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 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 *newcred = NULL, *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, *newsigacts;
372 struct vnode *tracevp = NULL;
373 struct ucred *tracecred = NULL;
375 struct vnode *textvp = NULL, *binvp = NULL;
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));
412 error = mac_execve_enter(imgp, mac_p);
418 * Translate the file name. namei() returns a vnode pointer
419 * in ni_vp amoung other things.
421 * XXXAUDIT: It would be desirable to also audit the name of the
422 * interpreter if this is an interpreted binary.
424 if (args->fname != NULL) {
425 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME
426 | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
429 SDT_PROBE1(proc, kernel, , exec, args->fname);
432 if (args->fname != NULL) {
433 #ifdef CAPABILITY_MODE
435 * While capability mode can't reach this point via direct
436 * path arguments to execve(), we also don't allow
437 * interpreters to be used in capability mode (for now).
438 * Catch indirect lookups and return a permissions error.
440 if (IN_CAPABILITY_MODE(td)) {
452 AUDIT_ARG_FD(args->fd);
454 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
456 error = fgetvp_exec(td, args->fd,
457 cap_rights_init(&rights, CAP_FEXECVE), &binvp);
460 vn_lock(binvp, LK_EXCLUSIVE | LK_RETRY);
461 AUDIT_ARG_VNODE1(binvp);
466 * Check file permissions (also 'opens' file)
468 error = exec_check_permissions(imgp);
470 goto exec_fail_dealloc;
472 imgp->object = imgp->vp->v_object;
473 if (imgp->object != NULL)
474 vm_object_reference(imgp->object);
477 * Set VV_TEXT now so no one can write to the executable while we're
480 * Remember if this was set before and unset it in case this is not
481 * actually an executable image.
483 textset = VOP_IS_TEXT(imgp->vp);
484 VOP_SET_TEXT(imgp->vp);
486 error = exec_map_first_page(imgp);
488 goto exec_fail_dealloc;
490 imgp->proc->p_osrel = 0;
492 * If the current process has a special image activator it
493 * wants to try first, call it. For example, emulating shell
494 * scripts differently.
497 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
498 error = img_first(imgp);
501 * Loop through the list of image activators, calling each one.
502 * An activator returns -1 if there is no match, 0 on success,
503 * and an error otherwise.
505 for (i = 0; error == -1 && execsw[i]; ++i) {
506 if (execsw[i]->ex_imgact == NULL ||
507 execsw[i]->ex_imgact == img_first) {
510 error = (*execsw[i]->ex_imgact)(imgp);
516 VOP_UNSET_TEXT(imgp->vp);
519 goto exec_fail_dealloc;
523 * Special interpreter operation, cleanup and loop up to try to
524 * activate the interpreter.
526 if (imgp->interpreted) {
527 exec_unmap_first_page(imgp);
529 * VV_TEXT needs to be unset for scripts. There is a short
530 * period before we determine that something is a script where
531 * VV_TEXT will be set. The vnode lock is held over this
532 * entire period so nothing should illegitimately be blocked.
534 VOP_UNSET_TEXT(imgp->vp);
535 /* free name buffer and old vnode */
536 if (args->fname != NULL)
537 NDFREE(&nd, NDF_ONLY_PNBUF);
539 mac_execve_interpreter_enter(binvp, &interpvplabel);
542 VOP_CLOSE(binvp, FREAD, td->td_ucred, td);
546 vm_object_deallocate(imgp->object);
548 /* set new name to that of the interpreter */
549 NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
550 UIO_SYSSPACE, imgp->interpreter_name, td);
551 args->fname = imgp->interpreter_name;
556 * NB: We unlock the vnode here because it is believed that none
557 * of the sv_copyout_strings/sv_fixup operations require the vnode.
559 VOP_UNLOCK(imgp->vp, 0);
562 * Do the best to calculate the full path to the image file.
564 if (imgp->auxargs != NULL &&
565 ((args->fname != NULL && args->fname[0] == '/') ||
566 vn_fullpath(td, imgp->vp, &imgp->execpath, &imgp->freepath) != 0))
567 imgp->execpath = args->fname;
569 if (disallow_high_osrel &&
570 P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
572 uprintf("Osrel %d for image %s too high\n", p->p_osrel,
573 imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
574 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
575 goto exec_fail_dealloc;
579 * Copy out strings (args and env) and initialize stack base
581 if (p->p_sysent->sv_copyout_strings)
582 stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
584 stack_base = exec_copyout_strings(imgp);
587 * If custom stack fixup routine present for this process
588 * let it do the stack setup.
