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>
41 #include <sys/capsicum.h>
42 #include <sys/eventhandler.h>
44 #include <sys/fcntl.h>
45 #include <sys/filedesc.h>
46 #include <sys/imgact.h>
47 #include <sys/imgact_elf.h>
48 #include <sys/kernel.h>
50 #include <sys/malloc.h>
52 #include <sys/mount.h>
53 #include <sys/mutex.h>
54 #include <sys/namei.h>
57 #include <sys/ptrace.h>
58 #include <sys/resourcevar.h>
59 #include <sys/rwlock.h>
60 #include <sys/sched.h>
62 #include <sys/sf_buf.h>
64 #include <sys/signalvar.h>
67 #include <sys/syscallsubr.h>
68 #include <sys/sysctl.h>
69 #include <sys/sysent.h>
70 #include <sys/sysproto.h>
71 #include <sys/timers.h>
73 #include <sys/vnode.h>
76 #include <sys/ktrace.h>
80 #include <vm/vm_param.h>
82 #include <vm/vm_page.h>
83 #include <vm/vm_map.h>
84 #include <vm/vm_kern.h>
85 #include <vm/vm_extern.h>
86 #include <vm/vm_object.h>
87 #include <vm/vm_pager.h>
90 #include <sys/pmckern.h>
93 #include <machine/reg.h>
95 #include <security/audit/audit.h>
96 #include <security/mac/mac_framework.h>
99 #include <sys/dtrace_bsd.h>
100 dtrace_execexit_func_t dtrace_fasttrap_exec;
103 SDT_PROVIDER_DECLARE(proc);
104 SDT_PROBE_DEFINE1(proc, , , exec, "char *");
105 SDT_PROBE_DEFINE1(proc, , , exec__failure, "int");
106 SDT_PROBE_DEFINE1(proc, , , exec__success, "char *");
108 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
110 int coredump_pack_fileinfo = 1;
111 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_fileinfo, CTLFLAG_RWTUN,
112 &coredump_pack_fileinfo, 0,
113 "Enable file path packing in 'procstat -f' coredump notes");
115 int coredump_pack_vmmapinfo = 1;
116 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_vmmapinfo, CTLFLAG_RWTUN,
117 &coredump_pack_vmmapinfo, 0,
118 "Enable file path packing in 'procstat -v' coredump notes");
120 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
121 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
122 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
123 static int do_execve(struct thread *td, struct image_args *args,
124 struct mac *mac_p, struct vmspace *oldvmspace);
126 /* XXX This should be vm_size_t. */
127 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD|
128 CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_ps_strings, "LU",
129 "Location of process' ps_strings structure");
131 /* XXX This should be vm_size_t. */
132 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD|
133 CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_usrstack, "LU",
134 "Top of process stack");
136 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD|CTLFLAG_MPSAFE,
137 NULL, 0, sysctl_kern_stackprot, "I",
138 "Stack memory permissions");
140 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
141 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
142 &ps_arg_cache_limit, 0,
143 "Process' command line characters cache limit");
145 static int disallow_high_osrel;
146 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW,
147 &disallow_high_osrel, 0,
148 "Disallow execution of binaries built for higher version of the world");
150 static int map_at_zero = 0;
151 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RWTUN, &map_at_zero, 0,
152 "Permit processes to map an object at virtual address 0.");
154 int core_dump_can_intr = 1;
155 SYSCTL_INT(_kern, OID_AUTO, core_dump_can_intr, CTLFLAG_RWTUN,
156 &core_dump_can_intr, 0,
157 "Core dumping interruptible with SIGKILL");
160 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
167 if (req->flags & SCTL_MASK32) {
169 val = (unsigned int)p->p_sysent->sv_psstrings;
170 error = SYSCTL_OUT(req, &val, sizeof(val));
173 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
174 sizeof(p->p_sysent->sv_psstrings));
179 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
186 if (req->flags & SCTL_MASK32) {
188 val = (unsigned int)p->p_sysent->sv_usrstack;
189 error = SYSCTL_OUT(req, &val, sizeof(val));
192 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
193 sizeof(p->p_sysent->sv_usrstack));
198 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
203 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
204 sizeof(p->p_sysent->sv_stackprot)));
208 * Each of the items is a pointer to a `const struct execsw', hence the
209 * double pointer here.
211 static const struct execsw **execsw;
213 #ifndef _SYS_SYSPROTO_H_
222 sys_execve(struct thread *td, struct execve_args *uap)
224 struct image_args args;
225 struct vmspace *oldvmspace;
228 error = pre_execve(td, &oldvmspace);
231 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
232 uap->argv, uap->envv);
234 error = kern_execve(td, &args, NULL, oldvmspace);
235 post_execve(td, error, oldvmspace);
236 AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
240 #ifndef _SYS_SYSPROTO_H_
241 struct fexecve_args {
248 sys_fexecve(struct thread *td, struct fexecve_args *uap)
250 struct image_args args;
251 struct vmspace *oldvmspace;
254 error = pre_execve(td, &oldvmspace);
257 error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
258 uap->argv, uap->envv);
261 error = kern_execve(td, &args, NULL, oldvmspace);
263 post_execve(td, error, oldvmspace);
264 AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
268 #ifndef _SYS_SYSPROTO_H_
269 struct __mac_execve_args {
278 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap)
281 struct image_args args;
282 struct vmspace *oldvmspace;
285 error = pre_execve(td, &oldvmspace);
288 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
289 uap->argv, uap->envv);
291 error = kern_execve(td, &args, uap->mac_p, oldvmspace);
292 post_execve(td, error, oldvmspace);
293 AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
301 pre_execve(struct thread *td, struct vmspace **oldvmspace)
306 KASSERT(td == curthread, ("non-current thread %p", td));
309 if ((p->p_flag & P_HADTHREADS) != 0) {
311 if (thread_single(p, SINGLE_BOUNDARY) != 0)
315 KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0,
317 *oldvmspace = p->p_vmspace;
322 post_execve(struct thread *td, int error, struct vmspace *oldvmspace)
326 KASSERT(td == curthread, ("non-current thread %p", td));
328 if ((p->p_flag & P_HADTHREADS) != 0) {
331 * If success, we upgrade to SINGLE_EXIT state to
332 * force other threads to suicide.
