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.
703 /* close files on exec */
708 * Malloc things before we need locks.
710 i = exec_args_get_begin_envv(imgp->args) - imgp->args->begin_argv;
711 /* Cache arguments if they fit inside our allowance */
712 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
713 newargs = pargs_alloc(i);
714 bcopy(imgp->args->begin_argv, newargs->ar_args, i);
718 * For security and other reasons, signal handlers cannot
719 * be shared after an exec. The new process gets a copy of the old
720 * handlers. In execsigs(), the new process will have its signals
723 if (sigacts_shared(p->p_sigacts)) {
724 oldsigacts = p->p_sigacts;
725 newsigacts = sigacts_alloc();
726 sigacts_copy(newsigacts, oldsigacts);
729 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
733 p->p_sigacts = newsigacts;
737 /* reset caught signals */
740 /* name this process - nameiexec(p, ndp) */
741 bzero(p->p_comm, sizeof(p->p_comm));
743 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
744 min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
745 else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0)
746 bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
747 bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
749 sched_clear_tdname(td);
753 * mark as execed, wakeup the process that vforked (if any) and tell
754 * it that it now has its own resources back
757 if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0)
758 p->p_flag2 &= ~P2_NOTRACE;
759 if ((p->p_flag2 & P2_STKGAP_DISABLE_EXEC) == 0)
760 p->p_flag2 &= ~P2_STKGAP_DISABLE;
761 if (p->p_flag & P_PPWAIT) {
762 p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
763 cv_broadcast(&p->p_pwait);
764 /* STOPs are no longer ignored, arrange for AST */
769 * Implement image setuid/setgid installation.
771 if (imgp->credential_setid) {
773 * Turn off syscall tracing for set-id programs, except for
774 * root. Record any set-id flags first to make sure that
775 * we do not regain any tracing during a possible block.
780 if (p->p_tracecred != NULL &&
781 priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED))
782 ktrprocexec(p, &tracecred, &tracevp);
785 * Close any file descriptors 0..2 that reference procfs,
786 * then make sure file descriptors 0..2 are in use.
788 * Both fdsetugidsafety() and fdcheckstd() may call functions
789 * taking sleepable locks, so temporarily drop our locks.
792 VOP_UNLOCK(imgp->vp);
794 error = fdcheckstd(td);
795 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
797 goto exec_fail_dealloc;
800 if (will_transition) {
801 mac_vnode_execve_transition(oldcred, imgp->newcred,
802 imgp->vp, interpvplabel, imgp);
806 if (oldcred->cr_uid == oldcred->cr_ruid &&
807 oldcred->cr_gid == oldcred->cr_rgid)
808 p->p_flag &= ~P_SUGID;
811 * Set the new credentials.
813 if (imgp->newcred != NULL) {
814 proc_set_cred(p, imgp->newcred);
820 * Store the vp for use in procfs. This vnode was referenced by namei
823 oldtextvp = p->p_textvp;
824 p->p_textvp = newtextvp;
828 * Tell the DTrace fasttrap provider about the exec if it
829 * has declared an interest.
831 if (dtrace_fasttrap_exec)
832 dtrace_fasttrap_exec(p);
836 * Notify others that we exec'd, and clear the P_INEXEC flag
837 * as we're now a bona fide freshly-execed process.
839 KNOTE_LOCKED(p->p_klist, NOTE_EXEC);
840 p->p_flag &= ~P_INEXEC;
842 /* clear "fork but no exec" flag, as we _are_ execing */
843 p->p_acflag &= ~AFORK;
846 * Free any previous argument cache and replace it with
847 * the new argument cache, if any.
857 * Check if system-wide sampling is in effect or if the
858 * current process is using PMCs. If so, do exec() time
859 * processing. This processing needs to happen AFTER the
860 * P_INEXEC flag is cleared.
