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",
127 "Location of process' ps_strings structure");
129 /* XXX This should be vm_size_t. */
130 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD|
131 CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_usrstack, "LU",
132 "Top of process stack");
134 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD|CTLFLAG_MPSAFE,
135 NULL, 0, sysctl_kern_stackprot, "I",
136 "Stack memory permissions");
138 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
139 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
140 &ps_arg_cache_limit, 0,
141 "Process' command line characters cache limit");
143 static int disallow_high_osrel;
144 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW,
145 &disallow_high_osrel, 0,
146 "Disallow execution of binaries built for higher version of the world");
148 static int map_at_zero = 0;
149 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RWTUN, &map_at_zero, 0,
150 "Permit processes to map an object at virtual address 0.");
153 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
160 if (req->flags & SCTL_MASK32) {
162 val = (unsigned int)p->p_sysent->sv_psstrings;
163 error = SYSCTL_OUT(req, &val, sizeof(val));
166 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
167 sizeof(p->p_sysent->sv_psstrings));
172 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
179 if (req->flags & SCTL_MASK32) {
181 val = (unsigned int)p->p_sysent->sv_usrstack;
182 error = SYSCTL_OUT(req, &val, sizeof(val));
185 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
186 sizeof(p->p_sysent->sv_usrstack));
191 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
196 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
197 sizeof(p->p_sysent->sv_stackprot)));
201 * Each of the items is a pointer to a `const struct execsw', hence the
202 * double pointer here.
204 static const struct execsw **execsw;
206 #ifndef _SYS_SYSPROTO_H_
215 sys_execve(struct thread *td, struct execve_args *uap)
217 struct image_args args;
218 struct vmspace *oldvmspace;
221 error = pre_execve(td, &oldvmspace);
224 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
225 uap->argv, uap->envv);
227 error = kern_execve(td, &args, NULL);
228 post_execve(td, error, oldvmspace);
232 #ifndef _SYS_SYSPROTO_H_
233 struct fexecve_args {
240 sys_fexecve(struct thread *td, struct fexecve_args *uap)
242 struct image_args args;
243 struct vmspace *oldvmspace;
246 error = pre_execve(td, &oldvmspace);
249 error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
250 uap->argv, uap->envv);
253 error = kern_execve(td, &args, NULL);
255 post_execve(td, error, oldvmspace);
259 #ifndef _SYS_SYSPROTO_H_
260 struct __mac_execve_args {
269 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap)
272 struct image_args args;
273 struct vmspace *oldvmspace;
276 error = pre_execve(td, &oldvmspace);
279 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
280 uap->argv, uap->envv);
282 error = kern_execve(td, &args, uap->mac_p);
283 post_execve(td, error, oldvmspace);
291 pre_execve(struct thread *td, struct vmspace **oldvmspace)
296 KASSERT(td == curthread, ("non-current thread %p", td));
299 if ((p->p_flag & P_HADTHREADS) != 0) {
301 if (thread_single(p, SINGLE_BOUNDARY) != 0)
305 KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0,
307 *oldvmspace = p->p_vmspace;
312 post_execve(struct thread *td, int error, struct vmspace *oldvmspace)
316 KASSERT(td == curthread, ("non-current thread %p", td));
318 if ((p->p_flag & P_HADTHREADS) != 0) {
321 * If success, we upgrade to SINGLE_EXIT state to
322 * force other threads to suicide.
324 if (error == EJUSTRETURN)
325 thread_single(p, SINGLE_EXIT);
327 thread_single_end(p, SINGLE_BOUNDARY);
330 if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
331 KASSERT(p->p_vmspace != oldvmspace,
332 ("oldvmspace still used"));
333 vmspace_free(oldvmspace);
334 td->td_pflags &= ~TDP_EXECVMSPC;
339 * XXX: kern_execve has the astonishing property of not always returning to
340 * the caller. If sufficiently bad things happen during the call to
341 * do_execve(), it can end up calling exit1(); as a result, callers must
342 * avoid doing anything which they might need to undo (e.g., allocating
346 kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p)
349 AUDIT_ARG_ARGV(args->begin_argv, args->argc,
350 exec_args_get_begin_envv(args) - args->begin_argv);
351 AUDIT_ARG_ENVV(exec_args_get_begin_envv(args), args->envc,
352 args->endp - exec_args_get_begin_envv(args));
353 return (do_execve(td, args, mac_p));
357 * In-kernel implementation of execve(). All arguments are assumed to be
358 * userspace pointers from the passed thread.
