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/capsicum.h>
38 #include <sys/systm.h>
39 #include <sys/capsicum.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, , , exec, "char *");
100 SDT_PROBE_DEFINE1(proc, , , exec__failure, "int");
101 SDT_PROBE_DEFINE1(proc, , , exec__success, "char *");
103 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
105 int coredump_pack_fileinfo = 1;
106 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_fileinfo, CTLFLAG_RWTUN,
107 &coredump_pack_fileinfo, 0,
108 "Enable file path packing in 'procstat -f' coredump notes");
110 int coredump_pack_vmmapinfo = 1;
111 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_vmmapinfo, CTLFLAG_RWTUN,
112 &coredump_pack_vmmapinfo, 0,
113 "Enable file path packing in 'procstat -v' coredump notes");
115 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
116 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
117 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
118 static int do_execve(struct thread *td, struct image_args *args,
121 /* XXX This should be vm_size_t. */
122 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD,
123 NULL, 0, sysctl_kern_ps_strings, "LU", "");
125 /* XXX This should be vm_size_t. */
126 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD|
127 CTLFLAG_CAPRD, NULL, 0, sysctl_kern_usrstack, "LU", "");
129 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD,
130 NULL, 0, sysctl_kern_stackprot, "I", "");
132 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
133 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
134 &ps_arg_cache_limit, 0, "");
136 static int disallow_high_osrel;
137 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW,
138 &disallow_high_osrel, 0,
139 "Disallow execution of binaries built for higher version of the world");
141 static int map_at_zero = 0;
142 TUNABLE_INT("security.bsd.map_at_zero", &map_at_zero);
143 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RW, &map_at_zero, 0,
144 "Permit processes to map an object at virtual address 0.");
147 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
154 if (req->flags & SCTL_MASK32) {
156 val = (unsigned int)p->p_sysent->sv_psstrings;
157 error = SYSCTL_OUT(req, &val, sizeof(val));
160 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
161 sizeof(p->p_sysent->sv_psstrings));
166 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
173 if (req->flags & SCTL_MASK32) {
175 val = (unsigned int)p->p_sysent->sv_usrstack;
176 error = SYSCTL_OUT(req, &val, sizeof(val));
179 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
180 sizeof(p->p_sysent->sv_usrstack));
185 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
190 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
191 sizeof(p->p_sysent->sv_stackprot)));
195 * Each of the items is a pointer to a `const struct execsw', hence the
196 * double pointer here.
198 static const struct execsw **execsw;
200 #ifndef _SYS_SYSPROTO_H_
209 sys_execve(struct thread *td, struct execve_args *uap)
211 struct image_args args;
212 struct vmspace *oldvmspace;
215 error = pre_execve(td, &oldvmspace);
218 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
219 uap->argv, uap->envv);
221 error = kern_execve(td, &args, NULL);
222 post_execve(td, error, oldvmspace);
226 #ifndef _SYS_SYSPROTO_H_
227 struct fexecve_args {
234 sys_fexecve(struct thread *td, struct fexecve_args *uap)
236 struct image_args args;
237 struct vmspace *oldvmspace;
240 error = pre_execve(td, &oldvmspace);
243 error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
244 uap->argv, uap->envv);
247 error = kern_execve(td, &args, NULL);
249 post_execve(td, error, oldvmspace);
253 #ifndef _SYS_SYSPROTO_H_
254 struct __mac_execve_args {
263 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap)
266 struct image_args args;
267 struct vmspace *oldvmspace;
270 error = pre_execve(td, &oldvmspace);
273 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
274 uap->argv, uap->envv);
276 error = kern_execve(td, &args, uap->mac_p);
277 post_execve(td, error, oldvmspace);
285 pre_execve(struct thread *td, struct vmspace **oldvmspace)
290 KASSERT(td == curthread, ("non-current thread %p", td));
293 if ((p->p_flag & P_HADTHREADS) != 0) {
295 if (thread_single(p, SINGLE_BOUNDARY) != 0)
299 KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0,
301 *oldvmspace = p->p_vmspace;
306 post_execve(struct thread *td, int error, struct vmspace *oldvmspace)
310 KASSERT(td == curthread, ("non-current thread %p", td));
312 if ((p->p_flag & P_HADTHREADS) != 0) {
315 * If success, we upgrade to SINGLE_EXIT state to
316 * force other threads to suicide.
