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, kernel, , exec, "char *");
100 SDT_PROBE_DEFINE1(proc, kernel, , exec__failure, "int");
101 SDT_PROBE_DEFINE1(proc, kernel, , 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_
199 sys_execve(struct thread *td, struct execve_args *uap)
201 struct image_args args;
202 struct vmspace *oldvmspace;
205 error = pre_execve(td, &oldvmspace);
208 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
209 uap->argv, uap->envv);
211 error = kern_execve(td, &args, NULL);
212 post_execve(td, error, oldvmspace);
216 #ifndef _SYS_SYSPROTO_H_
217 struct fexecve_args {
224 sys_fexecve(struct thread *td, struct fexecve_args *uap)
226 struct image_args args;
227 struct vmspace *oldvmspace;
230 error = pre_execve(td, &oldvmspace);
233 error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
234 uap->argv, uap->envv);
237 error = kern_execve(td, &args, NULL);
239 post_execve(td, error, oldvmspace);
243 #ifndef _SYS_SYSPROTO_H_
244 struct __mac_execve_args {
253 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap)
256 struct image_args args;
257 struct vmspace *oldvmspace;
260 error = pre_execve(td, &oldvmspace);
263 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
264 uap->argv, uap->envv);
266 error = kern_execve(td, &args, uap->mac_p);
267 post_execve(td, error, oldvmspace);
275 pre_execve(struct thread *td, struct vmspace **oldvmspace)
280 KASSERT(td == curthread, ("non-current thread %p", td));
283 if ((p->p_flag & P_HADTHREADS) != 0) {
285 if (thread_single(p, SINGLE_BOUNDARY) != 0)
289 KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0,
291 *oldvmspace = p->p_vmspace;
296 post_execve(struct thread *td, int error, struct vmspace *oldvmspace)
300 KASSERT(td == curthread, ("non-current thread %p", td));
302 if ((p->p_flag & P_HADTHREADS) != 0) {
305 * If success, we upgrade to SINGLE_EXIT state to
306 * force other threads to suicide.
309 thread_single(p, SINGLE_EXIT);
311 thread_single_end(p, SINGLE_BOUNDARY);
314 if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
315 KASSERT(p->p_vmspace != oldvmspace,
316 ("oldvmspace still used"));
317 vmspace_free(oldvmspace);
318 td->td_pflags &= ~TDP_EXECVMSPC;
323 * XXX: kern_execve has the astonishing property of not always returning to
324 * the caller. If sufficiently bad things happen during the call to
325 * do_execve(), it can end up calling exit1(); as a result, callers must
326 * avoid doing anything which they might need to undo (e.g., allocating
330 kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p)
333 AUDIT_ARG_ARGV(args->begin_argv, args->argc,
334 args->begin_envv - args->begin_argv);
335 AUDIT_ARG_ENVV(args->begin_envv, args->envc,
336 args->endp - args->begin_envv);
337 return (do_execve(td, args, mac_p));
341 * In-kernel implementation of execve(). All arguments are assumed to be
342 * userspace pointers from the passed thread.
345 do_execve(td, args, mac_p)
347 struct image_args *args;
350 struct proc *p = td->td_proc;
352 struct ucred *newcred = NULL, *oldcred;
353 struct uidinfo *euip = NULL;
354 register_t *stack_base;
356 struct image_params image_params, *imgp;
358 int (*img_first)(struct image_params *);
359 struct pargs *oldargs = NULL, *newargs = NULL;
360 struct sigacts *oldsigacts, *newsigacts;
362 struct vnode *tracevp = NULL;
363 struct ucred *tracecred = NULL;
365 struct vnode *textvp = NULL, *binvp = NULL;
367 int credential_changing;
370 struct label *interpvplabel = NULL;
374 struct pmckern_procexec pe;
376 static const char fexecv_proc_title[] = "(fexecv)";
378 imgp = &image_params;
381 * Lock the process and set the P_INEXEC flag to indicate that
382 * it should be left alone until we're done here. This is
383 * necessary to avoid race conditions - e.g. in ptrace() -
384 * that might allow a local user to illicitly obtain elevated
388 KASSERT((p->p_flag & P_INEXEC) == 0,
389 ("%s(): process already has P_INEXEC flag", __func__));
390 p->p_flag |= P_INEXEC;
394 * Initialize part of the common data
396 bzero(imgp, sizeof(*imgp));
402 error = mac_execve_enter(imgp, mac_p);
408 * Translate the file name. namei() returns a vnode pointer
409 * in ni_vp amoung other things.
