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>
56 #include <sys/ptrace.h>
57 #include <sys/resourcevar.h>
58 #include <sys/rwlock.h>
59 #include <sys/sched.h>
61 #include <sys/sf_buf.h>
63 #include <sys/signalvar.h>
66 #include <sys/syscallsubr.h>
67 #include <sys/sysctl.h>
68 #include <sys/sysent.h>
69 #include <sys/sysproto.h>
70 #include <sys/vnode.h>
73 #include <sys/ktrace.h>
77 #include <vm/vm_param.h>
79 #include <vm/vm_page.h>
80 #include <vm/vm_map.h>
81 #include <vm/vm_kern.h>
82 #include <vm/vm_extern.h>
83 #include <vm/vm_object.h>
84 #include <vm/vm_pager.h>
87 #include <sys/pmckern.h>
90 #include <machine/reg.h>
92 #include <security/audit/audit.h>
93 #include <security/mac/mac_framework.h>
96 #include <sys/dtrace_bsd.h>
97 dtrace_execexit_func_t dtrace_fasttrap_exec;
100 SDT_PROVIDER_DECLARE(proc);
101 SDT_PROBE_DEFINE1(proc, , , exec, "char *");
102 SDT_PROBE_DEFINE1(proc, , , exec__failure, "int");
103 SDT_PROBE_DEFINE1(proc, , , exec__success, "char *");
105 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
107 int coredump_pack_fileinfo = 1;
108 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_fileinfo, CTLFLAG_RWTUN,
109 &coredump_pack_fileinfo, 0,
110 "Enable file path packing in 'procstat -f' coredump notes");
112 int coredump_pack_vmmapinfo = 1;
113 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_vmmapinfo, CTLFLAG_RWTUN,
114 &coredump_pack_vmmapinfo, 0,
115 "Enable file path packing in 'procstat -v' coredump notes");
117 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
118 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
119 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
120 static int do_execve(struct thread *td, struct image_args *args,
121 struct mac *mac_p, struct vmspace *oldvmspace);
123 /* XXX This should be vm_size_t. */
124 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD|
125 CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_ps_strings, "LU",
126 "Location of process' ps_strings structure");
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",
131 "Top of process stack");
133 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD|CTLFLAG_MPSAFE,
134 NULL, 0, sysctl_kern_stackprot, "I",
135 "Stack memory permissions");
137 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
138 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
139 &ps_arg_cache_limit, 0,
140 "Process' command line characters cache limit");
142 static int disallow_high_osrel;
143 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW,
144 &disallow_high_osrel, 0,
145 "Disallow execution of binaries built for higher version of the world");
147 static int map_at_zero = 0;
148 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RWTUN, &map_at_zero, 0,
149 "Permit processes to map an object at virtual address 0.");
152 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
159 if (req->flags & SCTL_MASK32) {
161 val = (unsigned int)p->p_sysent->sv_psstrings;
162 error = SYSCTL_OUT(req, &val, sizeof(val));
165 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
166 sizeof(p->p_sysent->sv_psstrings));
171 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
178 if (req->flags & SCTL_MASK32) {
180 val = (unsigned int)p->p_sysent->sv_usrstack;
181 error = SYSCTL_OUT(req, &val, sizeof(val));
184 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
185 sizeof(p->p_sysent->sv_usrstack));
190 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
195 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
196 sizeof(p->p_sysent->sv_stackprot)));
200 * Each of the items is a pointer to a `const struct execsw', hence the
201 * double pointer here.
203 static const struct execsw **execsw;
205 #ifndef _SYS_SYSPROTO_H_
214 sys_execve(struct thread *td, struct execve_args *uap)
216 struct image_args args;
217 struct vmspace *oldvmspace;
220 error = pre_execve(td, &oldvmspace);
223 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
224 uap->argv, uap->envv);
226 error = kern_execve(td, &args, NULL, oldvmspace);
227 post_execve(td, error, oldvmspace);
228 AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
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, oldvmspace);
255 post_execve(td, error, oldvmspace);
256 AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
260 #ifndef _SYS_SYSPROTO_H_
261 struct __mac_execve_args {
270 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap)
273 struct image_args args;
274 struct vmspace *oldvmspace;
277 error = pre_execve(td, &oldvmspace);
280 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
281 uap->argv, uap->envv);
283 error = kern_execve(td, &args, uap->mac_p, oldvmspace);
284 post_execve(td, error, oldvmspace);
285 AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
293 pre_execve(struct thread *td, struct vmspace **oldvmspace)
298 KASSERT(td == curthread, ("non-current thread %p", td));
301 if ((p->p_flag & P_HADTHREADS) != 0) {
303 if (thread_single(p, SINGLE_BOUNDARY) != 0)
307 KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0,
309 *oldvmspace = p->p_vmspace;
314 post_execve(struct thread *td, int error, struct vmspace *oldvmspace)
318 KASSERT(td == curthread, ("non-current thread %p", td));
320 if ((p->p_flag & P_HADTHREADS) != 0) {
323 * If success, we upgrade to SINGLE_EXIT state to
324 * force other threads to suicide.
