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/timers.h>
72 #include <sys/vnode.h>
75 #include <sys/ktrace.h>
79 #include <vm/vm_param.h>
81 #include <vm/vm_page.h>
82 #include <vm/vm_map.h>
83 #include <vm/vm_kern.h>
84 #include <vm/vm_extern.h>
85 #include <vm/vm_object.h>
86 #include <vm/vm_pager.h>
89 #include <sys/pmckern.h>
92 #include <machine/reg.h>
94 #include <security/audit/audit.h>
95 #include <security/mac/mac_framework.h>
98 #include <sys/dtrace_bsd.h>
99 dtrace_execexit_func_t dtrace_fasttrap_exec;
102 SDT_PROVIDER_DECLARE(proc);
103 SDT_PROBE_DEFINE1(proc, , , exec, "char *");
104 SDT_PROBE_DEFINE1(proc, , , exec__failure, "int");
105 SDT_PROBE_DEFINE1(proc, , , exec__success, "char *");
107 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
109 int coredump_pack_fileinfo = 1;
110 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_fileinfo, CTLFLAG_RWTUN,
111 &coredump_pack_fileinfo, 0,
112 "Enable file path packing in 'procstat -f' coredump notes");
114 int coredump_pack_vmmapinfo = 1;
115 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_vmmapinfo, CTLFLAG_RWTUN,
116 &coredump_pack_vmmapinfo, 0,
117 "Enable file path packing in 'procstat -v' coredump notes");
119 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
120 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
121 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
122 static int do_execve(struct thread *td, struct image_args *args,
123 struct mac *mac_p, struct vmspace *oldvmspace);
125 /* XXX This should be vm_size_t. */
126 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD|
127 CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_ps_strings, "LU",
128 "Location of process' ps_strings structure");
130 /* XXX This should be vm_size_t. */
131 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD|
132 CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_usrstack, "LU",
133 "Top of process stack");
135 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD|CTLFLAG_MPSAFE,
136 NULL, 0, sysctl_kern_stackprot, "I",
137 "Stack memory permissions");
139 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
140 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
141 &ps_arg_cache_limit, 0,
142 "Process' command line characters cache limit");
144 static int disallow_high_osrel;
145 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW,
146 &disallow_high_osrel, 0,
147 "Disallow execution of binaries built for higher version of the world");
149 static int map_at_zero = 0;
150 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RWTUN, &map_at_zero, 0,
151 "Permit processes to map an object at virtual address 0.");
154 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
161 if (req->flags & SCTL_MASK32) {
163 val = (unsigned int)p->p_sysent->sv_psstrings;
164 error = SYSCTL_OUT(req, &val, sizeof(val));
167 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
168 sizeof(p->p_sysent->sv_psstrings));
173 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
180 if (req->flags & SCTL_MASK32) {
182 val = (unsigned int)p->p_sysent->sv_usrstack;
183 error = SYSCTL_OUT(req, &val, sizeof(val));
186 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
187 sizeof(p->p_sysent->sv_usrstack));
192 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
197 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
198 sizeof(p->p_sysent->sv_stackprot)));
202 * Each of the items is a pointer to a `const struct execsw', hence the
203 * double pointer here.
205 static const struct execsw **execsw;
207 #ifndef _SYS_SYSPROTO_H_
216 sys_execve(struct thread *td, struct execve_args *uap)
218 struct image_args args;
219 struct vmspace *oldvmspace;
222 error = pre_execve(td, &oldvmspace);
225 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
226 uap->argv, uap->envv);
228 error = kern_execve(td, &args, NULL, oldvmspace);
229 post_execve(td, error, oldvmspace);
230 AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
234 #ifndef _SYS_SYSPROTO_H_
235 struct fexecve_args {
242 sys_fexecve(struct thread *td, struct fexecve_args *uap)
244 struct image_args args;
245 struct vmspace *oldvmspace;
248 error = pre_execve(td, &oldvmspace);
251 error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
252 uap->argv, uap->envv);
255 error = kern_execve(td, &args, NULL, oldvmspace);
257 post_execve(td, error, oldvmspace);
258 AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
262 #ifndef _SYS_SYSPROTO_H_
263 struct __mac_execve_args {
272 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap)
275 struct image_args args;
276 struct vmspace *oldvmspace;
279 error = pre_execve(td, &oldvmspace);
282 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
283 uap->argv, uap->envv);
285 error = kern_execve(td, &args, uap->mac_p, oldvmspace);
286 post_execve(td, error, oldvmspace);
287 AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
295 pre_execve(struct thread *td, struct vmspace **oldvmspace)
300 KASSERT(td == curthread, ("non-current thread %p", td));
303 if ((p->p_flag & P_HADTHREADS) != 0) {
305 if (thread_single(p, SINGLE_BOUNDARY) != 0)
309 KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0,
311 *oldvmspace = p->p_vmspace;
316 post_execve(struct thread *td, int error, struct vmspace *oldvmspace)
320 KASSERT(td == curthread, ("non-current thread %p", td));
322 if ((p->p_flag & P_HADTHREADS) != 0) {
325 * If success, we upgrade to SINGLE_EXIT state to
326 * force other threads to suicide.