411 * XXXAUDIT: It would be desirable to also audit the name of the
412 * interpreter if this is an interpreted binary.
414 if (args->fname != NULL) {
415 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME
416 | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
419 SDT_PROBE(proc, kernel, , exec, args->fname, 0, 0, 0, 0 );
422 if (args->fname != NULL) {
423 #ifdef CAPABILITY_MODE
425 * While capability mode can't reach this point via direct
426 * path arguments to execve(), we also don't allow
427 * interpreters to be used in capability mode (for now).
428 * Catch indirect lookups and return a permissions error.
430 if (IN_CAPABILITY_MODE(td)) {
442 AUDIT_ARG_FD(args->fd);
444 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
446 error = fgetvp_exec(td, args->fd,
447 cap_rights_init(&rights, CAP_FEXECVE), &binvp);
450 vn_lock(binvp, LK_EXCLUSIVE | LK_RETRY);
451 AUDIT_ARG_VNODE1(binvp);
456 * Check file permissions (also 'opens' file)
458 error = exec_check_permissions(imgp);
460 goto exec_fail_dealloc;
462 imgp->object = imgp->vp->v_object;
463 if (imgp->object != NULL)
464 vm_object_reference(imgp->object);
467 * Set VV_TEXT now so no one can write to the executable while we're
470 * Remember if this was set before and unset it in case this is not
471 * actually an executable image.
473 textset = VOP_IS_TEXT(imgp->vp);
474 VOP_SET_TEXT(imgp->vp);
476 error = exec_map_first_page(imgp);
478 goto exec_fail_dealloc;
480 imgp->proc->p_osrel = 0;
482 * If the current process has a special image activator it
483 * wants to try first, call it. For example, emulating shell
484 * scripts differently.
487 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
488 error = img_first(imgp);
491 * Loop through the list of image activators, calling each one.
492 * An activator returns -1 if there is no match, 0 on success,
493 * and an error otherwise.
495 for (i = 0; error == -1 && execsw[i]; ++i) {
496 if (execsw[i]->ex_imgact == NULL ||
497 execsw[i]->ex_imgact == img_first) {
500 error = (*execsw[i]->ex_imgact)(imgp);
506 VOP_UNSET_TEXT(imgp->vp);
509 goto exec_fail_dealloc;
513 * Special interpreter operation, cleanup and loop up to try to
514 * activate the interpreter.
516 if (imgp->interpreted) {
517 exec_unmap_first_page(imgp);
519 * VV_TEXT needs to be unset for scripts. There is a short
520 * period before we determine that something is a script where
521 * VV_TEXT will be set. The vnode lock is held over this
522 * entire period so nothing should illegitimately be blocked.
524 VOP_UNSET_TEXT(imgp->vp);
525 /* free name buffer and old vnode */
526 if (args->fname != NULL)
527 NDFREE(&nd, NDF_ONLY_PNBUF);
529 mac_execve_interpreter_enter(binvp, &interpvplabel);
532 VOP_CLOSE(binvp, FREAD, td->td_ucred, td);
536 vm_object_deallocate(imgp->object);
538 /* set new name to that of the interpreter */
539 NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
540 UIO_SYSSPACE, imgp->interpreter_name, td);
541 args->fname = imgp->interpreter_name;
546 * NB: We unlock the vnode here because it is believed that none
547 * of the sv_copyout_strings/sv_fixup operations require the vnode.
549 VOP_UNLOCK(imgp->vp, 0);
552 * Do the best to calculate the full path to the image file.
