2 * Copyright (c) 1993, David Greenman
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD$");
30 #include "opt_capsicum.h"
31 #include "opt_hwpmc_hooks.h"
32 #include "opt_kdtrace.h"
33 #include "opt_ktrace.h"
36 #include <sys/param.h>
37 #include <sys/capability.h>
38 #include <sys/systm.h>
39 #include <sys/capability.h>
40 #include <sys/eventhandler.h>
42 #include <sys/mutex.h>
43 #include <sys/sysproto.h>
44 #include <sys/signalvar.h>
45 #include <sys/kernel.h>
46 #include <sys/mount.h>
47 #include <sys/filedesc.h>
48 #include <sys/fcntl.h>
51 #include <sys/imgact.h>
52 #include <sys/imgact_elf.h>
54 #include <sys/malloc.h>
57 #include <sys/pioctl.h>
58 #include <sys/namei.h>
59 #include <sys/resourcevar.h>
60 #include <sys/rwlock.h>
61 #include <sys/sched.h>
63 #include <sys/sf_buf.h>
64 #include <sys/syscallsubr.h>
65 #include <sys/sysent.h>
67 #include <sys/sysctl.h>
68 #include <sys/vnode.h>
71 #include <sys/ktrace.h>
75 #include <vm/vm_param.h>
77 #include <vm/vm_page.h>
78 #include <vm/vm_map.h>
79 #include <vm/vm_kern.h>
80 #include <vm/vm_extern.h>
81 #include <vm/vm_object.h>
82 #include <vm/vm_pager.h>
85 #include <sys/pmckern.h>
88 #include <machine/reg.h>
90 #include <security/audit/audit.h>
91 #include <security/mac/mac_framework.h>
94 #include <sys/dtrace_bsd.h>
95 dtrace_execexit_func_t dtrace_fasttrap_exec;
98 SDT_PROVIDER_DECLARE(proc);
99 SDT_PROBE_DEFINE1(proc, kernel, , exec, exec, "char *");
100 SDT_PROBE_DEFINE1(proc, kernel, , exec_failure, exec-failure, "int");
101 SDT_PROBE_DEFINE1(proc, kernel, , exec_success, exec-success, "char *");
103 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
105 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
106 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
107 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
108 static int do_execve(struct thread *td, struct image_args *args,
111 /* XXX This should be vm_size_t. */
112 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD,
113 NULL, 0, sysctl_kern_ps_strings, "LU", "");
115 /* XXX This should be vm_size_t. */
116 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD|
117 CTLFLAG_CAPRD, NULL, 0, sysctl_kern_usrstack, "LU", "");
119 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD,
120 NULL, 0, sysctl_kern_stackprot, "I", "");
122 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
123 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
124 &ps_arg_cache_limit, 0, "");
126 static int disallow_high_osrel;
127 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW,
128 &disallow_high_osrel, 0,
129 "Disallow execution of binaries built for higher version of the world");
131 static int map_at_zero = 0;
132 TUNABLE_INT("security.bsd.map_at_zero", &map_at_zero);
133 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RW, &map_at_zero, 0,
134 "Permit processes to map an object at virtual address 0.");
137 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
144 if (req->flags & SCTL_MASK32) {
146 val = (unsigned int)p->p_sysent->sv_psstrings;
147 error = SYSCTL_OUT(req, &val, sizeof(val));
150 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
151 sizeof(p->p_sysent->sv_psstrings));
156 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
163 if (req->flags & SCTL_MASK32) {
165 val = (unsigned int)p->p_sysent->sv_usrstack;
166 error = SYSCTL_OUT(req, &val, sizeof(val));
169 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
170 sizeof(p->p_sysent->sv_usrstack));
175 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
180 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
181 sizeof(p->p_sysent->sv_stackprot)));
185 * Each of the items is a pointer to a `const struct execsw', hence the
186 * double pointer here.