328 if (error == EJUSTRETURN)
329 thread_single(p, SINGLE_EXIT);
331 thread_single_end(p, SINGLE_BOUNDARY);
334 exec_cleanup(td, oldvmspace);
338 * kern_execve() has the astonishing property of not always returning to
339 * the caller. If sufficiently bad things happen during the call to
340 * do_execve(), it can end up calling exit1(); as a result, callers must
341 * avoid doing anything which they might need to undo (e.g., allocating
345 kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p,
346 struct vmspace *oldvmspace)
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, oldvmspace));
357 execve_nosetid(struct image_params *imgp)
359 imgp->credential_setid = false;
360 if (imgp->newcred != NULL) {
361 crfree(imgp->newcred);
362 imgp->newcred = NULL;
367 * In-kernel implementation of execve(). All arguments are assumed to be
368 * userspace pointers from the passed thread.
371 do_execve(struct thread *td, struct image_args *args, struct mac *mac_p,
372 struct vmspace *oldvmspace)
374 struct proc *p = td->td_proc;
376 struct ucred *oldcred;
377 struct uidinfo *euip = NULL;
378 uintptr_t stack_base;
379 struct image_params image_params, *imgp;
381 int (*img_first)(struct image_params *);
382 struct pargs *oldargs = NULL, *newargs = NULL;
383 struct sigacts *oldsigacts = NULL, *newsigacts = NULL;
385 struct ktr_io_params *kiop;
387 struct vnode *oldtextvp = NULL, *newtextvp;
388 int credential_changing;
390 struct label *interpvplabel = NULL;
394 struct pmckern_procexec pe;
396 int error, i, orig_osrel;
398 static const char fexecv_proc_title[] = "(fexecv)";
400 imgp = &image_params;
406 * Lock the process and set the P_INEXEC flag to indicate that
407 * it should be left alone until we're done here. This is
408 * necessary to avoid race conditions - e.g. in ptrace() -
409 * that might allow a local user to illicitly obtain elevated
413 KASSERT((p->p_flag & P_INEXEC) == 0,
414 ("%s(): process already has P_INEXEC flag", __func__));
415 p->p_flag |= P_INEXEC;
419 * Initialize part of the common data
421 bzero(imgp, sizeof(*imgp));
425 oldcred = p->p_ucred;
426 orig_osrel = p->p_osrel;
427 orig_fctl0 = p->p_fctl0;
430 error = mac_execve_enter(imgp, mac_p);
436 * Translate the file name. namei() returns a vnode pointer
437 * in ni_vp among other things.
439 * XXXAUDIT: It would be desirable to also audit the name of the
440 * interpreter if this is an interpreted binary.
442 if (args->fname != NULL) {
443 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW |
444 SAVENAME | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
447 SDT_PROBE1(proc, , , exec, args->fname);
450 if (args->fname != NULL) {
451 #ifdef CAPABILITY_MODE
453 * While capability mode can't reach this point via direct
454 * path arguments to execve(), we also don't allow
455 * interpreters to be used in capability mode (for now).
456 * Catch indirect lookups and return a permissions error.
458 if (IN_CAPABILITY_MODE(td)) {
467 newtextvp = nd.ni_vp;
468 imgp->vp = newtextvp;
470 AUDIT_ARG_FD(args->fd);
472 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
474 error = fgetvp_exec(td, args->fd, &cap_fexecve_rights, &newtextvp);
477 vn_lock(newtextvp, LK_SHARED | LK_RETRY);
478 AUDIT_ARG_VNODE1(newtextvp);
479 imgp->vp = newtextvp;
483 * Check file permissions. Also 'opens' file and sets its vnode to
486 error = exec_check_permissions(imgp);
488 goto exec_fail_dealloc;
490 imgp->object = imgp->vp->v_object;
491 if (imgp->object != NULL)
492 vm_object_reference(imgp->object);
494 error = exec_map_first_page(imgp);
496 goto exec_fail_dealloc;
498 imgp->proc->p_osrel = 0;
499 imgp->proc->p_fctl0 = 0;
502 * Implement image setuid/setgid.
504 * Determine new credentials before attempting image activators
505 * so that it can be used by process_exec handlers to determine
506 * credential/setid changes.
508 * Don't honor setuid/setgid if the filesystem prohibits it or if
509 * the process is being traced.
511 * We disable setuid/setgid/etc in capability mode on the basis
512 * that most setugid applications are not written with that
513 * environment in mind, and will therefore almost certainly operate
514 * incorrectly. In principle there's no reason that setugid
515 * applications might not be useful in capability mode, so we may want
516 * to reconsider this conservative design choice in the future.
518 * XXXMAC: For the time being, use NOSUID to also prohibit
519 * transitions on the file system.
521 credential_changing = 0;
522 credential_changing |= (attr.va_mode & S_ISUID) &&
523 oldcred->cr_uid != attr.va_uid;
524 credential_changing |= (attr.va_mode & S_ISGID) &&
525 oldcred->cr_gid != attr.va_gid;
527 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
528 interpvplabel, imgp);
529 credential_changing |= will_transition;
532 /* Don't inherit PROC_PDEATHSIG_CTL value if setuid/setgid. */
533 if (credential_changing)
534 imgp->proc->p_pdeathsig = 0;
536 if (credential_changing &&
537 #ifdef CAPABILITY_MODE
538 ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
540 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
541 (p->p_flag & P_TRACED) == 0) {
542 imgp->credential_setid = true;
543 VOP_UNLOCK(imgp->vp);
544 imgp->newcred = crdup(oldcred);
545 if (attr.va_mode & S_ISUID) {
546 euip = uifind(attr.va_uid);
547 change_euid(imgp->newcred, euip);
549 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
550 if (attr.va_mode & S_ISGID)
551 change_egid(imgp->newcred, attr.va_gid);
553 * Implement correct POSIX saved-id behavior.