319 thread_single(p, SINGLE_EXIT);
321 thread_single_end(p, SINGLE_BOUNDARY);
324 if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
325 KASSERT(p->p_vmspace != oldvmspace,
326 ("oldvmspace still used"));
327 vmspace_free(oldvmspace);
328 td->td_pflags &= ~TDP_EXECVMSPC;
333 * XXX: kern_execve has the astonishing property of not always returning to
334 * the caller. If sufficiently bad things happen during the call to
335 * do_execve(), it can end up calling exit1(); as a result, callers must
336 * avoid doing anything which they might need to undo (e.g., allocating
340 kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p)
343 AUDIT_ARG_ARGV(args->begin_argv, args->argc,
344 args->begin_envv - args->begin_argv);
345 AUDIT_ARG_ENVV(args->begin_envv, args->envc,
346 args->endp - args->begin_envv);
347 return (do_execve(td, args, mac_p));
351 * In-kernel implementation of execve(). All arguments are assumed to be
352 * userspace pointers from the passed thread.
355 do_execve(td, args, mac_p)
357 struct image_args *args;
360 struct proc *p = td->td_proc;
362 struct ucred *oldcred;
363 struct uidinfo *euip = NULL;
364 register_t *stack_base;
366 struct image_params image_params, *imgp;
368 int (*img_first)(struct image_params *);
369 struct pargs *oldargs = NULL, *newargs = NULL;
370 struct sigacts *oldsigacts = NULL, *newsigacts = NULL;
372 struct vnode *tracevp = NULL;
373 struct ucred *tracecred = NULL;
375 struct vnode *oldtextvp = NULL, *newtextvp;
377 int credential_changing;
380 struct label *interpvplabel = NULL;
384 struct pmckern_procexec pe;
386 static const char fexecv_proc_title[] = "(fexecv)";
388 imgp = &image_params;
391 * Lock the process and set the P_INEXEC flag to indicate that
392 * it should be left alone until we're done here. This is
393 * necessary to avoid race conditions - e.g. in ptrace() -
394 * that might allow a local user to illicitly obtain elevated
398 KASSERT((p->p_flag & P_INEXEC) == 0,
399 ("%s(): process already has P_INEXEC flag", __func__));
400 p->p_flag |= P_INEXEC;
404 * Initialize part of the common data
406 bzero(imgp, sizeof(*imgp));
410 oldcred = p->p_ucred;
413 error = mac_execve_enter(imgp, mac_p);
419 * Translate the file name. namei() returns a vnode pointer
420 * in ni_vp among other things.
422 * XXXAUDIT: It would be desirable to also audit the name of the
423 * interpreter if this is an interpreted binary.
425 if (args->fname != NULL) {
426 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME
427 | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
430 SDT_PROBE1(proc, , , exec, args->fname);
433 if (args->fname != NULL) {
434 #ifdef CAPABILITY_MODE
436 * While capability mode can't reach this point via direct
437 * path arguments to execve(), we also don't allow
438 * interpreters to be used in capability mode (for now).
439 * Catch indirect lookups and return a permissions error.
441 if (IN_CAPABILITY_MODE(td)) {
450 newtextvp = nd.ni_vp;
451 imgp->vp = newtextvp;
453 AUDIT_ARG_FD(args->fd);
455 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
457 error = fgetvp_exec(td, args->fd,
458 cap_rights_init(&rights, CAP_FEXECVE), &newtextvp);
461 vn_lock(newtextvp, LK_EXCLUSIVE | LK_RETRY);
462 AUDIT_ARG_VNODE1(newtextvp);
463 imgp->vp = newtextvp;
467 * Check file permissions (also 'opens' file)
469 error = exec_check_permissions(imgp);
471 goto exec_fail_dealloc;
473 imgp->object = imgp->vp->v_object;
474 if (imgp->object != NULL)
475 vm_object_reference(imgp->object);
478 * Set VV_TEXT now so no one can write to the executable while we're
481 * Remember if this was set before and unset it in case this is not
482 * actually an executable image.