554 if (imgp->auxargs != NULL &&
555 ((args->fname != NULL && args->fname[0] == '/') ||
556 vn_fullpath(td, imgp->vp, &imgp->execpath, &imgp->freepath) != 0))
557 imgp->execpath = args->fname;
559 if (disallow_high_osrel &&
560 P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
562 uprintf("Osrel %d for image %s too high\n", p->p_osrel,
563 imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
564 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
565 goto exec_fail_dealloc;
569 * Copy out strings (args and env) and initialize stack base
571 if (p->p_sysent->sv_copyout_strings)
572 stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
574 stack_base = exec_copyout_strings(imgp);
577 * If custom stack fixup routine present for this process
578 * let it do the stack setup.
579 * Else stuff argument count as first item on stack
581 if (p->p_sysent->sv_fixup != NULL)
582 (*p->p_sysent->sv_fixup)(&stack_base, imgp);
584 suword(--stack_base, imgp->args->argc);
587 * For security and other reasons, the file descriptor table cannot
588 * be shared after an exec.
591 /* close files on exec */
595 * Malloc things before we need locks.
597 i = imgp->args->begin_envv - imgp->args->begin_argv;
598 /* Cache arguments if they fit inside our allowance */
599 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
600 newargs = pargs_alloc(i);
601 bcopy(imgp->args->begin_argv, newargs->ar_args, i);
604 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
606 /* Get a reference to the vnode prior to locking the proc */
610 * For security and other reasons, signal handlers cannot
611 * be shared after an exec. The new process gets a copy of the old
612 * handlers. In execsigs(), the new process will have its signals
615 if (sigacts_shared(p->p_sigacts)) {
616 oldsigacts = p->p_sigacts;
617 newsigacts = sigacts_alloc();
618 sigacts_copy(newsigacts, oldsigacts);
621 newsigacts = NULL; /* satisfy gcc */
626 p->p_sigacts = newsigacts;
627 oldcred = p->p_ucred;
631 /* reset caught signals */
634 /* name this process - nameiexec(p, ndp) */
635 bzero(p->p_comm, sizeof(p->p_comm));
637 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
638 min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
639 else if (vn_commname(binvp, p->p_comm, sizeof(p->p_comm)) != 0)
640 bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
641 bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
643 sched_clear_tdname(td);
647 * mark as execed, wakeup the process that vforked (if any) and tell
648 * it that it now has its own resources back
651 if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0)
652 p->p_flag2 &= ~P2_NOTRACE;
653 if (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);
718 newcred = crdup(oldcred);
719 euip = uifind(attr.va_uid);
720 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
723 * Set the new credentials.
725 if (attr.va_mode & S_ISUID)
726 change_euid(newcred, euip);
727 if (attr.va_mode & S_ISGID)
728 change_egid(newcred, attr.va_gid);
730 if (will_transition) {
731 mac_vnode_execve_transition(oldcred, newcred, imgp->vp,
732 interpvplabel, imgp);
736 * Implement correct POSIX saved-id behavior.
738 * XXXMAC: Note that the current logic will save the
739 * uid and gid if a MAC domain transition occurs, even
740 * though maybe it shouldn't.
742 change_svuid(newcred, newcred->cr_uid);
743 change_svgid(newcred, newcred->cr_gid);
744 p->p_ucred = newcred;
746 if (oldcred->cr_uid == oldcred->cr_ruid &&
747 oldcred->cr_gid == oldcred->cr_rgid)
748 p->p_flag &= ~P_SUGID;
750 * Implement correct POSIX saved-id behavior.
752 * XXX: It's not clear that the existing behavior is
753 * POSIX-compliant. A number of sources indicate that the
754 * saved uid/gid should only be updated if the new ruid is
755 * not equal to the old ruid, or the new euid is not equal
756 * to the old euid and the new euid is not equal to the old
757 * ruid. The FreeBSD code always updates the saved uid/gid.
758 * Also, this code uses the new (replaced) euid and egid as
759 * the source, which may or may not be the right ones to use.
