach this point via direct
423 * path arguments to execve(), we also don't allow
424 * interpreters to be used in capability mode (for now).
425 * Catch indirect lookups and return a permissions error.
427 if (IN_CAPABILITY_MODE(td)) {
439 AUDIT_ARG_FD(args->fd);
441 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
443 error = fgetvp_exec(td, args->fd,
444 cap_rights_init(&rights, CAP_FEXECVE), &binvp);
447 vn_lock(binvp, LK_EXCLUSIVE | LK_RETRY);
448 AUDIT_ARG_VNODE1(binvp);
453 * Check file permissions (also 'opens' file)
455 error = exec_check_permissions(imgp);
457 goto exec_fail_dealloc;
459 imgp->object = imgp->vp->v_object;
460 if (imgp->object != NULL)
461 vm_object_reference(imgp->object);
464 * Set VV_TEXT now so no one can write to the executable while we're
467 * Remember if this was set before and unset it in case this is not
468 * actually an executable image.
470 textset = VOP_IS_TEXT(imgp->vp);
471 VOP_SET_TEXT(imgp->vp);
473 error = exec_map_first_page(imgp);
475 goto exec_fail_dealloc;
477 imgp->proc->p_osrel = 0;
479 * If the current process has a special image activator it
480 * wants to try first, call it. For example, emulating shell
481 * scripts differently.
484 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
485 error = img_first(imgp);
488 * Loop through the list of image activators, calling each one.
489 * An activator returns -1 if there is no match, 0 on success,
490 * and an error otherwise.
492 for (i = 0; error == -1 && execsw[i]; ++i) {
493 if (execsw[i]->ex_imgact == NULL ||
494 execsw[i]->ex_imgact == img_first) {
497 error = (*execsw[i]->ex_imgact)(imgp);
503 VOP_UNSET_TEXT(imgp->vp);
506 goto exec_fail_dealloc;
510 * Special interpreter operation, cleanup and loop up to try to
511 * activate the interpreter.
513 if (imgp->interpreted) {
514 exec_unmap_first_page(imgp);
516 * VV_TEXT needs to be unset for scripts. There is a short
517 * period before we determine that something is a script where
518 * VV_TEXT will be set. The vnode lock is held over this
519 * entire period so nothing should illegitimately be blocked.
521 VOP_UNSET_TEXT(imgp->vp);
522 /* free name buffer and old vnode */
523 if (args->fname != NULL)
524 NDFREE(&nd, NDF_ONLY_PNBUF);
526 mac_execve_interpreter_enter(binvp, &interpvplabel);
529 VOP_CLOSE(binvp, FREAD, td->td_ucred, td);
533 vm_object_deallocate(imgp->object);
535 /* set new name to that of the interpreter */
536 NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
537 UIO_SYSSPACE, imgp->interpreter_name, td);
538 args->fname = imgp->interpreter_name;
543 * NB: We unlock the vnode here because it is believed that none
544 * of the sv_copyout_strings/sv_fixup operations require the vnode.
546 VOP_UNLOCK(imgp->vp, 0);
549 * Do the best to calculate the full path to the image file.
551 if (imgp->auxargs != NULL &&
552 ((args->fname != NULL && args->fname[0] == '/') ||
553 vn_fullpath(td, imgp->vp, &imgp->execpath, &imgp->freepath) != 0))
554 imgp->execpath = args->fname;
556 if (disallow_high_osrel &&
557 P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
559 uprintf("Osrel %d for image %s too high\n", p->p_osrel,
560 imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
561 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
562 goto exec_fail_dealloc;
566 * Copy out strings (args and env) and initialize stack base
568 if (p->p_sysent->sv_copyout_strings)
569 stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
571 stack_base = exec_copyout_strings(imgp);
574 * If custom stack fixup routine present for this process
575 * let it do the stack setup.
576 * Else stuff argument count as first item on stack
578 if (p->p_sysent->sv_fixup != NULL)
579 (*p->p_sysent->sv_fixup)(&stack_base, imgp);
581 suword(--stack_base, imgp->args->argc);
584 * For security and other reasons, the file descriptor table cannot
585 * be shared after an exec.
