2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (c) 1993, David Greenman
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
32 #include "opt_capsicum.h"
33 #include "opt_hwpmc_hooks.h"
34 #include "opt_ktrace.h"
37 #include <sys/param.h>
38 #include <sys/systm.h>
40 #include <sys/capsicum.h>
41 #include <sys/eventhandler.h>
43 #include <sys/fcntl.h>
44 #include <sys/filedesc.h>
45 #include <sys/imgact.h>
46 #include <sys/imgact_elf.h>
47 #include <sys/kernel.h>
49 #include <sys/malloc.h>
51 #include <sys/mount.h>
52 #include <sys/mutex.h>
53 #include <sys/namei.h>
54 #include <sys/pioctl.h>
57 #include <sys/ptrace.h>
58 #include <sys/resourcevar.h>
59 #include <sys/rwlock.h>
60 #include <sys/sched.h>
62 #include <sys/sf_buf.h>
64 #include <sys/signalvar.h>
67 #include <sys/syscallsubr.h>
68 #include <sys/sysctl.h>
69 #include <sys/sysent.h>
70 #include <sys/sysproto.h>
71 #include <sys/vnode.h>
74 #include <sys/ktrace.h>
78 #include <vm/vm_param.h>
80 #include <vm/vm_page.h>
81 #include <vm/vm_map.h>
82 #include <vm/vm_kern.h>
83 #include <vm/vm_extern.h>
84 #include <vm/vm_object.h>
85 #include <vm/vm_pager.h>
88 #include <sys/pmckern.h>
91 #include <machine/reg.h>
93 #include <security/audit/audit.h>
94 #include <security/mac/mac_framework.h>
97 #include <sys/dtrace_bsd.h>
98 dtrace_execexit_func_t dtrace_fasttrap_exec;
101 SDT_PROVIDER_DECLARE(proc);
102 SDT_PROBE_DEFINE1(proc, , , exec, "char *");
103 SDT_PROBE_DEFINE1(proc, , , exec__failure, "int");
104 SDT_PROBE_DEFINE1(proc, , , exec__success, "char *");
106 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
108 int coredump_pack_fileinfo = 1;
109 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_fileinfo, CTLFLAG_RWTUN,
110 &coredump_pack_fileinfo, 0,
111 "Enable file path packing in 'procstat -f' coredump notes");
113 int coredump_pack_vmmapinfo = 1;
114 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_vmmapinfo, CTLFLAG_RWTUN,
115 &coredump_pack_vmmapinfo, 0,
116 "Enable file path packing in 'procstat -v' coredump notes");
118 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
119 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
120 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
121 static int do_execve(struct thread *td, struct image_args *args,
124 /* XXX This should be vm_size_t. */
125 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD|
126 CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_ps_strings, "LU", "");
128 /* XXX This should be vm_size_t. */
129 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD|
130 CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_usrstack, "LU", "");
132 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD|CTLFLAG_MPSAFE,
133 NULL, 0, sysctl_kern_stackprot, "I", "");
135 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
136 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
137 &ps_arg_cache_limit, 0, "");
139 static int disallow_high_osrel;
140 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW,
141 &disallow_high_osrel, 0,
142 "Disallow execution of binaries built for higher version of the world");
144 static int map_at_zero = 0;
145 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RWTUN, &map_at_zero, 0,
146 "Permit processes to map an object at virtual address 0.");
149 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
156 if (req->flags & SCTL_MASK32) {
158 val = (unsigned int)p->p_sysent->sv_psstrings;
159 error = SYSCTL_OUT(req, &val, sizeof(val));
162 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
163 sizeof(p->p_sysent->sv_psstrings));
168 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
175 if (req->flags & SCTL_MASK32) {
177 val = (unsigned int)p->p_sysent->sv_usrstack;
178 error = SYSCTL_OUT(req, &val, sizeof(val));
181 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
182 sizeof(p->p_sysent->sv_usrstack));
187 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
192 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
193 sizeof(p->p_sysent->sv_stackprot)));
197 * Each of the items is a pointer to a `const struct execsw', hence the
198 * double pointer here.
200 static const struct execsw **execsw;
202 #ifndef _SYS_SYSPROTO_H_
211 sys_execve(struct thread *td, struct execve_args *uap)
213 struct image_args args;
214 struct vmspace *oldvmspace;
217 error = pre_execve(td, &oldvmspace);
220 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
221 uap->argv, uap->envv);
223 error = kern_execve(td, &args, NULL);
224 post_execve(td, error, oldvmspace);
228 #ifndef _SYS_SYSPROTO_H_
229 struct fexecve_args {
236 sys_fexecve(struct thread *td, struct fexecve_args *uap)
238 struct image_args args;
239 struct vmspace *oldvmspace;
242 error = pre_execve(td, &oldvmspace);
245 error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
246 uap->argv, uap->envv);
249 error = kern_execve(td, &args, NULL);
251 post_execve(td, error, oldvmspace);
255 #ifndef _SYS_SYSPROTO_H_
256 struct __mac_execve_args {
265 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap)
268 struct image_args args;
269 struct vmspace *oldvmspace;
272 error = pre_execve(td, &oldvmspace);
275 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
276 uap->argv, uap->envv);
278 error = kern_execve(td, &args, uap->mac_p);
279 post_execve(td, error, oldvmspace);
287 pre_execve(struct thread *td, struct vmspace **oldvmspace)
292 KASSERT(td == curthread, ("non-current thread %p", td));
295 if ((p->p_flag & P_HADTHREADS) != 0) {
297 if (thread_single(p, SINGLE_BOUNDARY) != 0)
301 KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0,
303 *oldvmspace = p->p_vmspace;
308 post_execve(struct thread *td, int error, struct vmspace *oldvmspace)
312 KASSERT(td == curthread, ("non-current thread %p", td));
314 if ((p->p_flag & P_HADTHREADS) != 0) {
317 * If success, we upgrade to SINGLE_EXIT state to
318 * force other threads to suicide.