589 * Else stuff argument count as first item on stack
591 if (p->p_sysent->sv_fixup != NULL)
592 (*p->p_sysent->sv_fixup)(&stack_base, imgp);
594 suword(--stack_base, imgp->args->argc);
597 * For security and other reasons, the file descriptor table cannot
598 * be shared after an exec.
601 /* close files on exec */
605 * Malloc things before we need locks.
607 i = imgp->args->begin_envv - imgp->args->begin_argv;
608 /* Cache arguments if they fit inside our allowance */
609 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
610 newargs = pargs_alloc(i);
611 bcopy(imgp->args->begin_argv, newargs->ar_args, i);
614 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
616 /* Get a reference to the vnode prior to locking the proc */
620 * For security and other reasons, signal handlers cannot
621 * be shared after an exec. The new process gets a copy of the old
622 * handlers. In execsigs(), the new process will have its signals
625 if (sigacts_shared(p->p_sigacts)) {
626 oldsigacts = p->p_sigacts;
627 newsigacts = sigacts_alloc();
628 sigacts_copy(newsigacts, oldsigacts);
631 newsigacts = NULL; /* satisfy gcc */
636 p->p_sigacts = newsigacts;
637 oldcred = p->p_ucred;
641 /* reset caught signals */
644 /* name this process - nameiexec(p, ndp) */
645 bzero(p->p_comm, sizeof(p->p_comm));
647 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
648 min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
649 else if (vn_commname(binvp, p->p_comm, sizeof(p->p_comm)) != 0)
650 bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
651 bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
653 sched_clear_tdname(td);
657 * mark as execed, wakeup the process that vforked (if any) and tell
658 * it that it now has its own resources back
661 if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0)
662 p->p_flag2 &= ~P2_NOTRACE;
663 if (p->p_flag & P_PPWAIT) {
664 p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
665 cv_broadcast(&p->p_pwait);
669 * Implement image setuid/setgid.
671 * Don't honor setuid/setgid if the filesystem prohibits it or if
672 * the process is being traced.
674 * We disable setuid/setgid/etc in compatibility mode on the basis
675 * that most setugid applications are not written with that
676 * environment in mind, and will therefore almost certainly operate
677 * incorrectly. In principle there's no reason that setugid
678 * applications might not be useful in capability mode, so we may want
679 * to reconsider this conservative design choice in the future.
681 * XXXMAC: For the time being, use NOSUID to also prohibit
682 * transitions on the file system.
684 credential_changing = 0;
685 credential_changing |= (attr.va_mode & S_ISUID) && oldcred->cr_uid !=
687 credential_changing |= (attr.va_mode & S_ISGID) && oldcred->cr_gid !=
690 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
691 interpvplabel, imgp);
692 credential_changing |= will_transition;
695 if (credential_changing &&
696 #ifdef CAPABILITY_MODE
697 ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
699 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
700 (p->p_flag & P_TRACED) == 0) {
702 * Turn off syscall tracing for set-id programs, except for
703 * root. Record any set-id flags first to make sure that
704 * we do not regain any tracing during a possible block.
709 if (p->p_tracecred != NULL &&
710 priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED, 0))
711 ktrprocexec(p, &tracecred, &tracevp);
714 * Close any file descriptors 0..2 that reference procfs,
715 * then make sure file descriptors 0..2 are in use.
717 * setugidsafety() may call closef() and then pfind()
718 * which may grab the process lock.
719 * fdcheckstd() may call falloc() which may block to
720 * allocate memory, so temporarily drop the process lock.
723 VOP_UNLOCK(imgp->vp, 0);
725 error = fdcheckstd(td);
728 newcred = crdup(oldcred);
729 euip = uifind(attr.va_uid);
730 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
733 * Set the new credentials.
735 if (attr.va_mode & S_ISUID)
736 change_euid(newcred, euip);
737 if (attr.va_mode & S_ISGID)
738 change_egid(newcred, attr.va_gid);
740 if (will_transition) {
741 mac_vnode_execve_transition(oldcred, newcred, imgp->vp,
742 interpvplabel, imgp);
746 * Implement correct POSIX saved-id behavior.