555 * XXXMAC: Note that the current logic will save the
556 * uid and gid if a MAC domain transition occurs, even
557 * though maybe it shouldn't.
559 change_svuid(imgp->newcred, imgp->newcred->cr_uid);
560 change_svgid(imgp->newcred, imgp->newcred->cr_gid);
563 * Implement correct POSIX saved-id behavior.
565 * XXX: It's not clear that the existing behavior is
566 * POSIX-compliant. A number of sources indicate that the
567 * saved uid/gid should only be updated if the new ruid is
568 * not equal to the old ruid, or the new euid is not equal
569 * to the old euid and the new euid is not equal to the old
570 * ruid. The FreeBSD code always updates the saved uid/gid.
571 * Also, this code uses the new (replaced) euid and egid as
572 * the source, which may or may not be the right ones to use.
574 if (oldcred->cr_svuid != oldcred->cr_uid ||
575 oldcred->cr_svgid != oldcred->cr_gid) {
576 VOP_UNLOCK(imgp->vp);
577 imgp->newcred = crdup(oldcred);
578 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
579 change_svuid(imgp->newcred, imgp->newcred->cr_uid);
580 change_svgid(imgp->newcred, imgp->newcred->cr_gid);
583 /* The new credentials are installed into the process later. */
586 * Do the best to calculate the full path to the image file.
588 if (args->fname != NULL && args->fname[0] == '/')
589 imgp->execpath = args->fname;
591 VOP_UNLOCK(imgp->vp);
592 if (vn_fullpath(imgp->vp, &imgp->execpath, &imgp->freepath) != 0)
593 imgp->execpath = args->fname;
594 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
598 * If the current process has a special image activator it
599 * wants to try first, call it. For example, emulating shell
600 * scripts differently.
603 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
604 error = img_first(imgp);
607 * Loop through the list of image activators, calling each one.
608 * An activator returns -1 if there is no match, 0 on success,
609 * and an error otherwise.
611 for (i = 0; error == -1 && execsw[i]; ++i) {
612 if (execsw[i]->ex_imgact == NULL ||
613 execsw[i]->ex_imgact == img_first) {
616 error = (*execsw[i]->ex_imgact)(imgp);
622 goto exec_fail_dealloc;
626 * Special interpreter operation, cleanup and loop up to try to
627 * activate the interpreter.
629 if (imgp->interpreted) {
630 exec_unmap_first_page(imgp);
632 * The text reference needs to be removed for scripts.
633 * There is a short period before we determine that
634 * something is a script where text reference is active.
635 * The vnode lock is held over this entire period
636 * so nothing should illegitimately be blocked.
638 MPASS(imgp->textset);
639 VOP_UNSET_TEXT_CHECKED(newtextvp);
640 imgp->textset = false;
641 /* free name buffer and old vnode */
642 if (args->fname != NULL)
643 NDFREE(&nd, NDF_ONLY_PNBUF);
645 mac_execve_interpreter_enter(newtextvp, &interpvplabel);
648 VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td);
652 vm_object_deallocate(imgp->object);
654 execve_nosetid(imgp);
655 imgp->execpath = NULL;
656 free(imgp->freepath, M_TEMP);
657 imgp->freepath = NULL;
658 /* set new name to that of the interpreter */
659 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW |
660 SAVENAME, UIO_SYSSPACE, imgp->interpreter_name, td);
661 args->fname = imgp->interpreter_name;
666 * NB: We unlock the vnode here because it is believed that none
667 * of the sv_copyout_strings/sv_fixup operations require the vnode.
669 VOP_UNLOCK(imgp->vp);
671 if (disallow_high_osrel &&
672 P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
674 uprintf("Osrel %d for image %s too high\n", p->p_osrel,
675 imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
676 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
677 goto exec_fail_dealloc;
680 /* ABI enforces the use of Capsicum. Switch into capabilities mode. */
681 if (SV_PROC_FLAG(p, SV_CAPSICUM))
682 sys_cap_enter(td, NULL);
685 * Copy out strings (args and env) and initialize stack base.
687 error = (*p->p_sysent->sv_copyout_strings)(imgp, &stack_base);
689 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
690 goto exec_fail_dealloc;
696 error = (*p->p_sysent->sv_fixup)(&stack_base, imgp);
698 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
699 goto exec_fail_dealloc;
702 if (args->fdp != NULL) {
703 /* Install a brand new file descriptor table. */
704 fdinstall_remapped(td, args->fdp);
708 * Keep on using the existing file descriptor table. For
709 * security and other reasons, the file descriptor table
710 * cannot be shared after an exec.
714 /* close files on exec */
719 * Malloc things before we need locks.
721 i = exec_args_get_begin_envv(imgp->args) - imgp->args->begin_argv;
722 /* Cache arguments if they fit inside our allowance */
723 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
724 newargs = pargs_alloc(i);
725 bcopy(imgp->args->begin_argv, newargs->ar_args, i);
729 * For security and other reasons, signal handlers cannot
730 * be shared after an exec. The new process gets a copy of the old
731 * handlers. In execsigs(), the new process will have its signals
734 if (sigacts_shared(p->p_sigacts)) {
735 oldsigacts = p->p_sigacts;
736 newsigacts = sigacts_alloc();
737 sigacts_copy(newsigacts, oldsigacts);
740 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
744 p->p_sigacts = newsigacts;
748 /* reset caught signals */
751 /* name this process - nameiexec(p, ndp) */
752 bzero(p->p_comm, sizeof(p->p_comm));
754 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
755 min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
756 else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0)
757 bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
758 bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
760 sched_clear_tdname(td);
764 * mark as execed, wakeup the process that vforked (if any) and tell
765 * it that it now has its own resources back
768 if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0)
769 p->p_flag2 &= ~P2_NOTRACE;
770 if ((p->p_flag2 & P2_STKGAP_DISABLE_EXEC) == 0)
771 p->p_flag2 &= ~P2_STKGAP_DISABLE;
772 if (p->p_flag & P_PPWAIT) {
773 p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
774 cv_broadcast(&p->p_pwait);
775 /* STOPs are no longer ignored, arrange for AST */
779 if (imgp->sysent->sv_setid_allowed != NULL &&
780 !(*imgp->sysent->sv_setid_allowed)(td, imgp))
781 execve_nosetid(imgp);
784 * Implement image setuid/setgid installation.