334 if (error == EJUSTRETURN)
335 thread_single(p, SINGLE_EXIT);
337 thread_single_end(p, SINGLE_BOUNDARY);
340 exec_cleanup(td, oldvmspace);
344 * kern_execve() has the astonishing property of not always returning to
345 * the caller. If sufficiently bad things happen during the call to
346 * do_execve(), it can end up calling exit1(); as a result, callers must
347 * avoid doing anything which they might need to undo (e.g., allocating
351 kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p,
352 struct vmspace *oldvmspace)
355 AUDIT_ARG_ARGV(args->begin_argv, args->argc,
356 exec_args_get_begin_envv(args) - args->begin_argv);
357 AUDIT_ARG_ENVV(exec_args_get_begin_envv(args), args->envc,
358 args->endp - exec_args_get_begin_envv(args));
359 return (do_execve(td, args, mac_p, oldvmspace));
363 execve_nosetid(struct image_params *imgp)
365 imgp->credential_setid = false;
366 if (imgp->newcred != NULL) {
367 crfree(imgp->newcred);
368 imgp->newcred = NULL;
373 * In-kernel implementation of execve(). All arguments are assumed to be
374 * userspace pointers from the passed thread.
377 do_execve(struct thread *td, struct image_args *args, struct mac *mac_p,
378 struct vmspace *oldvmspace)
380 struct proc *p = td->td_proc;
382 struct ucred *oldcred;
383 struct uidinfo *euip = NULL;
384 uintptr_t stack_base;
385 struct image_params image_params, *imgp;
387 int (*img_first)(struct image_params *);
388 struct pargs *oldargs = NULL, *newargs = NULL;
389 struct sigacts *oldsigacts = NULL, *newsigacts = NULL;
391 struct ktr_io_params *kiop;
393 struct vnode *oldtextvp = NULL, *newtextvp;
394 bool credential_changing;
396 struct label *interpvplabel = NULL;
397 bool will_transition;
400 struct pmckern_procexec pe;
402 int error, i, orig_osrel;
404 static const char fexecv_proc_title[] = "(fexecv)";
406 imgp = &image_params;
412 * Lock the process and set the P_INEXEC flag to indicate that
413 * it should be left alone until we're done here. This is
414 * necessary to avoid race conditions - e.g. in ptrace() -
415 * that might allow a local user to illicitly obtain elevated
419 KASSERT((p->p_flag & P_INEXEC) == 0,
420 ("%s(): process already has P_INEXEC flag", __func__));
421 p->p_flag |= P_INEXEC;
425 * Initialize part of the common data
427 bzero(imgp, sizeof(*imgp));
431 oldcred = p->p_ucred;
432 orig_osrel = p->p_osrel;
433 orig_fctl0 = p->p_fctl0;
436 error = mac_execve_enter(imgp, mac_p);
442 * Translate the file name. namei() returns a vnode pointer
443 * in ni_vp among other things.
445 * XXXAUDIT: It would be desirable to also audit the name of the
446 * interpreter if this is an interpreted binary.
448 if (args->fname != NULL) {
449 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW |
450 SAVENAME | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
453 SDT_PROBE1(proc, , , exec, args->fname);
456 if (args->fname != NULL) {
457 #ifdef CAPABILITY_MODE
459 * While capability mode can't reach this point via direct
460 * path arguments to execve(), we also don't allow
461 * interpreters to be used in capability mode (for now).
462 * Catch indirect lookups and return a permissions error.
464 if (IN_CAPABILITY_MODE(td)) {
473 newtextvp = nd.ni_vp;
474 imgp->vp = newtextvp;
476 AUDIT_ARG_FD(args->fd);
478 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
480 error = fgetvp_exec(td, args->fd, &cap_fexecve_rights,
484 vn_lock(newtextvp, LK_SHARED | LK_RETRY);
485 AUDIT_ARG_VNODE1(newtextvp);
486 imgp->vp = newtextvp;
490 * Check file permissions. Also 'opens' file and sets its vnode to
493 error = exec_check_permissions(imgp);
495 goto exec_fail_dealloc;
497 imgp->object = imgp->vp->v_object;
498 if (imgp->object != NULL)
499 vm_object_reference(imgp->object);
501 error = exec_map_first_page(imgp);
503 goto exec_fail_dealloc;
505 imgp->proc->p_osrel = 0;
506 imgp->proc->p_fctl0 = 0;
509 * Implement image setuid/setgid.
511 * Determine new credentials before attempting image activators
512 * so that it can be used by process_exec handlers to determine
513 * credential/setid changes.
515 * Don't honor setuid/setgid if the filesystem prohibits it or if
516 * the process is being traced.
518 * We disable setuid/setgid/etc in capability mode on the basis
519 * that most setugid applications are not written with that
520 * environment in mind, and will therefore almost certainly operate
521 * incorrectly. In principle there's no reason that setugid
522 * applications might not be useful in capability mode, so we may want
523 * to reconsider this conservative design choice in the future.
525 * XXXMAC: For the time being, use NOSUID to also prohibit
526 * transitions on the file system.