188 static const struct execsw **execsw;
190 #ifndef _SYS_SYSPROTO_H_
201 struct execve_args /* {
208 struct image_args args;
210 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
211 uap->argv, uap->envv);
213 error = kern_execve(td, &args, NULL);
217 #ifndef _SYS_SYSPROTO_H_
218 struct fexecve_args {
225 sys_fexecve(struct thread *td, struct fexecve_args *uap)
228 struct image_args args;
230 error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
231 uap->argv, uap->envv);
234 error = kern_execve(td, &args, NULL);
239 #ifndef _SYS_SYSPROTO_H_
240 struct __mac_execve_args {
249 sys___mac_execve(td, uap)
251 struct __mac_execve_args /* {
260 struct image_args args;
262 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
263 uap->argv, uap->envv);
265 error = kern_execve(td, &args, uap->mac_p);
273 * XXX: kern_execve has the astonishing property of not always returning to
274 * the caller. If sufficiently bad things happen during the call to
275 * do_execve(), it can end up calling exit1(); as a result, callers must
276 * avoid doing anything which they might need to undo (e.g., allocating
280 kern_execve(td, args, mac_p)
282 struct image_args *args;
285 struct proc *p = td->td_proc;
288 AUDIT_ARG_ARGV(args->begin_argv, args->argc,
289 args->begin_envv - args->begin_argv);
290 AUDIT_ARG_ENVV(args->begin_envv, args->envc,
291 args->endp - args->begin_envv);
292 if (p->p_flag & P_HADTHREADS) {
294 if (thread_single(SINGLE_BOUNDARY)) {
296 exec_free_args(args);
297 return (ERESTART); /* Try again later. */
302 error = do_execve(td, args, mac_p);
304 if (p->p_flag & P_HADTHREADS) {
307 * If success, we upgrade to SINGLE_EXIT state to
308 * force other threads to suicide.
311 thread_single(SINGLE_EXIT);
321 * In-kernel implementation of execve(). All arguments are assumed to be
322 * userspace pointers from the passed thread.
325 do_execve(td, args, mac_p)
327 struct image_args *args;
330 struct proc *p = td->td_proc;
332 struct ucred *newcred = NULL, *oldcred;
333 struct uidinfo *euip;
334 register_t *stack_base;
336 struct image_params image_params, *imgp;
338 int (*img_first)(struct image_params *);
339 struct pargs *oldargs = NULL, *newargs = NULL;
340 struct sigacts *oldsigacts, *newsigacts;
342 struct vnode *tracevp = NULL;
343 struct ucred *tracecred = NULL;
345 struct vnode *textvp = NULL, *binvp = NULL;
347 int credential_changing;
350 struct label *interpvplabel = NULL;
354 struct pmckern_procexec pe;
356 static const char fexecv_proc_title[] = "(fexecv)";
358 imgp = &image_params;
361 * Lock the process and set the P_INEXEC flag to indicate that
362 * it should be left alone until we're done here. This is
363 * necessary to avoid race conditions - e.g. in ptrace() -
364 * that might allow a local user to illicitly obtain elevated
368 KASSERT((p->p_flag & P_INEXEC) == 0,
369 ("%s(): process already has P_INEXEC flag", __func__));
370 p->p_flag |= P_INEXEC;
374 * Initialize part of the common data
377 imgp->execlabel = NULL;
379 imgp->entry_addr = 0;
380 imgp->reloc_base = 0;
381 imgp->vmspace_destroyed = 0;
382 imgp->interpreted = 0;
384 imgp->interpreter_name = NULL;
385 imgp->auxargs = NULL;
388 imgp->firstpage = NULL;
389 imgp->ps_strings = 0;
390 imgp->auxarg_size = 0;
392 imgp->execpath = imgp->freepath = NULL;
397 imgp->pagesizeslen = 0;
398 imgp->stack_prot = 0;
401 error = mac_execve_enter(imgp, mac_p);
406 imgp->image_header = NULL;
409 * Translate the file name. namei() returns a vnode pointer
410 * in ni_vp amoung other things.
412 * XXXAUDIT: It would be desirable to also audit the name of the
413 * interpreter if this is an interpreted binary.
415 if (args->fname != NULL) {
416 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME
417 | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
420 SDT_PROBE(proc, kernel, , exec, args->fname, 0, 0, 0, 0 );
423 if (args->fname != NULL) {
424 #ifdef CAPABILITY_MODE
426 * While capability mode can't reach this point via direct
427 * path arguments to execve(), we also don't allow
428 * interpreters to be used in capability mode (for now).