484 textset = VOP_IS_TEXT(imgp->vp);
485 VOP_SET_TEXT(imgp->vp);
487 error = exec_map_first_page(imgp);
489 goto exec_fail_dealloc;
491 imgp->proc->p_osrel = 0;
494 * Implement image setuid/setgid.
496 * Determine new credentials before attempting image activators
497 * so that it can be used by process_exec handlers to determine
498 * credential/setid changes.
500 * Don't honor setuid/setgid if the filesystem prohibits it or if
501 * the process is being traced.
503 * We disable setuid/setgid/etc in capability mode on the basis
504 * that most setugid applications are not written with that
505 * environment in mind, and will therefore almost certainly operate
506 * incorrectly. In principle there's no reason that setugid
507 * applications might not be useful in capability mode, so we may want
508 * to reconsider this conservative design choice in the future.
510 * XXXMAC: For the time being, use NOSUID to also prohibit
511 * transitions on the file system.
513 credential_changing = 0;
514 credential_changing |= (attr.va_mode & S_ISUID) &&
515 oldcred->cr_uid != attr.va_uid;
516 credential_changing |= (attr.va_mode & S_ISGID) &&
517 oldcred->cr_gid != attr.va_gid;
519 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
520 interpvplabel, imgp);
521 credential_changing |= will_transition;
524 if (credential_changing &&
525 #ifdef CAPABILITY_MODE
526 ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
528 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
529 (p->p_flag & P_TRACED) == 0) {
530 imgp->credential_setid = true;
531 VOP_UNLOCK(imgp->vp, 0);
532 imgp->newcred = crdup(oldcred);
533 if (attr.va_mode & S_ISUID) {
534 euip = uifind(attr.va_uid);
535 change_euid(imgp->newcred, euip);
537 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
538 if (attr.va_mode & S_ISGID)
539 change_egid(imgp->newcred, attr.va_gid);
541 * Implement correct POSIX saved-id behavior.
543 * XXXMAC: Note that the current logic will save the
544 * uid and gid if a MAC domain transition occurs, even
545 * though maybe it shouldn't.
547 change_svuid(imgp->newcred, imgp->newcred->cr_uid);
548 change_svgid(imgp->newcred, imgp->newcred->cr_gid);
551 * Implement correct POSIX saved-id behavior.
553 * XXX: It's not clear that the existing behavior is
554 * POSIX-compliant. A number of sources indicate that the
555 * saved uid/gid should only be updated if the new ruid is
556 * not equal to the old ruid, or the new euid is not equal
557 * to the old euid and the new euid is not equal to the old
558 * ruid. The FreeBSD code always updates the saved uid/gid.
559 * Also, this code uses the new (replaced) euid and egid as
560 * the source, which may or may not be the right ones to use.
562 if (oldcred->cr_svuid != oldcred->cr_uid ||
563 oldcred->cr_svgid != oldcred->cr_gid) {
564 VOP_UNLOCK(imgp->vp, 0);
565 imgp->newcred = crdup(oldcred);
566 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
567 change_svuid(imgp->newcred, imgp->newcred->cr_uid);
568 change_svgid(imgp->newcred, imgp->newcred->cr_gid);
571 /* The new credentials are installed into the process later. */
574 * Do the best to calculate the full path to the image file.
576 if (args->fname != NULL && args->fname[0] == '/')
577 imgp->execpath = args->fname;
579 VOP_UNLOCK(imgp->vp, 0);
580 if (vn_fullpath(td, imgp->vp, &imgp->execpath,
581 &imgp->freepath) != 0)
582 imgp->execpath = args->fname;
583 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
587 * If the current process has a special image activator it
588 * wants to try first, call it. For example, emulating shell
589 * scripts differently.
592 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
593 error = img_first(imgp);
596 * Loop through the list of image activators, calling each one.
597 * An activator returns -1 if there is no match, 0 on success,
598 * and an error otherwise.