320 if (error == EJUSTRETURN)
321 thread_single(p, SINGLE_EXIT);
323 thread_single_end(p, SINGLE_BOUNDARY);
326 if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
327 KASSERT(p->p_vmspace != oldvmspace,
328 ("oldvmspace still used"));
329 vmspace_free(oldvmspace);
330 td->td_pflags &= ~TDP_EXECVMSPC;
335 * XXX: kern_execve has the astonishing property of not always returning to
336 * the caller. If sufficiently bad things happen during the call to
337 * do_execve(), it can end up calling exit1(); as a result, callers must
338 * avoid doing anything which they might need to undo (e.g., allocating
342 kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p)
345 AUDIT_ARG_ARGV(args->begin_argv, args->argc,
346 exec_args_get_begin_envv(args) - args->begin_argv);
347 AUDIT_ARG_ENVV(exec_args_get_begin_envv(args), args->envc,
348 args->endp - exec_args_get_begin_envv(args));
349 return (do_execve(td, args, mac_p));
353 * In-kernel implementation of execve(). All arguments are assumed to be
354 * userspace pointers from the passed thread.
357 do_execve(struct thread *td, struct image_args *args, struct mac *mac_p)
359 struct proc *p = td->td_proc;
361 struct ucred *oldcred;
362 struct uidinfo *euip = NULL;
363 register_t *stack_base;
365 struct image_params image_params, *imgp;
367 int (*img_first)(struct image_params *);
368 struct pargs *oldargs = NULL, *newargs = NULL;
369 struct sigacts *oldsigacts = NULL, *newsigacts = NULL;
371 struct vnode *tracevp = NULL;
372 struct ucred *tracecred = NULL;
374 struct vnode *oldtextvp = NULL, *newtextvp;
375 int credential_changing;
377 struct label *interpvplabel = NULL;
381 struct pmckern_procexec pe;
383 static const char fexecv_proc_title[] = "(fexecv)";
385 imgp = &image_params;
388 * Lock the process and set the P_INEXEC flag to indicate that
389 * it should be left alone until we're done here. This is
390 * necessary to avoid race conditions - e.g. in ptrace() -
391 * that might allow a local user to illicitly obtain elevated
395 KASSERT((p->p_flag & P_INEXEC) == 0,
396 ("%s(): process already has P_INEXEC flag", __func__));
397 p->p_flag |= P_INEXEC;
401 * Initialize part of the common data
403 bzero(imgp, sizeof(*imgp));
407 oldcred = p->p_ucred;
410 error = mac_execve_enter(imgp, mac_p);
416 * Translate the file name. namei() returns a vnode pointer
417 * in ni_vp among other things.
419 * XXXAUDIT: It would be desirable to also audit the name of the
420 * interpreter if this is an interpreted binary.
422 if (args->fname != NULL) {
423 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW |
424 SAVENAME | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
427 SDT_PROBE1(proc, , , exec, args->fname);
430 if (args->fname != NULL) {
431 #ifdef CAPABILITY_MODE
433 * While capability mode can't reach this point via direct
434 * path arguments to execve(), we also don't allow
435 * interpreters to be used in capability mode (for now).
436 * Catch indirect lookups and return a permissions error.
438 if (IN_CAPABILITY_MODE(td)) {
447 newtextvp = nd.ni_vp;
448 imgp->vp = newtextvp;
450 AUDIT_ARG_FD(args->fd);
452 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
454 error = fgetvp_exec(td, args->fd, &cap_fexecve_rights, &newtextvp);
457 vn_lock(newtextvp, LK_SHARED | LK_RETRY);
458 AUDIT_ARG_VNODE1(newtextvp);
459 imgp->vp = newtextvp;
463 * Check file permissions. Also 'opens' file and sets its vnode to
466 error = exec_check_permissions(imgp);
468 goto exec_fail_dealloc;
470 imgp->object = imgp->vp->v_object;
471 if (imgp->object != NULL)
472 vm_object_reference(imgp->object);
474 error = exec_map_first_page(imgp);
476 goto exec_fail_dealloc;
478 imgp->proc->p_osrel = 0;
479 imgp->proc->p_fctl0 = 0;
482 * Implement image setuid/setgid.
484 * Determine new credentials before attempting image activators
485 * so that it can be used by process_exec handlers to determine
486 * credential/setid changes.
488 * Don't honor setuid/setgid if the filesystem prohibits it or if
489 * the process is being traced.
491 * We disable setuid/setgid/etc in capability mode on the basis
492 * that most setugid applications are not written with that
493 * environment in mind, and will therefore almost certainly operate
494 * incorrectly. In principle there's no reason that setugid
495 * applications might not be useful in capability mode, so we may want
496 * to reconsider this conservative design choice in the future.
498 * XXXMAC: For the time being, use NOSUID to also prohibit
499 * transitions on the file system.