429 * Catch indirect lookups and return a permissions error.
431 if (IN_CAPABILITY_MODE(td)) {
443 AUDIT_ARG_FD(args->fd);
445 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
447 error = fgetvp_exec(td, args->fd,
448 cap_rights_init(&rights, CAP_FEXECVE), &binvp);
451 vn_lock(binvp, LK_EXCLUSIVE | LK_RETRY);
452 AUDIT_ARG_VNODE1(binvp);
457 * Check file permissions (also 'opens' file)
459 error = exec_check_permissions(imgp);
461 goto exec_fail_dealloc;
463 imgp->object = imgp->vp->v_object;
464 if (imgp->object != NULL)
465 vm_object_reference(imgp->object);
468 * Set VV_TEXT now so no one can write to the executable while we're
471 * Remember if this was set before and unset it in case this is not
472 * actually an executable image.
474 textset = VOP_IS_TEXT(imgp->vp);
475 VOP_SET_TEXT(imgp->vp);
477 error = exec_map_first_page(imgp);
479 goto exec_fail_dealloc;
481 imgp->proc->p_osrel = 0;
483 * If the current process has a special image activator it
484 * wants to try first, call it. For example, emulating shell
485 * scripts differently.
488 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
489 error = img_first(imgp);
492 * Loop through the list of image activators, calling each one.
493 * An activator returns -1 if there is no match, 0 on success,
494 * and an error otherwise.
496 for (i = 0; error == -1 && execsw[i]; ++i) {
497 if (execsw[i]->ex_imgact == NULL ||
498 execsw[i]->ex_imgact == img_first) {
501 error = (*execsw[i]->ex_imgact)(imgp);
507 VOP_UNSET_TEXT(imgp->vp);
510 goto exec_fail_dealloc;
514 * Special interpreter operation, cleanup and loop up to try to
515 * activate the interpreter.
517 if (imgp->interpreted) {
518 exec_unmap_first_page(imgp);
520 * VV_TEXT needs to be unset for scripts. There is a short
521 * period before we determine that something is a script where
522 * VV_TEXT will be set. The vnode lock is held over this
523 * entire period so nothing should illegitimately be blocked.
525 VOP_UNSET_TEXT(imgp->vp);
526 /* free name buffer and old vnode */
527 if (args->fname != NULL)
528 NDFREE(&nd, NDF_ONLY_PNBUF);
530 mac_execve_interpreter_enter(binvp, &interpvplabel);
533 VOP_CLOSE(binvp, FREAD, td->td_ucred, td);
537 vm_object_deallocate(imgp->object);
539 /* set new name to that of the interpreter */
540 NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
541 UIO_SYSSPACE, imgp->interpreter_name, td);
542 args->fname = imgp->interpreter_name;
547 * NB: We unlock the vnode here because it is believed that none
548 * of the sv_copyout_strings/sv_fixup operations require the vnode.
550 VOP_UNLOCK(imgp->vp, 0);
553 * Do the best to calculate the full path to the image file.
555 if (imgp->auxargs != NULL &&
556 ((args->fname != NULL && args->fname[0] == '/') ||
557 vn_fullpath(td, imgp->vp, &imgp->execpath, &imgp->freepath) != 0))
558 imgp->execpath = args->fname;
560 if (disallow_high_osrel &&
561 P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
563 uprintf("Osrel %d for image %s too high\n", p->p_osrel,
564 imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
565 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
566 goto exec_fail_dealloc;
570 * Copy out strings (args and env) and initialize stack base
572 if (p->p_sysent->sv_copyout_strings)
573 stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
575 stack_base = exec_copyout_strings(imgp);
578 * If custom stack fixup routine present for this process
579 * let it do the stack setup.
580 * Else stuff argument count as first item on stack
582 if (p->p_sysent->sv_fixup != NULL)
583 (*p->p_sysent->sv_fixup)(&stack_base, imgp);
585 suword(--stack_base, imgp->args->argc);
588 * For security and other reasons, the file descriptor table cannot
589 * be shared after an exec.