361 do_execve(struct thread *td, struct image_args *args, struct mac *mac_p)
363 struct proc *p = td->td_proc;
365 struct ucred *oldcred;
366 struct uidinfo *euip = NULL;
367 uintptr_t stack_base;
368 struct image_params image_params, *imgp;
370 int (*img_first)(struct image_params *);
371 struct pargs *oldargs = NULL, *newargs = NULL;
372 struct sigacts *oldsigacts = NULL, *newsigacts = NULL;
374 struct vnode *tracevp = NULL;
375 struct ucred *tracecred = NULL;
377 struct vnode *oldtextvp = NULL, *newtextvp;
378 int credential_changing;
380 struct label *interpvplabel = NULL;
384 struct pmckern_procexec pe;
386 int error, i, orig_osrel;
388 static const char fexecv_proc_title[] = "(fexecv)";
390 imgp = &image_params;
393 * Lock the process and set the P_INEXEC flag to indicate that
394 * it should be left alone until we're done here. This is
395 * necessary to avoid race conditions - e.g. in ptrace() -
396 * that might allow a local user to illicitly obtain elevated
400 KASSERT((p->p_flag & P_INEXEC) == 0,
401 ("%s(): process already has P_INEXEC flag", __func__));
402 p->p_flag |= P_INEXEC;
406 * Initialize part of the common data
408 bzero(imgp, sizeof(*imgp));
412 oldcred = p->p_ucred;
413 orig_osrel = p->p_osrel;
414 orig_fctl0 = p->p_fctl0;
417 error = mac_execve_enter(imgp, mac_p);
423 * Translate the file name. namei() returns a vnode pointer
424 * in ni_vp among other things.
426 * XXXAUDIT: It would be desirable to also audit the name of the
427 * interpreter if this is an interpreted binary.
429 if (args->fname != NULL) {
430 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW |
431 SAVENAME | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
434 SDT_PROBE1(proc, , , exec, args->fname);
437 if (args->fname != NULL) {
438 #ifdef CAPABILITY_MODE
440 * While capability mode can't reach this point via direct
441 * path arguments to execve(), we also don't allow
442 * interpreters to be used in capability mode (for now).
443 * Catch indirect lookups and return a permissions error.
445 if (IN_CAPABILITY_MODE(td)) {
454 newtextvp = nd.ni_vp;
455 imgp->vp = newtextvp;
457 AUDIT_ARG_FD(args->fd);
459 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
461 error = fgetvp_exec(td, args->fd, &cap_fexecve_rights, &newtextvp);
464 vn_lock(newtextvp, LK_SHARED | LK_RETRY);
465 AUDIT_ARG_VNODE1(newtextvp);
466 imgp->vp = newtextvp;
470 * Check file permissions. Also 'opens' file and sets its vnode to
473 error = exec_check_permissions(imgp);
475 goto exec_fail_dealloc;
477 imgp->object = imgp->vp->v_object;
478 if (imgp->object != NULL)
479 vm_object_reference(imgp->object);
481 error = exec_map_first_page(imgp);
483 goto exec_fail_dealloc;
485 imgp->proc->p_osrel = 0;
486 imgp->proc->p_fctl0 = 0;
489 * Implement image setuid/setgid.
491 * Determine new credentials before attempting image activators
492 * so that it can be used by process_exec handlers to determine
493 * credential/setid changes.
495 * Don't honor setuid/setgid if the filesystem prohibits it or if
496 * the process is being traced.
498 * We disable setuid/setgid/etc in capability mode on the basis
499 * that most setugid applications are not written with that
500 * environment in mind, and will therefore almost certainly operate
501 * incorrectly. In principle there's no reason that setugid
502 * applications might not be useful in capability mode, so we may want
503 * to reconsider this conservative design choice in the future.
505 * XXXMAC: For the time being, use NOSUID to also prohibit
506 * transitions on the file system.
508 credential_changing = 0;
509 credential_changing |= (attr.va_mode & S_ISUID) &&
510 oldcred->cr_uid != attr.va_uid;
511 credential_changing |= (attr.va_mode & S_ISGID) &&
512 oldcred->cr_gid != attr.va_gid;
514 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
515 interpvplabel, imgp);
516 credential_changing |= will_transition;
519 /* Don't inherit PROC_PDEATHSIG_CTL value if setuid/setgid. */
520 if (credential_changing)
521 imgp->proc->p_pdeathsig = 0;
523 if (credential_changing &&
524 #ifdef CAPABILITY_MODE
525 ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
527 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
528 (p->p_flag & P_TRACED) == 0) {
529 imgp->credential_setid = true;
530 VOP_UNLOCK(imgp->vp);
531 imgp->newcred = crdup(oldcred);
532 if (attr.va_mode & S_ISUID) {
533 euip = uifind(attr.va_uid);
534 change_euid(imgp->newcred, euip);
536 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
537 if (attr.va_mode & S_ISGID)
538 change_egid(imgp->newcred, attr.va_gid);
540 * Implement correct POSIX saved-id behavior.
542 * XXXMAC: Note that the current logic will save the
543 * uid and gid if a MAC domain transition occurs, even
544 * though maybe it shouldn't.
546 change_svuid(imgp->newcred, imgp->newcred->cr_uid);
547 change_svgid(imgp->newcred, imgp->newcred->cr_gid);
550 * Implement correct POSIX saved-id behavior.
552 * XXX: It's not clear that the existing behavior is
553 * POSIX-compliant. A number of sources indicate that the
554 * saved uid/gid should only be updated if the new ruid is
555 * not equal to the old ruid, or the new euid is not equal
556 * to the old euid and the new euid is not equal to the old
557 * ruid. The FreeBSD code always updates the saved uid/gid.