326 if (error == EJUSTRETURN)
327 thread_single(p, SINGLE_EXIT);
329 thread_single_end(p, SINGLE_BOUNDARY);
332 exec_cleanup(td, oldvmspace);
336 * kern_execve() has the astonishing property of not always returning to
337 * the caller. If sufficiently bad things happen during the call to
338 * do_execve(), it can end up calling exit1(); as a result, callers must
339 * avoid doing anything which they might need to undo (e.g., allocating
343 kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p,
344 struct vmspace *oldvmspace)
347 AUDIT_ARG_ARGV(args->begin_argv, args->argc,
348 exec_args_get_begin_envv(args) - args->begin_argv);
349 AUDIT_ARG_ENVV(exec_args_get_begin_envv(args), args->envc,
350 args->endp - exec_args_get_begin_envv(args));
351 return (do_execve(td, args, mac_p, oldvmspace));
355 * In-kernel implementation of execve(). All arguments are assumed to be
356 * userspace pointers from the passed thread.
359 do_execve(struct thread *td, struct image_args *args, struct mac *mac_p,
360 struct vmspace *oldvmspace)
362 struct proc *p = td->td_proc;
364 struct ucred *oldcred;
365 struct uidinfo *euip = NULL;
366 uintptr_t stack_base;
367 struct image_params image_params, *imgp;
369 int (*img_first)(struct image_params *);
370 struct pargs *oldargs = NULL, *newargs = NULL;
371 struct sigacts *oldsigacts = NULL, *newsigacts = NULL;
373 struct vnode *tracevp = NULL;
374 struct ucred *tracecred = NULL;
376 struct vnode *oldtextvp = NULL, *newtextvp;
377 int credential_changing;
379 struct label *interpvplabel = NULL;
383 struct pmckern_procexec pe;
385 int error, i, orig_osrel;
387 static const char fexecv_proc_title[] = "(fexecv)";
389 imgp = &image_params;
392 * Lock the process and set the P_INEXEC flag to indicate that
393 * it should be left alone until we're done here. This is
394 * necessary to avoid race conditions - e.g. in ptrace() -
395 * that might allow a local user to illicitly obtain elevated
399 KASSERT((p->p_flag & P_INEXEC) == 0,
400 ("%s(): process already has P_INEXEC flag", __func__));
401 p->p_flag |= P_INEXEC;
405 * Initialize part of the common data
407 bzero(imgp, sizeof(*imgp));
411 oldcred = p->p_ucred;
412 orig_osrel = p->p_osrel;
413 orig_fctl0 = p->p_fctl0;
416 error = mac_execve_enter(imgp, mac_p);
422 * Translate the file name. namei() returns a vnode pointer
423 * in ni_vp among other things.
425 * XXXAUDIT: It would be desirable to also audit the name of the
426 * interpreter if this is an interpreted binary.
428 if (args->fname != NULL) {
429 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW |
430 SAVENAME | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
433 SDT_PROBE1(proc, , , exec, args->fname);
436 if (args->fname != NULL) {
437 #ifdef CAPABILITY_MODE
439 * While capability mode can't reach this point via direct
440 * path arguments to execve(), we also don't allow
441 * interpreters to be used in capability mode (for now).
442 * Catch indirect lookups and return a permissions error.
444 if (IN_CAPABILITY_MODE(td)) {
453 newtextvp = nd.ni_vp;
454 imgp->vp = newtextvp;
456 AUDIT_ARG_FD(args->fd);
458 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
460 error = fgetvp_exec(td, args->fd, &cap_fexecve_rights, &newtextvp);
463 vn_lock(newtextvp, LK_SHARED | LK_RETRY);
464 AUDIT_ARG_VNODE1(newtextvp);
465 imgp->vp = newtextvp;
469 * Check file permissions. Also 'opens' file and sets its vnode to
472 error = exec_check_permissions(imgp);
474 goto exec_fail_dealloc;
476 imgp->object = imgp->vp->v_object;
477 if (imgp->object != NULL)
478 vm_object_reference(imgp->object);
480 error = exec_map_first_page(imgp);
482 goto exec_fail_dealloc;
484 imgp->proc->p_osrel = 0;
485 imgp->proc->p_fctl0 = 0;
488 * Implement image setuid/setgid.
490 * Determine new credentials before attempting image activators
491 * so that it can be used by process_exec handlers to determine
492 * credential/setid changes.
494 * Don't honor setuid/setgid if the filesystem prohibits it or if
495 * the process is being traced.
497 * We disable setuid/setgid/etc in capability mode on the basis
498 * that most setugid applications are not written with that
499 * environment in mind, and will therefore almost certainly operate
500 * incorrectly. In principle there's no reason that setugid
501 * applications might not be useful in capability mode, so we may want
502 * to reconsider this conservative design choice in the future.
504 * XXXMAC: For the time being, use NOSUID to also prohibit
505 * transitions on the file system.
507 credential_changing = 0;
508 credential_changing |= (attr.va_mode & S_ISUID) &&
509 oldcred->cr_uid != attr.va_uid;
510 credential_changing |= (attr.va_mode & S_ISGID) &&
511 oldcred->cr_gid != attr.va_gid;
513 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
514 interpvplabel, imgp);
515 credential_changing |= will_transition;
518 /* Don't inherit PROC_PDEATHSIG_CTL value if setuid/setgid. */
519 if (credential_changing)
520 imgp->proc->p_pdeathsig = 0;
522 if (credential_changing &&
523 #ifdef CAPABILITY_MODE
524 ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
526 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
527 (p->p_flag & P_TRACED) == 0) {
528 imgp->credential_setid = true;
529 VOP_UNLOCK(imgp->vp);
530 imgp->newcred = crdup(oldcred);
531 if (attr.va_mode & S_ISUID) {
532 euip = uifind(attr.va_uid);
533 change_euid(imgp->newcred, euip);
535 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
536 if (attr.va_mode & S_ISGID)
537 change_egid(imgp->newcred, attr.va_gid);
539 * Implement correct POSIX saved-id behavior.