600 for (i = 0; error == -1 && execsw[i]; ++i) {
601 if (execsw[i]->ex_imgact == NULL ||
602 execsw[i]->ex_imgact == img_first) {
605 error = (*execsw[i]->ex_imgact)(imgp);
611 VOP_UNSET_TEXT(imgp->vp);
614 goto exec_fail_dealloc;
618 * Special interpreter operation, cleanup and loop up to try to
619 * activate the interpreter.
621 if (imgp->interpreted) {
622 exec_unmap_first_page(imgp);
624 * VV_TEXT needs to be unset for scripts. There is a short
625 * period before we determine that something is a script where
626 * VV_TEXT will be set. The vnode lock is held over this
627 * entire period so nothing should illegitimately be blocked.
629 VOP_UNSET_TEXT(imgp->vp);
630 /* free name buffer and old vnode */
631 if (args->fname != NULL)
632 NDFREE(&nd, NDF_ONLY_PNBUF);
634 mac_execve_interpreter_enter(newtextvp, &interpvplabel);
637 VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td);
641 vm_object_deallocate(imgp->object);
643 imgp->credential_setid = false;
644 if (imgp->newcred != NULL) {
645 crfree(imgp->newcred);
646 imgp->newcred = NULL;
648 imgp->execpath = NULL;
649 free(imgp->freepath, M_TEMP);
650 imgp->freepath = NULL;
651 /* set new name to that of the interpreter */
652 NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
653 UIO_SYSSPACE, imgp->interpreter_name, td);
654 args->fname = imgp->interpreter_name;
659 * NB: We unlock the vnode here because it is believed that none
660 * of the sv_copyout_strings/sv_fixup operations require the vnode.
662 VOP_UNLOCK(imgp->vp, 0);
664 if (disallow_high_osrel &&
665 P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
667 uprintf("Osrel %d for image %s too high\n", p->p_osrel,
668 imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
669 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
670 goto exec_fail_dealloc;
674 * Copy out strings (args and env) and initialize stack base
676 if (p->p_sysent->sv_copyout_strings)
677 stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
679 stack_base = exec_copyout_strings(imgp);
682 * If custom stack fixup routine present for this process
683 * let it do the stack setup.
684 * Else stuff argument count as first item on stack
686 if (p->p_sysent->sv_fixup != NULL)
687 (*p->p_sysent->sv_fixup)(&stack_base, imgp);
689 suword(--stack_base, imgp->args->argc);
692 * For security and other reasons, the file descriptor table cannot
693 * be shared after an exec.
696 /* close files on exec */
700 * Malloc things before we need locks.
702 i = imgp->args->begin_envv - imgp->args->begin_argv;
703 /* Cache arguments if they fit inside our allowance */
704 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
705 newargs = pargs_alloc(i);
706 bcopy(imgp->args->begin_argv, newargs->ar_args, i);
710 * For security and other reasons, signal handlers cannot
711 * be shared after an exec. The new process gets a copy of the old
712 * handlers. In execsigs(), the new process will have its signals
715 if (sigacts_shared(p->p_sigacts)) {
716 oldsigacts = p->p_sigacts;
717 newsigacts = sigacts_alloc();
718 sigacts_copy(newsigacts, oldsigacts);
721 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
725 p->p_sigacts = newsigacts;
729 /* reset caught signals */
732 /* name this process - nameiexec(p, ndp) */
733 bzero(p->p_comm, sizeof(p->p_comm));
735 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
736 min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
737 else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0)
738 bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
739 bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
741 sched_clear_tdname(td);
745 * mark as execed, wakeup the process that vforked (if any) and tell
746 * it that it now has its own resources back
749 if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0)
750 p->p_flag2 &= ~P2_NOTRACE;
751 if (p->p_flag & P_PPWAIT) {
752 p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
753 cv_broadcast(&p->p_pwait);
757 * Implement image setuid/setgid installation.
759 if (imgp->credential_setid) {
761 * Turn off syscall tracing for set-id programs, except for
762 * root. Record any set-id flags first to make sure that
763 * we do not regain any tracing during a possible block.
768 if (p->p_tracecred != NULL &&
769 priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED, 0))
770 ktrprocexec(p, &tracecred, &tracevp);
773 * Close any file descriptors 0..2 that reference procfs,
774 * then make sure file descriptors 0..2 are in use.