588 /* close files on exec */
592 * Malloc things before we need locks.
594 i = imgp->args->begin_envv - imgp->args->begin_argv;
595 /* Cache arguments if they fit inside our allowance */
596 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
597 newargs = pargs_alloc(i);
598 bcopy(imgp->args->begin_argv, newargs->ar_args, i);
601 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
603 /* Get a reference to the vnode prior to locking the proc */
607 * For security and other reasons, signal handlers cannot
608 * be shared after an exec. The new process gets a copy of the old
609 * handlers. In execsigs(), the new process will have its signals
612 if (sigacts_shared(p->p_sigacts)) {
613 oldsigacts = p->p_sigacts;
614 newsigacts = sigacts_alloc();
615 sigacts_copy(newsigacts, oldsigacts);
618 newsigacts = NULL; /* satisfy gcc */
623 p->p_sigacts = newsigacts;
624 oldcred = p->p_ucred;
628 /* reset caught signals */
631 /* name this process - nameiexec(p, ndp) */
632 bzero(p->p_comm, sizeof(p->p_comm));
634 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
635 min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
636 else if (vn_commname(binvp, p->p_comm, sizeof(p->p_comm)) != 0)
637 bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
638 bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
640 sched_clear_tdname(td);
644 * mark as execed, wakeup the process that vforked (if any) and tell
645 * it that it now has its own resources back
648 if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0)
649 p->p_flag2 &= ~P2_NOTRACE;
650 if (p->p_flag & P_PPWAIT) {
651 p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
652 cv_broadcast(&p->p_pwait);
656 * Implement image setuid/setgid.
658 * Don't honor setuid/setgid if the filesystem prohibits it or if
659 * the process is being traced.
661 * We disable setuid/setgid/etc in compatibility mode on the basis
662 * that most setugid applications are not written with that
663 * environment in mind, and will therefore almost certainly operate
664 * incorrectly. In principle there's no reason that setugid
665 * applications might not be useful in capability mode, so we may want
666 * to reconsider this conservative design choice in the future.
668 * XXXMAC: For the time being, use NOSUID to also prohibit
669 * transitions on the file system.
671 credential_changing = 0;
672 credential_changing |= (attr.va_mode & S_ISUID) && oldcred->cr_uid !=
674 credential_changing |= (attr.va_mode & S_ISGID) && oldcred->cr_gid !=
677 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
678 interpvplabel, imgp);
679 credential_changing |= will_transition;
682 if (credential_changing &&
683 #ifdef CAPABILITY_MODE
684 ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
686 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
687 (p->p_flag & P_TRACED) == 0) {
689 * Turn off syscall tracing for set-id programs, except for
690 * root. Record any set-id flags first to make sure that
691 * we do not regain any tracing during a possible block.
696 if (p->p_tracecred != NULL &&
697 priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED, 0))
698 ktrprocexec(p, &tracecred, &tracevp);
701 * Close any file descriptors 0..2 that reference procfs,
702 * then make sure file descriptors 0..2 are in use.
704 * Both fdsetugidsafety() and fdcheckstd() may call functions
705 * taking sleepable locks, so temporarily drop our locks.
708 VOP_UNLOCK(imgp->vp, 0);
710 error = fdcheckstd(td);
713 newcred = crdup(oldcred);
714 euip = uifind(attr.va_uid);
715 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
718 * Set the new credentials.
720 if (attr.va_mode & S_ISUID)
721 change_euid(newcred, euip);
722 if (attr.va_mode & S_ISGID)
723 change_egid(newcred, attr.va_gid);
725 if (will_transition) {
726 mac_vnode_execve_transition(oldcred, newcred, imgp->vp,
727 interpvplabel, imgp);
731 * Implement correct POSIX saved-id behavior.
733 * XXXMAC: Note that the current logic will save the
734 * uid and gid if a MAC domain transition occurs, even
735 * though maybe it shouldn't.
737 change_svuid(newcred, newcred->cr_uid);
738 change_svgid(newcred, newcred->cr_gid);
739 proc_set_cred(p, newcred);
741 if (oldcred->cr_uid == oldcred->cr_ruid &&
742 oldcred->cr_gid == oldcred->cr_rgid)
743 p->p_flag &= ~P_SUGID;
745 * Implement correct POSIX saved-id behavior.