786 if (imgp->credential_setid) {
788 * Turn off syscall tracing for set-id programs, except for
789 * root. Record any set-id flags first to make sure that
790 * we do not regain any tracing during a possible block.
794 kiop = ktrprocexec(p);
797 * Close any file descriptors 0..2 that reference procfs,
798 * then make sure file descriptors 0..2 are in use.
800 * Both fdsetugidsafety() and fdcheckstd() may call functions
801 * taking sleepable locks, so temporarily drop our locks.
804 VOP_UNLOCK(imgp->vp);
806 error = fdcheckstd(td);
807 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
809 goto exec_fail_dealloc;
812 if (will_transition) {
813 mac_vnode_execve_transition(oldcred, imgp->newcred,
814 imgp->vp, interpvplabel, imgp);
818 if (oldcred->cr_uid == oldcred->cr_ruid &&
819 oldcred->cr_gid == oldcred->cr_rgid)
820 p->p_flag &= ~P_SUGID;
823 * Set the new credentials.
825 if (imgp->newcred != NULL) {
826 proc_set_cred(p, imgp->newcred);
832 * Store the vp for use in procfs. This vnode was referenced by namei
835 oldtextvp = p->p_textvp;
836 p->p_textvp = newtextvp;
840 * Tell the DTrace fasttrap provider about the exec if it
841 * has declared an interest.
843 if (dtrace_fasttrap_exec)
844 dtrace_fasttrap_exec(p);
848 * Notify others that we exec'd, and clear the P_INEXEC flag
849 * as we're now a bona fide freshly-execed process.
851 KNOTE_LOCKED(p->p_klist, NOTE_EXEC);
852 p->p_flag &= ~P_INEXEC;
854 /* clear "fork but no exec" flag, as we _are_ execing */
855 p->p_acflag &= ~AFORK;
858 * Free any previous argument cache and replace it with
859 * the new argument cache, if any.
869 * Check if system-wide sampling is in effect or if the
870 * current process is using PMCs. If so, do exec() time
871 * processing. This processing needs to happen AFTER the
872 * P_INEXEC flag is cleared.
874 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
875 VOP_UNLOCK(imgp->vp);
876 pe.pm_credentialschanged = credential_changing;
877 pe.pm_entryaddr = imgp->entry_addr;
879 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
880 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
884 /* Set values passed into the program in registers. */
885 (*p->p_sysent->sv_setregs)(td, imgp, stack_base);
887 VOP_MMAPPED(imgp->vp);
889 SDT_PROBE1(proc, , , exec__success, args->fname);
893 p->p_osrel = orig_osrel;
894 p->p_fctl0 = orig_fctl0;
897 if (imgp->firstpage != NULL)
898 exec_unmap_first_page(imgp);
900 if (imgp->vp != NULL) {
902 NDFREE(&nd, NDF_ONLY_PNBUF);
904 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
906 VOP_UNSET_TEXT_CHECKED(imgp->vp);
910 VOP_UNLOCK(imgp->vp);
913 if (imgp->object != NULL)
914 vm_object_deallocate(imgp->object);
916 free(imgp->freepath, M_TEMP);
919 if (p->p_ptevents & PTRACE_EXEC) {
921 if (p->p_ptevents & PTRACE_EXEC)
922 td->td_dbgflags |= TDB_EXEC;
927 /* we're done here, clear P_INEXEC */
929 p->p_flag &= ~P_INEXEC;
932 SDT_PROBE1(proc, , , exec__failure, error);
935 if (imgp->newcred != NULL && oldcred != NULL)
936 crfree(imgp->newcred);
939 mac_execve_exit(imgp);
940 mac_execve_interpreter_exit(interpvplabel);
942 exec_free_args(args);
945 * Handle deferred decrement of ref counts.
947 if (oldtextvp != NULL)
950 ktr_io_params_free(kiop);
954 if (oldsigacts != NULL)
955 sigacts_free(oldsigacts);
959 if (error && imgp->vmspace_destroyed) {
960 /* sorry, no more process anymore. exit gracefully */
961 exec_cleanup(td, oldvmspace);
962 exit1(td, 0, SIGABRT);
972 * We don't want cpu_set_syscall_retval() to overwrite any of
973 * the register values put in place by exec_setregs().
974 * Implementations of cpu_set_syscall_retval() will leave
975 * registers unmodified when returning EJUSTRETURN.