748 * XXXMAC: Note that the current logic will save the
749 * uid and gid if a MAC domain transition occurs, even
750 * though maybe it shouldn't.
752 change_svuid(newcred, newcred->cr_uid);
753 change_svgid(newcred, newcred->cr_gid);
754 p->p_ucred = newcred;
756 if (oldcred->cr_uid == oldcred->cr_ruid &&
757 oldcred->cr_gid == oldcred->cr_rgid)
758 p->p_flag &= ~P_SUGID;
760 * Implement correct POSIX saved-id behavior.
762 * XXX: It's not clear that the existing behavior is
763 * POSIX-compliant. A number of sources indicate that the
764 * saved uid/gid should only be updated if the new ruid is
765 * not equal to the old ruid, or the new euid is not equal
766 * to the old euid and the new euid is not equal to the old
767 * ruid. The FreeBSD code always updates the saved uid/gid.
768 * Also, this code uses the new (replaced) euid and egid as
769 * the source, which may or may not be the right ones to use.
771 if (oldcred->cr_svuid != oldcred->cr_uid ||
772 oldcred->cr_svgid != oldcred->cr_gid) {
774 VOP_UNLOCK(imgp->vp, 0);
775 newcred = crdup(oldcred);
776 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
778 change_svuid(newcred, newcred->cr_uid);
779 change_svgid(newcred, newcred->cr_gid);
780 p->p_ucred = newcred;
785 * Store the vp for use in procfs. This vnode was referenced prior
786 * to locking the proc lock.
788 textvp = p->p_textvp;
793 * Tell the DTrace fasttrap provider about the exec if it
794 * has declared an interest.
796 if (dtrace_fasttrap_exec)
797 dtrace_fasttrap_exec(p);
801 * Notify others that we exec'd, and clear the P_INEXEC flag
802 * as we're now a bona fide freshly-execed process.
804 KNOTE_LOCKED(&p->p_klist, NOTE_EXEC);
805 p->p_flag &= ~P_INEXEC;
807 /* clear "fork but no exec" flag, as we _are_ execing */
808 p->p_acflag &= ~AFORK;
811 * Free any previous argument cache and replace it with
812 * the new argument cache, if any.
820 * Check if system-wide sampling is in effect or if the
821 * current process is using PMCs. If so, do exec() time
822 * processing. This processing needs to happen AFTER the
823 * P_INEXEC flag is cleared.
825 * The proc lock needs to be released before taking the PMC
828 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
830 VOP_UNLOCK(imgp->vp, 0);
831 pe.pm_credentialschanged = credential_changing;
832 pe.pm_entryaddr = imgp->entry_addr;
834 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
835 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
838 #else /* !HWPMC_HOOKS */
842 /* Set values passed into the program in registers. */
843 if (p->p_sysent->sv_setregs)
844 (*p->p_sysent->sv_setregs)(td, imgp,
845 (u_long)(uintptr_t)stack_base);
847 exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base);
849 vfs_mark_atime(imgp->vp, td->td_ucred);
851 SDT_PROBE1(proc, kernel, , exec__success, args->fname);
853 VOP_UNLOCK(imgp->vp, 0);
856 * Free any resources malloc'd earlier that we didn't use.
864 * Handle deferred decrement of ref counts.
868 if (binvp && error != 0)
873 if (tracecred != NULL)
876 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
879 if (oldsigacts != NULL)
880 sigacts_free(oldsigacts);
885 * free various allocated resources
887 if (imgp->firstpage != NULL)
888 exec_unmap_first_page(imgp);
890 if (imgp->vp != NULL) {
892 NDFREE(&nd, NDF_ONLY_PNBUF);
894 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
898 if (imgp->object != NULL)
899 vm_object_deallocate(imgp->object);
901 free(imgp->freepath, M_TEMP);
905 td->td_dbgflags |= TDB_EXEC;
909 * Stop the process here if its stop event mask has
910 * the S_EXEC bit set.