558 * Also, this code uses the new (replaced) euid and egid as
559 * the source, which may or may not be the right ones to use.
561 if (oldcred->cr_svuid != oldcred->cr_uid ||
562 oldcred->cr_svgid != oldcred->cr_gid) {
563 VOP_UNLOCK(imgp->vp);
564 imgp->newcred = crdup(oldcred);
565 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
566 change_svuid(imgp->newcred, imgp->newcred->cr_uid);
567 change_svgid(imgp->newcred, imgp->newcred->cr_gid);
570 /* The new credentials are installed into the process later. */
573 * Do the best to calculate the full path to the image file.
575 if (args->fname != NULL && args->fname[0] == '/')
576 imgp->execpath = args->fname;
578 VOP_UNLOCK(imgp->vp);
579 if (vn_fullpath(td, imgp->vp, &imgp->execpath,
580 &imgp->freepath) != 0)
581 imgp->execpath = args->fname;
582 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
586 * If the current process has a special image activator it
587 * wants to try first, call it. For example, emulating shell
588 * scripts differently.
591 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
592 error = img_first(imgp);
595 * Loop through the list of image activators, calling each one.
596 * An activator returns -1 if there is no match, 0 on success,
597 * and an error otherwise.
599 for (i = 0; error == -1 && execsw[i]; ++i) {
600 if (execsw[i]->ex_imgact == NULL ||
601 execsw[i]->ex_imgact == img_first) {
604 error = (*execsw[i]->ex_imgact)(imgp);
610 goto exec_fail_dealloc;
614 * Special interpreter operation, cleanup and loop up to try to
615 * activate the interpreter.
617 if (imgp->interpreted) {
618 exec_unmap_first_page(imgp);
620 * The text reference needs to be removed for scripts.
621 * There is a short period before we determine that
622 * something is a script where text reference is active.
623 * The vnode lock is held over this entire period
624 * so nothing should illegitimately be blocked.
626 MPASS(imgp->textset);
627 VOP_UNSET_TEXT_CHECKED(newtextvp);
628 imgp->textset = false;
629 /* free name buffer and old vnode */
630 if (args->fname != NULL)
631 NDFREE(&nd, NDF_ONLY_PNBUF);
633 mac_execve_interpreter_enter(newtextvp, &interpvplabel);
636 VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td);
640 vm_object_deallocate(imgp->object);
642 imgp->credential_setid = false;
643 if (imgp->newcred != NULL) {
644 crfree(imgp->newcred);
645 imgp->newcred = NULL;
647 imgp->execpath = NULL;
648 free(imgp->freepath, M_TEMP);
649 imgp->freepath = NULL;
650 /* set new name to that of the interpreter */
651 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME,
652 UIO_SYSSPACE, imgp->interpreter_name, td);
653 args->fname = imgp->interpreter_name;
658 * NB: We unlock the vnode here because it is believed that none
659 * of the sv_copyout_strings/sv_fixup operations require the vnode.
661 VOP_UNLOCK(imgp->vp);
663 if (disallow_high_osrel &&
664 P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
666 uprintf("Osrel %d for image %s too high\n", p->p_osrel,
667 imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
668 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
669 goto exec_fail_dealloc;
672 /* ABI enforces the use of Capsicum. Switch into capabilities mode. */
673 if (SV_PROC_FLAG(p, SV_CAPSICUM))
674 sys_cap_enter(td, NULL);
677 * Copy out strings (args and env) and initialize stack base.
679 error = (*p->p_sysent->sv_copyout_strings)(imgp, &stack_base);
681 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
682 goto exec_fail_dealloc;
688 error = (*p->p_sysent->sv_fixup)(&stack_base, imgp);
690 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
691 goto exec_fail_dealloc;
694 if (args->fdp != NULL) {
695 /* Install a brand new file descriptor table. */
696 fdinstall_remapped(td, args->fdp);
700 * Keep on using the existing file descriptor table. For
701 * security and other reasons, the file descriptor table
702 * cannot be shared after an exec.
705 /* close files on exec */
710 * Malloc things before we need locks.
712 i = exec_args_get_begin_envv(imgp->args) - imgp->args->begin_argv;
713 /* Cache arguments if they fit inside our allowance */
714 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
715 newargs = pargs_alloc(i);
716 bcopy(imgp->args->begin_argv, newargs->ar_args, i);
720 * For security and other reasons, signal handlers cannot
721 * be shared after an exec. The new process gets a copy of the old
722 * handlers. In execsigs(), the new process will have its signals
725 if (sigacts_shared(p->p_sigacts)) {
726 oldsigacts = p->p_sigacts;
727 newsigacts = sigacts_alloc();
728 sigacts_copy(newsigacts, oldsigacts);
731 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
735 p->p_sigacts = newsigacts;
739 /* reset caught signals */
742 /* name this process - nameiexec(p, ndp) */
743 bzero(p->p_comm, sizeof(p->p_comm));
745 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
746 min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
747 else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0)
748 bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
749 bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
751 sched_clear_tdname(td);
755 * mark as execed, wakeup the process that vforked (if any) and tell
756 * it that it now has its own resources back
759 if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0)
760 p->p_flag2 &= ~P2_NOTRACE;
761 if ((p->p_flag2 & P2_STKGAP_DISABLE_EXEC) == 0)
762 p->p_flag2 &= ~P2_STKGAP_DISABLE;
763 if (p->p_flag & P_PPWAIT) {
764 p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
765 cv_broadcast(&p->p_pwait);
766 /* STOPs are no longer ignored, arrange for AST */
771 * Implement image setuid/setgid installation.