501 credential_changing = 0;
502 credential_changing |= (attr.va_mode & S_ISUID) &&
503 oldcred->cr_uid != attr.va_uid;
504 credential_changing |= (attr.va_mode & S_ISGID) &&
505 oldcred->cr_gid != attr.va_gid;
507 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
508 interpvplabel, imgp);
509 credential_changing |= will_transition;
512 /* Don't inherit PROC_PDEATHSIG_CTL value if setuid/setgid. */
513 if (credential_changing)
514 imgp->proc->p_pdeathsig = 0;
516 if (credential_changing &&
517 #ifdef CAPABILITY_MODE
518 ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
520 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
521 (p->p_flag & P_TRACED) == 0) {
522 imgp->credential_setid = true;
523 VOP_UNLOCK(imgp->vp, 0);
524 imgp->newcred = crdup(oldcred);
525 if (attr.va_mode & S_ISUID) {
526 euip = uifind(attr.va_uid);
527 change_euid(imgp->newcred, euip);
529 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
530 if (attr.va_mode & S_ISGID)
531 change_egid(imgp->newcred, attr.va_gid);
533 * Implement correct POSIX saved-id behavior.
535 * XXXMAC: Note that the current logic will save the
536 * uid and gid if a MAC domain transition occurs, even
537 * though maybe it shouldn't.
539 change_svuid(imgp->newcred, imgp->newcred->cr_uid);
540 change_svgid(imgp->newcred, imgp->newcred->cr_gid);
543 * Implement correct POSIX saved-id behavior.
545 * XXX: It's not clear that the existing behavior is
546 * POSIX-compliant. A number of sources indicate that the
547 * saved uid/gid should only be updated if the new ruid is
548 * not equal to the old ruid, or the new euid is not equal
549 * to the old euid and the new euid is not equal to the old
550 * ruid. The FreeBSD code always updates the saved uid/gid.
551 * Also, this code uses the new (replaced) euid and egid as
552 * the source, which may or may not be the right ones to use.
554 if (oldcred->cr_svuid != oldcred->cr_uid ||
555 oldcred->cr_svgid != oldcred->cr_gid) {
556 VOP_UNLOCK(imgp->vp, 0);
557 imgp->newcred = crdup(oldcred);
558 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
559 change_svuid(imgp->newcred, imgp->newcred->cr_uid);
560 change_svgid(imgp->newcred, imgp->newcred->cr_gid);
563 /* The new credentials are installed into the process later. */
566 * Do the best to calculate the full path to the image file.
568 if (args->fname != NULL && args->fname[0] == '/')
569 imgp->execpath = args->fname;
571 VOP_UNLOCK(imgp->vp, 0);
572 if (vn_fullpath(td, imgp->vp, &imgp->execpath,
573 &imgp->freepath) != 0)
574 imgp->execpath = args->fname;
575 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
579 * If the current process has a special image activator it
580 * wants to try first, call it. For example, emulating shell
581 * scripts differently.
584 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
585 error = img_first(imgp);
588 * Loop through the list of image activators, calling each one.
589 * An activator returns -1 if there is no match, 0 on success,
590 * and an error otherwise.
592 for (i = 0; error == -1 && execsw[i]; ++i) {
593 if (execsw[i]->ex_imgact == NULL ||
594 execsw[i]->ex_imgact == img_first) {
597 error = (*execsw[i]->ex_imgact)(imgp);
603 goto exec_fail_dealloc;
607 * Special interpreter operation, cleanup and loop up to try to
608 * activate the interpreter.
610 if (imgp->interpreted) {
611 exec_unmap_first_page(imgp);
613 * The text reference needs to be removed for scripts.
614 * There is a short period before we determine that
615 * something is a script where text reference is active.
616 * The vnode lock is held over this entire period
617 * so nothing should illegitimately be blocked.
619 MPASS(imgp->textset);
620 VOP_UNSET_TEXT_CHECKED(newtextvp);
621 imgp->textset = false;
622 /* free name buffer and old vnode */
623 if (args->fname != NULL)
624 NDFREE(&nd, NDF_ONLY_PNBUF);
626 mac_execve_interpreter_enter(newtextvp, &interpvplabel);
629 VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td);
633 vm_object_deallocate(imgp->object);
635 imgp->credential_setid = false;
636 if (imgp->newcred != NULL) {
637 crfree(imgp->newcred);
638 imgp->newcred = NULL;
640 imgp->execpath = NULL;
641 free(imgp->freepath, M_TEMP);
642 imgp->freepath = NULL;
643 /* set new name to that of the interpreter */
644 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME,
645 UIO_SYSSPACE, imgp->interpreter_name, td);
646 args->fname = imgp->interpreter_name;
651 * NB: We unlock the vnode here because it is believed that none
652 * of the sv_copyout_strings/sv_fixup operations require the vnode.
654 VOP_UNLOCK(imgp->vp, 0);
656 if (disallow_high_osrel &&
657 P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
659 uprintf("Osrel %d for image %s too high\n", p->p_osrel,
660 imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
661 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
662 goto exec_fail_dealloc;
665 /* ABI enforces the use of Capsicum. Switch into capabilities mode. */
666 if (SV_PROC_FLAG(p, SV_CAPSICUM))
667 sys_cap_enter(td, NULL);
670 * Copy out strings (args and env) and initialize stack base.
672 stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
677 error = (*p->p_sysent->sv_fixup)(&stack_base, imgp);
679 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
680 goto exec_fail_dealloc;
683 if (args->fdp != NULL) {
684 /* Install a brand new file descriptor table. */
685 fdinstall_remapped(td, args->fdp);
689 * Keep on using the existing file descriptor table. For
690 * security and other reasons, the file descriptor table
691 * cannot be shared after an exec.