862 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
863 VOP_UNLOCK(imgp->vp);
864 pe.pm_credentialschanged = credential_changing;
865 pe.pm_entryaddr = imgp->entry_addr;
867 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
868 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
872 /* Set values passed into the program in registers. */
873 (*p->p_sysent->sv_setregs)(td, imgp, stack_base);
875 VOP_MMAPPED(imgp->vp);
877 SDT_PROBE1(proc, , , exec__success, args->fname);
881 p->p_osrel = orig_osrel;
882 p->p_fctl0 = orig_fctl0;
885 if (imgp->firstpage != NULL)
886 exec_unmap_first_page(imgp);
888 if (imgp->vp != NULL) {
890 NDFREE(&nd, NDF_ONLY_PNBUF);
892 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
894 VOP_UNSET_TEXT_CHECKED(imgp->vp);
898 VOP_UNLOCK(imgp->vp);
901 if (imgp->object != NULL)
902 vm_object_deallocate(imgp->object);
904 free(imgp->freepath, M_TEMP);
907 if (p->p_ptevents & PTRACE_EXEC) {
909 if (p->p_ptevents & PTRACE_EXEC)
910 td->td_dbgflags |= TDB_EXEC;
915 /* we're done here, clear P_INEXEC */
917 p->p_flag &= ~P_INEXEC;
920 SDT_PROBE1(proc, , , exec__failure, error);
923 if (imgp->newcred != NULL && oldcred != NULL)
924 crfree(imgp->newcred);
927 mac_execve_exit(imgp);
928 mac_execve_interpreter_exit(interpvplabel);
930 exec_free_args(args);
933 * Handle deferred decrement of ref counts.
935 if (oldtextvp != NULL)
940 if (tracecred != NULL)
945 if (oldsigacts != NULL)
946 sigacts_free(oldsigacts);
950 if (error && imgp->vmspace_destroyed) {
951 /* sorry, no more process anymore. exit gracefully */
952 exec_cleanup(td, oldvmspace);
953 exit1(td, 0, SIGABRT);
963 * We don't want cpu_set_syscall_retval() to overwrite any of
964 * the register values put in place by exec_setregs().
965 * Implementations of cpu_set_syscall_retval() will leave
966 * registers unmodified when returning EJUSTRETURN.
968 return (error == 0 ? EJUSTRETURN : error);
972 exec_cleanup(struct thread *td, struct vmspace *oldvmspace)
974 if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
975 KASSERT(td->td_proc->p_vmspace != oldvmspace,
976 ("oldvmspace still used"));
977 vmspace_free(oldvmspace);
978 td->td_pflags &= ~TDP_EXECVMSPC;
983 exec_map_first_page(struct image_params *imgp)
989 if (imgp->firstpage != NULL)
990 exec_unmap_first_page(imgp);
992 object = imgp->vp->v_object;
995 #if VM_NRESERVLEVEL > 0
996 if ((object->flags & OBJ_COLORED) == 0) {
997 VM_OBJECT_WLOCK(object);
998 vm_object_color(object, 0);
999 VM_OBJECT_WUNLOCK(object);
1002 error = vm_page_grab_valid_unlocked(&m, object, 0,
1003 VM_ALLOC_COUNT(VM_INITIAL_PAGEIN) |
1004 VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED);
1006 if (error != VM_PAGER_OK)
1008 imgp->firstpage = sf_buf_alloc(m, 0);
1009 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
1015 exec_unmap_first_page(struct image_params *imgp)
1019 if (imgp->firstpage != NULL) {
1020 m = sf_buf_page(imgp->firstpage);
1021 sf_buf_free(imgp->firstpage);
1022 imgp->firstpage = NULL;
1023 vm_page_unwire(m, PQ_ACTIVE);
1028 * Destroy old address space, and allocate a new stack.
1029 * The new stack is only sgrowsiz large because it is grown
1030 * automatically on a page fault.