773 if (imgp->credential_setid) {
775 * Turn off syscall tracing for set-id programs, except for
776 * root. Record any set-id flags first to make sure that
777 * we do not regain any tracing during a possible block.
782 if (p->p_tracecred != NULL &&
783 priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED))
784 ktrprocexec(p, &tracecred, &tracevp);
787 * Close any file descriptors 0..2 that reference procfs,
788 * then make sure file descriptors 0..2 are in use.
790 * Both fdsetugidsafety() and fdcheckstd() may call functions
791 * taking sleepable locks, so temporarily drop our locks.
794 VOP_UNLOCK(imgp->vp);
796 error = fdcheckstd(td);
797 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
799 goto exec_fail_dealloc;
802 if (will_transition) {
803 mac_vnode_execve_transition(oldcred, imgp->newcred,
804 imgp->vp, interpvplabel, imgp);
808 if (oldcred->cr_uid == oldcred->cr_ruid &&
809 oldcred->cr_gid == oldcred->cr_rgid)
810 p->p_flag &= ~P_SUGID;
813 * Set the new credentials.
815 if (imgp->newcred != NULL) {
816 proc_set_cred(p, imgp->newcred);
822 * Store the vp for use in procfs. This vnode was referenced by namei
825 oldtextvp = p->p_textvp;
826 p->p_textvp = newtextvp;
830 * Tell the DTrace fasttrap provider about the exec if it
831 * has declared an interest.
833 if (dtrace_fasttrap_exec)
834 dtrace_fasttrap_exec(p);
838 * Notify others that we exec'd, and clear the P_INEXEC flag
839 * as we're now a bona fide freshly-execed process.
841 KNOTE_LOCKED(p->p_klist, NOTE_EXEC);
842 p->p_flag &= ~P_INEXEC;
844 /* clear "fork but no exec" flag, as we _are_ execing */
845 p->p_acflag &= ~AFORK;
848 * Free any previous argument cache and replace it with
849 * the new argument cache, if any.
859 * Check if system-wide sampling is in effect or if the
860 * current process is using PMCs. If so, do exec() time
861 * processing. This processing needs to happen AFTER the
862 * P_INEXEC flag is cleared.
864 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
865 VOP_UNLOCK(imgp->vp);
866 pe.pm_credentialschanged = credential_changing;
867 pe.pm_entryaddr = imgp->entry_addr;
869 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
870 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
874 /* Set values passed into the program in registers. */
875 (*p->p_sysent->sv_setregs)(td, imgp, stack_base);
877 VOP_MMAPPED(imgp->vp);
879 SDT_PROBE1(proc, , , exec__success, args->fname);
883 p->p_osrel = orig_osrel;
884 p->p_fctl0 = orig_fctl0;
887 if (imgp->firstpage != NULL)
888 exec_unmap_first_page(imgp);
890 if (imgp->vp != NULL) {
892 NDFREE(&nd, NDF_ONLY_PNBUF);
894 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
896 VOP_UNSET_TEXT_CHECKED(imgp->vp);
900 VOP_UNLOCK(imgp->vp);
903 if (imgp->object != NULL)
904 vm_object_deallocate(imgp->object);
906 free(imgp->freepath, M_TEMP);
909 if (p->p_ptevents & PTRACE_EXEC) {
911 if (p->p_ptevents & PTRACE_EXEC)
912 td->td_dbgflags |= TDB_EXEC;
917 * Stop the process here if its stop event mask has
918 * the S_EXEC bit set.
920 STOPEVENT(p, S_EXEC, 0);
923 /* we're done here, clear P_INEXEC */
925 p->p_flag &= ~P_INEXEC;
928 SDT_PROBE1(proc, , , exec__failure, error);
931 if (imgp->newcred != NULL && oldcred != NULL)
932 crfree(imgp->newcred);
935 mac_execve_exit(imgp);
936 mac_execve_interpreter_exit(interpvplabel);
938 exec_free_args(args);
941 * Handle deferred decrement of ref counts.
943 if (oldtextvp != NULL)
948 if (tracecred != NULL)
953 if (oldsigacts != NULL)
954 sigacts_free(oldsigacts);
958 if (error && imgp->vmspace_destroyed) {
959 /* sorry, no more process anymore. exit gracefully */
960 exit1(td, 0, SIGABRT);
970 * We don't want cpu_set_syscall_retval() to overwrite any of
971 * the register values put in place by exec_setregs().
972 * Implementations of cpu_set_syscall_retval() will leave
973 * registers unmodified when returning EJUSTRETURN.