594 * Malloc things before we need locks.
597 euip = uifind(attr.va_uid);
598 i = imgp->args->begin_envv - imgp->args->begin_argv;
599 /* Cache arguments if they fit inside our allowance */
600 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
601 newargs = pargs_alloc(i);
602 bcopy(imgp->args->begin_argv, newargs->ar_args, i);
605 /* close files on exec */
607 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
609 /* Get a reference to the vnode prior to locking the proc */
613 * For security and other reasons, signal handlers cannot
614 * be shared after an exec. The new process gets a copy of the old
615 * handlers. In execsigs(), the new process will have its signals
619 oldcred = crcopysafe(p, newcred);
620 if (sigacts_shared(p->p_sigacts)) {
621 oldsigacts = p->p_sigacts;
623 newsigacts = sigacts_alloc();
624 sigacts_copy(newsigacts, oldsigacts);
626 p->p_sigacts = newsigacts;
633 /* reset caught signals */
636 /* name this process - nameiexec(p, ndp) */
637 bzero(p->p_comm, sizeof(p->p_comm));
639 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
640 min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
641 else if (vn_commname(binvp, p->p_comm, sizeof(p->p_comm)) != 0)
642 bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
643 bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
645 sched_clear_tdname(td);
649 * mark as execed, wakeup the process that vforked (if any) and tell
650 * it that it now has its own resources back
653 if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
654 p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
655 cv_broadcast(&p->p_pwait);
659 * Implement image setuid/setgid.
661 * Don't honor setuid/setgid if the filesystem prohibits it or if
662 * the process is being traced.
664 * We disable setuid/setgid/etc in compatibility mode on the basis
665 * that most setugid applications are not written with that
666 * environment in mind, and will therefore almost certainly operate
667 * incorrectly. In principle there's no reason that setugid
668 * applications might not be useful in capability mode, so we may want
669 * to reconsider this conservative design choice in the future.
671 * XXXMAC: For the time being, use NOSUID to also prohibit
672 * transitions on the file system.
674 credential_changing = 0;
675 credential_changing |= (attr.va_mode & S_ISUID) && oldcred->cr_uid !=
677 credential_changing |= (attr.va_mode & S_ISGID) && oldcred->cr_gid !=
680 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
681 interpvplabel, imgp);
682 credential_changing |= will_transition;
685 if (credential_changing &&
686 #ifdef CAPABILITY_MODE
687 ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
689 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
690 (p->p_flag & P_TRACED) == 0) {
692 * Turn off syscall tracing for set-id programs, except for
693 * root. Record any set-id flags first to make sure that
694 * we do not regain any tracing during a possible block.
699 if (p->p_tracecred != NULL &&
700 priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED, 0))
701 ktrprocexec(p, &tracecred, &tracevp);
704 * Close any file descriptors 0..2 that reference procfs,
705 * then make sure file descriptors 0..2 are in use.
707 * setugidsafety() may call closef() and then pfind()
708 * which may grab the process lock.
709 * fdcheckstd() may call falloc() which may block to
710 * allocate memory, so temporarily drop the process lock.
713 VOP_UNLOCK(imgp->vp, 0);
715 error = fdcheckstd(td);
716 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
721 * Set the new credentials.
723 if (attr.va_mode & S_ISUID)
724 change_euid(newcred, euip);
725 if (attr.va_mode & S_ISGID)
726 change_egid(newcred, attr.va_gid);
728 if (will_transition) {
729 mac_vnode_execve_transition(oldcred, newcred, imgp->vp,
730 interpvplabel, imgp);
734 * Implement correct POSIX saved-id behavior.
736 * XXXMAC: Note that the current logic will save the
737 * uid and gid if a MAC domain transition occurs, even
738 * though maybe it shouldn't.
740 change_svuid(newcred, newcred->cr_uid);
741 change_svgid(newcred, newcred->cr_gid);
742 p->p_ucred = newcred;
745 if (oldcred->cr_uid == oldcred->cr_ruid &&
746 oldcred->cr_gid == oldcred->cr_rgid)
747 p->p_flag &= ~P_SUGID;
749 * Implement correct POSIX saved-id behavior.