541 * XXXMAC: Note that the current logic will save the
542 * uid and gid if a MAC domain transition occurs, even
543 * though maybe it shouldn't.
545 change_svuid(imgp->newcred, imgp->newcred->cr_uid);
546 change_svgid(imgp->newcred, imgp->newcred->cr_gid);
549 * Implement correct POSIX saved-id behavior.
551 * XXX: It's not clear that the existing behavior is
552 * POSIX-compliant. A number of sources indicate that the
553 * saved uid/gid should only be updated if the new ruid is
554 * not equal to the old ruid, or the new euid is not equal
555 * to the old euid and the new euid is not equal to the old
556 * ruid. The FreeBSD code always updates the saved uid/gid.
557 * Also, this code uses the new (replaced) euid and egid as
558 * the source, which may or may not be the right ones to use.
560 if (oldcred->cr_svuid != oldcred->cr_uid ||
561 oldcred->cr_svgid != oldcred->cr_gid) {
562 VOP_UNLOCK(imgp->vp);
563 imgp->newcred = crdup(oldcred);
564 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
565 change_svuid(imgp->newcred, imgp->newcred->cr_uid);
566 change_svgid(imgp->newcred, imgp->newcred->cr_gid);
569 /* The new credentials are installed into the process later. */
572 * Do the best to calculate the full path to the image file.
574 if (args->fname != NULL && args->fname[0] == '/')
575 imgp->execpath = args->fname;
577 VOP_UNLOCK(imgp->vp);
578 if (vn_fullpath(imgp->vp, &imgp->execpath, &imgp->freepath) != 0)
579 imgp->execpath = args->fname;
580 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
584 * If the current process has a special image activator it
585 * wants to try first, call it. For example, emulating shell
586 * scripts differently.
589 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
590 error = img_first(imgp);
593 * Loop through the list of image activators, calling each one.
594 * An activator returns -1 if there is no match, 0 on success,
595 * and an error otherwise.
597 for (i = 0; error == -1 && execsw[i]; ++i) {
598 if (execsw[i]->ex_imgact == NULL ||
599 execsw[i]->ex_imgact == img_first) {
602 error = (*execsw[i]->ex_imgact)(imgp);
608 goto exec_fail_dealloc;
612 * Special interpreter operation, cleanup and loop up to try to
613 * activate the interpreter.
615 if (imgp->interpreted) {
616 exec_unmap_first_page(imgp);
618 * The text reference needs to be removed for scripts.
619 * There is a short period before we determine that
620 * something is a script where text reference is active.
621 * The vnode lock is held over this entire period
622 * so nothing should illegitimately be blocked.
624 MPASS(imgp->textset);
625 VOP_UNSET_TEXT_CHECKED(newtextvp);
626 imgp->textset = false;
627 /* free name buffer and old vnode */
628 if (args->fname != NULL)
629 NDFREE(&nd, NDF_ONLY_PNBUF);
631 mac_execve_interpreter_enter(newtextvp, &interpvplabel);
634 VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td);
638 vm_object_deallocate(imgp->object);
640 imgp->credential_setid = false;
641 if (imgp->newcred != NULL) {
642 crfree(imgp->newcred);
643 imgp->newcred = NULL;
645 imgp->execpath = NULL;
646 free(imgp->freepath, M_TEMP);
647 imgp->freepath = NULL;
648 /* set new name to that of the interpreter */
649 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW |
650 SAVENAME, UIO_SYSSPACE, imgp->interpreter_name, td);
651 args->fname = imgp->interpreter_name;
656 * NB: We unlock the vnode here because it is believed that none
657 * of the sv_copyout_strings/sv_fixup operations require the vnode.
659 VOP_UNLOCK(imgp->vp);
661 if (disallow_high_osrel &&
662 P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
664 uprintf("Osrel %d for image %s too high\n", p->p_osrel,
665 imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
666 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
667 goto exec_fail_dealloc;
670 /* ABI enforces the use of Capsicum. Switch into capabilities mode. */
671 if (SV_PROC_FLAG(p, SV_CAPSICUM))
672 sys_cap_enter(td, NULL);
675 * Copy out strings (args and env) and initialize stack base.
677 error = (*p->p_sysent->sv_copyout_strings)(imgp, &stack_base);
679 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
680 goto exec_fail_dealloc;
686 error = (*p->p_sysent->sv_fixup)(&stack_base, imgp);
688 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
689 goto exec_fail_dealloc;
692 if (args->fdp != NULL) {
693 /* Install a brand new file descriptor table. */
694 fdinstall_remapped(td, args->fdp);
698 * Keep on using the existing file descriptor table. For
699 * security and other reasons, the file descriptor table
700 * cannot be shared after an exec.
704 /* close files on exec */
709 * Malloc things before we need locks.