975 return (error == 0 ? EJUSTRETURN : error);
979 exec_map_first_page(struct image_params *imgp)
985 if (imgp->firstpage != NULL)
986 exec_unmap_first_page(imgp);
988 object = imgp->vp->v_object;
991 #if VM_NRESERVLEVEL > 0
992 if ((object->flags & OBJ_COLORED) == 0) {
993 VM_OBJECT_WLOCK(object);
994 vm_object_color(object, 0);
995 VM_OBJECT_WUNLOCK(object);
998 error = vm_page_grab_valid_unlocked(&m, object, 0,
999 VM_ALLOC_COUNT(VM_INITIAL_PAGEIN) |
1000 VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED);
1002 if (error != VM_PAGER_OK)
1004 imgp->firstpage = sf_buf_alloc(m, 0);
1005 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
1011 exec_unmap_first_page(struct image_params *imgp)
1015 if (imgp->firstpage != NULL) {
1016 m = sf_buf_page(imgp->firstpage);
1017 sf_buf_free(imgp->firstpage);
1018 imgp->firstpage = NULL;
1019 vm_page_unwire(m, PQ_ACTIVE);
1024 * Destroy old address space, and allocate a new stack.
1025 * The new stack is only sgrowsiz large because it is grown
1026 * automatically on a page fault.
1029 exec_new_vmspace(struct image_params *imgp, struct sysentvec *sv)
1032 struct proc *p = imgp->proc;
1033 struct vmspace *vmspace = p->p_vmspace;
1034 struct thread *td = curthread;
1036 struct rlimit rlim_stack;
1037 vm_offset_t sv_minuser, stack_addr;
1041 imgp->vmspace_destroyed = 1;
1044 sigfastblock_clear(td);
1046 /* May be called with Giant held */
1047 EVENTHANDLER_DIRECT_INVOKE(process_exec, p, imgp);
1050 * Blow away entire process VM, if address space not shared,
1051 * otherwise, create a new VM space so that other threads are
1054 map = &vmspace->vm_map;
1056 sv_minuser = sv->sv_minuser;
1058 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1059 if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser &&
1060 vm_map_max(map) == sv->sv_maxuser &&
1061 cpu_exec_vmspace_reuse(p, map)) {
1063 pmap_remove_pages(vmspace_pmap(vmspace));
1064 vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1066 * An exec terminates mlockall(MCL_FUTURE), ASLR state
1067 * must be re-evaluated.
1070 vm_map_modflags(map, 0, MAP_WIREFUTURE | MAP_ASLR |
1074 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1077 vmspace = p->p_vmspace;
1078 map = &vmspace->vm_map;
1080 map->flags |= imgp->map_flags;
1082 /* Map a shared page */
1083 obj = sv->sv_shared_page_obj;
1085 vm_object_reference(obj);
1086 error = vm_map_fixed(map, obj, 0,
1087 sv->sv_shared_page_base, sv->sv_shared_page_len,
1088 VM_PROT_READ | VM_PROT_EXECUTE,
1089 VM_PROT_READ | VM_PROT_EXECUTE,
1090 MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1091 if (error != KERN_SUCCESS) {
1092 vm_object_deallocate(obj);
1093 return (vm_mmap_to_errno(error));
1097 /* Allocate a new stack */
1098 if (imgp->stack_sz != 0) {
1099 ssiz = trunc_page(imgp->stack_sz);
1101 lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack);
1103 if (ssiz > rlim_stack.rlim_max)
1104 ssiz = rlim_stack.rlim_max;
1105 if (ssiz > rlim_stack.rlim_cur) {
1106 rlim_stack.rlim_cur = ssiz;
1107 kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack);
1109 } else if (sv->sv_maxssiz != NULL) {
1110 ssiz = *sv->sv_maxssiz;
1114 imgp->eff_stack_sz = lim_cur(curthread, RLIMIT_STACK);
1115 if (ssiz < imgp->eff_stack_sz)
1116 imgp->eff_stack_sz = ssiz;
1117 stack_addr = sv->sv_usrstack - ssiz;
1118 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1119 obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1120 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1121 if (error != KERN_SUCCESS)
1122 return (vm_mmap_to_errno(error));
1125 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they
1126 * are still used to enforce the stack rlimit on the process stack.
1128 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1129 vmspace->vm_maxsaddr = (char *)stack_addr;
1135 * Copy out argument and environment strings from the old process address
1136 * space into the temporary string buffer.