747 * XXX: It's not clear that the existing behavior is
748 * POSIX-compliant. A number of sources indicate that the
749 * saved uid/gid should only be updated if the new ruid is
750 * not equal to the old ruid, or the new euid is not equal
751 * to the old euid and the new euid is not equal to the old
752 * ruid. The FreeBSD code always updates the saved uid/gid.
753 * Also, this code uses the new (replaced) euid and egid as
754 * the source, which may or may not be the right ones to use.
756 if (oldcred->cr_svuid != oldcred->cr_uid ||
757 oldcred->cr_svgid != oldcred->cr_gid) {
759 VOP_UNLOCK(imgp->vp, 0);
760 newcred = crdup(oldcred);
761 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
763 change_svuid(newcred, newcred->cr_uid);
764 change_svgid(newcred, newcred->cr_gid);
765 proc_set_cred(p, newcred);
770 * Store the vp for use in procfs. This vnode was referenced prior
771 * to locking the proc lock.
773 textvp = p->p_textvp;
778 * Tell the DTrace fasttrap provider about the exec if it
779 * has declared an interest.
781 if (dtrace_fasttrap_exec)
782 dtrace_fasttrap_exec(p);
786 * Notify others that we exec'd, and clear the P_INEXEC flag
787 * as we're now a bona fide freshly-execed process.
789 KNOTE_LOCKED(&p->p_klist, NOTE_EXEC);
790 p->p_flag &= ~P_INEXEC;
792 /* clear "fork but no exec" flag, as we _are_ execing */
793 p->p_acflag &= ~AFORK;
796 * Free any previous argument cache and replace it with
797 * the new argument cache, if any.
805 * Check if system-wide sampling is in effect or if the
806 * current process is using PMCs. If so, do exec() time
807 * processing. This processing needs to happen AFTER the
808 * P_INEXEC flag is cleared.
810 * The proc lock needs to be released before taking the PMC
813 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
815 VOP_UNLOCK(imgp->vp, 0);
816 pe.pm_credentialschanged = credential_changing;
817 pe.pm_entryaddr = imgp->entry_addr;
819 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
820 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
823 #else /* !HWPMC_HOOKS */
827 /* Set values passed into the program in registers. */
828 if (p->p_sysent->sv_setregs)
829 (*p->p_sysent->sv_setregs)(td, imgp,
830 (u_long)(uintptr_t)stack_base);
832 exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base);
834 vfs_mark_atime(imgp->vp, td->td_ucred);
836 SDT_PROBE(proc, kernel, , exec__success, args->fname, 0, 0, 0, 0);
838 VOP_UNLOCK(imgp->vp, 0);
841 * Free any resources malloc'd earlier that we didn't use.
849 * Handle deferred decrement of ref counts.
858 if (tracecred != NULL)
861 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
864 if (oldsigacts != NULL)
865 sigacts_free(oldsigacts);
870 * free various allocated resources
872 if (imgp->firstpage != NULL)
873 exec_unmap_first_page(imgp);
875 if (imgp->vp != NULL) {
877 NDFREE(&nd, NDF_ONLY_PNBUF);
879 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
883 if (imgp->object != NULL)
884 vm_object_deallocate(imgp->object);
886 free(imgp->freepath, M_TEMP);
890 td->td_dbgflags |= TDB_EXEC;
894 * Stop the process here if its stop event mask has
895 * the S_EXEC bit set.