1033 exec_new_vmspace(struct image_params *imgp, struct sysentvec *sv)
1036 struct proc *p = imgp->proc;
1037 struct vmspace *vmspace = p->p_vmspace;
1038 struct thread *td = curthread;
1040 struct rlimit rlim_stack;
1041 vm_offset_t sv_minuser, stack_addr;
1045 imgp->vmspace_destroyed = 1;
1048 sigfastblock_clear(td);
1050 /* May be called with Giant held */
1051 EVENTHANDLER_DIRECT_INVOKE(process_exec, p, imgp);
1054 * Blow away entire process VM, if address space not shared,
1055 * otherwise, create a new VM space so that other threads are
1058 map = &vmspace->vm_map;
1060 sv_minuser = sv->sv_minuser;
1062 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1063 if (refcount_load(&vmspace->vm_refcnt) == 1 &&
1064 vm_map_min(map) == sv_minuser &&
1065 vm_map_max(map) == sv->sv_maxuser &&
1066 cpu_exec_vmspace_reuse(p, map)) {
1068 pmap_remove_pages(vmspace_pmap(vmspace));
1069 vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1071 * An exec terminates mlockall(MCL_FUTURE), ASLR state
1072 * must be re-evaluated.
1075 vm_map_modflags(map, 0, MAP_WIREFUTURE | MAP_ASLR |
1079 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1082 vmspace = p->p_vmspace;
1083 map = &vmspace->vm_map;
1085 map->flags |= imgp->map_flags;
1087 /* Map a shared page */
1088 obj = sv->sv_shared_page_obj;
1090 vm_object_reference(obj);
1091 error = vm_map_fixed(map, obj, 0,
1092 sv->sv_shared_page_base, sv->sv_shared_page_len,
1093 VM_PROT_READ | VM_PROT_EXECUTE,
1094 VM_PROT_READ | VM_PROT_EXECUTE,
1095 MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1096 if (error != KERN_SUCCESS) {
1097 vm_object_deallocate(obj);
1098 return (vm_mmap_to_errno(error));
1102 /* Allocate a new stack */
1103 if (imgp->stack_sz != 0) {
1104 ssiz = trunc_page(imgp->stack_sz);
1106 lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack);
1108 if (ssiz > rlim_stack.rlim_max)
1109 ssiz = rlim_stack.rlim_max;
1110 if (ssiz > rlim_stack.rlim_cur) {
1111 rlim_stack.rlim_cur = ssiz;
1112 kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack);
1114 } else if (sv->sv_maxssiz != NULL) {
1115 ssiz = *sv->sv_maxssiz;
1119 imgp->eff_stack_sz = lim_cur(curthread, RLIMIT_STACK);
1120 if (ssiz < imgp->eff_stack_sz)
1121 imgp->eff_stack_sz = ssiz;
1122 stack_addr = sv->sv_usrstack - ssiz;
1123 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1124 obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1125 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1126 if (error != KERN_SUCCESS)
1127 return (vm_mmap_to_errno(error));
1130 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they
1131 * are still used to enforce the stack rlimit on the process stack.
1133 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1134 vmspace->vm_maxsaddr = (char *)stack_addr;
1140 * Copy out argument and environment strings from the old process address
1141 * space into the temporary string buffer.
1144 exec_copyin_args(struct image_args *args, const char *fname,
1145 enum uio_seg segflg, char **argv, char **envv)
1150 bzero(args, sizeof(*args));
1155 * Allocate demand-paged memory for the file name, argument, and
1156 * environment strings.
1158 error = exec_alloc_args(args);
1163 * Copy the file name.