711 i = exec_args_get_begin_envv(imgp->args) - imgp->args->begin_argv;
712 /* Cache arguments if they fit inside our allowance */
713 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
714 newargs = pargs_alloc(i);
715 bcopy(imgp->args->begin_argv, newargs->ar_args, i);
719 * For security and other reasons, signal handlers cannot
720 * be shared after an exec. The new process gets a copy of the old
721 * handlers. In execsigs(), the new process will have its signals
724 if (sigacts_shared(p->p_sigacts)) {
725 oldsigacts = p->p_sigacts;
726 newsigacts = sigacts_alloc();
727 sigacts_copy(newsigacts, oldsigacts);
730 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
734 p->p_sigacts = newsigacts;
738 /* reset caught signals */
741 /* name this process - nameiexec(p, ndp) */
742 bzero(p->p_comm, sizeof(p->p_comm));
744 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
745 min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
746 else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0)
747 bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
748 bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
750 sched_clear_tdname(td);
754 * mark as execed, wakeup the process that vforked (if any) and tell
755 * it that it now has its own resources back
758 if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0)
759 p->p_flag2 &= ~P2_NOTRACE;
760 if ((p->p_flag2 & P2_STKGAP_DISABLE_EXEC) == 0)
761 p->p_flag2 &= ~P2_STKGAP_DISABLE;
762 if (p->p_flag & P_PPWAIT) {
763 p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
764 cv_broadcast(&p->p_pwait);
765 /* STOPs are no longer ignored, arrange for AST */
770 * Implement image setuid/setgid installation.
772 if (imgp->credential_setid) {
774 * Turn off syscall tracing for set-id programs, except for
775 * root. Record any set-id flags first to make sure that
776 * we do not regain any tracing during a possible block.
781 if (p->p_tracecred != NULL &&
782 priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED))
783 ktrprocexec(p, &tracecred, &tracevp);
786 * Close any file descriptors 0..2 that reference procfs,
787 * then make sure file descriptors 0..2 are in use.
789 * Both fdsetugidsafety() and fdcheckstd() may call functions
790 * taking sleepable locks, so temporarily drop our locks.
793 VOP_UNLOCK(imgp->vp);
795 error = fdcheckstd(td);
796 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
798 goto exec_fail_dealloc;
801 if (will_transition) {
802 mac_vnode_execve_transition(oldcred, imgp->newcred,
803 imgp->vp, interpvplabel, imgp);
807 if (oldcred->cr_uid == oldcred->cr_ruid &&
808 oldcred->cr_gid == oldcred->cr_rgid)
809 p->p_flag &= ~P_SUGID;
812 * Set the new credentials.
814 if (imgp->newcred != NULL) {
815 proc_set_cred(p, imgp->newcred);
821 * Store the vp for use in procfs. This vnode was referenced by namei
824 oldtextvp = p->p_textvp;
825 p->p_textvp = newtextvp;
829 * Tell the DTrace fasttrap provider about the exec if it
830 * has declared an interest.
832 if (dtrace_fasttrap_exec)
833 dtrace_fasttrap_exec(p);
837 * Notify others that we exec'd, and clear the P_INEXEC flag
838 * as we're now a bona fide freshly-execed process.
840 KNOTE_LOCKED(p->p_klist, NOTE_EXEC);
841 p->p_flag &= ~P_INEXEC;
843 /* clear "fork but no exec" flag, as we _are_ execing */
844 p->p_acflag &= ~AFORK;
847 * Free any previous argument cache and replace it with
848 * the new argument cache, if any.
858 * Check if system-wide sampling is in effect or if the
859 * current process is using PMCs. If so, do exec() time
860 * processing. This processing needs to happen AFTER the
861 * P_INEXEC flag is cleared.
863 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
864 VOP_UNLOCK(imgp->vp);
865 pe.pm_credentialschanged = credential_changing;
866 pe.pm_entryaddr = imgp->entry_addr;
868 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
869 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
873 /* Set values passed into the program in registers. */
874 (*p->p_sysent->sv_setregs)(td, imgp, stack_base);
876 VOP_MMAPPED(imgp->vp);
878 SDT_PROBE1(proc, , , exec__success, args->fname);
882 p->p_osrel = orig_osrel;
883 p->p_fctl0 = orig_fctl0;
886 if (imgp->firstpage != NULL)
887 exec_unmap_first_page(imgp);
889 if (imgp->vp != NULL) {
891 NDFREE(&nd, NDF_ONLY_PNBUF);
893 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
895 VOP_UNSET_TEXT_CHECKED(imgp->vp);
899 VOP_UNLOCK(imgp->vp);
902 if (imgp->object != NULL)
903 vm_object_deallocate(imgp->object);
905 free(imgp->freepath, M_TEMP);
908 if (p->p_ptevents & PTRACE_EXEC) {
910 if (p->p_ptevents & PTRACE_EXEC)
911 td->td_dbgflags |= TDB_EXEC;
916 /* we're done here, clear P_INEXEC */
918 p->p_flag &= ~P_INEXEC;
921 SDT_PROBE1(proc, , , exec__failure, error);
924 if (imgp->newcred != NULL && oldcred != NULL)
925 crfree(imgp->newcred);
928 mac_execve_exit(imgp);
929 mac_execve_interpreter_exit(interpvplabel);
931 exec_free_args(args);
934 * Handle deferred decrement of ref counts.
936 if (oldtextvp != NULL)
941 if (tracecred != NULL)
946 if (oldsigacts != NULL)
947 sigacts_free(oldsigacts);
951 if (error && imgp->vmspace_destroyed) {
952 /* sorry, no more process anymore. exit gracefully */
953 exec_cleanup(td, oldvmspace);
954 exit1(td, 0, SIGABRT);
964 * We don't want cpu_set_syscall_retval() to overwrite any of
965 * the register values put in place by exec_setregs().