528 credential_changing = false;
529 credential_changing |= (attr.va_mode & S_ISUID) &&
530 oldcred->cr_uid != attr.va_uid;
531 credential_changing |= (attr.va_mode & S_ISGID) &&
532 oldcred->cr_gid != attr.va_gid;
534 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
535 interpvplabel, imgp) != 0;
536 credential_changing |= will_transition;
539 /* Don't inherit PROC_PDEATHSIG_CTL value if setuid/setgid. */
540 if (credential_changing)
541 imgp->proc->p_pdeathsig = 0;
543 if (credential_changing &&
544 #ifdef CAPABILITY_MODE
545 ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
547 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
548 (p->p_flag & P_TRACED) == 0) {
549 imgp->credential_setid = true;
550 VOP_UNLOCK(imgp->vp);
551 imgp->newcred = crdup(oldcred);
552 if (attr.va_mode & S_ISUID) {
553 euip = uifind(attr.va_uid);
554 change_euid(imgp->newcred, euip);
556 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
557 if (attr.va_mode & S_ISGID)
558 change_egid(imgp->newcred, attr.va_gid);
560 * Implement correct POSIX saved-id behavior.
562 * XXXMAC: Note that the current logic will save the
563 * uid and gid if a MAC domain transition occurs, even
564 * though maybe it shouldn't.
566 change_svuid(imgp->newcred, imgp->newcred->cr_uid);
567 change_svgid(imgp->newcred, imgp->newcred->cr_gid);
570 * Implement correct POSIX saved-id behavior.
572 * XXX: It's not clear that the existing behavior is
573 * POSIX-compliant. A number of sources indicate that the
574 * saved uid/gid should only be updated if the new ruid is
575 * not equal to the old ruid, or the new euid is not equal
576 * to the old euid and the new euid is not equal to the old
577 * ruid. The FreeBSD code always updates the saved uid/gid.
578 * Also, this code uses the new (replaced) euid and egid as
579 * the source, which may or may not be the right ones to use.
581 if (oldcred->cr_svuid != oldcred->cr_uid ||
582 oldcred->cr_svgid != oldcred->cr_gid) {
583 VOP_UNLOCK(imgp->vp);
584 imgp->newcred = crdup(oldcred);
585 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
586 change_svuid(imgp->newcred, imgp->newcred->cr_uid);
587 change_svgid(imgp->newcred, imgp->newcred->cr_gid);
590 /* The new credentials are installed into the process later. */
593 * Do the best to calculate the full path to the image file.
595 if (args->fname != NULL && args->fname[0] == '/')
596 imgp->execpath = args->fname;
598 VOP_UNLOCK(imgp->vp);
599 if (vn_fullpath(imgp->vp, &imgp->execpath,
600 &imgp->freepath) != 0)
601 imgp->execpath = args->fname;
602 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
606 * If the current process has a special image activator it
607 * wants to try first, call it. For example, emulating shell
608 * scripts differently.
611 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
612 error = img_first(imgp);
615 * Loop through the list of image activators, calling each one.
616 * An activator returns -1 if there is no match, 0 on success,
617 * and an error otherwise.
619 for (i = 0; error == -1 && execsw[i]; ++i) {
620 if (execsw[i]->ex_imgact == NULL ||
621 execsw[i]->ex_imgact == img_first) {
624 error = (*execsw[i]->ex_imgact)(imgp);
630 goto exec_fail_dealloc;
634 * Special interpreter operation, cleanup and loop up to try to
635 * activate the interpreter.
637 if (imgp->interpreted) {
638 exec_unmap_first_page(imgp);
640 * The text reference needs to be removed for scripts.
641 * There is a short period before we determine that
642 * something is a script where text reference is active.
643 * The vnode lock is held over this entire period
644 * so nothing should illegitimately be blocked.
646 MPASS(imgp->textset);
647 VOP_UNSET_TEXT_CHECKED(newtextvp);
648 imgp->textset = false;
649 /* free name buffer and old vnode */
650 if (args->fname != NULL)
651 NDFREE(&nd, NDF_ONLY_PNBUF);
653 mac_execve_interpreter_enter(newtextvp, &interpvplabel);
656 VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td);
657 imgp->opened = false;
660 vm_object_deallocate(imgp->object);
662 execve_nosetid(imgp);
663 imgp->execpath = NULL;
664 free(imgp->freepath, M_TEMP);
665 imgp->freepath = NULL;
666 /* set new name to that of the interpreter */
667 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW |
668 SAVENAME, UIO_SYSSPACE, imgp->interpreter_name, td);
669 args->fname = imgp->interpreter_name;
674 * NB: We unlock the vnode here because it is believed that none
675 * of the sv_copyout_strings/sv_fixup operations require the vnode.
677 VOP_UNLOCK(imgp->vp);
679 if (disallow_high_osrel &&
680 P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
682 uprintf("Osrel %d for image %s too high\n", p->p_osrel,
683 imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
684 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
685 goto exec_fail_dealloc;
688 /* ABI enforces the use of Capsicum. Switch into capabilities mode. */
689 if (SV_PROC_FLAG(p, SV_CAPSICUM))
690 sys_cap_enter(td, NULL);
693 * Copy out strings (args and env) and initialize stack base.
695 error = (*p->p_sysent->sv_copyout_strings)(imgp, &stack_base);
697 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
698 goto exec_fail_dealloc;
704 error = (*p->p_sysent->sv_fixup)(&stack_base, imgp);
706 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
707 goto exec_fail_dealloc;
710 if (args->fdp != NULL) {
711 /* Install a brand new file descriptor table. */
712 fdinstall_remapped(td, args->fdp);
716 * Keep on using the existing file descriptor table. For
717 * security and other reasons, the file descriptor table
718 * cannot be shared after an exec.
722 /* close files on exec */
727 * Malloc things before we need locks.