761 if (oldcred->cr_svuid != oldcred->cr_uid ||
762 oldcred->cr_svgid != oldcred->cr_gid) {
764 VOP_UNLOCK(imgp->vp, 0);
765 newcred = crdup(oldcred);
766 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
768 change_svuid(newcred, newcred->cr_uid);
769 change_svgid(newcred, newcred->cr_gid);
770 p->p_ucred = newcred;
775 * Store the vp for use in procfs. This vnode was referenced prior
776 * to locking the proc lock.
778 textvp = p->p_textvp;
783 * Tell the DTrace fasttrap provider about the exec if it
784 * has declared an interest.
786 if (dtrace_fasttrap_exec)
787 dtrace_fasttrap_exec(p);
791 * Notify others that we exec'd, and clear the P_INEXEC flag
792 * as we're now a bona fide freshly-execed process.
794 KNOTE_LOCKED(&p->p_klist, NOTE_EXEC);
795 p->p_flag &= ~P_INEXEC;
797 /* clear "fork but no exec" flag, as we _are_ execing */
798 p->p_acflag &= ~AFORK;
801 * Free any previous argument cache and replace it with
802 * the new argument cache, if any.
810 * Check if system-wide sampling is in effect or if the
811 * current process is using PMCs. If so, do exec() time
812 * processing. This processing needs to happen AFTER the
813 * P_INEXEC flag is cleared.
815 * The proc lock needs to be released before taking the PMC
818 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
820 VOP_UNLOCK(imgp->vp, 0);
821 pe.pm_credentialschanged = credential_changing;
822 pe.pm_entryaddr = imgp->entry_addr;
824 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
825 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
828 #else /* !HWPMC_HOOKS */
832 /* Set values passed into the program in registers. */
833 if (p->p_sysent->sv_setregs)
834 (*p->p_sysent->sv_setregs)(td, imgp,
835 (u_long)(uintptr_t)stack_base);
837 exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base);
839 vfs_mark_atime(imgp->vp, td->td_ucred);
841 SDT_PROBE(proc, kernel, , exec__success, args->fname, 0, 0, 0, 0);
843 VOP_UNLOCK(imgp->vp, 0);
846 * Free any resources malloc'd earlier that we didn't use.
854 * Handle deferred decrement of ref counts.
858 if (binvp && error != 0)
863 if (tracecred != NULL)
866 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
869 if (oldsigacts != NULL)
870 sigacts_free(oldsigacts);
875 * free various allocated resources
877 if (imgp->firstpage != NULL)
878 exec_unmap_first_page(imgp);
880 if (imgp->vp != NULL) {
882 NDFREE(&nd, NDF_ONLY_PNBUF);
884 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
888 if (imgp->object != NULL)
889 vm_object_deallocate(imgp->object);
891 free(imgp->freepath, M_TEMP);
895 td->td_dbgflags |= TDB_EXEC;
899 * Stop the process here if its stop event mask has
900 * the S_EXEC bit set.