694 /* close files on exec */
699 * Malloc things before we need locks.
701 i = exec_args_get_begin_envv(imgp->args) - imgp->args->begin_argv;
702 /* Cache arguments if they fit inside our allowance */
703 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
704 newargs = pargs_alloc(i);
705 bcopy(imgp->args->begin_argv, newargs->ar_args, i);
709 * For security and other reasons, signal handlers cannot
710 * be shared after an exec. The new process gets a copy of the old
711 * handlers. In execsigs(), the new process will have its signals
714 if (sigacts_shared(p->p_sigacts)) {
715 oldsigacts = p->p_sigacts;
716 newsigacts = sigacts_alloc();
717 sigacts_copy(newsigacts, oldsigacts);
720 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
724 p->p_sigacts = newsigacts;
728 /* reset caught signals */
731 /* name this process - nameiexec(p, ndp) */
732 bzero(p->p_comm, sizeof(p->p_comm));
734 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
735 min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
736 else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0)
737 bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
738 bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
740 sched_clear_tdname(td);
744 * mark as execed, wakeup the process that vforked (if any) and tell
745 * it that it now has its own resources back
748 if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0)
749 p->p_flag2 &= ~P2_NOTRACE;
750 if ((p->p_flag2 & P2_STKGAP_DISABLE_EXEC) == 0)
751 p->p_flag2 &= ~P2_STKGAP_DISABLE;
752 if (p->p_flag & P_PPWAIT) {
753 p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
754 cv_broadcast(&p->p_pwait);
755 /* STOPs are no longer ignored, arrange for AST */
760 * Implement image setuid/setgid installation.
762 if (imgp->credential_setid) {
764 * Turn off syscall tracing for set-id programs, except for
765 * root. Record any set-id flags first to make sure that
766 * we do not regain any tracing during a possible block.
771 if (p->p_tracecred != NULL &&
772 priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED))
773 ktrprocexec(p, &tracecred, &tracevp);
776 * Close any file descriptors 0..2 that reference procfs,
777 * then make sure file descriptors 0..2 are in use.
779 * Both fdsetugidsafety() and fdcheckstd() may call functions
780 * taking sleepable locks, so temporarily drop our locks.
783 VOP_UNLOCK(imgp->vp, 0);
785 error = fdcheckstd(td);
786 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
788 goto exec_fail_dealloc;
791 if (will_transition) {
792 mac_vnode_execve_transition(oldcred, imgp->newcred,
793 imgp->vp, interpvplabel, imgp);
797 if (oldcred->cr_uid == oldcred->cr_ruid &&
798 oldcred->cr_gid == oldcred->cr_rgid)
799 p->p_flag &= ~P_SUGID;
802 * Set the new credentials.
804 if (imgp->newcred != NULL) {
805 proc_set_cred(p, imgp->newcred);
811 * Store the vp for use in procfs. This vnode was referenced by namei
814 oldtextvp = p->p_textvp;
815 p->p_textvp = newtextvp;
819 * Tell the DTrace fasttrap provider about the exec if it
820 * has declared an interest.
822 if (dtrace_fasttrap_exec)
823 dtrace_fasttrap_exec(p);
827 * Notify others that we exec'd, and clear the P_INEXEC flag
828 * as we're now a bona fide freshly-execed process.
830 KNOTE_LOCKED(p->p_klist, NOTE_EXEC);
831 p->p_flag &= ~P_INEXEC;
833 /* clear "fork but no exec" flag, as we _are_ execing */
834 p->p_acflag &= ~AFORK;
837 * Free any previous argument cache and replace it with
838 * the new argument cache, if any.
848 * Check if system-wide sampling is in effect or if the
849 * current process is using PMCs. If so, do exec() time
850 * processing. This processing needs to happen AFTER the
851 * P_INEXEC flag is cleared.
853 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
854 VOP_UNLOCK(imgp->vp, 0);
855 pe.pm_credentialschanged = credential_changing;
856 pe.pm_entryaddr = imgp->entry_addr;
858 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
859 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
863 /* Set values passed into the program in registers. */
864 (*p->p_sysent->sv_setregs)(td, imgp, (u_long)(uintptr_t)stack_base);
866 vfs_mark_atime(imgp->vp, td->td_ucred);
868 SDT_PROBE1(proc, , , exec__success, args->fname);
871 if (imgp->firstpage != NULL)
872 exec_unmap_first_page(imgp);
874 if (imgp->vp != NULL) {
876 NDFREE(&nd, NDF_ONLY_PNBUF);
878 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
880 VOP_UNSET_TEXT_CHECKED(imgp->vp);
884 VOP_UNLOCK(imgp->vp, 0);
887 if (imgp->object != NULL)
888 vm_object_deallocate(imgp->object);
890 free(imgp->freepath, M_TEMP);
893 if (p->p_ptevents & PTRACE_EXEC) {
895 if (p->p_ptevents & PTRACE_EXEC)
896 td->td_dbgflags |= TDB_EXEC;
901 * Stop the process here if its stop event mask has
902 * the S_EXEC bit set.
904 STOPEVENT(p, S_EXEC, 0);
907 /* we're done here, clear P_INEXEC */
909 p->p_flag &= ~P_INEXEC;
912 SDT_PROBE1(proc, , , exec__failure, error);
915 if (imgp->newcred != NULL && oldcred != NULL)
916 crfree(imgp->newcred);
919 mac_execve_exit(imgp);
920 mac_execve_interpreter_exit(interpvplabel);
922 exec_free_args(args);
925 * Handle deferred decrement of ref counts.