729 i = exec_args_get_begin_envv(imgp->args) - imgp->args->begin_argv;
730 /* Cache arguments if they fit inside our allowance */
731 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
732 newargs = pargs_alloc(i);
733 bcopy(imgp->args->begin_argv, newargs->ar_args, i);
737 * For security and other reasons, signal handlers cannot
738 * be shared after an exec. The new process gets a copy of the old
739 * handlers. In execsigs(), the new process will have its signals
742 if (sigacts_shared(p->p_sigacts)) {
743 oldsigacts = p->p_sigacts;
744 newsigacts = sigacts_alloc();
745 sigacts_copy(newsigacts, oldsigacts);
748 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
752 p->p_sigacts = newsigacts;
756 /* reset caught signals */
759 /* name this process - nameiexec(p, ndp) */
760 bzero(p->p_comm, sizeof(p->p_comm));
762 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
763 min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
764 else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0)
765 bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
766 bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
768 sched_clear_tdname(td);
772 * mark as execed, wakeup the process that vforked (if any) and tell
773 * it that it now has its own resources back
776 if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0)
777 p->p_flag2 &= ~P2_NOTRACE;
778 if ((p->p_flag2 & P2_STKGAP_DISABLE_EXEC) == 0)
779 p->p_flag2 &= ~P2_STKGAP_DISABLE;
780 if (p->p_flag & P_PPWAIT) {
781 p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
782 cv_broadcast(&p->p_pwait);
783 /* STOPs are no longer ignored, arrange for AST */
787 if ((imgp->sysent->sv_setid_allowed != NULL &&
788 !(*imgp->sysent->sv_setid_allowed)(td, imgp)) ||
789 (p->p_flag2 & P2_NO_NEW_PRIVS) != 0)
790 execve_nosetid(imgp);
793 * Implement image setuid/setgid installation.
795 if (imgp->credential_setid) {
797 * Turn off syscall tracing for set-id programs, except for
798 * root. Record any set-id flags first to make sure that
799 * we do not regain any tracing during a possible block.
803 kiop = ktrprocexec(p);
806 * Close any file descriptors 0..2 that reference procfs,
807 * then make sure file descriptors 0..2 are in use.
809 * Both fdsetugidsafety() and fdcheckstd() may call functions
810 * taking sleepable locks, so temporarily drop our locks.
813 VOP_UNLOCK(imgp->vp);
815 error = fdcheckstd(td);
816 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
818 goto exec_fail_dealloc;
821 if (will_transition) {
822 mac_vnode_execve_transition(oldcred, imgp->newcred,
823 imgp->vp, interpvplabel, imgp);
827 if (oldcred->cr_uid == oldcred->cr_ruid &&
828 oldcred->cr_gid == oldcred->cr_rgid)
829 p->p_flag &= ~P_SUGID;
832 * Set the new credentials.
834 if (imgp->newcred != NULL) {
835 proc_set_cred(p, imgp->newcred);
841 * Store the vp for use in procfs. This vnode was referenced by namei
844 oldtextvp = p->p_textvp;
845 p->p_textvp = newtextvp;
849 * Tell the DTrace fasttrap provider about the exec if it
850 * has declared an interest.
852 if (dtrace_fasttrap_exec)
853 dtrace_fasttrap_exec(p);
857 * Notify others that we exec'd, and clear the P_INEXEC flag
858 * as we're now a bona fide freshly-execed process.
860 KNOTE_LOCKED(p->p_klist, NOTE_EXEC);
861 p->p_flag &= ~P_INEXEC;
863 /* clear "fork but no exec" flag, as we _are_ execing */
864 p->p_acflag &= ~AFORK;
867 * Free any previous argument cache and replace it with
868 * the new argument cache, if any.
878 * Check if system-wide sampling is in effect or if the
879 * current process is using PMCs. If so, do exec() time
880 * processing. This processing needs to happen AFTER the
881 * P_INEXEC flag is cleared.
883 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
884 VOP_UNLOCK(imgp->vp);
885 pe.pm_credentialschanged = credential_changing;
886 pe.pm_entryaddr = imgp->entry_addr;
888 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
889 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
893 /* Set values passed into the program in registers. */
894 (*p->p_sysent->sv_setregs)(td, imgp, stack_base);
896 VOP_MMAPPED(imgp->vp);
898 SDT_PROBE1(proc, , , exec__success, args->fname);
902 p->p_osrel = orig_osrel;
903 p->p_fctl0 = orig_fctl0;
906 if (imgp->firstpage != NULL)
907 exec_unmap_first_page(imgp);
909 if (imgp->vp != NULL) {
911 NDFREE(&nd, NDF_ONLY_PNBUF);
913 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
915 VOP_UNSET_TEXT_CHECKED(imgp->vp);
919 VOP_UNLOCK(imgp->vp);
922 if (imgp->object != NULL)
923 vm_object_deallocate(imgp->object);
925 free(imgp->freepath, M_TEMP);
928 if (p->p_ptevents & PTRACE_EXEC) {
930 if (p->p_ptevents & PTRACE_EXEC)
931 td->td_dbgflags |= TDB_EXEC;
936 /* we're done here, clear P_INEXEC */
938 p->p_flag &= ~P_INEXEC;
941 SDT_PROBE1(proc, , , exec__failure, error);
944 if (imgp->newcred != NULL && oldcred != NULL)
945 crfree(imgp->newcred);
948 mac_execve_exit(imgp);
949 mac_execve_interpreter_exit(interpvplabel);
951 exec_free_args(args);
954 * Handle deferred decrement of ref counts.
956 if (oldtextvp != NULL)
959 ktr_io_params_free(kiop);
963 if (oldsigacts != NULL)
964 sigacts_free(oldsigacts);
968 if (error && imgp->vmspace_destroyed) {
969 /* sorry, no more process anymore. exit gracefully */
970 exec_cleanup(td, oldvmspace);
971 exit1(td, 0, SIGABRT);
981 * We don't want cpu_set_syscall_retval() to overwrite any of
982 * the register values put in place by exec_setregs().
983 * Implementations of cpu_set_syscall_retval() will leave
984 * registers unmodified when returning EJUSTRETURN.