897 STOPEVENT(p, S_EXEC, 0);
902 /* we're done here, clear P_INEXEC */
904 p->p_flag &= ~P_INEXEC;
907 SDT_PROBE(proc, kernel, , exec__failure, error, 0, 0, 0, 0);
911 mac_execve_exit(imgp);
912 mac_execve_interpreter_exit(interpvplabel);
914 exec_free_args(args);
916 if (error && imgp->vmspace_destroyed) {
917 /* sorry, no more process anymore. exit gracefully */
918 exit1(td, W_EXITCODE(0, SIGABRT));
931 exec_map_first_page(imgp)
932 struct image_params *imgp;
936 vm_page_t ma[VM_INITIAL_PAGEIN];
939 if (imgp->firstpage != NULL)
940 exec_unmap_first_page(imgp);
942 object = imgp->vp->v_object;
945 VM_OBJECT_WLOCK(object);
946 #if VM_NRESERVLEVEL > 0
947 vm_object_color(object, 0);
949 ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL);
950 if (ma[0]->valid != VM_PAGE_BITS_ALL) {
951 initial_pagein = VM_INITIAL_PAGEIN;
952 if (initial_pagein > object->size)
953 initial_pagein = object->size;
954 for (i = 1; i < initial_pagein; i++) {
955 if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) {
958 if (vm_page_tryxbusy(ma[i]))
961 ma[i] = vm_page_alloc(object, i,
962 VM_ALLOC_NORMAL | VM_ALLOC_IFNOTCACHED);
968 rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
969 ma[0] = vm_page_lookup(object, 0);
970 if ((rv != VM_PAGER_OK) || (ma[0] == NULL)) {
974 vm_page_unlock(ma[0]);
976 VM_OBJECT_WUNLOCK(object);
980 vm_page_xunbusy(ma[0]);
983 vm_page_activate(ma[0]);
984 vm_page_unlock(ma[0]);
985 VM_OBJECT_WUNLOCK(object);
987 imgp->firstpage = sf_buf_alloc(ma[0], 0);
988 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
994 exec_unmap_first_page(imgp)
995 struct image_params *imgp;
999 if (imgp->firstpage != NULL) {
1000 m = sf_buf_page(imgp->firstpage);
1001 sf_buf_free(imgp->firstpage);
1002 imgp->firstpage = NULL;
1010 * Destroy old address space, and allocate a new stack
1011 * The new stack is only SGROWSIZ large because it is grown
1012 * automatically in trap.c.
1015 exec_new_vmspace(imgp, sv)
1016 struct image_params *imgp;
1017 struct sysentvec *sv;
1020 struct proc *p = imgp->proc;
1021 struct vmspace *vmspace = p->p_vmspace;
1023 struct rlimit rlim_stack;
1024 vm_offset_t sv_minuser, stack_addr;
1028 imgp->vmspace_destroyed = 1;
1031 /* May be called with Giant held */
1032 EVENTHANDLER_INVOKE(process_exec, p, imgp);
1035 * Blow away entire process VM, if address space not shared,
1036 * otherwise, create a new VM space so that other threads are
1039 map = &vmspace->vm_map;
1041 sv_minuser = sv->sv_minuser;
1043 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1044 if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser &&
1045 vm_map_max(map) == sv->sv_maxuser) {
1047 pmap_remove_pages(vmspace_pmap(vmspace));
1048 vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1050 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1053 vmspace = p->p_vmspace;
1054 map = &vmspace->vm_map;
1057 /* Map a shared page */
1058 obj = sv->sv_shared_page_obj;
1060 vm_object_reference(obj);
1061 error = vm_map_fixed(map, obj, 0,
1062 sv->sv_shared_page_base, sv->sv_shared_page_len,
1063 VM_PROT_READ | VM_PROT_EXECUTE,
1064 VM_PROT_READ | VM_PROT_EXECUTE,
1065 MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1067 vm_object_deallocate(obj);
1072 /* Allocate a new stack */
1073 if (imgp->stack_sz != 0) {
1074 ssiz = trunc_page(imgp->stack_sz);
1076 lim_rlimit(p, RLIMIT_STACK, &rlim_stack);
1078 if (ssiz > rlim_stack.rlim_max)
1079 ssiz = rlim_stack.rlim_max;
1080 if (ssiz > rlim_stack.rlim_cur) {
1081 rlim_stack.rlim_cur = ssiz;
1082 kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack);
1084 } else if (sv->sv_maxssiz != NULL) {
1085 ssiz = *sv->sv_maxssiz;
1089 stack_addr = sv->sv_usrstack - ssiz;
1090 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1091 obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1093 VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1098 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they
1099 * are still used to enforce the stack rlimit on the process stack.