927 if (oldtextvp != NULL)
932 if (tracecred != NULL)
937 if (oldsigacts != NULL)
938 sigacts_free(oldsigacts);
942 if (error && imgp->vmspace_destroyed) {
943 /* sorry, no more process anymore. exit gracefully */
944 exit1(td, 0, SIGABRT);
954 * We don't want cpu_set_syscall_retval() to overwrite any of
955 * the register values put in place by exec_setregs().
956 * Implementations of cpu_set_syscall_retval() will leave
957 * registers unmodified when returning EJUSTRETURN.
959 return (error == 0 ? EJUSTRETURN : error);
963 exec_map_first_page(struct image_params *imgp)
965 int rv, i, after, initial_pagein;
966 vm_page_t ma[VM_INITIAL_PAGEIN];
969 if (imgp->firstpage != NULL)
970 exec_unmap_first_page(imgp);
972 object = imgp->vp->v_object;
975 VM_OBJECT_WLOCK(object);
976 #if VM_NRESERVLEVEL > 0
977 vm_object_color(object, 0);
979 ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY |
981 if (ma[0]->valid != VM_PAGE_BITS_ALL) {
982 vm_page_xbusy(ma[0]);
983 if (!vm_pager_has_page(object, 0, NULL, &after)) {
984 vm_page_unwire_noq(ma[0]);
986 VM_OBJECT_WUNLOCK(object);
989 initial_pagein = min(after, VM_INITIAL_PAGEIN);
990 KASSERT(initial_pagein <= object->size,
991 ("%s: initial_pagein %d object->size %ju",
992 __func__, initial_pagein, (uintmax_t )object->size));
993 for (i = 1; i < initial_pagein; i++) {
994 if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) {
997 if (!vm_page_tryxbusy(ma[i]))
1000 ma[i] = vm_page_alloc(object, i,
1007 rv = vm_pager_get_pages(object, ma, initial_pagein, NULL, NULL);
1008 if (rv != VM_PAGER_OK) {
1009 vm_page_unwire_noq(ma[0]);
1010 for (i = 0; i < initial_pagein; i++)
1011 vm_page_free(ma[i]);
1012 VM_OBJECT_WUNLOCK(object);
1015 vm_page_xunbusy(ma[0]);
1016 for (i = 1; i < initial_pagein; i++)
1017 vm_page_readahead_finish(ma[i]);
1019 VM_OBJECT_WUNLOCK(object);
1021 imgp->firstpage = sf_buf_alloc(ma[0], 0);
1022 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
1028 exec_unmap_first_page(struct image_params *imgp)
1032 if (imgp->firstpage != NULL) {
1033 m = sf_buf_page(imgp->firstpage);
1034 sf_buf_free(imgp->firstpage);
1035 imgp->firstpage = NULL;
1036 vm_page_unwire(m, PQ_ACTIVE);
1041 * Destroy old address space, and allocate a new stack.
1042 * The new stack is only sgrowsiz large because it is grown
1043 * automatically on a page fault.
1046 exec_new_vmspace(struct image_params *imgp, struct sysentvec *sv)
1049 struct proc *p = imgp->proc;
1050 struct vmspace *vmspace = p->p_vmspace;
1052 struct rlimit rlim_stack;
1053 vm_offset_t sv_minuser, stack_addr;
1057 imgp->vmspace_destroyed = 1;
1060 /* May be called with Giant held */
1061 EVENTHANDLER_DIRECT_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 &&
1075 cpu_exec_vmspace_reuse(p, map)) {
1077 pmap_remove_pages(vmspace_pmap(vmspace));
1078 vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1080 * An exec terminates mlockall(MCL_FUTURE), ASLR state
1081 * must be re-evaluated.
1084 vm_map_modflags(map, 0, MAP_WIREFUTURE | MAP_ASLR |
1088 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1091 vmspace = p->p_vmspace;
1092 map = &vmspace->vm_map;
1094 map->flags |= imgp->map_flags;
1096 /* Map a shared page */
1097 obj = sv->sv_shared_page_obj;
1099 vm_object_reference(obj);
1100 error = vm_map_fixed(map, obj, 0,
1101 sv->sv_shared_page_base, sv->sv_shared_page_len,
1102 VM_PROT_READ | VM_PROT_EXECUTE,
1103 VM_PROT_READ | VM_PROT_EXECUTE,
1104 MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1105 if (error != KERN_SUCCESS) {
1106 vm_object_deallocate(obj);
1107 return (vm_mmap_to_errno(error));
1111 /* Allocate a new stack */
1112 if (imgp->stack_sz != 0) {
1113 ssiz = trunc_page(imgp->stack_sz);
1115 lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack);
1117 if (ssiz > rlim_stack.rlim_max)
1118 ssiz = rlim_stack.rlim_max;
1119 if (ssiz > rlim_stack.rlim_cur) {
1120 rlim_stack.rlim_cur = ssiz;
1121 kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack);
1123 } else if (sv->sv_maxssiz != NULL) {
1124 ssiz = *sv->sv_maxssiz;
1128 imgp->eff_stack_sz = lim_cur(curthread, RLIMIT_STACK);
1129 if (ssiz < imgp->eff_stack_sz)
1130 imgp->eff_stack_sz = ssiz;
1131 stack_addr = sv->sv_usrstack - ssiz;
1132 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1133 obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1134 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1135 if (error != KERN_SUCCESS)
1136 return (vm_mmap_to_errno(error));
1139 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they
1140 * are still used to enforce the stack rlimit on the process stack.