1139 exec_copyin_args(struct image_args *args, const char *fname,
1140 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 error = exec_args_add_fname(args, fname, segflg);
1165 * extract arguments first
1168 error = fueword(argv++, &arg);
1175 error = exec_args_add_arg(args, (char *)(uintptr_t)arg,
1182 * extract environment strings
1186 error = fueword(envv++, &env);
1193 error = exec_args_add_env(args,
1194 (char *)(uintptr_t)env, UIO_USERSPACE);
1203 exec_free_args(args);
1208 exec_copyin_data_fds(struct thread *td, struct image_args *args,
1209 const void *data, size_t datalen, const int *fds, size_t fdslen)
1211 struct filedesc *ofdp;
1216 memset(args, '\0', sizeof(*args));
1217 ofdp = td->td_proc->p_fd;
1218 if (datalen >= ARG_MAX || fdslen > ofdp->fd_lastfile + 1)
1220 error = exec_alloc_args(args);
1224 args->begin_argv = args->buf;
1225 args->stringspace = ARG_MAX;
1229 * Argument buffer has been provided. Copy it into the
1230 * kernel as a single string and add a terminating null
1233 error = copyin(data, args->begin_argv, datalen);
1236 args->begin_argv[datalen] = '\0';
1237 args->endp = args->begin_argv + datalen + 1;
1238 args->stringspace -= datalen + 1;
1241 * Traditional argument counting. Count the number of
1244 for (p = args->begin_argv; p < args->endp; ++p)
1248 /* No argument buffer provided. */
1249 args->endp = args->begin_argv;
1252 /* Create new file descriptor table. */
1253 kfds = malloc(fdslen * sizeof(int), M_TEMP, M_WAITOK);
1254 error = copyin(fds, kfds, fdslen * sizeof(int));
1259 error = fdcopy_remapped(ofdp, kfds, fdslen, &args->fdp);
1266 exec_free_args(args);
1270 struct exec_args_kva {
1273 SLIST_ENTRY(exec_args_kva) next;
1276 DPCPU_DEFINE_STATIC(struct exec_args_kva *, exec_args_kva);
1278 static SLIST_HEAD(, exec_args_kva) exec_args_kva_freelist;
1279 static struct mtx exec_args_kva_mtx;
1280 static u_int exec_args_gen;
1283 exec_prealloc_args_kva(void *arg __unused)
1285 struct exec_args_kva *argkva;
1288 SLIST_INIT(&exec_args_kva_freelist);
1289 mtx_init(&exec_args_kva_mtx, "exec args kva", NULL, MTX_DEF);
1290 for (i = 0; i < exec_map_entries; i++) {
1291 argkva = malloc(sizeof(*argkva), M_PARGS, M_WAITOK);
1292 argkva->addr = kmap_alloc_wait(exec_map, exec_map_entry_size);
1293 argkva->gen = exec_args_gen;
1294 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1297 SYSINIT(exec_args_kva, SI_SUB_EXEC, SI_ORDER_ANY, exec_prealloc_args_kva, NULL);
1300 exec_alloc_args_kva(void **cookie)
1302 struct exec_args_kva *argkva;
1304 argkva = (void *)atomic_readandclear_ptr(
1305 (uintptr_t *)DPCPU_PTR(exec_args_kva));
1306 if (argkva == NULL) {
1307 mtx_lock(&exec_args_kva_mtx);
1308 while ((argkva = SLIST_FIRST(&exec_args_kva_freelist)) == NULL)
1309 (void)mtx_sleep(&exec_args_kva_freelist,
1310 &exec_args_kva_mtx, 0, "execkva", 0);
1311 SLIST_REMOVE_HEAD(&exec_args_kva_freelist, next);
1312 mtx_unlock(&exec_args_kva_mtx);
1314 *(struct exec_args_kva **)cookie = argkva;
1315 return (argkva->addr);
1319 exec_release_args_kva(struct exec_args_kva *argkva, u_int gen)
1323 base = argkva->addr;
1324 if (argkva->gen != gen) {
1325 (void)vm_map_madvise(exec_map, base, base + exec_map_entry_size,
1329 if (!atomic_cmpset_ptr((uintptr_t *)DPCPU_PTR(exec_args_kva),
1330 (uintptr_t)NULL, (uintptr_t)argkva)) {
1331 mtx_lock(&exec_args_kva_mtx);
1332 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1333 wakeup_one(&exec_args_kva_freelist);
1334 mtx_unlock(&exec_args_kva_mtx);
1339 exec_free_args_kva(void *cookie)
1342 exec_release_args_kva(cookie, exec_args_gen);
1346 exec_args_kva_lowmem(void *arg __unused)
1348 SLIST_HEAD(, exec_args_kva) head;
1349 struct exec_args_kva *argkva;
1353 gen = atomic_fetchadd_int(&exec_args_gen, 1) + 1;
1356 * Force an madvise of each KVA range. Any currently allocated ranges
1357 * will have MADV_FREE applied once they are freed.
1360 mtx_lock(&exec_args_kva_mtx);
1361 SLIST_SWAP(&head, &exec_args_kva_freelist, exec_args_kva);
1362 mtx_unlock(&exec_args_kva_mtx);
1363 while ((argkva = SLIST_FIRST(&head)) != NULL) {
1364 SLIST_REMOVE_HEAD(&head, next);
1365 exec_release_args_kva(argkva, gen);
1369 argkva = (void *)atomic_readandclear_ptr(
1370 (uintptr_t *)DPCPU_ID_PTR(i, exec_args_kva));
1372 exec_release_args_kva(argkva, gen);
1375 EVENTHANDLER_DEFINE(vm_lowmem, exec_args_kva_lowmem, NULL,
1376 EVENTHANDLER_PRI_ANY);
1379 * Allocate temporary demand-paged, zero-filled memory for the file name,
1380 * argument, and environment strings.