912 STOPEVENT(p, S_EXEC, 0);
917 /* we're done here, clear P_INEXEC */
919 p->p_flag &= ~P_INEXEC;
922 SDT_PROBE1(proc, kernel, , exec__failure, error);
926 mac_execve_exit(imgp);
927 mac_execve_interpreter_exit(interpvplabel);
929 exec_free_args(args);
931 if (error && imgp->vmspace_destroyed) {
932 /* sorry, no more process anymore. exit gracefully */
933 exit1(td, W_EXITCODE(0, SIGABRT));
946 exec_map_first_page(imgp)
947 struct image_params *imgp;
951 vm_page_t ma[VM_INITIAL_PAGEIN];
954 if (imgp->firstpage != NULL)
955 exec_unmap_first_page(imgp);
957 object = imgp->vp->v_object;
960 VM_OBJECT_WLOCK(object);
961 #if VM_NRESERVLEVEL > 0
962 if ((object->flags & OBJ_COLORED) == 0) {
963 object->flags |= OBJ_COLORED;
964 object->pg_color = 0;
967 ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL);
968 if (ma[0]->valid != VM_PAGE_BITS_ALL) {
969 initial_pagein = VM_INITIAL_PAGEIN;
970 if (initial_pagein > object->size)
971 initial_pagein = object->size;
972 for (i = 1; i < initial_pagein; i++) {
973 if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) {
976 if (vm_page_tryxbusy(ma[i]))
979 ma[i] = vm_page_alloc(object, i,
980 VM_ALLOC_NORMAL | VM_ALLOC_IFNOTCACHED);
986 rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
987 ma[0] = vm_page_lookup(object, 0);
988 if ((rv != VM_PAGER_OK) || (ma[0] == NULL)) {
992 vm_page_unlock(ma[0]);
994 VM_OBJECT_WUNLOCK(object);
998 vm_page_xunbusy(ma[0]);
1000 vm_page_hold(ma[0]);
1001 vm_page_activate(ma[0]);
1002 vm_page_unlock(ma[0]);
1003 VM_OBJECT_WUNLOCK(object);
1005 imgp->firstpage = sf_buf_alloc(ma[0], 0);
1006 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
1012 exec_unmap_first_page(imgp)
1013 struct image_params *imgp;
1017 if (imgp->firstpage != NULL) {
1018 m = sf_buf_page(imgp->firstpage);
1019 sf_buf_free(imgp->firstpage);
1020 imgp->firstpage = NULL;
1028 * Destroy old address space, and allocate a new stack
1029 * The new stack is only SGROWSIZ large because it is grown
1030 * automatically in trap.c.
1033 exec_new_vmspace(imgp, sv)
1034 struct image_params *imgp;
1035 struct sysentvec *sv;
1038 struct proc *p = imgp->proc;
1039 struct vmspace *vmspace = p->p_vmspace;
1041 struct rlimit rlim_stack;
1042 vm_offset_t sv_minuser, stack_addr;
1046 imgp->vmspace_destroyed = 1;
1049 /* May be called with Giant held */
1050 EVENTHANDLER_INVOKE(process_exec, p, imgp);
1053 * Blow away entire process VM, if address space not shared,
1054 * otherwise, create a new VM space so that other threads are
1057 map = &vmspace->vm_map;
1059 sv_minuser = sv->sv_minuser;
1061 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1062 if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser &&
1063 vm_map_max(map) == sv->sv_maxuser) {
1065 pmap_remove_pages(vmspace_pmap(vmspace));
1066 vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1068 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1071 vmspace = p->p_vmspace;
1072 map = &vmspace->vm_map;
1075 /* Map a shared page */
1076 obj = sv->sv_shared_page_obj;
1078 vm_object_reference(obj);
1079 error = vm_map_fixed(map, obj, 0,
1080 sv->sv_shared_page_base, sv->sv_shared_page_len,
1081 VM_PROT_READ | VM_PROT_EXECUTE,
1082 VM_PROT_READ | VM_PROT_EXECUTE,
1083 MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1085 vm_object_deallocate(obj);
1090 /* Allocate a new stack */
1091 if (imgp->stack_sz != 0) {
1092 ssiz = trunc_page(imgp->stack_sz);
1094 lim_rlimit(p, RLIMIT_STACK, &rlim_stack);
1096 if (ssiz > rlim_stack.rlim_max)
1097 ssiz = rlim_stack.rlim_max;
1098 if (ssiz > rlim_stack.rlim_cur) {
1099 rlim_stack.rlim_cur = ssiz;
1100 kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack);
1102 } else if (sv->sv_maxssiz != NULL) {
1103 ssiz = *sv->sv_maxssiz;
1107 stack_addr = sv->sv_usrstack - ssiz;
1108 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1109 obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1111 VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1116 /* Allocate a new register stack */
1117 error = vm_map_stack(map, IA64_BACKINGSTORE, (vm_size_t)ssiz,
1118 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_UP);
1123 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
1124 * VM_STACK case, but they are still used to monitor the size of the
1125 * process stack so we can check the stack rlimit.