977 return (error == 0 ? EJUSTRETURN : error);
981 exec_cleanup(struct thread *td, struct vmspace *oldvmspace)
983 if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
984 KASSERT(td->td_proc->p_vmspace != oldvmspace,
985 ("oldvmspace still used"));
986 vmspace_free(oldvmspace);
987 td->td_pflags &= ~TDP_EXECVMSPC;
992 exec_map_first_page(struct image_params *imgp)
998 if (imgp->firstpage != NULL)
999 exec_unmap_first_page(imgp);
1001 object = imgp->vp->v_object;
1004 #if VM_NRESERVLEVEL > 0
1005 if ((object->flags & OBJ_COLORED) == 0) {
1006 VM_OBJECT_WLOCK(object);
1007 vm_object_color(object, 0);
1008 VM_OBJECT_WUNLOCK(object);
1011 error = vm_page_grab_valid_unlocked(&m, object, 0,
1012 VM_ALLOC_COUNT(VM_INITIAL_PAGEIN) |
1013 VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED);
1015 if (error != VM_PAGER_OK)
1017 imgp->firstpage = sf_buf_alloc(m, 0);
1018 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
1024 exec_unmap_first_page(struct image_params *imgp)
1028 if (imgp->firstpage != NULL) {
1029 m = sf_buf_page(imgp->firstpage);
1030 sf_buf_free(imgp->firstpage);
1031 imgp->firstpage = NULL;
1032 vm_page_unwire(m, PQ_ACTIVE);
1037 * Destroy old address space, and allocate a new stack.
1038 * The new stack is only sgrowsiz large because it is grown
1039 * automatically on a page fault.
1042 exec_new_vmspace(struct image_params *imgp, struct sysentvec *sv)
1045 struct proc *p = imgp->proc;
1046 struct vmspace *vmspace = p->p_vmspace;
1047 struct thread *td = curthread;
1049 struct rlimit rlim_stack;
1050 vm_offset_t sv_minuser, stack_addr;
1052 vm_prot_t stack_prot;
1055 imgp->vmspace_destroyed = 1;
1058 sigfastblock_clear(td);
1061 if (sv->sv_onexec != NULL)
1062 sv->sv_onexec(p, imgp);
1064 EVENTHANDLER_DIRECT_INVOKE(process_exec, p, imgp);
1067 * Blow away entire process VM, if address space not shared,
1068 * otherwise, create a new VM space so that other threads are
1071 map = &vmspace->vm_map;
1073 sv_minuser = sv->sv_minuser;
1075 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1076 if (refcount_load(&vmspace->vm_refcnt) == 1 &&
1077 vm_map_min(map) == sv_minuser &&
1078 vm_map_max(map) == sv->sv_maxuser &&
1079 cpu_exec_vmspace_reuse(p, map)) {
1081 pmap_remove_pages(vmspace_pmap(vmspace));
1082 vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1084 * An exec terminates mlockall(MCL_FUTURE).
1085 * ASLR and W^X states must be re-evaluated.
1088 vm_map_modflags(map, 0, MAP_WIREFUTURE | MAP_ASLR |
1089 MAP_ASLR_IGNSTART | MAP_WXORX);
1092 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1095 vmspace = p->p_vmspace;
1096 map = &vmspace->vm_map;
1098 map->flags |= imgp->map_flags;
1100 /* Map a shared page */
1101 obj = sv->sv_shared_page_obj;
1103 vm_object_reference(obj);
1104 error = vm_map_fixed(map, obj, 0,
1105 sv->sv_shared_page_base, sv->sv_shared_page_len,
1106 VM_PROT_READ | VM_PROT_EXECUTE,
1107 VM_PROT_READ | VM_PROT_EXECUTE,
1108 MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1109 if (error != KERN_SUCCESS) {
1110 vm_object_deallocate(obj);
1111 return (vm_mmap_to_errno(error));
1115 /* Allocate a new stack */
1116 if (imgp->stack_sz != 0) {
1117 ssiz = trunc_page(imgp->stack_sz);
1119 lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack);
1121 if (ssiz > rlim_stack.rlim_max)
1122 ssiz = rlim_stack.rlim_max;
1123 if (ssiz > rlim_stack.rlim_cur) {
1124 rlim_stack.rlim_cur = ssiz;
1125 kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack);
1127 } else if (sv->sv_maxssiz != NULL) {
1128 ssiz = *sv->sv_maxssiz;
1132 imgp->eff_stack_sz = lim_cur(curthread, RLIMIT_STACK);
1133 if (ssiz < imgp->eff_stack_sz)
1134 imgp->eff_stack_sz = ssiz;
1135 stack_addr = sv->sv_usrstack - ssiz;
1136 stack_prot = obj != NULL && imgp->stack_prot != 0 ?
1137 imgp->stack_prot : sv->sv_stackprot;
1138 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz, stack_prot,
1139 VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1140 if (error != KERN_SUCCESS) {
1141 uprintf("exec_new_vmspace: mapping stack size %#jx prot %#x "
1142 "failed mach error %d errno %d\n", (uintmax_t)ssiz,
1143 stack_prot, error, vm_mmap_to_errno(error));
1144 return (vm_mmap_to_errno(error));
1148 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they
1149 * are still used to enforce the stack rlimit on the process stack.
1151 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1152 vmspace->vm_maxsaddr = (char *)stack_addr;
1158 * Copy out argument and environment strings from the old process address
1159 * space into the temporary string buffer.
1162 exec_copyin_args(struct image_args *args, const char *fname,
1163 enum uio_seg segflg, char **argv, char **envv)
1168 bzero(args, sizeof(*args));
1173 * Allocate demand-paged memory for the file name, argument, and
1174 * environment strings.
1176 error = exec_alloc_args(args);
1181 * Copy the file name.