1101 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1102 vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - ssiz;
1108 * Copy out argument and environment strings from the old process address
1109 * space into the temporary string buffer.
1112 exec_copyin_args(struct image_args *args, char *fname,
1113 enum uio_seg segflg, char **argv, char **envv)
1119 bzero(args, sizeof(*args));
1124 * Allocate demand-paged memory for the file name, argument, and
1125 * environment strings.
1127 error = exec_alloc_args(args);
1132 * Copy the file name.
1134 if (fname != NULL) {
1135 args->fname = args->buf;
1136 error = (segflg == UIO_SYSSPACE) ?
1137 copystr(fname, args->fname, PATH_MAX, &length) :
1138 copyinstr(fname, args->fname, PATH_MAX, &length);
1144 args->begin_argv = args->buf + length;
1145 args->endp = args->begin_argv;
1146 args->stringspace = ARG_MAX;
1149 * extract arguments first
1152 error = fueword(argv++, &argp);
1159 error = copyinstr((void *)(uintptr_t)argp, args->endp,
1160 args->stringspace, &length);
1162 if (error == ENAMETOOLONG)
1166 args->stringspace -= length;
1167 args->endp += length;
1171 args->begin_envv = args->endp;
1174 * extract environment strings
1178 error = fueword(envv++, &envp);
1185 error = copyinstr((void *)(uintptr_t)envp,
1186 args->endp, args->stringspace, &length);
1188 if (error == ENAMETOOLONG)
1192 args->stringspace -= length;
1193 args->endp += length;
1201 exec_free_args(args);
1206 * Allocate temporary demand-paged, zero-filled memory for the file name,
1207 * argument, and environment strings. Returns zero if the allocation succeeds
1208 * and ENOMEM otherwise.
1211 exec_alloc_args(struct image_args *args)
1214 args->buf = (char *)kmap_alloc_wait(exec_map, PATH_MAX + ARG_MAX);
1215 return (args->buf != NULL ? 0 : ENOMEM);
1219 exec_free_args(struct image_args *args)
1222 if (args->buf != NULL) {
1223 kmap_free_wakeup(exec_map, (vm_offset_t)args->buf,
1224 PATH_MAX + ARG_MAX);
1227 if (args->fname_buf != NULL) {
1228 free(args->fname_buf, M_TEMP);
1229 args->fname_buf = NULL;
1234 * Copy strings out to the new process address space, constructing new arg
1235 * and env vector tables. Return a pointer to the base so that it can be used
1236 * as the initial stack pointer.
1239 exec_copyout_strings(imgp)
1240 struct image_params *imgp;
1246 register_t *stack_base;
1247 struct ps_strings *arginfo;
1249 size_t execpath_len;
1250 int szsigcode, szps;
1251 char canary[sizeof(long) * 8];
1253 szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1255 * Calculate string base and vector table pointers.
1256 * Also deal with signal trampoline code for this exec type.
1258 if (imgp->execpath != NULL && imgp->auxargs != NULL)
1259 execpath_len = strlen(imgp->execpath) + 1;
1264 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1265 if (p->p_sysent->sv_sigcode_base == 0) {
1266 if (p->p_sysent->sv_szsigcode != NULL)
1267 szsigcode = *(p->p_sysent->sv_szsigcode);
1269 destp = (uintptr_t)arginfo;
1274 if (szsigcode != 0) {
1276 destp = rounddown2(destp, sizeof(void *));
1277 copyout(p->p_sysent->sv_sigcode, (void *)destp, szsigcode);
1281 * Copy the image path for the rtld.
1283 if (execpath_len != 0) {
1284 destp -= execpath_len;
1285 imgp->execpathp = destp;
1286 copyout(imgp->execpath, (void *)destp, execpath_len);
1290 * Prepare the canary for SSP.
1292 arc4rand(canary, sizeof(canary), 0);
1293 destp -= sizeof(canary);
1294 imgp->canary = destp;
1295 copyout(canary, (void *)destp, sizeof(canary));
1296 imgp->canarylen = sizeof(canary);
1299 * Prepare the pagesizes array.