1142 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1143 vmspace->vm_maxsaddr = (char *)stack_addr;
1149 * Copy out argument and environment strings from the old process address
1150 * space into the temporary string buffer.
1153 exec_copyin_args(struct image_args *args, const char *fname,
1154 enum uio_seg segflg, char **argv, char **envv)
1159 bzero(args, sizeof(*args));
1164 * Allocate demand-paged memory for the file name, argument, and
1165 * environment strings.
1167 error = exec_alloc_args(args);
1172 * Copy the file name.
1174 error = exec_args_add_fname(args, fname, segflg);
1179 * extract arguments first
1182 error = fueword(argv++, &arg);
1189 error = exec_args_add_arg(args, (char *)(uintptr_t)arg,
1196 * extract environment strings
1200 error = fueword(envv++, &env);
1207 error = exec_args_add_env(args,
1208 (char *)(uintptr_t)env, UIO_USERSPACE);
1217 exec_free_args(args);
1222 exec_copyin_data_fds(struct thread *td, struct image_args *args,
1223 const void *data, size_t datalen, const int *fds, size_t fdslen)
1225 struct filedesc *ofdp;
1230 memset(args, '\0', sizeof(*args));
1231 ofdp = td->td_proc->p_fd;
1232 if (datalen >= ARG_MAX || fdslen > ofdp->fd_lastfile + 1)
1234 error = exec_alloc_args(args);
1238 args->begin_argv = args->buf;
1239 args->stringspace = ARG_MAX;
1243 * Argument buffer has been provided. Copy it into the
1244 * kernel as a single string and add a terminating null
1247 error = copyin(data, args->begin_argv, datalen);
1250 args->begin_argv[datalen] = '\0';
1251 args->endp = args->begin_argv + datalen + 1;
1252 args->stringspace -= datalen + 1;
1255 * Traditional argument counting. Count the number of
1258 for (p = args->begin_argv; p < args->endp; ++p)
1262 /* No argument buffer provided. */
1263 args->endp = args->begin_argv;
1266 /* Create new file descriptor table. */
1267 kfds = malloc(fdslen * sizeof(int), M_TEMP, M_WAITOK);
1268 error = copyin(fds, kfds, fdslen * sizeof(int));
1273 error = fdcopy_remapped(ofdp, kfds, fdslen, &args->fdp);
1280 exec_free_args(args);
1284 struct exec_args_kva {
1287 SLIST_ENTRY(exec_args_kva) next;
1290 DPCPU_DEFINE_STATIC(struct exec_args_kva *, exec_args_kva);
1292 static SLIST_HEAD(, exec_args_kva) exec_args_kva_freelist;
1293 static struct mtx exec_args_kva_mtx;
1294 static u_int exec_args_gen;
1297 exec_prealloc_args_kva(void *arg __unused)
1299 struct exec_args_kva *argkva;
1302 SLIST_INIT(&exec_args_kva_freelist);
1303 mtx_init(&exec_args_kva_mtx, "exec args kva", NULL, MTX_DEF);
1304 for (i = 0; i < exec_map_entries; i++) {
1305 argkva = malloc(sizeof(*argkva), M_PARGS, M_WAITOK);
1306 argkva->addr = kmap_alloc_wait(exec_map, exec_map_entry_size);
1307 argkva->gen = exec_args_gen;
1308 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1311 SYSINIT(exec_args_kva, SI_SUB_EXEC, SI_ORDER_ANY, exec_prealloc_args_kva, NULL);
1314 exec_alloc_args_kva(void **cookie)
1316 struct exec_args_kva *argkva;
1318 argkva = (void *)atomic_readandclear_ptr(
1319 (uintptr_t *)DPCPU_PTR(exec_args_kva));
1320 if (argkva == NULL) {
1321 mtx_lock(&exec_args_kva_mtx);
1322 while ((argkva = SLIST_FIRST(&exec_args_kva_freelist)) == NULL)
1323 (void)mtx_sleep(&exec_args_kva_freelist,
1324 &exec_args_kva_mtx, 0, "execkva", 0);
1325 SLIST_REMOVE_HEAD(&exec_args_kva_freelist, next);
1326 mtx_unlock(&exec_args_kva_mtx);
1328 *(struct exec_args_kva **)cookie = argkva;
1329 return (argkva->addr);
1333 exec_release_args_kva(struct exec_args_kva *argkva, u_int gen)
1337 base = argkva->addr;
1338 if (argkva->gen != gen) {
1339 (void)vm_map_madvise(exec_map, base, base + exec_map_entry_size,
1343 if (!atomic_cmpset_ptr((uintptr_t *)DPCPU_PTR(exec_args_kva),
1344 (uintptr_t)NULL, (uintptr_t)argkva)) {
1345 mtx_lock(&exec_args_kva_mtx);
1346 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1347 wakeup_one(&exec_args_kva_freelist);
1348 mtx_unlock(&exec_args_kva_mtx);
1353 exec_free_args_kva(void *cookie)
1356 exec_release_args_kva(cookie, exec_args_gen);
1360 exec_args_kva_lowmem(void *arg __unused)
1362 SLIST_HEAD(, exec_args_kva) head;
1363 struct exec_args_kva *argkva;
1367 gen = atomic_fetchadd_int(&exec_args_gen, 1) + 1;
1370 * Force an madvise of each KVA range. Any currently allocated ranges
1371 * will have MADV_FREE applied once they are freed.