776 * setugidsafety() may call closef() and then pfind()
777 * which may grab the process lock.
778 * fdcheckstd() may call falloc() which may block to
779 * allocate memory, so temporarily drop the process lock.
782 VOP_UNLOCK(imgp->vp, 0);
784 error = fdcheckstd(td);
785 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
787 goto exec_fail_dealloc;
790 if (will_transition) {
791 mac_vnode_execve_transition(oldcred, imgp->newcred,
792 imgp->vp, interpvplabel, imgp);
796 if (oldcred->cr_uid == oldcred->cr_ruid &&
797 oldcred->cr_gid == oldcred->cr_rgid)
798 p->p_flag &= ~P_SUGID;
801 * Set the new credentials.
803 if (imgp->newcred != NULL) {
804 proc_set_cred(p, imgp->newcred);
810 * Store the vp for use in procfs. This vnode was referenced by namei
813 oldtextvp = p->p_textvp;
814 p->p_textvp = newtextvp;
818 * Tell the DTrace fasttrap provider about the exec if it
819 * has declared an interest.
821 if (dtrace_fasttrap_exec)
822 dtrace_fasttrap_exec(p);
826 * Notify others that we exec'd, and clear the P_INEXEC flag
827 * as we're now a bona fide freshly-execed process.
829 KNOTE_LOCKED(&p->p_klist, NOTE_EXEC);
830 p->p_flag &= ~P_INEXEC;
832 /* clear "fork but no exec" flag, as we _are_ execing */
833 p->p_acflag &= ~AFORK;
836 * Free any previous argument cache and replace it with
837 * the new argument cache, if any.
845 * Check if system-wide sampling is in effect or if the
846 * current process is using PMCs. If so, do exec() time
847 * processing. This processing needs to happen AFTER the
848 * P_INEXEC flag is cleared.
850 * The proc lock needs to be released before taking the PMC
853 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
855 VOP_UNLOCK(imgp->vp, 0);
856 pe.pm_credentialschanged = credential_changing;
857 pe.pm_entryaddr = imgp->entry_addr;
859 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
860 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
863 #else /* !HWPMC_HOOKS */
867 /* Set values passed into the program in registers. */
868 if (p->p_sysent->sv_setregs)
869 (*p->p_sysent->sv_setregs)(td, imgp,
870 (u_long)(uintptr_t)stack_base);
872 exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base);
874 vfs_mark_atime(imgp->vp, td->td_ucred);
876 SDT_PROBE1(proc, , , exec__success, args->fname);
879 if (imgp->firstpage != NULL)
880 exec_unmap_first_page(imgp);
882 if (imgp->vp != NULL) {
884 NDFREE(&nd, NDF_ONLY_PNBUF);
886 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
890 VOP_UNLOCK(imgp->vp, 0);
893 if (imgp->object != NULL)
894 vm_object_deallocate(imgp->object);
896 free(imgp->freepath, M_TEMP);
900 td->td_dbgflags |= TDB_EXEC;
904 * Stop the process here if its stop event mask has
905 * the S_EXEC bit set.
907 STOPEVENT(p, S_EXEC, 0);
910 /* we're done here, clear P_INEXEC */
912 p->p_flag &= ~P_INEXEC;
915 SDT_PROBE1(proc, , , exec__failure, error);
918 if (imgp->newcred != NULL && oldcred != NULL)
919 crfree(imgp->newcred);
922 mac_execve_exit(imgp);
923 mac_execve_interpreter_exit(interpvplabel);
925 exec_free_args(args);
928 * Handle deferred decrement of ref counts.
