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_compat.h"
32 #include "opt_hwpmc_hooks.h"
33 #include "opt_ktrace.h"
36 #include <sys/param.h>
37 #include <sys/systm.h>
39 #include <sys/capsicum.h>
40 #include <sys/eventhandler.h>
42 #include <sys/fcntl.h>
43 #include <sys/filedesc.h>
44 #include <sys/imgact.h>
45 #include <sys/imgact_elf.h>
46 #include <sys/kernel.h>
48 #include <sys/malloc.h>
50 #include <sys/mount.h>
51 #include <sys/mutex.h>
52 #include <sys/namei.h>
53 #include <sys/pioctl.h>
56 #include <sys/ptrace.h>
57 #include <sys/resourcevar.h>
58 #include <sys/rwlock.h>
59 #include <sys/sched.h>
61 #include <sys/sf_buf.h>
63 #include <sys/signalvar.h>
66 #include <sys/syscallsubr.h>
67 #include <sys/sysctl.h>
68 #include <sys/sysent.h>
69 #include <sys/sysproto.h>
70 #include <sys/vnode.h>
73 #include <sys/ktrace.h>
77 #include <vm/vm_param.h>
79 #include <vm/vm_page.h>
80 #include <vm/vm_map.h>
81 #include <vm/vm_kern.h>
82 #include <vm/vm_extern.h>
83 #include <vm/vm_object.h>
84 #include <vm/vm_pager.h>
87 #include <sys/pmckern.h>
90 #include <machine/reg.h>
92 #include <security/audit/audit.h>
93 #include <security/mac/mac_framework.h>
96 #include <sys/dtrace_bsd.h>
97 dtrace_execexit_func_t dtrace_fasttrap_exec;
100 SDT_PROVIDER_DECLARE(proc);
101 SDT_PROBE_DEFINE1(proc, , , exec, "char *");
102 SDT_PROBE_DEFINE1(proc, , , exec__failure, "int");
103 SDT_PROBE_DEFINE1(proc, , , exec__success, "char *");
105 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
107 int coredump_pack_fileinfo = 1;
108 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_fileinfo, CTLFLAG_RWTUN,
109 &coredump_pack_fileinfo, 0,
110 "Enable file path packing in 'procstat -f' coredump notes");
112 int coredump_pack_vmmapinfo = 1;
113 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_vmmapinfo, CTLFLAG_RWTUN,
114 &coredump_pack_vmmapinfo, 0,
115 "Enable file path packing in 'procstat -v' coredump notes");
117 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
118 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
119 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
120 static int do_execve(struct thread *td, struct image_args *args,
123 /* XXX This should be vm_size_t. */
124 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD|
125 CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_ps_strings, "LU", "");
127 /* XXX This should be vm_size_t. */
128 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD|
129 CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_usrstack, "LU", "");
131 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD|CTLFLAG_MPSAFE,
132 NULL, 0, sysctl_kern_stackprot, "I", "");
134 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
135 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
136 &ps_arg_cache_limit, 0, "");
138 static int disallow_high_osrel;
139 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW,
140 &disallow_high_osrel, 0,
141 "Disallow execution of binaries built for higher version of the world");
143 static int map_at_zero = 0;
144 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RWTUN, &map_at_zero, 0,
145 "Permit processes to map an object at virtual address 0.");
147 EVENTHANDLER_LIST_DECLARE(process_exec);
150 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
157 if (req->flags & SCTL_MASK32) {
159 val = (unsigned int)p->p_sysent->sv_psstrings;
160 error = SYSCTL_OUT(req, &val, sizeof(val));
163 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
164 sizeof(p->p_sysent->sv_psstrings));
169 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
176 if (req->flags & SCTL_MASK32) {
178 val = (unsigned int)p->p_sysent->sv_usrstack;
179 error = SYSCTL_OUT(req, &val, sizeof(val));
182 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
183 sizeof(p->p_sysent->sv_usrstack));
188 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
193 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
194 sizeof(p->p_sysent->sv_stackprot)));
198 * Each of the items is a pointer to a `const struct execsw', hence the
199 * double pointer here.