986 return (error == 0 ? EJUSTRETURN : error);
990 exec_cleanup(struct thread *td, struct vmspace *oldvmspace)
992 if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
993 KASSERT(td->td_proc->p_vmspace != oldvmspace,
994 ("oldvmspace still used"));
995 vmspace_free(oldvmspace);
996 td->td_pflags &= ~TDP_EXECVMSPC;
1001 exec_map_first_page(struct image_params *imgp)
1007 if (imgp->firstpage != NULL)
1008 exec_unmap_first_page(imgp);
1010 object = imgp->vp->v_object;
1013 #if VM_NRESERVLEVEL > 0
1014 if ((object->flags & OBJ_COLORED) == 0) {
1015 VM_OBJECT_WLOCK(object);
1016 vm_object_color(object, 0);
1017 VM_OBJECT_WUNLOCK(object);
1020 error = vm_page_grab_valid_unlocked(&m, object, 0,
1021 VM_ALLOC_COUNT(VM_INITIAL_PAGEIN) |
1022 VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED);
1024 if (error != VM_PAGER_OK)
1026 imgp->firstpage = sf_buf_alloc(m, 0);
1027 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
1033 exec_unmap_first_page(struct image_params *imgp)
1037 if (imgp->firstpage != NULL) {
1038 m = sf_buf_page(imgp->firstpage);
1039 sf_buf_free(imgp->firstpage);
1040 imgp->firstpage = NULL;
1041 vm_page_unwire(m, PQ_ACTIVE);
1046 exec_onexec_old(struct thread *td)
1048 sigfastblock_clear(td);
1049 umtx_exec(td->td_proc);
1053 * This is an optimization which removes the unmanaged shared page
1054 * mapping. In combination with pmap_remove_pages(), which cleans all
1055 * managed mappings in the process' vmspace pmap, no work will be left
1056 * for pmap_remove(min, max).
1059 exec_free_abi_mappings(struct proc *p)
1061 struct vmspace *vmspace;
1062 struct sysentvec *sv;
1064 vmspace = p->p_vmspace;
1065 if (refcount_load(&vmspace->vm_refcnt) != 1)
1069 if (sv->sv_shared_page_obj == NULL)
1072 pmap_remove(vmspace_pmap(vmspace), sv->sv_shared_page_base,
1073 sv->sv_shared_page_base + sv->sv_shared_page_len);
1077 * Destroy old address space, and allocate a new stack.
1078 * The new stack is only sgrowsiz large because it is grown
1079 * automatically on a page fault.
1082 exec_new_vmspace(struct image_params *imgp, struct sysentvec *sv)
1085 struct proc *p = imgp->proc;
1086 struct vmspace *vmspace = p->p_vmspace;
1087 struct thread *td = curthread;
1089 struct rlimit rlim_stack;
1090 vm_offset_t sv_minuser, stack_addr;
1092 vm_prot_t stack_prot;
1095 imgp->vmspace_destroyed = true;
1098 if (p->p_sysent->sv_onexec_old != NULL)
1099 p->p_sysent->sv_onexec_old(td);
1101 if (sv->sv_onexec != NULL)
1102 sv->sv_onexec(p, imgp);
1104 EVENTHANDLER_DIRECT_INVOKE(process_exec, p, imgp);
1107 * Blow away entire process VM, if address space not shared,
1108 * otherwise, create a new VM space so that other threads are
1111 map = &vmspace->vm_map;
1113 sv_minuser = sv->sv_minuser;
1115 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1116 if (refcount_load(&vmspace->vm_refcnt) == 1 &&
1117 vm_map_min(map) == sv_minuser &&
1118 vm_map_max(map) == sv->sv_maxuser &&
1119 cpu_exec_vmspace_reuse(p, map)) {
1120 exec_free_abi_mappings(p);
1122 pmap_remove_pages(vmspace_pmap(vmspace));
1123 vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1125 * An exec terminates mlockall(MCL_FUTURE).
1126 * ASLR and W^X states must be re-evaluated.
1129 vm_map_modflags(map, 0, MAP_WIREFUTURE | MAP_ASLR |
1130 MAP_ASLR_IGNSTART | MAP_WXORX);
1133 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1136 vmspace = p->p_vmspace;
1137 map = &vmspace->vm_map;
1139 map->flags |= imgp->map_flags;
1141 /* Map a shared page */
1142 obj = sv->sv_shared_page_obj;
1144 vm_object_reference(obj);
1145 error = vm_map_fixed(map, obj, 0,
1146 sv->sv_shared_page_base, sv->sv_shared_page_len,
1147 VM_PROT_READ | VM_PROT_EXECUTE,
1148 VM_PROT_READ | VM_PROT_EXECUTE,
1149 MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1150 if (error != KERN_SUCCESS) {
1151 vm_object_deallocate(obj);
1152 return (vm_mmap_to_errno(error));
1156 /* Allocate a new stack */
1157 if (imgp->stack_sz != 0) {
1158 ssiz = trunc_page(imgp->stack_sz);
1160 lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack);
1162 if (ssiz > rlim_stack.rlim_max)
1163 ssiz = rlim_stack.rlim_max;
1164 if (ssiz > rlim_stack.rlim_cur) {
1165 rlim_stack.rlim_cur = ssiz;
1166 kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack);
1168 } else if (sv->sv_maxssiz != NULL) {
1169 ssiz = *sv->sv_maxssiz;
1173 imgp->eff_stack_sz = lim_cur(curthread, RLIMIT_STACK);
1174 if (ssiz < imgp->eff_stack_sz)
1175 imgp->eff_stack_sz = ssiz;
1176 stack_addr = sv->sv_usrstack - ssiz;
1177 stack_prot = obj != NULL && imgp->stack_prot != 0 ?
1178 imgp->stack_prot : sv->sv_stackprot;
1179 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz, stack_prot,
1180 VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1181 if (error != KERN_SUCCESS) {
1182 uprintf("exec_new_vmspace: mapping stack size %#jx prot %#x "
1183 "failed mach error %d errno %d\n", (uintmax_t)ssiz,
1184 stack_prot, error, vm_mmap_to_errno(error));
1185 return (vm_mmap_to_errno(error));
1189 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they
1190 * are still used to enforce the stack rlimit on the process stack.