930 if (oldtextvp != NULL)
935 if (tracecred != NULL)
940 if (oldsigacts != NULL)
941 sigacts_free(oldsigacts);
945 if (error && imgp->vmspace_destroyed) {
946 /* sorry, no more process anymore. exit gracefully */
947 exit1(td, W_EXITCODE(0, SIGABRT));
960 exec_map_first_page(imgp)
961 struct image_params *imgp;
965 vm_page_t ma[VM_INITIAL_PAGEIN];
968 if (imgp->firstpage != NULL)
969 exec_unmap_first_page(imgp);
971 object = imgp->vp->v_object;
974 VM_OBJECT_WLOCK(object);
975 #if VM_NRESERVLEVEL > 0
976 if ((object->flags & OBJ_COLORED) == 0) {
977 object->flags |= OBJ_COLORED;
978 object->pg_color = 0;
981 ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL);
982 if (ma[0]->valid != VM_PAGE_BITS_ALL) {
983 initial_pagein = VM_INITIAL_PAGEIN;
984 if (initial_pagein > object->size)
985 initial_pagein = object->size;
986 for (i = 1; i < initial_pagein; i++) {
987 if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) {
990 if (vm_page_tryxbusy(ma[i]))
993 ma[i] = vm_page_alloc(object, i,
994 VM_ALLOC_NORMAL | VM_ALLOC_IFNOTCACHED);
1000 rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
1001 ma[0] = vm_page_lookup(object, 0);
1002 if ((rv != VM_PAGER_OK) || (ma[0] == NULL)) {
1003 if (ma[0] != NULL) {
1004 vm_page_lock(ma[0]);
1005 vm_page_free(ma[0]);
1006 vm_page_unlock(ma[0]);
1008 VM_OBJECT_WUNLOCK(object);
1012 vm_page_xunbusy(ma[0]);
1013 vm_page_lock(ma[0]);
1014 vm_page_hold(ma[0]);
1015 vm_page_activate(ma[0]);
1016 vm_page_unlock(ma[0]);
1017 VM_OBJECT_WUNLOCK(object);
1019 imgp->firstpage = sf_buf_alloc(ma[0], 0);
1020 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
1026 exec_unmap_first_page(imgp)
1027 struct image_params *imgp;
1031 if (imgp->firstpage != NULL) {
1032 m = sf_buf_page(imgp->firstpage);
1033 sf_buf_free(imgp->firstpage);
1034 imgp->firstpage = NULL;
1042 * Destroy old address space, and allocate a new stack
1043 * The new stack is only SGROWSIZ large because it is grown
1044 * automatically in trap.c.
1047 exec_new_vmspace(imgp, sv)
1048 struct image_params *imgp;
1049 struct sysentvec *sv;
1052 struct proc *p = imgp->proc;
1053 struct vmspace *vmspace = p->p_vmspace;
1055 struct rlimit rlim_stack;
1056 vm_offset_t sv_minuser, stack_addr;
1060 imgp->vmspace_destroyed = 1;
1063 /* May be called with Giant held */
1064 EVENTHANDLER_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 (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser &&
1077 vm_map_max(map) == sv->sv_maxuser) {
1079 pmap_remove_pages(vmspace_pmap(vmspace));
1080 vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1082 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1085 vmspace = p->p_vmspace;
1086 map = &vmspace->vm_map;
1089 /* Map a shared page */
1090 obj = sv->sv_shared_page_obj;
1092 vm_object_reference(obj);
1093 error = vm_map_fixed(map, obj, 0,
1094 sv->sv_shared_page_base, sv->sv_shared_page_len,
1095 VM_PROT_READ | VM_PROT_EXECUTE,
1096 VM_PROT_READ | VM_PROT_EXECUTE,
1097 MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1099 vm_object_deallocate(obj);
1104 /* Allocate a new stack */
1105 if (imgp->stack_sz != 0) {
1106 ssiz = trunc_page(imgp->stack_sz);
1108 lim_rlimit(p, RLIMIT_STACK, &rlim_stack);
1110 if (ssiz > rlim_stack.rlim_max)
1111 ssiz = rlim_stack.rlim_max;
1112 if (ssiz > rlim_stack.rlim_cur) {
1113 rlim_stack.rlim_cur = ssiz;
1114 kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack);
1116 } else if (sv->sv_maxssiz != NULL) {
1117 ssiz = *sv->sv_maxssiz;
1121 stack_addr = sv->sv_usrstack - ssiz;
1122 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1123 obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1125 VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1130 /* Allocate a new register stack */
1131 error = vm_map_stack(map, IA64_BACKINGSTORE, (vm_size_t)ssiz,
1132 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_UP);
1137 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
1138 * VM_STACK case, but they are still used to monitor the size of the
1139 * process stack so we can check the stack rlimit.