751 * XXX: It's not clear that the existing behavior is
752 * POSIX-compliant. A number of sources indicate that the
753 * saved uid/gid should only be updated if the new ruid is
754 * not equal to the old ruid, or the new euid is not equal
755 * to the old euid and the new euid is not equal to the old
756 * ruid. The FreeBSD code always updates the saved uid/gid.
757 * Also, this code uses the new (replaced) euid and egid as
758 * the source, which may or may not be the right ones to use.
760 if (oldcred->cr_svuid != oldcred->cr_uid ||
761 oldcred->cr_svgid != oldcred->cr_gid) {
762 change_svuid(newcred, newcred->cr_uid);
763 change_svgid(newcred, newcred->cr_gid);
764 p->p_ucred = newcred;
770 * Store the vp for use in procfs. This vnode was referenced prior
771 * to locking the proc lock.
773 textvp = p->p_textvp;
778 * Tell the DTrace fasttrap provider about the exec if it
779 * has declared an interest.
781 if (dtrace_fasttrap_exec)
782 dtrace_fasttrap_exec(p);
786 * Notify others that we exec'd, and clear the P_INEXEC flag
787 * as we're now a bona fide freshly-execed process.
789 KNOTE_LOCKED(&p->p_klist, NOTE_EXEC);
790 p->p_flag &= ~P_INEXEC;
792 /* clear "fork but no exec" flag, as we _are_ execing */
793 p->p_acflag &= ~AFORK;
796 * Free any previous argument cache and replace it with
797 * the new argument cache, if any.
805 * Check if system-wide sampling is in effect or if the
806 * current process is using PMCs. If so, do exec() time
807 * processing. This processing needs to happen AFTER the
808 * P_INEXEC flag is cleared.
810 * The proc lock needs to be released before taking the PMC
813 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
815 VOP_UNLOCK(imgp->vp, 0);
816 pe.pm_credentialschanged = credential_changing;
817 pe.pm_entryaddr = imgp->entry_addr;
819 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
820 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
823 #else /* !HWPMC_HOOKS */
827 /* Set values passed into the program in registers. */
828 if (p->p_sysent->sv_setregs)
829 (*p->p_sysent->sv_setregs)(td, imgp,
830 (u_long)(uintptr_t)stack_base);
832 exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base);
834 vfs_mark_atime(imgp->vp, td->td_ucred);
836 SDT_PROBE(proc, kernel, , exec_success, args->fname, 0, 0, 0, 0);
840 * Free any resources malloc'd earlier that we didn't use.
847 VOP_UNLOCK(imgp->vp, 0);
850 * Handle deferred decrement of ref counts.
854 if (binvp && error != 0)
859 if (tracecred != NULL)
862 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
865 if (oldsigacts != NULL)
866 sigacts_free(oldsigacts);
871 * free various allocated resources
873 if (imgp->firstpage != NULL)
874 exec_unmap_first_page(imgp);
876 if (imgp->vp != NULL) {
878 NDFREE(&nd, NDF_ONLY_PNBUF);
880 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
884 if (imgp->object != NULL)
885 vm_object_deallocate(imgp->object);
887 free(imgp->freepath, M_TEMP);
891 td->td_dbgflags |= TDB_EXEC;
895 * Stop the process here if its stop event mask has
896 * the S_EXEC bit set.