1127 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1128 vmspace->vm_maxsaddr = (char *)stack_addr;
1134 * Copy out argument and environment strings from the old process address
1135 * space into the temporary string buffer.
1138 exec_copyin_args(struct image_args *args, char *fname,
1139 enum uio_seg segflg, char **argv, char **envv)
1145 bzero(args, sizeof(*args));
1150 * Allocate demand-paged memory for the file name, argument, and
1151 * environment strings.
1153 error = exec_alloc_args(args);
1158 * Copy the file name.
1160 if (fname != NULL) {
1161 args->fname = args->buf;
1162 error = (segflg == UIO_SYSSPACE) ?
1163 copystr(fname, args->fname, PATH_MAX, &length) :
1164 copyinstr(fname, args->fname, PATH_MAX, &length);
1170 args->begin_argv = args->buf + length;
1171 args->endp = args->begin_argv;
1172 args->stringspace = ARG_MAX;
1175 * extract arguments first
1178 error = fueword(argv++, &argp);
1185 error = copyinstr((void *)(uintptr_t)argp, args->endp,
1186 args->stringspace, &length);
1188 if (error == ENAMETOOLONG)
1192 args->stringspace -= length;
1193 args->endp += length;
1197 args->begin_envv = args->endp;
1200 * extract environment strings
1204 error = fueword(envv++, &envp);
1211 error = copyinstr((void *)(uintptr_t)envp,
1212 args->endp, args->stringspace, &length);
1214 if (error == ENAMETOOLONG)
1218 args->stringspace -= length;
1219 args->endp += length;
1227 exec_free_args(args);
1232 * Allocate temporary demand-paged, zero-filled memory for the file name,
1233 * argument, and environment strings. Returns zero if the allocation succeeds
1234 * and ENOMEM otherwise.
1237 exec_alloc_args(struct image_args *args)
1240 args->buf = (char *)kmap_alloc_wait(exec_map, PATH_MAX + ARG_MAX);
1241 return (args->buf != NULL ? 0 : ENOMEM);
1245 exec_free_args(struct image_args *args)
1248 if (args->buf != NULL) {
1249 kmap_free_wakeup(exec_map, (vm_offset_t)args->buf,
1250 PATH_MAX + ARG_MAX);
1253 if (args->fname_buf != NULL) {
1254 free(args->fname_buf, M_TEMP);
1255 args->fname_buf = NULL;
1260 * Copy strings out to the new process address space, constructing new arg
1261 * and env vector tables. Return a pointer to the base so that it can be used
1262 * as the initial stack pointer.
1265 exec_copyout_strings(imgp)
1266 struct image_params *imgp;
1272 register_t *stack_base;
1273 struct ps_strings *arginfo;
1275 size_t execpath_len;
1276 int szsigcode, szps;
1277 char canary[sizeof(long) * 8];
1279 szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1281 * Calculate string base and vector table pointers.
1282 * Also deal with signal trampoline code for this exec type.
1284 if (imgp->execpath != NULL && imgp->auxargs != NULL)
1285 execpath_len = strlen(imgp->execpath) + 1;
1290 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1291 if (p->p_sysent->sv_sigcode_base == 0) {
1292 if (p->p_sysent->sv_szsigcode != NULL)
1293 szsigcode = *(p->p_sysent->sv_szsigcode);
1295 destp = (uintptr_t)arginfo;
1300 if (szsigcode != 0) {
1302 destp = rounddown2(destp, sizeof(void *));
1303 copyout(p->p_sysent->sv_sigcode, (void *)destp, szsigcode);
1307 * Copy the image path for the rtld.
1309 if (execpath_len != 0) {
1310 destp -= execpath_len;
1311 imgp->execpathp = destp;
1312 copyout(imgp->execpath, (void *)destp, execpath_len);
1316 * Prepare the canary for SSP.
1318 arc4rand(canary, sizeof(canary), 0);
1319 destp -= sizeof(canary);
1320 imgp->canary = destp;
1321 copyout(canary, (void *)destp, sizeof(canary));
1322 imgp->canarylen = sizeof(canary);
1325 * Prepare the pagesizes array.