1165 error = exec_args_add_fname(args, fname, segflg);
1170 * extract arguments first
1173 error = fueword(argv++, &arg);
1180 error = exec_args_add_arg(args, (char *)(uintptr_t)arg,
1187 * extract environment strings
1191 error = fueword(envv++, &env);
1198 error = exec_args_add_env(args,
1199 (char *)(uintptr_t)env, UIO_USERSPACE);
1208 exec_free_args(args);
1213 exec_copyin_data_fds(struct thread *td, struct image_args *args,
1214 const void *data, size_t datalen, const int *fds, size_t fdslen)
1216 struct filedesc *ofdp;
1221 memset(args, '\0', sizeof(*args));
1222 ofdp = td->td_proc->p_fd;
1223 if (datalen >= ARG_MAX || fdslen >= ofdp->fd_nfiles)
1225 error = exec_alloc_args(args);
1229 args->begin_argv = args->buf;
1230 args->stringspace = ARG_MAX;
1234 * Argument buffer has been provided. Copy it into the
1235 * kernel as a single string and add a terminating null
1238 error = copyin(data, args->begin_argv, datalen);
1241 args->begin_argv[datalen] = '\0';
1242 args->endp = args->begin_argv + datalen + 1;
1243 args->stringspace -= datalen + 1;
1246 * Traditional argument counting. Count the number of
1249 for (p = args->begin_argv; p < args->endp; ++p)
1253 /* No argument buffer provided. */
1254 args->endp = args->begin_argv;
1257 /* Create new file descriptor table. */
1258 kfds = malloc(fdslen * sizeof(int), M_TEMP, M_WAITOK);
1259 error = copyin(fds, kfds, fdslen * sizeof(int));
1264 error = fdcopy_remapped(ofdp, kfds, fdslen, &args->fdp);
1271 exec_free_args(args);
1275 struct exec_args_kva {
1278 SLIST_ENTRY(exec_args_kva) next;
1281 DPCPU_DEFINE_STATIC(struct exec_args_kva *, exec_args_kva);
1283 static SLIST_HEAD(, exec_args_kva) exec_args_kva_freelist;
1284 static struct mtx exec_args_kva_mtx;
1285 static u_int exec_args_gen;
1288 exec_prealloc_args_kva(void *arg __unused)
1290 struct exec_args_kva *argkva;
1293 SLIST_INIT(&exec_args_kva_freelist);
1294 mtx_init(&exec_args_kva_mtx, "exec args kva", NULL, MTX_DEF);
1295 for (i = 0; i < exec_map_entries; i++) {
1296 argkva = malloc(sizeof(*argkva), M_PARGS, M_WAITOK);
1297 argkva->addr = kmap_alloc_wait(exec_map, exec_map_entry_size);
1298 argkva->gen = exec_args_gen;
1299 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1302 SYSINIT(exec_args_kva, SI_SUB_EXEC, SI_ORDER_ANY, exec_prealloc_args_kva, NULL);
1305 exec_alloc_args_kva(void **cookie)
1307 struct exec_args_kva *argkva;
1309 argkva = (void *)atomic_readandclear_ptr(
1310 (uintptr_t *)DPCPU_PTR(exec_args_kva));
1311 if (argkva == NULL) {
1312 mtx_lock(&exec_args_kva_mtx);
1313 while ((argkva = SLIST_FIRST(&exec_args_kva_freelist)) == NULL)
1314 (void)mtx_sleep(&exec_args_kva_freelist,
1315 &exec_args_kva_mtx, 0, "execkva", 0);
1316 SLIST_REMOVE_HEAD(&exec_args_kva_freelist, next);
1317 mtx_unlock(&exec_args_kva_mtx);
1319 *(struct exec_args_kva **)cookie = argkva;
1320 return (argkva->addr);
1324 exec_release_args_kva(struct exec_args_kva *argkva, u_int gen)
1328 base = argkva->addr;
1329 if (argkva->gen != gen) {
1330 (void)vm_map_madvise(exec_map, base, base + exec_map_entry_size,
1334 if (!atomic_cmpset_ptr((uintptr_t *)DPCPU_PTR(exec_args_kva),
1335 (uintptr_t)NULL, (uintptr_t)argkva)) {
1336 mtx_lock(&exec_args_kva_mtx);
1337 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1338 wakeup_one(&exec_args_kva_freelist);
1339 mtx_unlock(&exec_args_kva_mtx);
1344 exec_free_args_kva(void *cookie)
1347 exec_release_args_kva(cookie, exec_args_gen);
1351 exec_args_kva_lowmem(void *arg __unused)
1353 SLIST_HEAD(, exec_args_kva) head;
1354 struct exec_args_kva *argkva;
1358 gen = atomic_fetchadd_int(&exec_args_gen, 1) + 1;
1361 * Force an madvise of each KVA range. Any currently allocated ranges
1362 * will have MADV_FREE applied once they are freed.