1383 exec_alloc_args(struct image_args *args)
1386 args->buf = (char *)exec_alloc_args_kva(&args->bufkva);
1391 exec_free_args(struct image_args *args)
1394 if (args->buf != NULL) {
1395 exec_free_args_kva(args->bufkva);
1398 if (args->fname_buf != NULL) {
1399 free(args->fname_buf, M_TEMP);
1400 args->fname_buf = NULL;
1402 if (args->fdp != NULL)
1403 fdescfree_remapped(args->fdp);
1407 * A set to functions to fill struct image args.
1409 * NOTE: exec_args_add_fname() must be called (possibly with a NULL
1410 * fname) before the other functions. All exec_args_add_arg() calls must
1411 * be made before any exec_args_add_env() calls. exec_args_adjust_args()
1412 * may be called any time after exec_args_add_fname().
1414 * exec_args_add_fname() - install path to be executed
1415 * exec_args_add_arg() - append an argument string
1416 * exec_args_add_env() - append an env string
1417 * exec_args_adjust_args() - adjust location of the argument list to
1418 * allow new arguments to be prepended
1421 exec_args_add_fname(struct image_args *args, const char *fname,
1422 enum uio_seg segflg)
1427 KASSERT(args->fname == NULL, ("fname already appended"));
1428 KASSERT(args->endp == NULL, ("already appending to args"));
1430 if (fname != NULL) {
1431 args->fname = args->buf;
1432 error = segflg == UIO_SYSSPACE ?
1433 copystr(fname, args->fname, PATH_MAX, &length) :
1434 copyinstr(fname, args->fname, PATH_MAX, &length);
1436 return (error == ENAMETOOLONG ? E2BIG : error);
1440 /* Set up for _arg_*()/_env_*() */
1441 args->endp = args->buf + length;
1442 /* begin_argv must be set and kept updated */
1443 args->begin_argv = args->endp;
1444 KASSERT(exec_map_entry_size - length >= ARG_MAX,
1445 ("too little space remaining for arguments %zu < %zu",
1446 exec_map_entry_size - length, (size_t)ARG_MAX));
1447 args->stringspace = ARG_MAX;
1453 exec_args_add_str(struct image_args *args, const char *str,
1454 enum uio_seg segflg, int *countp)
1459 KASSERT(args->endp != NULL, ("endp not initialized"));
1460 KASSERT(args->begin_argv != NULL, ("begin_argp not initialized"));
1462 error = (segflg == UIO_SYSSPACE) ?
1463 copystr(str, args->endp, args->stringspace, &length) :
1464 copyinstr(str, args->endp, args->stringspace, &length);
1466 return (error == ENAMETOOLONG ? E2BIG : error);
1467 args->stringspace -= length;
1468 args->endp += length;
1475 exec_args_add_arg(struct image_args *args, const char *argp,
1476 enum uio_seg segflg)
1479 KASSERT(args->envc == 0, ("appending args after env"));
1481 return (exec_args_add_str(args, argp, segflg, &args->argc));
1485 exec_args_add_env(struct image_args *args, const char *envp,
1486 enum uio_seg segflg)
1489 if (args->envc == 0)
1490 args->begin_envv = args->endp;
1492 return (exec_args_add_str(args, envp, segflg, &args->envc));
1496 exec_args_adjust_args(struct image_args *args, size_t consume, ssize_t extend)
1500 KASSERT(args->endp != NULL, ("endp not initialized"));
1501 KASSERT(args->begin_argv != NULL, ("begin_argp not initialized"));
1503 offset = extend - consume;
1504 if (args->stringspace < offset)
1506 memmove(args->begin_argv + extend, args->begin_argv + consume,
1507 args->endp - args->begin_argv + consume);
1509 args->begin_envv += offset;
1510 args->endp += offset;
1511 args->stringspace -= offset;
1516 exec_args_get_begin_envv(struct image_args *args)
1519 KASSERT(args->endp != NULL, ("endp not initialized"));
1522 return (args->begin_envv);
1523 return (args->endp);
1527 * Copy strings out to the new process address space, constructing new arg
1528 * and env vector tables. Return a pointer to the base so that it can be used
1529 * as the initial stack pointer.
1532 exec_copyout_strings(struct image_params *imgp, uintptr_t *stack_base)
1537 uintptr_t destp, ustringp;
1538 struct ps_strings *arginfo;
1540 size_t execpath_len;
1541 int error, szsigcode, szps;
1542 char canary[sizeof(long) * 8];
1544 szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1546 * Calculate string base and vector table pointers.
1547 * Also deal with signal trampoline code for this exec type.
1549 if (imgp->execpath != NULL && imgp->auxargs != NULL)
1550 execpath_len = strlen(imgp->execpath) + 1;
1555 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1556 if (p->p_sysent->sv_sigcode_base == 0) {
1557 if (p->p_sysent->sv_szsigcode != NULL)
1558 szsigcode = *(p->p_sysent->sv_szsigcode);
1560 destp = (uintptr_t)arginfo;
1565 if (szsigcode != 0) {
1567 destp = rounddown2(destp, sizeof(void *));
1568 error = copyout(p->p_sysent->sv_sigcode, (void *)destp,
1575 * Copy the image path for the rtld.