898 STOPEVENT(p, S_EXEC, 0);
903 /* we're done here, clear P_INEXEC */
905 p->p_flag &= ~P_INEXEC;
908 SDT_PROBE(proc, kernel, , exec_failure, error, 0, 0, 0, 0);
912 mac_execve_exit(imgp);
913 mac_execve_interpreter_exit(interpvplabel);
915 exec_free_args(args);
917 if (error && imgp->vmspace_destroyed) {
918 /* sorry, no more process anymore. exit gracefully */
919 exit1(td, W_EXITCODE(0, SIGABRT));
932 exec_map_first_page(imgp)
933 struct image_params *imgp;
937 vm_page_t ma[VM_INITIAL_PAGEIN];
940 if (imgp->firstpage != NULL)
941 exec_unmap_first_page(imgp);
943 object = imgp->vp->v_object;
946 VM_OBJECT_WLOCK(object);
947 #if VM_NRESERVLEVEL > 0
948 if ((object->flags & OBJ_COLORED) == 0) {
949 object->flags |= OBJ_COLORED;
950 object->pg_color = 0;
953 ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL);
954 if (ma[0]->valid != VM_PAGE_BITS_ALL) {
955 initial_pagein = VM_INITIAL_PAGEIN;
956 if (initial_pagein > object->size)
957 initial_pagein = object->size;
958 for (i = 1; i < initial_pagein; i++) {
959 if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) {
962 if (vm_page_tryxbusy(ma[i]))
965 ma[i] = vm_page_alloc(object, i,
966 VM_ALLOC_NORMAL | VM_ALLOC_IFNOTCACHED);
972 rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
973 ma[0] = vm_page_lookup(object, 0);
974 if ((rv != VM_PAGER_OK) || (ma[0] == NULL)) {
978 vm_page_unlock(ma[0]);
980 VM_OBJECT_WUNLOCK(object);
984 vm_page_xunbusy(ma[0]);
987 vm_page_unlock(ma[0]);
988 VM_OBJECT_WUNLOCK(object);
990 imgp->firstpage = sf_buf_alloc(ma[0], 0);
991 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
997 exec_unmap_first_page(imgp)
998 struct image_params *imgp;
1002 if (imgp->firstpage != NULL) {
1003 m = sf_buf_page(imgp->firstpage);
1004 sf_buf_free(imgp->firstpage);
1005 imgp->firstpage = NULL;
1013 * Destroy old address space, and allocate a new stack
1014 * The new stack is only SGROWSIZ large because it is grown
1015 * automatically in trap.c.
1018 exec_new_vmspace(imgp, sv)
1019 struct image_params *imgp;
1020 struct sysentvec *sv;
1023 struct proc *p = imgp->proc;
1024 struct vmspace *vmspace = p->p_vmspace;
1026 vm_offset_t sv_minuser, stack_addr;
1030 imgp->vmspace_destroyed = 1;
1033 /* May be called with Giant held */
1034 EVENTHANDLER_INVOKE(process_exec, p, imgp);
1037 * Blow away entire process VM, if address space not shared,
1038 * otherwise, create a new VM space so that other threads are
1041 map = &vmspace->vm_map;
1043 sv_minuser = sv->sv_minuser;
1045 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1046 if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser &&
1047 vm_map_max(map) == sv->sv_maxuser) {
1049 pmap_remove_pages(vmspace_pmap(vmspace));
1050 vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1052 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1055 vmspace = p->p_vmspace;
1056 map = &vmspace->vm_map;
1059 /* Map a shared page */
1060 obj = sv->sv_shared_page_obj;
1062 vm_object_reference(obj);
1063 error = vm_map_fixed(map, obj, 0,
1064 sv->sv_shared_page_base, sv->sv_shared_page_len,
1065 VM_PROT_READ | VM_PROT_EXECUTE,
1066 VM_PROT_READ | VM_PROT_EXECUTE,
1067 MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1069 vm_object_deallocate(obj);
1074 /* Allocate a new stack */
1075 if (sv->sv_maxssiz != NULL)
1076 ssiz = *sv->sv_maxssiz;
1079 stack_addr = sv->sv_usrstack - ssiz;
1080 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1081 obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1083 VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1088 /* Allocate a new register stack */
1089 stack_addr = IA64_BACKINGSTORE;
1090 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1091 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_UP);
1096 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
1097 * VM_STACK case, but they are still used to monitor the size of the
1098 * process stack so we can check the stack rlimit.
1100 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1101 vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - ssiz;
1107 * Copy out argument and environment strings from the old process address
1108 * space into the temporary string buffer.
1111 exec_copyin_args(struct image_args *args, char *fname,
1112 enum uio_seg segflg, char **argv, char **envv)
1118 bzero(args, sizeof(*args));
1123 * Allocate demand-paged memory for the file name, argument, and
1124 * environment strings.