902 STOPEVENT(p, S_EXEC, 0);
907 /* we're done here, clear P_INEXEC */
909 p->p_flag &= ~P_INEXEC;
912 SDT_PROBE(proc, kernel, , exec__failure, error, 0, 0, 0, 0);
916 mac_execve_exit(imgp);
917 mac_execve_interpreter_exit(interpvplabel);
919 exec_free_args(args);
921 if (error && imgp->vmspace_destroyed) {
922 /* sorry, no more process anymore. exit gracefully */
923 exit1(td, W_EXITCODE(0, SIGABRT));
936 exec_map_first_page(imgp)
937 struct image_params *imgp;
941 vm_page_t ma[VM_INITIAL_PAGEIN];
944 if (imgp->firstpage != NULL)
945 exec_unmap_first_page(imgp);
947 object = imgp->vp->v_object;
950 VM_OBJECT_WLOCK(object);
951 #if VM_NRESERVLEVEL > 0
952 if ((object->flags & OBJ_COLORED) == 0) {
953 object->flags |= OBJ_COLORED;
954 object->pg_color = 0;
957 ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL);
958 if (ma[0]->valid != VM_PAGE_BITS_ALL) {
959 initial_pagein = VM_INITIAL_PAGEIN;
960 if (initial_pagein > object->size)
961 initial_pagein = object->size;
962 for (i = 1; i < initial_pagein; i++) {
963 if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) {
966 if (vm_page_tryxbusy(ma[i]))
969 ma[i] = vm_page_alloc(object, i,
970 VM_ALLOC_NORMAL | VM_ALLOC_IFNOTCACHED);
976 rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
977 ma[0] = vm_page_lookup(object, 0);
978 if ((rv != VM_PAGER_OK) || (ma[0] == NULL)) {
982 vm_page_unlock(ma[0]);
984 VM_OBJECT_WUNLOCK(object);
988 vm_page_xunbusy(ma[0]);
991 vm_page_activate(ma[0]);
992 vm_page_unlock(ma[0]);
993 VM_OBJECT_WUNLOCK(object);
995 imgp->firstpage = sf_buf_alloc(ma[0], 0);
996 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
1002 exec_unmap_first_page(imgp)
1003 struct image_params *imgp;
1007 if (imgp->firstpage != NULL) {
1008 m = sf_buf_page(imgp->firstpage);
1009 sf_buf_free(imgp->firstpage);
1010 imgp->firstpage = NULL;
1018 * Destroy old address space, and allocate a new stack
1019 * The new stack is only SGROWSIZ large because it is grown
1020 * automatically in trap.c.
1023 exec_new_vmspace(imgp, sv)
1024 struct image_params *imgp;
1025 struct sysentvec *sv;
1028 struct proc *p = imgp->proc;
1029 struct vmspace *vmspace = p->p_vmspace;
1031 struct rlimit rlim_stack;
1032 vm_offset_t sv_minuser, stack_addr;
1036 imgp->vmspace_destroyed = 1;
1039 /* May be called with Giant held */
1040 EVENTHANDLER_INVOKE(process_exec, p, imgp);
1043 * Blow away entire process VM, if address space not shared,
1044 * otherwise, create a new VM space so that other threads are
1047 map = &vmspace->vm_map;
1049 sv_minuser = sv->sv_minuser;
1051 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1052 if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser &&
1053 vm_map_max(map) == sv->sv_maxuser) {
1055 pmap_remove_pages(vmspace_pmap(vmspace));
1056 vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1058 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1061 vmspace = p->p_vmspace;
1062 map = &vmspace->vm_map;
1065 /* Map a shared page */
1066 obj = sv->sv_shared_page_obj;
1068 vm_object_reference(obj);
1069 error = vm_map_fixed(map, obj, 0,
1070 sv->sv_shared_page_base, sv->sv_shared_page_len,
1071 VM_PROT_READ | VM_PROT_EXECUTE,
1072 VM_PROT_READ | VM_PROT_EXECUTE,
1073 MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1075 vm_object_deallocate(obj);
1080 /* Allocate a new stack */
1081 if (imgp->stack_sz != 0) {
1082 ssiz = trunc_page(imgp->stack_sz);
1084 lim_rlimit(p, RLIMIT_STACK, &rlim_stack);
1086 if (ssiz > rlim_stack.rlim_max)
1087 ssiz = rlim_stack.rlim_max;
1088 if (ssiz > rlim_stack.rlim_cur) {
1089 rlim_stack.rlim_cur = ssiz;
1090 kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack);
1092 } else if (sv->sv_maxssiz != NULL) {
1093 ssiz = *sv->sv_maxssiz;
1097 stack_addr = sv->sv_usrstack - ssiz;
1098 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1099 obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1101 VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1106 /* Allocate a new register stack */
1107 stack_addr = IA64_BACKINGSTORE;
1108 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1109 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_UP);
1114 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
1115 * VM_STACK case, but they are still used to monitor the size of the
1116 * process stack so we can check the stack rlimit.
1118 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1119 vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - ssiz;
1125 * Copy out argument and environment strings from the old process address
1126 * space into the temporary string buffer.