1192 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1193 vmspace->vm_maxsaddr = (char *)stack_addr;
1199 * Copy out argument and environment strings from the old process address
1200 * space into the temporary string buffer.
1203 exec_copyin_args(struct image_args *args, const char *fname,
1204 enum uio_seg segflg, char **argv, char **envv)
1209 bzero(args, sizeof(*args));
1214 * Allocate demand-paged memory for the file name, argument, and
1215 * environment strings.
1217 error = exec_alloc_args(args);
1222 * Copy the file name.
1224 error = exec_args_add_fname(args, fname, segflg);
1229 * extract arguments first
1232 error = fueword(argv++, &arg);
1239 error = exec_args_add_arg(args, (char *)(uintptr_t)arg,
1246 * extract environment strings
1250 error = fueword(envv++, &env);
1257 error = exec_args_add_env(args,
1258 (char *)(uintptr_t)env, UIO_USERSPACE);
1267 exec_free_args(args);
1272 exec_copyin_data_fds(struct thread *td, struct image_args *args,
1273 const void *data, size_t datalen, const int *fds, size_t fdslen)
1275 struct filedesc *ofdp;
1280 memset(args, '\0', sizeof(*args));
1281 ofdp = td->td_proc->p_fd;
1282 if (datalen >= ARG_MAX || fdslen >= ofdp->fd_nfiles)
1284 error = exec_alloc_args(args);
1288 args->begin_argv = args->buf;
1289 args->stringspace = ARG_MAX;
1293 * Argument buffer has been provided. Copy it into the
1294 * kernel as a single string and add a terminating null
1297 error = copyin(data, args->begin_argv, datalen);
1300 args->begin_argv[datalen] = '\0';
1301 args->endp = args->begin_argv + datalen + 1;
1302 args->stringspace -= datalen + 1;
1305 * Traditional argument counting. Count the number of
1308 for (p = args->begin_argv; p < args->endp; ++p)
1312 /* No argument buffer provided. */
1313 args->endp = args->begin_argv;
1316 /* Create new file descriptor table. */
1317 kfds = malloc(fdslen * sizeof(int), M_TEMP, M_WAITOK);
1318 error = copyin(fds, kfds, fdslen * sizeof(int));
1323 error = fdcopy_remapped(ofdp, kfds, fdslen, &args->fdp);
1330 exec_free_args(args);
1334 struct exec_args_kva {
1337 SLIST_ENTRY(exec_args_kva) next;
1340 DPCPU_DEFINE_STATIC(struct exec_args_kva *, exec_args_kva);
1342 static SLIST_HEAD(, exec_args_kva) exec_args_kva_freelist;
1343 static struct mtx exec_args_kva_mtx;
1344 static u_int exec_args_gen;
1347 exec_prealloc_args_kva(void *arg __unused)
1349 struct exec_args_kva *argkva;
1352 SLIST_INIT(&exec_args_kva_freelist);
1353 mtx_init(&exec_args_kva_mtx, "exec args kva", NULL, MTX_DEF);
1354 for (i = 0; i < exec_map_entries; i++) {
1355 argkva = malloc(sizeof(*argkva), M_PARGS, M_WAITOK);
1356 argkva->addr = kmap_alloc_wait(exec_map, exec_map_entry_size);
1357 argkva->gen = exec_args_gen;
1358 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1361 SYSINIT(exec_args_kva, SI_SUB_EXEC, SI_ORDER_ANY, exec_prealloc_args_kva, NULL);
1364 exec_alloc_args_kva(void **cookie)
1366 struct exec_args_kva *argkva;
1368 argkva = (void *)atomic_readandclear_ptr(
1369 (uintptr_t *)DPCPU_PTR(exec_args_kva));
1370 if (argkva == NULL) {
1371 mtx_lock(&exec_args_kva_mtx);
1372 while ((argkva = SLIST_FIRST(&exec_args_kva_freelist)) == NULL)
1373 (void)mtx_sleep(&exec_args_kva_freelist,
1374 &exec_args_kva_mtx, 0, "execkva", 0);
1375 SLIST_REMOVE_HEAD(&exec_args_kva_freelist, next);
1376 mtx_unlock(&exec_args_kva_mtx);
1378 kasan_mark((void *)argkva->addr, exec_map_entry_size,
1379 exec_map_entry_size, 0);
1380 *(struct exec_args_kva **)cookie = argkva;
1381 return (argkva->addr);
1385 exec_release_args_kva(struct exec_args_kva *argkva, u_int gen)
1389 base = argkva->addr;
1390 kasan_mark((void *)argkva->addr, 0, exec_map_entry_size,
1391 KASAN_EXEC_ARGS_FREED);
1392 if (argkva->gen != gen) {
1393 (void)vm_map_madvise(exec_map, base, base + exec_map_entry_size,
1397 if (!atomic_cmpset_ptr((uintptr_t *)DPCPU_PTR(exec_args_kva),
1398 (uintptr_t)NULL, (uintptr_t)argkva)) {
1399 mtx_lock(&exec_args_kva_mtx);
1400 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1401 wakeup_one(&exec_args_kva_freelist);
1402 mtx_unlock(&exec_args_kva_mtx);
1407 exec_free_args_kva(void *cookie)
1410 exec_release_args_kva(cookie, exec_args_gen);
1414 exec_args_kva_lowmem(void *arg __unused)
1416 SLIST_HEAD(, exec_args_kva) head;
1417 struct exec_args_kva *argkva;
1421 gen = atomic_fetchadd_int(&exec_args_gen, 1) + 1;
1424 * Force an madvise of each KVA range. Any currently allocated ranges
1425 * will have MADV_FREE applied once they are freed.