1374 mtx_lock(&exec_args_kva_mtx);
1375 SLIST_SWAP(&head, &exec_args_kva_freelist, exec_args_kva);
1376 mtx_unlock(&exec_args_kva_mtx);
1377 while ((argkva = SLIST_FIRST(&head)) != NULL) {
1378 SLIST_REMOVE_HEAD(&head, next);
1379 exec_release_args_kva(argkva, gen);
1383 argkva = (void *)atomic_readandclear_ptr(
1384 (uintptr_t *)DPCPU_ID_PTR(i, exec_args_kva));
1386 exec_release_args_kva(argkva, gen);
1389 EVENTHANDLER_DEFINE(vm_lowmem, exec_args_kva_lowmem, NULL,
1390 EVENTHANDLER_PRI_ANY);
1393 * Allocate temporary demand-paged, zero-filled memory for the file name,
1394 * argument, and environment strings.
1397 exec_alloc_args(struct image_args *args)
1400 args->buf = (char *)exec_alloc_args_kva(&args->bufkva);
1405 exec_free_args(struct image_args *args)
1408 if (args->buf != NULL) {
1409 exec_free_args_kva(args->bufkva);
1412 if (args->fname_buf != NULL) {
1413 free(args->fname_buf, M_TEMP);
1414 args->fname_buf = NULL;
1416 if (args->fdp != NULL)
1417 fdescfree_remapped(args->fdp);
1421 * A set to functions to fill struct image args.
1423 * NOTE: exec_args_add_fname() must be called (possibly with a NULL
1424 * fname) before the other functions. All exec_args_add_arg() calls must
1425 * be made before any exec_args_add_env() calls. exec_args_adjust_args()
1426 * may be called any time after exec_args_add_fname().
1428 * exec_args_add_fname() - install path to be executed
1429 * exec_args_add_arg() - append an argument string
1430 * exec_args_add_env() - append an env string
1431 * exec_args_adjust_args() - adjust location of the argument list to
1432 * allow new arguments to be prepended
1435 exec_args_add_fname(struct image_args *args, const char *fname,
1436 enum uio_seg segflg)
1441 KASSERT(args->fname == NULL, ("fname already appended"));
1442 KASSERT(args->endp == NULL, ("already appending to args"));
1444 if (fname != NULL) {
1445 args->fname = args->buf;
1446 error = segflg == UIO_SYSSPACE ?
1447 copystr(fname, args->fname, PATH_MAX, &length) :
1448 copyinstr(fname, args->fname, PATH_MAX, &length);
1450 return (error == ENAMETOOLONG ? E2BIG : error);
1454 /* Set up for _arg_*()/_env_*() */
1455 args->endp = args->buf + length;
1456 /* begin_argv must be set and kept updated */
1457 args->begin_argv = args->endp;
1458 KASSERT(exec_map_entry_size - length >= ARG_MAX,
1459 ("too little space remaining for arguments %zu < %zu",
1460 exec_map_entry_size - length, (size_t)ARG_MAX));
1461 args->stringspace = ARG_MAX;
1467 exec_args_add_str(struct image_args *args, const char *str,
1468 enum uio_seg segflg, int *countp)
1473 KASSERT(args->endp != NULL, ("endp not initialized"));
1474 KASSERT(args->begin_argv != NULL, ("begin_argp not initialized"));
1476 error = (segflg == UIO_SYSSPACE) ?
1477 copystr(str, args->endp, args->stringspace, &length) :
1478 copyinstr(str, args->endp, args->stringspace, &length);
1480 return (error == ENAMETOOLONG ? E2BIG : error);
1481 args->stringspace -= length;
1482 args->endp += length;
1489 exec_args_add_arg(struct image_args *args, const char *argp,
1490 enum uio_seg segflg)
1493 KASSERT(args->envc == 0, ("appending args after env"));
1495 return (exec_args_add_str(args, argp, segflg, &args->argc));
1499 exec_args_add_env(struct image_args *args, const char *envp,
1500 enum uio_seg segflg)
1503 if (args->envc == 0)
1504 args->begin_envv = args->endp;
1506 return (exec_args_add_str(args, envp, segflg, &args->envc));
1510 exec_args_adjust_args(struct image_args *args, size_t consume, ssize_t extend)
1514 KASSERT(args->endp != NULL, ("endp not initialized"));
1515 KASSERT(args->begin_argv != NULL, ("begin_argp not initialized"));
1517 offset = extend - consume;
1518 if (args->stringspace < offset)
1520 memmove(args->begin_argv + extend, args->begin_argv + consume,
1521 args->endp - args->begin_argv + consume);
1523 args->begin_envv += offset;
1524 args->endp += offset;
1525 args->stringspace -= offset;
1530 exec_args_get_begin_envv(struct image_args *args)
1533 KASSERT(args->endp != NULL, ("endp not initialized"));
1536 return (args->begin_envv);
1537 return (args->endp);
1541 * Copy strings out to the new process address space, constructing new arg
1542 * and env vector tables. Return a pointer to the base so that it can be used
1543 * as the initial stack pointer.
1546 exec_copyout_strings(struct image_params *imgp)
1552 register_t *stack_base;
1553 struct ps_strings *arginfo;
1555 size_t execpath_len;
1556 int szsigcode, szps;
1557 char canary[sizeof(long) * 8];
1559 szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1561 * Calculate string base and vector table pointers.