966 * Implementations of cpu_set_syscall_retval() will leave
967 * registers unmodified when returning EJUSTRETURN.
969 return (error == 0 ? EJUSTRETURN : error);
973 exec_cleanup(struct thread *td, struct vmspace *oldvmspace)
975 if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
976 KASSERT(td->td_proc->p_vmspace != oldvmspace,
977 ("oldvmspace still used"));
978 vmspace_free(oldvmspace);
979 td->td_pflags &= ~TDP_EXECVMSPC;
984 exec_map_first_page(struct image_params *imgp)
990 if (imgp->firstpage != NULL)
991 exec_unmap_first_page(imgp);
993 object = imgp->vp->v_object;
996 #if VM_NRESERVLEVEL > 0
997 if ((object->flags & OBJ_COLORED) == 0) {
998 VM_OBJECT_WLOCK(object);
999 vm_object_color(object, 0);
1000 VM_OBJECT_WUNLOCK(object);
1003 error = vm_page_grab_valid_unlocked(&m, object, 0,
1004 VM_ALLOC_COUNT(VM_INITIAL_PAGEIN) |
1005 VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED);
1007 if (error != VM_PAGER_OK)
1009 imgp->firstpage = sf_buf_alloc(m, 0);
1010 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
1016 exec_unmap_first_page(struct image_params *imgp)
1020 if (imgp->firstpage != NULL) {
1021 m = sf_buf_page(imgp->firstpage);
1022 sf_buf_free(imgp->firstpage);
1023 imgp->firstpage = NULL;
1024 vm_page_unwire(m, PQ_ACTIVE);
1029 * Destroy old address space, and allocate a new stack.
1030 * The new stack is only sgrowsiz large because it is grown
1031 * automatically on a page fault.
1034 exec_new_vmspace(struct image_params *imgp, struct sysentvec *sv)
1037 struct proc *p = imgp->proc;
1038 struct vmspace *vmspace = p->p_vmspace;
1039 struct thread *td = curthread;
1041 struct rlimit rlim_stack;
1042 vm_offset_t sv_minuser, stack_addr;
1046 imgp->vmspace_destroyed = 1;
1049 sigfastblock_clear(td);
1051 /* May be called with Giant held */
1052 EVENTHANDLER_DIRECT_INVOKE(process_exec, p, imgp);
1055 * Blow away entire process VM, if address space not shared,
1056 * otherwise, create a new VM space so that other threads are
1059 map = &vmspace->vm_map;
1061 sv_minuser = sv->sv_minuser;
1063 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1064 if (refcount_load(&vmspace->vm_refcnt) == 1 &&
1065 vm_map_min(map) == sv_minuser &&
1066 vm_map_max(map) == sv->sv_maxuser &&
1067 cpu_exec_vmspace_reuse(p, map)) {
1069 pmap_remove_pages(vmspace_pmap(vmspace));
1070 vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1072 * An exec terminates mlockall(MCL_FUTURE), ASLR state
1073 * must be re-evaluated.
1076 vm_map_modflags(map, 0, MAP_WIREFUTURE | MAP_ASLR |
1080 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1083 vmspace = p->p_vmspace;
1084 map = &vmspace->vm_map;
1086 map->flags |= imgp->map_flags;
1088 /* Map a shared page */
1089 obj = sv->sv_shared_page_obj;
1091 vm_object_reference(obj);
1092 error = vm_map_fixed(map, obj, 0,
1093 sv->sv_shared_page_base, sv->sv_shared_page_len,
1094 VM_PROT_READ | VM_PROT_EXECUTE,
1095 VM_PROT_READ | VM_PROT_EXECUTE,
1096 MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1097 if (error != KERN_SUCCESS) {
1098 vm_object_deallocate(obj);
1099 return (vm_mmap_to_errno(error));
1103 /* Allocate a new stack */
1104 if (imgp->stack_sz != 0) {
1105 ssiz = trunc_page(imgp->stack_sz);
1107 lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack);
1109 if (ssiz > rlim_stack.rlim_max)
1110 ssiz = rlim_stack.rlim_max;
1111 if (ssiz > rlim_stack.rlim_cur) {
1112 rlim_stack.rlim_cur = ssiz;
1113 kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack);
1115 } else if (sv->sv_maxssiz != NULL) {
1116 ssiz = *sv->sv_maxssiz;
1120 imgp->eff_stack_sz = lim_cur(curthread, RLIMIT_STACK);
1121 if (ssiz < imgp->eff_stack_sz)
1122 imgp->eff_stack_sz = ssiz;
1123 stack_addr = sv->sv_usrstack - ssiz;
1124 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1125 obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1126 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1127 if (error != KERN_SUCCESS)
1128 return (vm_mmap_to_errno(error));
1131 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they
1132 * are still used to enforce the stack rlimit on the process stack.
1134 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1135 vmspace->vm_maxsaddr = (char *)stack_addr;
1141 * Copy out argument and environment strings from the old process address
1142 * space into the temporary string buffer.
1145 exec_copyin_args(struct image_args *args, const char *fname,
1146 enum uio_seg segflg, char **argv, char **envv)
1151 bzero(args, sizeof(*args));
1156 * Allocate demand-paged memory for the file name, argument, and
1157 * environment strings.