1428 mtx_lock(&exec_args_kva_mtx);
1429 SLIST_SWAP(&head, &exec_args_kva_freelist, exec_args_kva);
1430 mtx_unlock(&exec_args_kva_mtx);
1431 while ((argkva = SLIST_FIRST(&head)) != NULL) {
1432 SLIST_REMOVE_HEAD(&head, next);
1433 exec_release_args_kva(argkva, gen);
1437 argkva = (void *)atomic_readandclear_ptr(
1438 (uintptr_t *)DPCPU_ID_PTR(i, exec_args_kva));
1440 exec_release_args_kva(argkva, gen);
1443 EVENTHANDLER_DEFINE(vm_lowmem, exec_args_kva_lowmem, NULL,
1444 EVENTHANDLER_PRI_ANY);
1447 * Allocate temporary demand-paged, zero-filled memory for the file name,
1448 * argument, and environment strings.
1451 exec_alloc_args(struct image_args *args)
1454 args->buf = (char *)exec_alloc_args_kva(&args->bufkva);
1459 exec_free_args(struct image_args *args)
1462 if (args->buf != NULL) {
1463 exec_free_args_kva(args->bufkva);
1466 if (args->fname_buf != NULL) {
1467 free(args->fname_buf, M_TEMP);
1468 args->fname_buf = NULL;
1470 if (args->fdp != NULL)
1471 fdescfree_remapped(args->fdp);
1475 * A set to functions to fill struct image args.
1477 * NOTE: exec_args_add_fname() must be called (possibly with a NULL
1478 * fname) before the other functions. All exec_args_add_arg() calls must
1479 * be made before any exec_args_add_env() calls. exec_args_adjust_args()
1480 * may be called any time after exec_args_add_fname().
1482 * exec_args_add_fname() - install path to be executed
1483 * exec_args_add_arg() - append an argument string
1484 * exec_args_add_env() - append an env string
1485 * exec_args_adjust_args() - adjust location of the argument list to
1486 * allow new arguments to be prepended
1489 exec_args_add_fname(struct image_args *args, const char *fname,
1490 enum uio_seg segflg)
1495 KASSERT(args->fname == NULL, ("fname already appended"));
1496 KASSERT(args->endp == NULL, ("already appending to args"));
1498 if (fname != NULL) {
1499 args->fname = args->buf;
1500 error = segflg == UIO_SYSSPACE ?
1501 copystr(fname, args->fname, PATH_MAX, &length) :
1502 copyinstr(fname, args->fname, PATH_MAX, &length);
1504 return (error == ENAMETOOLONG ? E2BIG : error);
1508 /* Set up for _arg_*()/_env_*() */
1509 args->endp = args->buf + length;
1510 /* begin_argv must be set and kept updated */
1511 args->begin_argv = args->endp;
1512 KASSERT(exec_map_entry_size - length >= ARG_MAX,
1513 ("too little space remaining for arguments %zu < %zu",
1514 exec_map_entry_size - length, (size_t)ARG_MAX));
1515 args->stringspace = ARG_MAX;
1521 exec_args_add_str(struct image_args *args, const char *str,
1522 enum uio_seg segflg, int *countp)
1527 KASSERT(args->endp != NULL, ("endp not initialized"));
1528 KASSERT(args->begin_argv != NULL, ("begin_argp not initialized"));
1530 error = (segflg == UIO_SYSSPACE) ?
1531 copystr(str, args->endp, args->stringspace, &length) :
1532 copyinstr(str, args->endp, args->stringspace, &length);
1534 return (error == ENAMETOOLONG ? E2BIG : error);
1535 args->stringspace -= length;
1536 args->endp += length;
1543 exec_args_add_arg(struct image_args *args, const char *argp,
1544 enum uio_seg segflg)
1547 KASSERT(args->envc == 0, ("appending args after env"));
1549 return (exec_args_add_str(args, argp, segflg, &args->argc));
1553 exec_args_add_env(struct image_args *args, const char *envp,
1554 enum uio_seg segflg)
1557 if (args->envc == 0)
1558 args->begin_envv = args->endp;
1560 return (exec_args_add_str(args, envp, segflg, &args->envc));
1564 exec_args_adjust_args(struct image_args *args, size_t consume, ssize_t extend)
1568 KASSERT(args->endp != NULL, ("endp not initialized"));
1569 KASSERT(args->begin_argv != NULL, ("begin_argp not initialized"));
1571 offset = extend - consume;
1572 if (args->stringspace < offset)
1574 memmove(args->begin_argv + extend, args->begin_argv + consume,
1575 args->endp - args->begin_argv + consume);
1577 args->begin_envv += offset;
1578 args->endp += offset;
1579 args->stringspace -= offset;
1584 exec_args_get_begin_envv(struct image_args *args)
1587 KASSERT(args->endp != NULL, ("endp not initialized"));
1590 return (args->begin_envv);
1591 return (args->endp);
1595 exec_stackgap(struct image_params *imgp, uintptr_t *dp)
1597 if (imgp->sysent->sv_stackgap == NULL ||
1598 (imgp->proc->p_fctl0 & (NT_FREEBSD_FCTL_ASLR_DISABLE |
1599 NT_FREEBSD_FCTL_ASG_DISABLE)) != 0 ||
1600 (imgp->map_flags & MAP_ASLR) == 0)
1602 imgp->sysent->sv_stackgap(imgp, dp);
1606 * Copy strings out to the new process address space, constructing new arg
1607 * and env vector tables. Return a pointer to the base so that it can be used
1608 * as the initial stack pointer.
1611 exec_copyout_strings(struct image_params *imgp, uintptr_t *stack_base)
1616 uintptr_t destp, ustringp;
1617 struct ps_strings *arginfo;
1619 size_t execpath_len;
1620 int error, szsigcode, szps;
1621 char canary[sizeof(long) * 8];
1623 szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1625 * Calculate string base and vector table pointers.