1129 exec_copyin_args(struct image_args *args, char *fname,
1130 enum uio_seg segflg, char **argv, char **envv)
1136 bzero(args, sizeof(*args));
1141 * Allocate demand-paged memory for the file name, argument, and
1142 * environment strings.
1144 error = exec_alloc_args(args);
1149 * Copy the file name.
1151 if (fname != NULL) {
1152 args->fname = args->buf;
1153 error = (segflg == UIO_SYSSPACE) ?
1154 copystr(fname, args->fname, PATH_MAX, &length) :
1155 copyinstr(fname, args->fname, PATH_MAX, &length);
1161 args->begin_argv = args->buf + length;
1162 args->endp = args->begin_argv;
1163 args->stringspace = ARG_MAX;
1166 * extract arguments first
1169 error = fueword(argv++, &argp);
1176 error = copyinstr((void *)(uintptr_t)argp, args->endp,
1177 args->stringspace, &length);
1179 if (error == ENAMETOOLONG)
1183 args->stringspace -= length;
1184 args->endp += length;
1188 args->begin_envv = args->endp;
1191 * extract environment strings
1195 error = fueword(envv++, &envp);
1202 error = copyinstr((void *)(uintptr_t)envp,
1203 args->endp, args->stringspace, &length);
1205 if (error == ENAMETOOLONG)
1209 args->stringspace -= length;
1210 args->endp += length;
1218 exec_free_args(args);
1223 * Allocate temporary demand-paged, zero-filled memory for the file name,
1224 * argument, and environment strings. Returns zero if the allocation succeeds
1225 * and ENOMEM otherwise.
1228 exec_alloc_args(struct image_args *args)
1231 args->buf = (char *)kmap_alloc_wait(exec_map, PATH_MAX + ARG_MAX);
1232 return (args->buf != NULL ? 0 : ENOMEM);
1236 exec_free_args(struct image_args *args)
1239 if (args->buf != NULL) {
1240 kmap_free_wakeup(exec_map, (vm_offset_t)args->buf,
1241 PATH_MAX + ARG_MAX);
1244 if (args->fname_buf != NULL) {
1245 free(args->fname_buf, M_TEMP);
1246 args->fname_buf = NULL;
1251 * Copy strings out to the new process address space, constructing new arg
1252 * and env vector tables. Return a pointer to the base so that it can be used
1253 * as the initial stack pointer.
1256 exec_copyout_strings(imgp)
1257 struct image_params *imgp;
1263 register_t *stack_base;
1264 struct ps_strings *arginfo;
1266 size_t execpath_len;
1267 int szsigcode, szps;
1268 char canary[sizeof(long) * 8];
1270 szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1272 * Calculate string base and vector table pointers.
1273 * Also deal with signal trampoline code for this exec type.
1275 if (imgp->execpath != NULL && imgp->auxargs != NULL)
1276 execpath_len = strlen(imgp->execpath) + 1;
1281 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1282 if (p->p_sysent->sv_sigcode_base == 0) {
1283 if (p->p_sysent->sv_szsigcode != NULL)
1284 szsigcode = *(p->p_sysent->sv_szsigcode);
1286 destp = (uintptr_t)arginfo;
1291 if (szsigcode != 0) {
1293 destp = rounddown2(destp, sizeof(void *));
1294 copyout(p->p_sysent->sv_sigcode, (void *)destp, szsigcode);
1298 * Copy the image path for the rtld.
1300 if (execpath_len != 0) {
1301 destp -= execpath_len;
1302 imgp->execpathp = destp;
1303 copyout(imgp->execpath, (void *)destp, execpath_len);
1307 * Prepare the canary for SSP.
1309 arc4rand(canary, sizeof(canary), 0);
1310 destp -= sizeof(canary);
1311 imgp->canary = destp;
1312 copyout(canary, (void *)destp, sizeof(canary));
1313 imgp->canarylen = sizeof(canary);
1316 * Prepare the pagesizes array.