1562 * Also deal with signal trampoline code for this exec type.
1564 if (imgp->execpath != NULL && imgp->auxargs != NULL)
1565 execpath_len = strlen(imgp->execpath) + 1;
1570 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1571 if (p->p_sysent->sv_sigcode_base == 0) {
1572 if (p->p_sysent->sv_szsigcode != NULL)
1573 szsigcode = *(p->p_sysent->sv_szsigcode);
1575 destp = (uintptr_t)arginfo;
1580 if (szsigcode != 0) {
1582 destp = rounddown2(destp, sizeof(void *));
1583 copyout(p->p_sysent->sv_sigcode, (void *)destp, szsigcode);
1587 * Copy the image path for the rtld.
1589 if (execpath_len != 0) {
1590 destp -= execpath_len;
1591 destp = rounddown2(destp, sizeof(void *));
1592 imgp->execpathp = destp;
1593 copyout(imgp->execpath, (void *)destp, execpath_len);
1597 * Prepare the canary for SSP.
1599 arc4rand(canary, sizeof(canary), 0);
1600 destp -= sizeof(canary);
1601 imgp->canary = destp;
1602 copyout(canary, (void *)destp, sizeof(canary));
1603 imgp->canarylen = sizeof(canary);
1606 * Prepare the pagesizes array.
1609 destp = rounddown2(destp, sizeof(void *));
1610 imgp->pagesizes = destp;
1611 copyout(pagesizes, (void *)destp, szps);
1612 imgp->pagesizeslen = szps;
1614 destp -= ARG_MAX - imgp->args->stringspace;
1615 destp = rounddown2(destp, sizeof(void *));
1617 vectp = (char **)destp;
1618 if (imgp->sysent->sv_stackgap != NULL)
1619 imgp->sysent->sv_stackgap(imgp, (u_long *)&vectp);
1621 if (imgp->auxargs) {
1623 * Allocate room on the stack for the ELF auxargs
1624 * array. It has up to AT_COUNT entries.
1626 vectp -= howmany(AT_COUNT * sizeof(Elf_Auxinfo),
1631 * Allocate room for the argv[] and env vectors including the
1632 * terminating NULL pointers.
1634 vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
1637 * vectp also becomes our initial stack base
1639 stack_base = (register_t *)vectp;
1641 stringp = imgp->args->begin_argv;
1642 argc = imgp->args->argc;
1643 envc = imgp->args->envc;
1646 * Copy out strings - arguments and environment.
1648 copyout(stringp, (void *)destp, ARG_MAX - imgp->args->stringspace);
1651 * Fill in "ps_strings" struct for ps, w, etc.
1653 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1654 suword32(&arginfo->ps_nargvstr, argc);
1657 * Fill in argument portion of vector table.
1659 for (; argc > 0; --argc) {
1660 suword(vectp++, (long)(intptr_t)destp);
1661 while (*stringp++ != 0)
1666 /* a null vector table pointer separates the argp's from the envp's */
1669 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1670 suword32(&arginfo->ps_nenvstr, envc);
1673 * Fill in environment portion of vector table.
1675 for (; envc > 0; --envc) {
1676 suword(vectp++, (long)(intptr_t)destp);
1677 while (*stringp++ != 0)
1682 /* end of vector table is a null pointer */
1685 return (stack_base);
1689 * Check permissions of file to execute.
1690 * Called with imgp->vp locked.
1691 * Return 0 for success or error code on failure.
1694 exec_check_permissions(struct image_params *imgp)
1696 struct vnode *vp = imgp->vp;
1697 struct vattr *attr = imgp->attr;
1703 /* Get file attributes */
1704 error = VOP_GETATTR(vp, attr, td->td_ucred);
1709 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1715 * 1) Check if file execution is disabled for the filesystem that
1716 * this file resides on.
1717 * 2) Ensure that at least one execute bit is on. Otherwise, a
1718 * privileged user will always succeed, and we don't want this
1719 * to happen unless the file really is executable.
1720 * 3) Ensure that the file is a regular file.
1722 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1723 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1724 (attr->va_type != VREG))
1728 * Zero length files can't be exec'd
1730 if (attr->va_size == 0)
1734 * Check for execute permission to file based on current credentials.
1736 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1741 * Check number of open-for-writes on the file and deny execution
1744 * Add a text reference now so no one can write to the
1745 * executable while we're activating it.
1747 * Remember if this was set before and unset it in case this is not
1748 * actually an executable image.
1750 error = VOP_SET_TEXT(vp);
1753 imgp->textset = true;
1756 * Call filesystem specific open routine (which does nothing in the
1759 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1766 * Exec handler registration
1769 exec_register(const struct execsw *execsw_arg)
1771 const struct execsw **es, **xs, **newexecsw;
1772 u_int count = 2; /* New slot and trailing NULL */
1775 for (es = execsw; *es; es++)
1777 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1780 for (es = execsw; *es; es++)
1785 free(execsw, M_TEMP);
1791 exec_unregister(const struct execsw *execsw_arg)
1793 const struct execsw **es, **xs, **newexecsw;
1797 panic("unregister with no handlers left?\n");
1799 for (es = execsw; *es; es++) {
1800 if (*es == execsw_arg)
1805 for (es = execsw; *es; es++)
1806 if (*es != execsw_arg)
1808 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1810 for (es = execsw; *es; es++)
1811 if (*es != execsw_arg)
1815 free(execsw, M_TEMP);