1159 error = exec_alloc_args(args);
1164 * Copy the file name.
1166 error = exec_args_add_fname(args, fname, segflg);
1171 * extract arguments first
1174 error = fueword(argv++, &arg);
1181 error = exec_args_add_arg(args, (char *)(uintptr_t)arg,
1188 * extract environment strings
1192 error = fueword(envv++, &env);
1199 error = exec_args_add_env(args,
1200 (char *)(uintptr_t)env, UIO_USERSPACE);
1209 exec_free_args(args);
1214 exec_copyin_data_fds(struct thread *td, struct image_args *args,
1215 const void *data, size_t datalen, const int *fds, size_t fdslen)
1217 struct filedesc *ofdp;
1222 memset(args, '\0', sizeof(*args));
1223 ofdp = td->td_proc->p_fd;
1224 if (datalen >= ARG_MAX || fdslen >= ofdp->fd_nfiles)
1226 error = exec_alloc_args(args);
1230 args->begin_argv = args->buf;
1231 args->stringspace = ARG_MAX;
1235 * Argument buffer has been provided. Copy it into the
1236 * kernel as a single string and add a terminating null
1239 error = copyin(data, args->begin_argv, datalen);
1242 args->begin_argv[datalen] = '\0';
1243 args->endp = args->begin_argv + datalen + 1;
1244 args->stringspace -= datalen + 1;
1247 * Traditional argument counting. Count the number of
1250 for (p = args->begin_argv; p < args->endp; ++p)
1254 /* No argument buffer provided. */
1255 args->endp = args->begin_argv;
1258 /* Create new file descriptor table. */
1259 kfds = malloc(fdslen * sizeof(int), M_TEMP, M_WAITOK);
1260 error = copyin(fds, kfds, fdslen * sizeof(int));
1265 error = fdcopy_remapped(ofdp, kfds, fdslen, &args->fdp);
1272 exec_free_args(args);
1276 struct exec_args_kva {
1279 SLIST_ENTRY(exec_args_kva) next;
1282 DPCPU_DEFINE_STATIC(struct exec_args_kva *, exec_args_kva);
1284 static SLIST_HEAD(, exec_args_kva) exec_args_kva_freelist;
1285 static struct mtx exec_args_kva_mtx;
1286 static u_int exec_args_gen;
1289 exec_prealloc_args_kva(void *arg __unused)
1291 struct exec_args_kva *argkva;
1294 SLIST_INIT(&exec_args_kva_freelist);
1295 mtx_init(&exec_args_kva_mtx, "exec args kva", NULL, MTX_DEF);
1296 for (i = 0; i < exec_map_entries; i++) {
1297 argkva = malloc(sizeof(*argkva), M_PARGS, M_WAITOK);
1298 argkva->addr = kmap_alloc_wait(exec_map, exec_map_entry_size);
1299 argkva->gen = exec_args_gen;
1300 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1303 SYSINIT(exec_args_kva, SI_SUB_EXEC, SI_ORDER_ANY, exec_prealloc_args_kva, NULL);
1306 exec_alloc_args_kva(void **cookie)
1308 struct exec_args_kva *argkva;
1310 argkva = (void *)atomic_readandclear_ptr(
1311 (uintptr_t *)DPCPU_PTR(exec_args_kva));
1312 if (argkva == NULL) {
1313 mtx_lock(&exec_args_kva_mtx);
1314 while ((argkva = SLIST_FIRST(&exec_args_kva_freelist)) == NULL)
1315 (void)mtx_sleep(&exec_args_kva_freelist,
1316 &exec_args_kva_mtx, 0, "execkva", 0);
1317 SLIST_REMOVE_HEAD(&exec_args_kva_freelist, next);
1318 mtx_unlock(&exec_args_kva_mtx);
1320 *(struct exec_args_kva **)cookie = argkva;
1321 return (argkva->addr);
1325 exec_release_args_kva(struct exec_args_kva *argkva, u_int gen)
1329 base = argkva->addr;
1330 if (argkva->gen != gen) {
1331 (void)vm_map_madvise(exec_map, base, base + exec_map_entry_size,
1335 if (!atomic_cmpset_ptr((uintptr_t *)DPCPU_PTR(exec_args_kva),
1336 (uintptr_t)NULL, (uintptr_t)argkva)) {
1337 mtx_lock(&exec_args_kva_mtx);
1338 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1339 wakeup_one(&exec_args_kva_freelist);
1340 mtx_unlock(&exec_args_kva_mtx);
1345 exec_free_args_kva(void *cookie)
1348 exec_release_args_kva(cookie, exec_args_gen);
1352 exec_args_kva_lowmem(void *arg __unused)
1354 SLIST_HEAD(, exec_args_kva) head;
1355 struct exec_args_kva *argkva;
1359 gen = atomic_fetchadd_int(&exec_args_gen, 1) + 1;
1362 * Force an madvise of each KVA range. Any currently allocated ranges
1363 * will have MADV_FREE applied once they are freed.
1366 mtx_lock(&exec_args_kva_mtx);
1367 SLIST_SWAP(&head, &exec_args_kva_freelist, exec_args_kva);
1368 mtx_unlock(&exec_args_kva_mtx);
1369 while ((argkva = SLIST_FIRST(&head)) != NULL) {
1370 SLIST_REMOVE_HEAD(&head, next);
1371 exec_release_args_kva(argkva, gen);
1375 argkva = (void *)atomic_readandclear_ptr(
1376 (uintptr_t *)DPCPU_ID_PTR(i, exec_args_kva));
1378 exec_release_args_kva(argkva, gen);
1381 EVENTHANDLER_DEFINE(vm_lowmem, exec_args_kva_lowmem, NULL,
1382 EVENTHANDLER_PRI_ANY);
1385 * Allocate temporary demand-paged, zero-filled memory for the file name,
1386 * argument, and environment strings.