1183 error = exec_args_add_fname(args, fname, segflg);
1188 * extract arguments first
1191 error = fueword(argv++, &arg);
1198 error = exec_args_add_arg(args, (char *)(uintptr_t)arg,
1205 * extract environment strings
1209 error = fueword(envv++, &env);
1216 error = exec_args_add_env(args,
1217 (char *)(uintptr_t)env, UIO_USERSPACE);
1226 exec_free_args(args);
1231 exec_copyin_data_fds(struct thread *td, struct image_args *args,
1232 const void *data, size_t datalen, const int *fds, size_t fdslen)
1234 struct filedesc *ofdp;
1239 memset(args, '\0', sizeof(*args));
1240 ofdp = td->td_proc->p_fd;
1241 if (datalen >= ARG_MAX || fdslen >= ofdp->fd_nfiles)
1243 error = exec_alloc_args(args);
1247 args->begin_argv = args->buf;
1248 args->stringspace = ARG_MAX;
1252 * Argument buffer has been provided. Copy it into the
1253 * kernel as a single string and add a terminating null
1256 error = copyin(data, args->begin_argv, datalen);
1259 args->begin_argv[datalen] = '\0';
1260 args->endp = args->begin_argv + datalen + 1;
1261 args->stringspace -= datalen + 1;
1264 * Traditional argument counting. Count the number of
1267 for (p = args->begin_argv; p < args->endp; ++p)
1271 /* No argument buffer provided. */
1272 args->endp = args->begin_argv;
1275 /* Create new file descriptor table. */
1276 kfds = malloc(fdslen * sizeof(int), M_TEMP, M_WAITOK);
1277 error = copyin(fds, kfds, fdslen * sizeof(int));
1282 error = fdcopy_remapped(ofdp, kfds, fdslen, &args->fdp);
1289 exec_free_args(args);
1293 struct exec_args_kva {
1296 SLIST_ENTRY(exec_args_kva) next;
1299 DPCPU_DEFINE_STATIC(struct exec_args_kva *, exec_args_kva);
1301 static SLIST_HEAD(, exec_args_kva) exec_args_kva_freelist;
1302 static struct mtx exec_args_kva_mtx;
1303 static u_int exec_args_gen;
1306 exec_prealloc_args_kva(void *arg __unused)
1308 struct exec_args_kva *argkva;
1311 SLIST_INIT(&exec_args_kva_freelist);
1312 mtx_init(&exec_args_kva_mtx, "exec args kva", NULL, MTX_DEF);
1313 for (i = 0; i < exec_map_entries; i++) {
1314 argkva = malloc(sizeof(*argkva), M_PARGS, M_WAITOK);
1315 argkva->addr = kmap_alloc_wait(exec_map, exec_map_entry_size);
1316 argkva->gen = exec_args_gen;
1317 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1320 SYSINIT(exec_args_kva, SI_SUB_EXEC, SI_ORDER_ANY, exec_prealloc_args_kva, NULL);
1323 exec_alloc_args_kva(void **cookie)
1325 struct exec_args_kva *argkva;
1327 argkva = (void *)atomic_readandclear_ptr(
1328 (uintptr_t *)DPCPU_PTR(exec_args_kva));
1329 if (argkva == NULL) {
1330 mtx_lock(&exec_args_kva_mtx);
1331 while ((argkva = SLIST_FIRST(&exec_args_kva_freelist)) == NULL)
1332 (void)mtx_sleep(&exec_args_kva_freelist,
1333 &exec_args_kva_mtx, 0, "execkva", 0);
1334 SLIST_REMOVE_HEAD(&exec_args_kva_freelist, next);
1335 mtx_unlock(&exec_args_kva_mtx);
1337 *(struct exec_args_kva **)cookie = argkva;
1338 return (argkva->addr);
1342 exec_release_args_kva(struct exec_args_kva *argkva, u_int gen)
1346 base = argkva->addr;
1347 if (argkva->gen != gen) {
1348 (void)vm_map_madvise(exec_map, base, base + exec_map_entry_size,
1352 if (!atomic_cmpset_ptr((uintptr_t *)DPCPU_PTR(exec_args_kva),
1353 (uintptr_t)NULL, (uintptr_t)argkva)) {
1354 mtx_lock(&exec_args_kva_mtx);
1355 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1356 wakeup_one(&exec_args_kva_freelist);
1357 mtx_unlock(&exec_args_kva_mtx);
1362 exec_free_args_kva(void *cookie)
1365 exec_release_args_kva(cookie, exec_args_gen);
1369 exec_args_kva_lowmem(void *arg __unused)
1371 SLIST_HEAD(, exec_args_kva) head;
1372 struct exec_args_kva *argkva;
1376 gen = atomic_fetchadd_int(&exec_args_gen, 1) + 1;
1379 * Force an madvise of each KVA range. Any currently allocated ranges
1380 * will have MADV_FREE applied once they are freed.
1383 mtx_lock(&exec_args_kva_mtx);
1384 SLIST_SWAP(&head, &exec_args_kva_freelist, exec_args_kva);
1385 mtx_unlock(&exec_args_kva_mtx);
1386 while ((argkva = SLIST_FIRST(&head)) != NULL) {
1387 SLIST_REMOVE_HEAD(&head, next);
1388 exec_release_args_kva(argkva, gen);
1392 argkva = (void *)atomic_readandclear_ptr(
1393 (uintptr_t *)DPCPU_ID_PTR(i, exec_args_kva));
1395 exec_release_args_kva(argkva, gen);
1398 EVENTHANDLER_DEFINE(vm_lowmem, exec_args_kva_lowmem, NULL,
1399 EVENTHANDLER_PRI_ANY);
1402 * Allocate temporary demand-paged, zero-filled memory for the file name,
1403 * argument, and environment strings.