1141 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1142 vmspace->vm_maxsaddr = (char *)stack_addr;
1148 * Copy out argument and environment strings from the old process address
1149 * space into the temporary string buffer.
1152 exec_copyin_args(struct image_args *args, char *fname,
1153 enum uio_seg segflg, char **argv, char **envv)
1159 bzero(args, sizeof(*args));
1164 * Allocate demand-paged memory for the file name, argument, and
1165 * environment strings.
1167 error = exec_alloc_args(args);
1172 * Copy the file name.
1174 if (fname != NULL) {
1175 args->fname = args->buf;
1176 error = (segflg == UIO_SYSSPACE) ?
1177 copystr(fname, args->fname, PATH_MAX, &length) :
1178 copyinstr(fname, args->fname, PATH_MAX, &length);
1184 args->begin_argv = args->buf + length;
1185 args->endp = args->begin_argv;
1186 args->stringspace = ARG_MAX;
1189 * extract arguments first
1192 error = fueword(argv++, &argp);
1199 error = copyinstr((void *)(uintptr_t)argp, args->endp,
1200 args->stringspace, &length);
1202 if (error == ENAMETOOLONG)
1206 args->stringspace -= length;
1207 args->endp += length;
1211 args->begin_envv = args->endp;
1214 * extract environment strings
1218 error = fueword(envv++, &envp);
1225 error = copyinstr((void *)(uintptr_t)envp,
1226 args->endp, args->stringspace, &length);
1228 if (error == ENAMETOOLONG)
1232 args->stringspace -= length;
1233 args->endp += length;
1241 exec_free_args(args);
1246 * Allocate temporary demand-paged, zero-filled memory for the file name,
1247 * argument, and environment strings. Returns zero if the allocation succeeds
1248 * and ENOMEM otherwise.
1251 exec_alloc_args(struct image_args *args)
1254 args->buf = (char *)kmap_alloc_wait(exec_map, PATH_MAX + ARG_MAX);
1255 return (args->buf != NULL ? 0 : ENOMEM);
1259 exec_free_args(struct image_args *args)
1262 if (args->buf != NULL) {
1263 kmap_free_wakeup(exec_map, (vm_offset_t)args->buf,
1264 PATH_MAX + ARG_MAX);
1267 if (args->fname_buf != NULL) {
1268 free(args->fname_buf, M_TEMP);
1269 args->fname_buf = NULL;
1274 * Copy strings out to the new process address space, constructing new arg
1275 * and env vector tables. Return a pointer to the base so that it can be used
1276 * as the initial stack pointer.
1279 exec_copyout_strings(imgp)
1280 struct image_params *imgp;
1286 register_t *stack_base;
1287 struct ps_strings *arginfo;
1289 size_t execpath_len;
1290 int szsigcode, szps;
1291 char canary[sizeof(long) * 8];
1293 szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1295 * Calculate string base and vector table pointers.
1296 * Also deal with signal trampoline code for this exec type.
1298 if (imgp->execpath != NULL && imgp->auxargs != NULL)
1299 execpath_len = strlen(imgp->execpath) + 1;
1304 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1305 if (p->p_sysent->sv_sigcode_base == 0) {
1306 if (p->p_sysent->sv_szsigcode != NULL)
1307 szsigcode = *(p->p_sysent->sv_szsigcode);
1309 destp = (uintptr_t)arginfo;
1314 if (szsigcode != 0) {
1316 destp = rounddown2(destp, sizeof(void *));
1317 copyout(p->p_sysent->sv_sigcode, (void *)destp, szsigcode);
1321 * Copy the image path for the rtld.
1323 if (execpath_len != 0) {
1324 destp -= execpath_len;
1325 imgp->execpathp = destp;
1326 copyout(imgp->execpath, (void *)destp, execpath_len);
1330 * Prepare the canary for SSP.
1332 arc4rand(canary, sizeof(canary), 0);
1333 destp -= sizeof(canary);
1334 imgp->canary = destp;
1335 copyout(canary, (void *)destp, sizeof(canary));
1336 imgp->canarylen = sizeof(canary);
1339 * Prepare the pagesizes array.