1389 exec_alloc_args(struct image_args *args)
1392 args->buf = (char *)exec_alloc_args_kva(&args->bufkva);
1397 exec_free_args(struct image_args *args)
1400 if (args->buf != NULL) {
1401 exec_free_args_kva(args->bufkva);
1404 if (args->fname_buf != NULL) {
1405 free(args->fname_buf, M_TEMP);
1406 args->fname_buf = NULL;
1408 if (args->fdp != NULL)
1409 fdescfree_remapped(args->fdp);
1413 * A set to functions to fill struct image args.
1415 * NOTE: exec_args_add_fname() must be called (possibly with a NULL
1416 * fname) before the other functions. All exec_args_add_arg() calls must
1417 * be made before any exec_args_add_env() calls. exec_args_adjust_args()
1418 * may be called any time after exec_args_add_fname().
1420 * exec_args_add_fname() - install path to be executed
1421 * exec_args_add_arg() - append an argument string
1422 * exec_args_add_env() - append an env string
1423 * exec_args_adjust_args() - adjust location of the argument list to
1424 * allow new arguments to be prepended
1427 exec_args_add_fname(struct image_args *args, const char *fname,
1428 enum uio_seg segflg)
1433 KASSERT(args->fname == NULL, ("fname already appended"));
1434 KASSERT(args->endp == NULL, ("already appending to args"));
1436 if (fname != NULL) {
1437 args->fname = args->buf;
1438 error = segflg == UIO_SYSSPACE ?
1439 copystr(fname, args->fname, PATH_MAX, &length) :
1440 copyinstr(fname, args->fname, PATH_MAX, &length);
1442 return (error == ENAMETOOLONG ? E2BIG : error);
1446 /* Set up for _arg_*()/_env_*() */
1447 args->endp = args->buf + length;
1448 /* begin_argv must be set and kept updated */
1449 args->begin_argv = args->endp;
1450 KASSERT(exec_map_entry_size - length >= ARG_MAX,
1451 ("too little space remaining for arguments %zu < %zu",
1452 exec_map_entry_size - length, (size_t)ARG_MAX));
1453 args->stringspace = ARG_MAX;
1459 exec_args_add_str(struct image_args *args, const char *str,
1460 enum uio_seg segflg, int *countp)
1465 KASSERT(args->endp != NULL, ("endp not initialized"));
1466 KASSERT(args->begin_argv != NULL, ("begin_argp not initialized"));
1468 error = (segflg == UIO_SYSSPACE) ?
1469 copystr(str, args->endp, args->stringspace, &length) :
1470 copyinstr(str, args->endp, args->stringspace, &length);
1472 return (error == ENAMETOOLONG ? E2BIG : error);
1473 args->stringspace -= length;
1474 args->endp += length;
1481 exec_args_add_arg(struct image_args *args, const char *argp,
1482 enum uio_seg segflg)
1485 KASSERT(args->envc == 0, ("appending args after env"));
1487 return (exec_args_add_str(args, argp, segflg, &args->argc));
1491 exec_args_add_env(struct image_args *args, const char *envp,
1492 enum uio_seg segflg)
1495 if (args->envc == 0)
1496 args->begin_envv = args->endp;
1498 return (exec_args_add_str(args, envp, segflg, &args->envc));
1502 exec_args_adjust_args(struct image_args *args, size_t consume, ssize_t extend)
1506 KASSERT(args->endp != NULL, ("endp not initialized"));
1507 KASSERT(args->begin_argv != NULL, ("begin_argp not initialized"));
1509 offset = extend - consume;
1510 if (args->stringspace < offset)
1512 memmove(args->begin_argv + extend, args->begin_argv + consume,
1513 args->endp - args->begin_argv + consume);
1515 args->begin_envv += offset;
1516 args->endp += offset;
1517 args->stringspace -= offset;
1522 exec_args_get_begin_envv(struct image_args *args)
1525 KASSERT(args->endp != NULL, ("endp not initialized"));
1528 return (args->begin_envv);
1529 return (args->endp);
1533 * Copy strings out to the new process address space, constructing new arg
1534 * and env vector tables. Return a pointer to the base so that it can be used
1535 * as the initial stack pointer.
1538 exec_copyout_strings(struct image_params *imgp, uintptr_t *stack_base)
1543 uintptr_t destp, ustringp;
1544 struct ps_strings *arginfo;
1546 size_t execpath_len;
1547 int error, szsigcode, szps;
1548 char canary[sizeof(long) * 8];
1550 szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1552 * Calculate string base and vector table pointers.
1553 * Also deal with signal trampoline code for this exec type.
1555 if (imgp->execpath != NULL && imgp->auxargs != NULL)
1556 execpath_len = strlen(imgp->execpath) + 1;
1561 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1562 imgp->ps_strings = arginfo;
1563 if (p->p_sysent->sv_sigcode_base == 0) {
1564 if (p->p_sysent->sv_szsigcode != NULL)
1565 szsigcode = *(p->p_sysent->sv_szsigcode);
1567 destp = (uintptr_t)arginfo;
1572 if (szsigcode != 0) {
1574 destp = rounddown2(destp, sizeof(void *));
1575 error = copyout(p->p_sysent->sv_sigcode, (void *)destp,
1582 * Copy the image path for the rtld.