1365 mtx_lock(&exec_args_kva_mtx);
1366 SLIST_SWAP(&head, &exec_args_kva_freelist, exec_args_kva);
1367 mtx_unlock(&exec_args_kva_mtx);
1368 while ((argkva = SLIST_FIRST(&head)) != NULL) {
1369 SLIST_REMOVE_HEAD(&head, next);
1370 exec_release_args_kva(argkva, gen);
1374 argkva = (void *)atomic_readandclear_ptr(
1375 (uintptr_t *)DPCPU_ID_PTR(i, exec_args_kva));
1377 exec_release_args_kva(argkva, gen);
1380 EVENTHANDLER_DEFINE(vm_lowmem, exec_args_kva_lowmem, NULL,
1381 EVENTHANDLER_PRI_ANY);
1384 * Allocate temporary demand-paged, zero-filled memory for the file name,
1385 * argument, and environment strings.
1388 exec_alloc_args(struct image_args *args)
1391 args->buf = (char *)exec_alloc_args_kva(&args->bufkva);
1396 exec_free_args(struct image_args *args)
1399 if (args->buf != NULL) {
1400 exec_free_args_kva(args->bufkva);
1403 if (args->fname_buf != NULL) {
1404 free(args->fname_buf, M_TEMP);
1405 args->fname_buf = NULL;
1407 if (args->fdp != NULL)
1408 fdescfree_remapped(args->fdp);
1412 * A set to functions to fill struct image args.
1414 * NOTE: exec_args_add_fname() must be called (possibly with a NULL
1415 * fname) before the other functions. All exec_args_add_arg() calls must
1416 * be made before any exec_args_add_env() calls. exec_args_adjust_args()
1417 * may be called any time after exec_args_add_fname().
1419 * exec_args_add_fname() - install path to be executed
1420 * exec_args_add_arg() - append an argument string
1421 * exec_args_add_env() - append an env string
1422 * exec_args_adjust_args() - adjust location of the argument list to
1423 * allow new arguments to be prepended
1426 exec_args_add_fname(struct image_args *args, const char *fname,
1427 enum uio_seg segflg)
1432 KASSERT(args->fname == NULL, ("fname already appended"));
1433 KASSERT(args->endp == NULL, ("already appending to args"));
1435 if (fname != NULL) {
1436 args->fname = args->buf;
1437 error = segflg == UIO_SYSSPACE ?
1438 copystr(fname, args->fname, PATH_MAX, &length) :
1439 copyinstr(fname, args->fname, PATH_MAX, &length);
1441 return (error == ENAMETOOLONG ? E2BIG : error);
1445 /* Set up for _arg_*()/_env_*() */
1446 args->endp = args->buf + length;
1447 /* begin_argv must be set and kept updated */
1448 args->begin_argv = args->endp;
1449 KASSERT(exec_map_entry_size - length >= ARG_MAX,
1450 ("too little space remaining for arguments %zu < %zu",
1451 exec_map_entry_size - length, (size_t)ARG_MAX));
1452 args->stringspace = ARG_MAX;
1458 exec_args_add_str(struct image_args *args, const char *str,
1459 enum uio_seg segflg, int *countp)
1464 KASSERT(args->endp != NULL, ("endp not initialized"));
1465 KASSERT(args->begin_argv != NULL, ("begin_argp not initialized"));
1467 error = (segflg == UIO_SYSSPACE) ?