1319 destp = rounddown2(destp, sizeof(void *));
1320 imgp->pagesizes = destp;
1321 copyout(pagesizes, (void *)destp, szps);
1322 imgp->pagesizeslen = szps;
1324 destp -= ARG_MAX - imgp->args->stringspace;
1325 destp = rounddown2(destp, sizeof(void *));
1328 * If we have a valid auxargs ptr, prepare some room
1331 if (imgp->auxargs) {
1333 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1334 * lower compatibility.
1336 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1339 * The '+ 2' is for the null pointers at the end of each of
1340 * the arg and env vector sets,and imgp->auxarg_size is room
1341 * for argument of Runtime loader.
1343 vectp = (char **)(destp - (imgp->args->argc +
1344 imgp->args->envc + 2 + imgp->auxarg_size)
1348 * The '+ 2' is for the null pointers at the end of each of
1349 * the arg and env vector sets
1351 vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc
1352 + 2) * sizeof(char *));
1356 * vectp also becomes our initial stack base
1358 stack_base = (register_t *)vectp;
1360 stringp = imgp->args->begin_argv;
1361 argc = imgp->args->argc;
1362 envc = imgp->args->envc;
1365 * Copy out strings - arguments and environment.
1367 copyout(stringp, (void *)destp, ARG_MAX - imgp->args->stringspace);
1370 * Fill in "ps_strings" struct for ps, w, etc.
1372 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1373 suword32(&arginfo->ps_nargvstr, argc);
1376 * Fill in argument portion of vector table.
1378 for (; argc > 0; --argc) {
1379 suword(vectp++, (long)(intptr_t)destp);
1380 while (*stringp++ != 0)
1385 /* a null vector table pointer separates the argp's from the envp's */
1388 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1389 suword32(&arginfo->ps_nenvstr, envc);
1392 * Fill in environment portion of vector table.
1394 for (; envc > 0; --envc) {
1395 suword(vectp++, (long)(intptr_t)destp);
1396 while (*stringp++ != 0)
1401 /* end of vector table is a null pointer */
1404 return (stack_base);
1408 * Check permissions of file to execute.
1409 * Called with imgp->vp locked.
1410 * Return 0 for success or error code on failure.
1413 exec_check_permissions(imgp)
1414 struct image_params *imgp;
1416 struct vnode *vp = imgp->vp;
1417 struct vattr *attr = imgp->attr;
1419 int error, writecount;
1423 /* Get file attributes */
1424 error = VOP_GETATTR(vp, attr, td->td_ucred);
1429 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1435 * 1) Check if file execution is disabled for the filesystem that
1436 * this file resides on.
1437 * 2) Ensure that at least one execute bit is on. Otherwise, a
1438 * privileged user will always succeed, and we don't want this
1439 * to happen unless the file really is executable.
1440 * 3) Ensure that the file is a regular file.
1442 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1443 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1444 (attr->va_type != VREG))
1448 * Zero length files can't be exec'd
1450 if (attr->va_size == 0)
1454 * Check for execute permission to file based on current credentials.
1456 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1461 * Check number of open-for-writes on the file and deny execution
1464 error = VOP_GET_WRITECOUNT(vp, &writecount);
1467 if (writecount != 0)
1471 * Call filesystem specific open routine (which does nothing in the
1474 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1481 * Exec handler registration
1484 exec_register(execsw_arg)
1485 const struct execsw *execsw_arg;
1487 const struct execsw **es, **xs, **newexecsw;
1488 int count = 2; /* New slot and trailing NULL */
1491 for (es = execsw; *es; es++)
1493 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1494 if (newexecsw == NULL)
1498 for (es = execsw; *es; es++)
1503 free(execsw, M_TEMP);
1509 exec_unregister(execsw_arg)
1510 const struct execsw *execsw_arg;
1512 const struct execsw **es, **xs, **newexecsw;
1516 panic("unregister with no handlers left?\n");
1518 for (es = execsw; *es; es++) {
1519 if (*es == execsw_arg)
1524 for (es = execsw; *es; es++)
1525 if (*es != execsw_arg)
1527 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1528 if (newexecsw == NULL)
1531 for (es = execsw; *es; es++)
1532 if (*es != execsw_arg)
1536 free(execsw, M_TEMP);