1584 if (execpath_len != 0) {
1585 destp -= execpath_len;
1586 destp = rounddown2(destp, sizeof(void *));
1587 imgp->execpathp = (void *)destp;
1588 error = copyout(imgp->execpath, imgp->execpathp, execpath_len);
1594 * Prepare the canary for SSP.
1596 arc4rand(canary, sizeof(canary), 0);
1597 destp -= sizeof(canary);
1598 imgp->canary = (void *)destp;
1599 error = copyout(canary, imgp->canary, sizeof(canary));
1602 imgp->canarylen = sizeof(canary);
1605 * Prepare the pagesizes array.
1608 destp = rounddown2(destp, sizeof(void *));
1609 imgp->pagesizes = (void *)destp;
1610 error = copyout(pagesizes, imgp->pagesizes, szps);
1613 imgp->pagesizeslen = szps;
1616 * Allocate room for the argument and environment strings.
1618 destp -= ARG_MAX - imgp->args->stringspace;
1619 destp = rounddown2(destp, sizeof(void *));
1622 if (imgp->sysent->sv_stackgap != NULL)
1623 imgp->sysent->sv_stackgap(imgp, &destp);
1625 if (imgp->auxargs) {
1627 * Allocate room on the stack for the ELF auxargs
1628 * array. It has up to AT_COUNT entries.
1630 destp -= AT_COUNT * sizeof(Elf_Auxinfo);
1631 destp = rounddown2(destp, sizeof(void *));
1634 vectp = (char **)destp;
1637 * Allocate room for the argv[] and env vectors including the
1638 * terminating NULL pointers.
1640 vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
1643 * vectp also becomes our initial stack base
1645 *stack_base = (uintptr_t)vectp;
1647 stringp = imgp->args->begin_argv;
1648 argc = imgp->args->argc;
1649 envc = imgp->args->envc;
1652 * Copy out strings - arguments and environment.
1654 error = copyout(stringp, (void *)ustringp,
1655 ARG_MAX - imgp->args->stringspace);
1660 * Fill in "ps_strings" struct for ps, w, etc.
1663 if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 ||
1664 suword32(&arginfo->ps_nargvstr, argc) != 0)
1668 * Fill in argument portion of vector table.
1670 for (; argc > 0; --argc) {
1671 if (suword(vectp++, ustringp) != 0)
1673 while (*stringp++ != 0)
1678 /* a null vector table pointer separates the argp's from the envp's */
1679 if (suword(vectp++, 0) != 0)
1683 if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 ||
1684 suword32(&arginfo->ps_nenvstr, envc) != 0)
1688 * Fill in environment portion of vector table.
1690 for (; envc > 0; --envc) {
1691 if (suword(vectp++, ustringp) != 0)
1693 while (*stringp++ != 0)
1698 /* end of vector table is a null pointer */
1699 if (suword(vectp, 0) != 0)
1702 if (imgp->auxargs) {
1704 error = imgp->sysent->sv_copyout_auxargs(imgp,
1714 * Check permissions of file to execute.
1715 * Called with imgp->vp locked.
1716 * Return 0 for success or error code on failure.
1719 exec_check_permissions(struct image_params *imgp)
1721 struct vnode *vp = imgp->vp;
1722 struct vattr *attr = imgp->attr;
1728 /* Get file attributes */
1729 error = VOP_GETATTR(vp, attr, td->td_ucred);
1734 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1740 * 1) Check if file execution is disabled for the filesystem that
1741 * this file resides on.
1742 * 2) Ensure that at least one execute bit is on. Otherwise, a
1743 * privileged user will always succeed, and we don't want this
1744 * to happen unless the file really is executable.
1745 * 3) Ensure that the file is a regular file.
1747 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1748 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1749 (attr->va_type != VREG))
1753 * Zero length files can't be exec'd
1755 if (attr->va_size == 0)
1759 * Check for execute permission to file based on current credentials.
1761 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1766 * Check number of open-for-writes on the file and deny execution
1769 * Add a text reference now so no one can write to the
1770 * executable while we're activating it.
1772 * Remember if this was set before and unset it in case this is not
1773 * actually an executable image.
1775 error = VOP_SET_TEXT(vp);
1778 imgp->textset = true;
1781 * Call filesystem specific open routine (which does nothing in the
1784 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1791 * Exec handler registration
1794 exec_register(const struct execsw *execsw_arg)
1796 const struct execsw **es, **xs, **newexecsw;
1797 u_int count = 2; /* New slot and trailing NULL */
1800 for (es = execsw; *es; es++)
1802 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1805 for (es = execsw; *es; es++)
1810 free(execsw, M_TEMP);
1816 exec_unregister(const struct execsw *execsw_arg)
1818 const struct execsw **es, **xs, **newexecsw;
1822 panic("unregister with no handlers left?\n");
1824 for (es = execsw; *es; es++) {
1825 if (*es == execsw_arg)
1830 for (es = execsw; *es; es++)
1831 if (*es != execsw_arg)
1833 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1835 for (es = execsw; *es; es++)
1836 if (*es != execsw_arg)
1840 free(execsw, M_TEMP);