1577 if (execpath_len != 0) {
1578 destp -= execpath_len;
1579 destp = rounddown2(destp, sizeof(void *));
1580 imgp->execpathp = destp;
1581 error = copyout(imgp->execpath, (void *)destp, execpath_len);
1587 * Prepare the canary for SSP.
1589 arc4rand(canary, sizeof(canary), 0);
1590 destp -= sizeof(canary);
1591 imgp->canary = destp;
1592 error = copyout(canary, (void *)destp, sizeof(canary));
1595 imgp->canarylen = sizeof(canary);
1598 * Prepare the pagesizes array.
1601 destp = rounddown2(destp, sizeof(void *));
1602 imgp->pagesizes = destp;
1603 error = copyout(pagesizes, (void *)destp, szps);
1606 imgp->pagesizeslen = szps;
1609 * Allocate room for the argument and environment strings.
1611 destp -= ARG_MAX - imgp->args->stringspace;
1612 destp = rounddown2(destp, sizeof(void *));
1615 if (imgp->sysent->sv_stackgap != NULL)
1616 imgp->sysent->sv_stackgap(imgp, &destp);
1618 if (imgp->auxargs) {
1620 * Allocate room on the stack for the ELF auxargs
1621 * array. It has up to AT_COUNT entries.
1623 destp -= AT_COUNT * sizeof(Elf_Auxinfo);
1624 destp = rounddown2(destp, sizeof(void *));
1627 vectp = (char **)destp;
1630 * Allocate room for the argv[] and env vectors including the
1631 * terminating NULL pointers.
1633 vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
1636 * vectp also becomes our initial stack base
1638 *stack_base = (uintptr_t)vectp;
1640 stringp = imgp->args->begin_argv;
1641 argc = imgp->args->argc;
1642 envc = imgp->args->envc;
1645 * Copy out strings - arguments and environment.
1647 error = copyout(stringp, (void *)ustringp,
1648 ARG_MAX - imgp->args->stringspace);
1653 * Fill in "ps_strings" struct for ps, w, etc.
1655 if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 ||
1656 suword32(&arginfo->ps_nargvstr, argc) != 0)
1660 * Fill in argument portion of vector table.
1662 for (; argc > 0; --argc) {
1663 if (suword(vectp++, ustringp) != 0)
1665 while (*stringp++ != 0)
1670 /* a null vector table pointer separates the argp's from the envp's */
1671 if (suword(vectp++, 0) != 0)
1674 if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 ||
1675 suword32(&arginfo->ps_nenvstr, envc) != 0)
1679 * Fill in environment portion of vector table.
1681 for (; envc > 0; --envc) {
1682 if (suword(vectp++, ustringp) != 0)
1684 while (*stringp++ != 0)
1689 /* end of vector table is a null pointer */
1690 if (suword(vectp, 0) != 0)
1693 if (imgp->auxargs) {
1695 error = imgp->sysent->sv_copyout_auxargs(imgp,
1705 * Check permissions of file to execute.
1706 * Called with imgp->vp locked.
1707 * Return 0 for success or error code on failure.
1710 exec_check_permissions(struct image_params *imgp)
1712 struct vnode *vp = imgp->vp;
1713 struct vattr *attr = imgp->attr;
1719 /* Get file attributes */
1720 error = VOP_GETATTR(vp, attr, td->td_ucred);
1725 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1731 * 1) Check if file execution is disabled for the filesystem that
1732 * this file resides on.
1733 * 2) Ensure that at least one execute bit is on. Otherwise, a
1734 * privileged user will always succeed, and we don't want this
1735 * to happen unless the file really is executable.
1736 * 3) Ensure that the file is a regular file.
1738 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1739 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1740 (attr->va_type != VREG))
1744 * Zero length files can't be exec'd
1746 if (attr->va_size == 0)
1750 * Check for execute permission to file based on current credentials.
1752 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1757 * Check number of open-for-writes on the file and deny execution
1760 * Add a text reference now so no one can write to the
1761 * executable while we're activating it.
1763 * Remember if this was set before and unset it in case this is not
1764 * actually an executable image.
1766 error = VOP_SET_TEXT(vp);
1769 imgp->textset = true;
1772 * Call filesystem specific open routine (which does nothing in the
1775 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1782 * Exec handler registration
1785 exec_register(const struct execsw *execsw_arg)
1787 const struct execsw **es, **xs, **newexecsw;
1788 u_int count = 2; /* New slot and trailing NULL */
1791 for (es = execsw; *es; es++)
1793 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1796 for (es = execsw; *es; es++)
1801 free(execsw, M_TEMP);
1807 exec_unregister(const struct execsw *execsw_arg)
1809 const struct execsw **es, **xs, **newexecsw;
1813 panic("unregister with no handlers left?\n");
1815 for (es = execsw; *es; es++) {
1816 if (*es == execsw_arg)
1821 for (es = execsw; *es; es++)
1822 if (*es != execsw_arg)
1824 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1826 for (es = execsw; *es; es++)
1827 if (*es != execsw_arg)
1831 free(execsw, M_TEMP);