1468 copystr(str, args->endp, args->stringspace, &length) :
1469 copyinstr(str, args->endp, args->stringspace, &length);
1471 return (error == ENAMETOOLONG ? E2BIG : error);
1472 args->stringspace -= length;
1473 args->endp += length;
1480 exec_args_add_arg(struct image_args *args, const char *argp,
1481 enum uio_seg segflg)
1484 KASSERT(args->envc == 0, ("appending args after env"));
1486 return (exec_args_add_str(args, argp, segflg, &args->argc));
1490 exec_args_add_env(struct image_args *args, const char *envp,
1491 enum uio_seg segflg)
1494 if (args->envc == 0)
1495 args->begin_envv = args->endp;
1497 return (exec_args_add_str(args, envp, segflg, &args->envc));
1501 exec_args_adjust_args(struct image_args *args, size_t consume, ssize_t extend)
1505 KASSERT(args->endp != NULL, ("endp not initialized"));
1506 KASSERT(args->begin_argv != NULL, ("begin_argp not initialized"));
1508 offset = extend - consume;
1509 if (args->stringspace < offset)
1511 memmove(args->begin_argv + extend, args->begin_argv + consume,
1512 args->endp - args->begin_argv + consume);
1514 args->begin_envv += offset;
1515 args->endp += offset;
1516 args->stringspace -= offset;
1521 exec_args_get_begin_envv(struct image_args *args)
1524 KASSERT(args->endp != NULL, ("endp not initialized"));
1527 return (args->begin_envv);
1528 return (args->endp);
1532 * Copy strings out to the new process address space, constructing new arg
1533 * and env vector tables. Return a pointer to the base so that it can be used
1534 * as the initial stack pointer.
1537 exec_copyout_strings(struct image_params *imgp, uintptr_t *stack_base)
1542 uintptr_t destp, ustringp;
1543 struct ps_strings *arginfo;
1545 size_t execpath_len;
1546 int error, szsigcode, szps;
1547 char canary[sizeof(long) * 8];
1549 szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1551 * Calculate string base and vector table pointers.
1552 * Also deal with signal trampoline code for this exec type.
1554 if (imgp->execpath != NULL && imgp->auxargs != NULL)
1555 execpath_len = strlen(imgp->execpath) + 1;
1560 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1561 imgp->ps_strings = arginfo;
1562 if (p->p_sysent->sv_sigcode_base == 0) {
1563 if (p->p_sysent->sv_szsigcode != NULL)
1564 szsigcode = *(p->p_sysent->sv_szsigcode);
1566 destp = (uintptr_t)arginfo;
1571 if (szsigcode != 0) {
1573 destp = rounddown2(destp, sizeof(void *));
1574 error = copyout(p->p_sysent->sv_sigcode, (void *)destp,
1581 * Copy the image path for the rtld.
1583 if (execpath_len != 0) {
1584 destp -= execpath_len;
1585 destp = rounddown2(destp, sizeof(void *));
1586 imgp->execpathp = (void *)destp;
1587 error = copyout(imgp->execpath, imgp->execpathp, execpath_len);
1593 * Prepare the canary for SSP.
1595 arc4rand(canary, sizeof(canary), 0);
1596 destp -= sizeof(canary);
1597 imgp->canary = (void *)destp;
1598 error = copyout(canary, imgp->canary, sizeof(canary));
1601 imgp->canarylen = sizeof(canary);
1604 * Prepare the pagesizes array.
1607 destp = rounddown2(destp, sizeof(void *));
1608 imgp->pagesizes = (void *)destp;
1609 error = copyout(pagesizes, imgp->pagesizes, szps);
1612 imgp->pagesizeslen = szps;
1615 * Allocate room for the argument and environment strings.
1617 destp -= ARG_MAX - imgp->args->stringspace;
1618 destp = rounddown2(destp, sizeof(void *));
1621 if (imgp->sysent->sv_stackgap != NULL)
1622 imgp->sysent->sv_stackgap(imgp, &destp);
1624 if (imgp->auxargs) {
1626 * Allocate room on the stack for the ELF auxargs
1627 * array. It has up to AT_COUNT entries.