1328 destp = rounddown2(destp, sizeof(void *));
1329 imgp->pagesizes = destp;
1330 copyout(pagesizes, (void *)destp, szps);
1331 imgp->pagesizeslen = szps;
1333 destp -= ARG_MAX - imgp->args->stringspace;
1334 destp = rounddown2(destp, sizeof(void *));
1337 * If we have a valid auxargs ptr, prepare some room
1340 if (imgp->auxargs) {
1342 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1343 * lower compatibility.
1345 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1348 * The '+ 2' is for the null pointers at the end of each of
1349 * the arg and env vector sets,and imgp->auxarg_size is room
1350 * for argument of Runtime loader.
1352 vectp = (char **)(destp - (imgp->args->argc +
1353 imgp->args->envc + 2 + imgp->auxarg_size)
1357 * The '+ 2' is for the null pointers at the end of each of
1358 * the arg and env vector sets
1360 vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc
1361 + 2) * sizeof(char *));
1365 * vectp also becomes our initial stack base
1367 stack_base = (register_t *)vectp;
1369 stringp = imgp->args->begin_argv;
1370 argc = imgp->args->argc;
1371 envc = imgp->args->envc;
1374 * Copy out strings - arguments and environment.
1376 copyout(stringp, (void *)destp, ARG_MAX - imgp->args->stringspace);
1379 * Fill in "ps_strings" struct for ps, w, etc.
1381 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1382 suword32(&arginfo->ps_nargvstr, argc);
1385 * Fill in argument portion of vector table.
1387 for (; argc > 0; --argc) {
1388 suword(vectp++, (long)(intptr_t)destp);
1389 while (*stringp++ != 0)
1394 /* a null vector table pointer separates the argp's from the envp's */
1397 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1398 suword32(&arginfo->ps_nenvstr, envc);
1401 * Fill in environment portion of vector table.
1403 for (; envc > 0; --envc) {
1404 suword(vectp++, (long)(intptr_t)destp);
1405 while (*stringp++ != 0)
1410 /* end of vector table is a null pointer */
1413 return (stack_base);
1417 * Check permissions of file to execute.
1418 * Called with imgp->vp locked.
1419 * Return 0 for success or error code on failure.
1422 exec_check_permissions(imgp)
1423 struct image_params *imgp;
1425 struct vnode *vp = imgp->vp;
1426 struct vattr *attr = imgp->attr;
1428 int error, writecount;
1432 /* Get file attributes */
1433 error = VOP_GETATTR(vp, attr, td->td_ucred);
1438 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1444 * 1) Check if file execution is disabled for the filesystem that
1445 * this file resides on.
1446 * 2) Ensure that at least one execute bit is on. Otherwise, a
1447 * privileged user will always succeed, and we don't want this
1448 * to happen unless the file really is executable.
1449 * 3) Ensure that the file is a regular file.
1451 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1452 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1453 (attr->va_type != VREG))
1457 * Zero length files can't be exec'd
1459 if (attr->va_size == 0)
1463 * Check for execute permission to file based on current credentials.
1465 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1470 * Check number of open-for-writes on the file and deny execution
1473 error = VOP_GET_WRITECOUNT(vp, &writecount);
1476 if (writecount != 0)
1480 * Call filesystem specific open routine (which does nothing in the
1483 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1490 * Exec handler registration
1493 exec_register(execsw_arg)
1494 const struct execsw *execsw_arg;
1496 const struct execsw **es, **xs, **newexecsw;
1497 int count = 2; /* New slot and trailing NULL */
1500 for (es = execsw; *es; es++)
1502 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1503 if (newexecsw == NULL)
1507 for (es = execsw; *es; es++)
1512 free(execsw, M_TEMP);
1518 exec_unregister(execsw_arg)
1519 const struct execsw *execsw_arg;
1521 const struct execsw **es, **xs, **newexecsw;
1525 panic("unregister with no handlers left?\n");
1527 for (es = execsw; *es; es++) {
1528 if (*es == execsw_arg)
1533 for (es = execsw; *es; es++)
1534 if (*es != execsw_arg)
1536 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1537 if (newexecsw == NULL)
1540 for (es = execsw; *es; es++)
1541 if (*es != execsw_arg)
1545 free(execsw, M_TEMP);