1406 exec_alloc_args(struct image_args *args)
1409 args->buf = (char *)exec_alloc_args_kva(&args->bufkva);
1414 exec_free_args(struct image_args *args)
1417 if (args->buf != NULL) {
1418 exec_free_args_kva(args->bufkva);
1421 if (args->fname_buf != NULL) {
1422 free(args->fname_buf, M_TEMP);
1423 args->fname_buf = NULL;
1425 if (args->fdp != NULL)
1426 fdescfree_remapped(args->fdp);
1430 * A set to functions to fill struct image args.
1432 * NOTE: exec_args_add_fname() must be called (possibly with a NULL
1433 * fname) before the other functions. All exec_args_add_arg() calls must
1434 * be made before any exec_args_add_env() calls. exec_args_adjust_args()
1435 * may be called any time after exec_args_add_fname().
1437 * exec_args_add_fname() - install path to be executed
1438 * exec_args_add_arg() - append an argument string
1439 * exec_args_add_env() - append an env string
1440 * exec_args_adjust_args() - adjust location of the argument list to
1441 * allow new arguments to be prepended
1444 exec_args_add_fname(struct image_args *args, const char *fname,
1445 enum uio_seg segflg)
1450 KASSERT(args->fname == NULL, ("fname already appended"));
1451 KASSERT(args->endp == NULL, ("already appending to args"));
1453 if (fname != NULL) {
1454 args->fname = args->buf;
1455 error = segflg == UIO_SYSSPACE ?
1456 copystr(fname, args->fname, PATH_MAX, &length) :
1457 copyinstr(fname, args->fname, PATH_MAX, &length);
1459 return (error == ENAMETOOLONG ? E2BIG : error);
1463 /* Set up for _arg_*()/_env_*() */
1464 args->endp = args->buf + length;
1465 /* begin_argv must be set and kept updated */
1466 args->begin_argv = args->endp;
1467 KASSERT(exec_map_entry_size - length >= ARG_MAX,
1468 ("too little space remaining for arguments %zu < %zu",
1469 exec_map_entry_size - length, (size_t)ARG_MAX));
1470 args->stringspace = ARG_MAX;
1476 exec_args_add_str(struct image_args *args, const char *str,
1477 enum uio_seg segflg, int *countp)
1482 KASSERT(args->endp != NULL, ("endp not initialized"));
1483 KASSERT(args->begin_argv != NULL, ("begin_argp not initialized"));
1485 error = (segflg == UIO_SYSSPACE) ?
1486 copystr(str, args->endp, args->stringspace, &length) :
1487 copyinstr(str, args->endp, args->stringspace, &length);
1489 return (error == ENAMETOOLONG ? E2BIG : error);
1490 args->stringspace -= length;
1491 args->endp += length;
1498 exec_args_add_arg(struct image_args *args, const char *argp,
1499 enum uio_seg segflg)
1502 KASSERT(args->envc == 0, ("appending args after env"));
1504 return (exec_args_add_str(args, argp, segflg, &args->argc));
1508 exec_args_add_env(struct image_args *args, const char *envp,
1509 enum uio_seg segflg)
1512 if (args->envc == 0)
1513 args->begin_envv = args->endp;
1515 return (exec_args_add_str(args, envp, segflg, &args->envc));
1519 exec_args_adjust_args(struct image_args *args, size_t consume, ssize_t extend)
1523 KASSERT(args->endp != NULL, ("endp not initialized"));
1524 KASSERT(args->begin_argv != NULL, ("begin_argp not initialized"));
1526 offset = extend - consume;
1527 if (args->stringspace < offset)
1529 memmove(args->begin_argv + extend, args->begin_argv + consume,
1530 args->endp - args->begin_argv + consume);
1532 args->begin_envv += offset;
1533 args->endp += offset;
1534 args->stringspace -= offset;
1539 exec_args_get_begin_envv(struct image_args *args)
1542 KASSERT(args->endp != NULL, ("endp not initialized"));
1545 return (args->begin_envv);
1546 return (args->endp);
1550 exec_stackgap(struct image_params *imgp, uintptr_t *dp)
1552 if (imgp->sysent->sv_stackgap == NULL ||
1553 (imgp->proc->p_fctl0 & (NT_FREEBSD_FCTL_ASLR_DISABLE |
1554 NT_FREEBSD_FCTL_ASG_DISABLE)) != 0 ||
1555 (imgp->map_flags & MAP_ASLR) == 0)
1557 imgp->sysent->sv_stackgap(imgp, dp);
1561 * Copy strings out to the new process address space, constructing new arg
1562 * and env vector tables. Return a pointer to the base so that it can be used
1563 * as the initial stack pointer.
1566 exec_copyout_strings(struct image_params *imgp, uintptr_t *stack_base)
1571 uintptr_t destp, ustringp;
1572 struct ps_strings *arginfo;
1574 size_t execpath_len;
1575 int error, szsigcode, szps;
1576 char canary[sizeof(long) * 8];
1578 szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1580 * Calculate string base and vector table pointers.
1581 * Also deal with signal trampoline code for this exec type.