1126 error = exec_alloc_args(args);
1131 * Copy the file name.
1133 if (fname != NULL) {
1134 args->fname = args->buf;
1135 error = (segflg == UIO_SYSSPACE) ?
1136 copystr(fname, args->fname, PATH_MAX, &length) :
1137 copyinstr(fname, args->fname, PATH_MAX, &length);
1143 args->begin_argv = args->buf + length;
1144 args->endp = args->begin_argv;
1145 args->stringspace = ARG_MAX;
1148 * extract arguments first
1150 while ((argp = (caddr_t) (intptr_t) fuword(argv++))) {
1151 if (argp == (caddr_t) -1) {
1155 if ((error = copyinstr(argp, args->endp,
1156 args->stringspace, &length))) {
1157 if (error == ENAMETOOLONG)
1161 args->stringspace -= length;
1162 args->endp += length;
1166 args->begin_envv = args->endp;
1169 * extract environment strings
1172 while ((envp = (caddr_t)(intptr_t)fuword(envv++))) {
1173 if (envp == (caddr_t)-1) {
1177 if ((error = copyinstr(envp, args->endp,
1178 args->stringspace, &length))) {
1179 if (error == ENAMETOOLONG)
1183 args->stringspace -= length;
1184 args->endp += length;
1192 exec_free_args(args);
1197 * Allocate temporary demand-paged, zero-filled memory for the file name,
1198 * argument, and environment strings. Returns zero if the allocation succeeds
1199 * and ENOMEM otherwise.
1202 exec_alloc_args(struct image_args *args)
1205 args->buf = (char *)kmap_alloc_wait(exec_map, PATH_MAX + ARG_MAX);
1206 return (args->buf != NULL ? 0 : ENOMEM);
1210 exec_free_args(struct image_args *args)
1213 if (args->buf != NULL) {
1214 kmap_free_wakeup(exec_map, (vm_offset_t)args->buf,
1215 PATH_MAX + ARG_MAX);
1218 if (args->fname_buf != NULL) {
1219 free(args->fname_buf, M_TEMP);
1220 args->fname_buf = NULL;
1225 * Copy strings out to the new process address space, constructing new arg
1226 * and env vector tables. Return a pointer to the base so that it can be used
1227 * as the initial stack pointer.
1230 exec_copyout_strings(imgp)
1231 struct image_params *imgp;
1235 char *stringp, *destp;
1236 register_t *stack_base;
1237 struct ps_strings *arginfo;
1239 size_t execpath_len;
1240 int szsigcode, szps;
1241 char canary[sizeof(long) * 8];
1243 szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1245 * Calculate string base and vector table pointers.
1246 * Also deal with signal trampoline code for this exec type.
1248 if (imgp->execpath != NULL && imgp->auxargs != NULL)
1249 execpath_len = strlen(imgp->execpath) + 1;
1254 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1255 if (p->p_sysent->sv_sigcode_base == 0) {
1256 if (p->p_sysent->sv_szsigcode != NULL)
1257 szsigcode = *(p->p_sysent->sv_szsigcode);
1259 destp = (caddr_t)arginfo - szsigcode - SPARE_USRSPACE -
1260 roundup(execpath_len, sizeof(char *)) -
1261 roundup(sizeof(canary), sizeof(char *)) -
1262 roundup(szps, sizeof(char *)) -
1263 roundup((ARG_MAX - imgp->args->stringspace), sizeof(char *));
1269 copyout(p->p_sysent->sv_sigcode, ((caddr_t)arginfo -
1270 szsigcode), szsigcode);
1273 * Copy the image path for the rtld.
1275 if (execpath_len != 0) {
1276 imgp->execpathp = (uintptr_t)arginfo - szsigcode - execpath_len;
1277 copyout(imgp->execpath, (void *)imgp->execpathp,
1282 * Prepare the canary for SSP.
1284 arc4rand(canary, sizeof(canary), 0);
1285 imgp->canary = (uintptr_t)arginfo - szsigcode - execpath_len -
1287 copyout(canary, (void *)imgp->canary, sizeof(canary));
1288 imgp->canarylen = sizeof(canary);
1291 * Prepare the pagesizes array.