1302 destp = rounddown2(destp, sizeof(void *));
1303 imgp->pagesizes = destp;
1304 copyout(pagesizes, (void *)destp, szps);
1305 imgp->pagesizeslen = szps;
1307 destp -= ARG_MAX - imgp->args->stringspace;
1308 destp = rounddown2(destp, sizeof(void *));
1311 * If we have a valid auxargs ptr, prepare some room
1314 if (imgp->auxargs) {
1316 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1317 * lower compatibility.
1319 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1322 * The '+ 2' is for the null pointers at the end of each of
1323 * the arg and env vector sets,and imgp->auxarg_size is room
1324 * for argument of Runtime loader.
1326 vectp = (char **)(destp - (imgp->args->argc +
1327 imgp->args->envc + 2 + imgp->auxarg_size)
1331 * The '+ 2' is for the null pointers at the end of each of
1332 * the arg and env vector sets
1334 vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc
1335 + 2) * sizeof(char *));
1339 * vectp also becomes our initial stack base
1341 stack_base = (register_t *)vectp;
1343 stringp = imgp->args->begin_argv;
1344 argc = imgp->args->argc;
1345 envc = imgp->args->envc;
1348 * Copy out strings - arguments and environment.
1350 copyout(stringp, (void *)destp, ARG_MAX - imgp->args->stringspace);
1353 * Fill in "ps_strings" struct for ps, w, etc.
1355 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1356 suword32(&arginfo->ps_nargvstr, argc);
1359 * Fill in argument portion of vector table.
1361 for (; argc > 0; --argc) {
1362 suword(vectp++, (long)(intptr_t)destp);
1363 while (*stringp++ != 0)
1368 /* a null vector table pointer separates the argp's from the envp's */
1371 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1372 suword32(&arginfo->ps_nenvstr, envc);
1375 * Fill in environment portion of vector table.
1377 for (; envc > 0; --envc) {
1378 suword(vectp++, (long)(intptr_t)destp);
1379 while (*stringp++ != 0)
1384 /* end of vector table is a null pointer */
1387 return (stack_base);
1391 * Check permissions of file to execute.
1392 * Called with imgp->vp locked.
1393 * Return 0 for success or error code on failure.
1396 exec_check_permissions(imgp)
1397 struct image_params *imgp;
1399 struct vnode *vp = imgp->vp;
1400 struct vattr *attr = imgp->attr;
1402 int error, writecount;
1406 /* Get file attributes */
1407 error = VOP_GETATTR(vp, attr, td->td_ucred);
1412 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1418 * 1) Check if file execution is disabled for the filesystem that
1419 * this file resides on.
1420 * 2) Ensure that at least one execute bit is on. Otherwise, a
1421 * privileged user will always succeed, and we don't want this
1422 * to happen unless the file really is executable.
1423 * 3) Ensure that the file is a regular file.
1425 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1426 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1427 (attr->va_type != VREG))
1431 * Zero length files can't be exec'd
1433 if (attr->va_size == 0)
1437 * Check for execute permission to file based on current credentials.
1439 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1444 * Check number of open-for-writes on the file and deny execution
1447 error = VOP_GET_WRITECOUNT(vp, &writecount);
1450 if (writecount != 0)
1454 * Call filesystem specific open routine (which does nothing in the
1457 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1464 * Exec handler registration
1467 exec_register(execsw_arg)
1468 const struct execsw *execsw_arg;
1470 const struct execsw **es, **xs, **newexecsw;
1471 int count = 2; /* New slot and trailing NULL */
1474 for (es = execsw; *es; es++)
1476 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1477 if (newexecsw == NULL)
1481 for (es = execsw; *es; es++)
1486 free(execsw, M_TEMP);
1492 exec_unregister(execsw_arg)
1493 const struct execsw *execsw_arg;
1495 const struct execsw **es, **xs, **newexecsw;
1499 panic("unregister with no handlers left?\n");
1501 for (es = execsw; *es; es++) {
1502 if (*es == execsw_arg)
1507 for (es = execsw; *es; es++)
1508 if (*es != execsw_arg)
1510 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1511 if (newexecsw == NULL)
1514 for (es = execsw; *es; es++)
1515 if (*es != execsw_arg)
1519 free(execsw, M_TEMP);