1342 destp = rounddown2(destp, sizeof(void *));
1343 imgp->pagesizes = destp;
1344 copyout(pagesizes, (void *)destp, szps);
1345 imgp->pagesizeslen = szps;
1347 destp -= ARG_MAX - imgp->args->stringspace;
1348 destp = rounddown2(destp, sizeof(void *));
1351 * If we have a valid auxargs ptr, prepare some room
1354 if (imgp->auxargs) {
1356 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1357 * lower compatibility.
1359 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1362 * The '+ 2' is for the null pointers at the end of each of
1363 * the arg and env vector sets,and imgp->auxarg_size is room
1364 * for argument of Runtime loader.
1366 vectp = (char **)(destp - (imgp->args->argc +
1367 imgp->args->envc + 2 + imgp->auxarg_size)
1371 * The '+ 2' is for the null pointers at the end of each of
1372 * the arg and env vector sets
1374 vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc
1375 + 2) * sizeof(char *));
1379 * vectp also becomes our initial stack base
1381 stack_base = (register_t *)vectp;
1383 stringp = imgp->args->begin_argv;
1384 argc = imgp->args->argc;
1385 envc = imgp->args->envc;
1388 * Copy out strings - arguments and environment.
1390 copyout(stringp, (void *)destp, ARG_MAX - imgp->args->stringspace);
1393 * Fill in "ps_strings" struct for ps, w, etc.
1395 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1396 suword32(&arginfo->ps_nargvstr, argc);
1399 * Fill in argument portion of vector table.
1401 for (; argc > 0; --argc) {
1402 suword(vectp++, (long)(intptr_t)destp);
1403 while (*stringp++ != 0)
1408 /* a null vector table pointer separates the argp's from the envp's */
1411 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1412 suword32(&arginfo->ps_nenvstr, envc);
1415 * Fill in environment portion of vector table.
1417 for (; envc > 0; --envc) {
1418 suword(vectp++, (long)(intptr_t)destp);
1419 while (*stringp++ != 0)
1424 /* end of vector table is a null pointer */
1427 return (stack_base);
1431 * Check permissions of file to execute.
1432 * Called with imgp->vp locked.
1433 * Return 0 for success or error code on failure.
1436 exec_check_permissions(imgp)
1437 struct image_params *imgp;
1439 struct vnode *vp = imgp->vp;
1440 struct vattr *attr = imgp->attr;
1442 int error, writecount;
1446 /* Get file attributes */
1447 error = VOP_GETATTR(vp, attr, td->td_ucred);
1452 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1458 * 1) Check if file execution is disabled for the filesystem that
1459 * this file resides on.
1460 * 2) Ensure that at least one execute bit is on. Otherwise, a
1461 * privileged user will always succeed, and we don't want this
1462 * to happen unless the file really is executable.
1463 * 3) Ensure that the file is a regular file.
1465 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1466 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1467 (attr->va_type != VREG))
1471 * Zero length files can't be exec'd
1473 if (attr->va_size == 0)
1477 * Check for execute permission to file based on current credentials.
1479 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1484 * Check number of open-for-writes on the file and deny execution
1487 error = VOP_GET_WRITECOUNT(vp, &writecount);
1490 if (writecount != 0)
1494 * Call filesystem specific open routine (which does nothing in the
1497 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1504 * Exec handler registration
1507 exec_register(execsw_arg)
1508 const struct execsw *execsw_arg;
1510 const struct execsw **es, **xs, **newexecsw;
1511 int count = 2; /* New slot and trailing NULL */
1514 for (es = execsw; *es; es++)
1516 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1519 for (es = execsw; *es; es++)
1524 free(execsw, M_TEMP);
1530 exec_unregister(execsw_arg)
1531 const struct execsw *execsw_arg;
1533 const struct execsw **es, **xs, **newexecsw;
1537 panic("unregister with no handlers left?\n");
1539 for (es = execsw; *es; es++) {
1540 if (*es == execsw_arg)
1545 for (es = execsw; *es; es++)
1546 if (*es != execsw_arg)
1548 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1550 for (es = execsw; *es; es++)
1551 if (*es != execsw_arg)
1555 free(execsw, M_TEMP);