1583 if (imgp->execpath != NULL && imgp->auxargs != NULL)
1584 execpath_len = strlen(imgp->execpath) + 1;
1589 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1590 imgp->ps_strings = arginfo;
1591 if (p->p_sysent->sv_sigcode_base == 0) {
1592 if (p->p_sysent->sv_szsigcode != NULL)
1593 szsigcode = *(p->p_sysent->sv_szsigcode);
1595 destp = (uintptr_t)arginfo;
1600 if (szsigcode != 0) {
1602 destp = rounddown2(destp, sizeof(void *));
1603 error = copyout(p->p_sysent->sv_sigcode, (void *)destp,
1610 * Copy the image path for the rtld.
1612 if (execpath_len != 0) {
1613 destp -= execpath_len;
1614 destp = rounddown2(destp, sizeof(void *));
1615 imgp->execpathp = (void *)destp;
1616 error = copyout(imgp->execpath, imgp->execpathp, execpath_len);
1622 * Prepare the canary for SSP.
1624 arc4rand(canary, sizeof(canary), 0);
1625 destp -= sizeof(canary);
1626 imgp->canary = (void *)destp;
1627 error = copyout(canary, imgp->canary, sizeof(canary));
1630 imgp->canarylen = sizeof(canary);
1633 * Prepare the pagesizes array.
1636 destp = rounddown2(destp, sizeof(void *));
1637 imgp->pagesizes = (void *)destp;
1638 error = copyout(pagesizes, imgp->pagesizes, szps);
1641 imgp->pagesizeslen = szps;
1644 * Allocate room for the argument and environment strings.
1646 destp -= ARG_MAX - imgp->args->stringspace;
1647 destp = rounddown2(destp, sizeof(void *));
1650 exec_stackgap(imgp, &destp);
1652 if (imgp->auxargs) {
1654 * Allocate room on the stack for the ELF auxargs
1655 * array. It has up to AT_COUNT entries.
1657 destp -= AT_COUNT * sizeof(Elf_Auxinfo);
1658 destp = rounddown2(destp, sizeof(void *));
1661 vectp = (char **)destp;
1664 * Allocate room for the argv[] and env vectors including the
1665 * terminating NULL pointers.
1667 vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
1670 * vectp also becomes our initial stack base
1672 *stack_base = (uintptr_t)vectp;
1674 stringp = imgp->args->begin_argv;
1675 argc = imgp->args->argc;
1676 envc = imgp->args->envc;
1679 * Copy out strings - arguments and environment.
1681 error = copyout(stringp, (void *)ustringp,
1682 ARG_MAX - imgp->args->stringspace);
1687 * Fill in "ps_strings" struct for ps, w, etc.
1690 if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 ||
1691 suword32(&arginfo->ps_nargvstr, argc) != 0)
1695 * Fill in argument portion of vector table.
1697 for (; argc > 0; --argc) {
1698 if (suword(vectp++, ustringp) != 0)
1700 while (*stringp++ != 0)
1705 /* a null vector table pointer separates the argp's from the envp's */
1706 if (suword(vectp++, 0) != 0)
1710 if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 ||
1711 suword32(&arginfo->ps_nenvstr, envc) != 0)
1715 * Fill in environment portion of vector table.
1717 for (; envc > 0; --envc) {
1718 if (suword(vectp++, ustringp) != 0)
1720 while (*stringp++ != 0)
1725 /* end of vector table is a null pointer */
1726 if (suword(vectp, 0) != 0)
1729 if (imgp->auxargs) {
1731 error = imgp->sysent->sv_copyout_auxargs(imgp,
1741 * Check permissions of file to execute.
1742 * Called with imgp->vp locked.
1743 * Return 0 for success or error code on failure.
1746 exec_check_permissions(struct image_params *imgp)
1748 struct vnode *vp = imgp->vp;
1749 struct vattr *attr = imgp->attr;
1755 /* Get file attributes */
1756 error = VOP_GETATTR(vp, attr, td->td_ucred);
1761 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1767 * 1) Check if file execution is disabled for the filesystem that
1768 * this file resides on.
1769 * 2) Ensure that at least one execute bit is on. Otherwise, a
1770 * privileged user will always succeed, and we don't want this
1771 * to happen unless the file really is executable.
1772 * 3) Ensure that the file is a regular file.
1774 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1775 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1776 (attr->va_type != VREG))
1780 * Zero length files can't be exec'd
1782 if (attr->va_size == 0)
1786 * Check for execute permission to file based on current credentials.
1788 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1793 * Check number of open-for-writes on the file and deny execution
1796 * Add a text reference now so no one can write to the
1797 * executable while we're activating it.
1799 * Remember if this was set before and unset it in case this is not
1800 * actually an executable image.
1802 error = VOP_SET_TEXT(vp);
1805 imgp->textset = true;
1808 * Call filesystem specific open routine (which does nothing in the
1811 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1818 * Exec handler registration
1821 exec_register(const struct execsw *execsw_arg)
1823 const struct execsw **es, **xs, **newexecsw;
1824 u_int count = 2; /* New slot and trailing NULL */
1827 for (es = execsw; *es; es++)
1829 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1832 for (es = execsw; *es; es++)
1837 free(execsw, M_TEMP);
1843 exec_unregister(const struct execsw *execsw_arg)
1845 const struct execsw **es, **xs, **newexecsw;
1849 panic("unregister with no handlers left?\n");
1851 for (es = execsw; *es; es++) {
1852 if (*es == execsw_arg)
1857 for (es = execsw; *es; es++)
1858 if (*es != execsw_arg)
1860 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1862 for (es = execsw; *es; es++)
1863 if (*es != execsw_arg)
1867 free(execsw, M_TEMP);