1629 destp -= AT_COUNT * sizeof(Elf_Auxinfo);
1630 destp = rounddown2(destp, sizeof(void *));
1633 vectp = (char **)destp;
1636 * Allocate room for the argv[] and env vectors including the
1637 * terminating NULL pointers.
1639 vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
1642 * vectp also becomes our initial stack base
1644 *stack_base = (uintptr_t)vectp;
1646 stringp = imgp->args->begin_argv;
1647 argc = imgp->args->argc;
1648 envc = imgp->args->envc;
1651 * Copy out strings - arguments and environment.
1653 error = copyout(stringp, (void *)ustringp,
1654 ARG_MAX - imgp->args->stringspace);
1659 * Fill in "ps_strings" struct for ps, w, etc.
1662 if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 ||
1663 suword32(&arginfo->ps_nargvstr, argc) != 0)
1667 * Fill in argument portion of vector table.
1669 for (; argc > 0; --argc) {
1670 if (suword(vectp++, ustringp) != 0)
1672 while (*stringp++ != 0)
1677 /* a null vector table pointer separates the argp's from the envp's */
1678 if (suword(vectp++, 0) != 0)
1682 if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 ||
1683 suword32(&arginfo->ps_nenvstr, envc) != 0)
1687 * Fill in environment portion of vector table.
1689 for (; envc > 0; --envc) {
1690 if (suword(vectp++, ustringp) != 0)
1692 while (*stringp++ != 0)
1697 /* end of vector table is a null pointer */
1698 if (suword(vectp, 0) != 0)
1701 if (imgp->auxargs) {
1703 error = imgp->sysent->sv_copyout_auxargs(imgp,
1713 * Check permissions of file to execute.
1714 * Called with imgp->vp locked.
1715 * Return 0 for success or error code on failure.
1718 exec_check_permissions(struct image_params *imgp)
1720 struct vnode *vp = imgp->vp;
1721 struct vattr *attr = imgp->attr;
1727 /* Get file attributes */
1728 error = VOP_GETATTR(vp, attr, td->td_ucred);
1733 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1739 * 1) Check if file execution is disabled for the filesystem that
1740 * this file resides on.
1741 * 2) Ensure that at least one execute bit is on. Otherwise, a
1742 * privileged user will always succeed, and we don't want this
1743 * to happen unless the file really is executable.
1744 * 3) Ensure that the file is a regular file.
1746 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1747 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1748 (attr->va_type != VREG))
1752 * Zero length files can't be exec'd
1754 if (attr->va_size == 0)
1758 * Check for execute permission to file based on current credentials.
1760 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1765 * Check number of open-for-writes on the file and deny execution
1768 * Add a text reference now so no one can write to the
1769 * executable while we're activating it.
1771 * Remember if this was set before and unset it in case this is not
1772 * actually an executable image.
1774 error = VOP_SET_TEXT(vp);
1777 imgp->textset = true;
1780 * Call filesystem specific open routine (which does nothing in the
1783 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1790 * Exec handler registration
1793 exec_register(const struct execsw *execsw_arg)
1795 const struct execsw **es, **xs, **newexecsw;
1796 u_int count = 2; /* New slot and trailing NULL */
1799 for (es = execsw; *es; es++)
1801 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1804 for (es = execsw; *es; es++)
1809 free(execsw, M_TEMP);
1815 exec_unregister(const struct execsw *execsw_arg)
1817 const struct execsw **es, **xs, **newexecsw;
1821 panic("unregister with no handlers left?\n");
1823 for (es = execsw; *es; es++) {
1824 if (*es == execsw_arg)
1829 for (es = execsw; *es; es++)
1830 if (*es != execsw_arg)
1832 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1834 for (es = execsw; *es; es++)
1835 if (*es != execsw_arg)
1839 free(execsw, M_TEMP);