1293 imgp->pagesizes = (uintptr_t)arginfo - szsigcode - execpath_len -
1294 roundup(sizeof(canary), sizeof(char *)) - szps;
1295 copyout(pagesizes, (void *)imgp->pagesizes, szps);
1296 imgp->pagesizeslen = szps;
1299 * If we have a valid auxargs ptr, prepare some room
1302 if (imgp->auxargs) {
1304 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1305 * lower compatibility.
1307 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1310 * The '+ 2' is for the null pointers at the end of each of
1311 * the arg and env vector sets,and imgp->auxarg_size is room
1312 * for argument of Runtime loader.
1314 vectp = (char **)(destp - (imgp->args->argc +
1315 imgp->args->envc + 2 + imgp->auxarg_size)
1319 * The '+ 2' is for the null pointers at the end of each of
1320 * the arg and env vector sets
1322 vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc + 2) *
1327 * vectp also becomes our initial stack base
1329 stack_base = (register_t *)vectp;
1331 stringp = imgp->args->begin_argv;
1332 argc = imgp->args->argc;
1333 envc = imgp->args->envc;
1336 * Copy out strings - arguments and environment.
1338 copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
1341 * Fill in "ps_strings" struct for ps, w, etc.
1343 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1344 suword32(&arginfo->ps_nargvstr, argc);
1347 * Fill in argument portion of vector table.
1349 for (; argc > 0; --argc) {
1350 suword(vectp++, (long)(intptr_t)destp);
1351 while (*stringp++ != 0)
1356 /* a null vector table pointer separates the argp's from the envp's */
1359 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1360 suword32(&arginfo->ps_nenvstr, envc);
1363 * Fill in environment portion of vector table.
1365 for (; envc > 0; --envc) {
1366 suword(vectp++, (long)(intptr_t)destp);
1367 while (*stringp++ != 0)
1372 /* end of vector table is a null pointer */
1375 return (stack_base);
1379 * Check permissions of file to execute.
1380 * Called with imgp->vp locked.
1381 * Return 0 for success or error code on failure.
1384 exec_check_permissions(imgp)
1385 struct image_params *imgp;
1387 struct vnode *vp = imgp->vp;
1388 struct vattr *attr = imgp->attr;
1390 int error, writecount;
1394 /* Get file attributes */
1395 error = VOP_GETATTR(vp, attr, td->td_ucred);
1400 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1406 * 1) Check if file execution is disabled for the filesystem that
1407 * this file resides on.
1408 * 2) Ensure that at least one execute bit is on. Otherwise, a
1409 * privileged user will always succeed, and we don't want this
1410 * to happen unless the file really is executable.
1411 * 3) Ensure that the file is a regular file.
1413 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1414 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1415 (attr->va_type != VREG))
1419 * Zero length files can't be exec'd
1421 if (attr->va_size == 0)
1425 * Check for execute permission to file based on current credentials.
1427 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1432 * Check number of open-for-writes on the file and deny execution
1435 error = VOP_GET_WRITECOUNT(vp, &writecount);
1438 if (writecount != 0)
1442 * Call filesystem specific open routine (which does nothing in the
1445 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1452 * Exec handler registration
1455 exec_register(execsw_arg)
1456 const struct execsw *execsw_arg;
1458 const struct execsw **es, **xs, **newexecsw;
1459 int count = 2; /* New slot and trailing NULL */
1462 for (es = execsw; *es; es++)
1464 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1465 if (newexecsw == NULL)
1469 for (es = execsw; *es; es++)
1474 free(execsw, M_TEMP);
1480 exec_unregister(execsw_arg)
1481 const struct execsw *execsw_arg;
1483 const struct execsw **es, **xs, **newexecsw;
1487 panic("unregister with no handlers left?\n");
1489 for (es = execsw; *es; es++) {
1490 if (*es == execsw_arg)
1495 for (es = execsw; *es; es++)
1496 if (*es != execsw_arg)
1498 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1499 if (newexecsw == NULL)
1502 for (es = execsw; *es; es++)
1503 if (*es != execsw_arg)
1507 free(execsw, M_TEMP);