1626 * Also deal with signal trampoline code for this exec type.
1628 if (imgp->execpath != NULL && imgp->auxargs != NULL)
1629 execpath_len = strlen(imgp->execpath) + 1;
1634 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1635 imgp->ps_strings = arginfo;
1636 if (p->p_sysent->sv_sigcode_base == 0) {
1637 if (p->p_sysent->sv_szsigcode != NULL)
1638 szsigcode = *(p->p_sysent->sv_szsigcode);
1640 destp = (uintptr_t)arginfo;
1645 if (szsigcode != 0) {
1647 destp = rounddown2(destp, sizeof(void *));
1648 error = copyout(p->p_sysent->sv_sigcode, (void *)destp,
1655 * Copy the image path for the rtld.
1657 if (execpath_len != 0) {
1658 destp -= execpath_len;
1659 destp = rounddown2(destp, sizeof(void *));
1660 imgp->execpathp = (void *)destp;
1661 error = copyout(imgp->execpath, imgp->execpathp, execpath_len);
1667 * Prepare the canary for SSP.
1669 arc4rand(canary, sizeof(canary), 0);
1670 destp -= sizeof(canary);
1671 imgp->canary = (void *)destp;
1672 error = copyout(canary, imgp->canary, sizeof(canary));
1675 imgp->canarylen = sizeof(canary);
1678 * Prepare the pagesizes array.
1681 destp = rounddown2(destp, sizeof(void *));
1682 imgp->pagesizes = (void *)destp;
1683 error = copyout(pagesizes, imgp->pagesizes, szps);
1686 imgp->pagesizeslen = szps;
1689 * Allocate room for the argument and environment strings.
1691 destp -= ARG_MAX - imgp->args->stringspace;
1692 destp = rounddown2(destp, sizeof(void *));
1695 exec_stackgap(imgp, &destp);
1697 if (imgp->auxargs) {
1699 * Allocate room on the stack for the ELF auxargs
1700 * array. It has up to AT_COUNT entries.
1702 destp -= AT_COUNT * sizeof(Elf_Auxinfo);
1703 destp = rounddown2(destp, sizeof(void *));
1706 vectp = (char **)destp;
1709 * Allocate room for the argv[] and env vectors including the
1710 * terminating NULL pointers.
1712 vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
1715 * vectp also becomes our initial stack base
1717 *stack_base = (uintptr_t)vectp;
1719 stringp = imgp->args->begin_argv;
1720 argc = imgp->args->argc;
1721 envc = imgp->args->envc;
1724 * Copy out strings - arguments and environment.
1726 error = copyout(stringp, (void *)ustringp,
1727 ARG_MAX - imgp->args->stringspace);
1732 * Fill in "ps_strings" struct for ps, w, etc.
1735 if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 ||
1736 suword32(&arginfo->ps_nargvstr, argc) != 0)
1740 * Fill in argument portion of vector table.
1742 for (; argc > 0; --argc) {
1743 if (suword(vectp++, ustringp) != 0)
1745 while (*stringp++ != 0)
1750 /* a null vector table pointer separates the argp's from the envp's */
1751 if (suword(vectp++, 0) != 0)
1755 if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 ||
1756 suword32(&arginfo->ps_nenvstr, envc) != 0)
1760 * Fill in environment portion of vector table.
1762 for (; envc > 0; --envc) {
1763 if (suword(vectp++, ustringp) != 0)
1765 while (*stringp++ != 0)
1770 /* end of vector table is a null pointer */
1771 if (suword(vectp, 0) != 0)
1774 if (imgp->auxargs) {
1776 error = imgp->sysent->sv_copyout_auxargs(imgp,
1786 * Check permissions of file to execute.
1787 * Called with imgp->vp locked.
1788 * Return 0 for success or error code on failure.
1791 exec_check_permissions(struct image_params *imgp)
1793 struct vnode *vp = imgp->vp;
1794 struct vattr *attr = imgp->attr;
1800 /* Get file attributes */
1801 error = VOP_GETATTR(vp, attr, td->td_ucred);
1806 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1812 * 1) Check if file execution is disabled for the filesystem that
1813 * this file resides on.
1814 * 2) Ensure that at least one execute bit is on. Otherwise, a
1815 * privileged user will always succeed, and we don't want this
1816 * to happen unless the file really is executable.
1817 * 3) Ensure that the file is a regular file.
1819 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1820 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1821 (attr->va_type != VREG))
1825 * Zero length files can't be exec'd
1827 if (attr->va_size == 0)
1831 * Check for execute permission to file based on current credentials.
1833 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1838 * Check number of open-for-writes on the file and deny execution
1841 * Add a text reference now so no one can write to the
1842 * executable while we're activating it.
1844 * Remember if this was set before and unset it in case this is not
1845 * actually an executable image.
1847 error = VOP_SET_TEXT(vp);
1850 imgp->textset = true;
1853 * Call filesystem specific open routine (which does nothing in the
1856 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1858 imgp->opened = true;
1863 * Exec handler registration
1866 exec_register(const struct execsw *execsw_arg)
1868 const struct execsw **es, **xs, **newexecsw;
1869 u_int count = 2; /* New slot and trailing NULL */
1872 for (es = execsw; *es; es++)
1874 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1877 for (es = execsw; *es; es++)
1882 free(execsw, M_TEMP);
1888 exec_unregister(const struct execsw *execsw_arg)
1890 const struct execsw **es, **xs, **newexecsw;
1894 panic("unregister with no handlers left?\n");
1896 for (es = execsw; *es; es++) {
1897 if (*es == execsw_arg)
1902 for (es = execsw; *es; es++)
1903 if (*es != execsw_arg)
1905 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1907 for (es = execsw; *es; es++)
1908 if (*es != execsw_arg)
1912 free(execsw, M_TEMP);