201 static const struct execsw **execsw;
203 #ifndef _SYS_SYSPROTO_H_
212 sys_execve(struct thread *td, struct execve_args *uap)
214 struct image_args args;
215 struct vmspace *oldvmspace;
218 error = pre_execve(td, &oldvmspace);
221 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
222 uap->argv, uap->envv);
224 error = kern_execve(td, &args, NULL);
225 post_execve(td, error, oldvmspace);
229 #ifndef _SYS_SYSPROTO_H_
230 struct fexecve_args {
237 sys_fexecve(struct thread *td, struct fexecve_args *uap)
239 struct image_args args;
240 struct vmspace *oldvmspace;
243 error = pre_execve(td, &oldvmspace);
246 error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
247 uap->argv, uap->envv);
250 error = kern_execve(td, &args, NULL);
252 post_execve(td, error, oldvmspace);
256 #ifndef _SYS_SYSPROTO_H_
257 struct __mac_execve_args {
266 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap)
269 struct image_args args;
270 struct vmspace *oldvmspace;
273 error = pre_execve(td, &oldvmspace);
276 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
277 uap->argv, uap->envv);
279 error = kern_execve(td, &args, uap->mac_p);
280 post_execve(td, error, oldvmspace);
288 pre_execve(struct thread *td, struct vmspace **oldvmspace)
293 KASSERT(td == curthread, ("non-current thread %p", td));
296 if ((p->p_flag & P_HADTHREADS) != 0) {
298 if (thread_single(p, SINGLE_BOUNDARY) != 0)
302 KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0,
304 *oldvmspace = p->p_vmspace;
309 post_execve(struct thread *td, int error, struct vmspace *oldvmspace)
313 KASSERT(td == curthread, ("non-current thread %p", td));
315 if ((p->p_flag & P_HADTHREADS) != 0) {
318 * If success, we upgrade to SINGLE_EXIT state to
319 * force other threads to suicide.
322 thread_single(p, SINGLE_EXIT);
324 thread_single_end(p, SINGLE_BOUNDARY);
327 if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
328 KASSERT(p->p_vmspace != oldvmspace,
329 ("oldvmspace still used"));
330 vmspace_free(oldvmspace);
331 td->td_pflags &= ~TDP_EXECVMSPC;
336 * XXX: kern_execve has the astonishing property of not always returning to
337 * the caller. If sufficiently bad things happen during the call to
338 * do_execve(), it can end up calling exit1(); as a result, callers must
339 * avoid doing anything which they might need to undo (e.g., allocating
343 kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p)
346 AUDIT_ARG_ARGV(args->begin_argv, args->argc,
347 args->begin_envv - args->begin_argv);
348 AUDIT_ARG_ENVV(args->begin_envv, args->envc,
349 args->endp - args->begin_envv);
350 return (do_execve(td, args, mac_p));
354 * In-kernel implementation of execve(). All arguments are assumed to be
355 * userspace pointers from the passed thread.
358 do_execve(struct thread *td, struct image_args *args, struct mac *mac_p)
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;
673 /* ABI enforces the use of Capsicum. Switch into capabilities mode. */
674 if (SV_PROC_FLAG(p, SV_CAPSICUM))
675 sys_cap_enter(td, NULL);
678 * Copy out strings (args and env) and initialize stack base
680 if (p->p_sysent->sv_copyout_strings)
681 stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
683 stack_base = exec_copyout_strings(imgp);
686 * If custom stack fixup routine present for this process
687 * let it do the stack setup.
688 * Else stuff argument count as first item on stack
690 if (p->p_sysent->sv_fixup != NULL)
691 (*p->p_sysent->sv_fixup)(&stack_base, imgp);
693 suword(--stack_base, imgp->args->argc);
695 if (args->fdp != NULL) {
696 /* Install a brand new file descriptor table. */
697 fdinstall_remapped(td, args->fdp);
701 * Keep on using the existing file descriptor table. For
702 * security and other reasons, the file descriptor table
703 * cannot be shared after an exec.
706 /* close files on exec */
711 * Malloc things before we need locks.
713 i = imgp->args->begin_envv - imgp->args->begin_argv;
714 /* Cache arguments if they fit inside our allowance */
715 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
716 newargs = pargs_alloc(i);
717 bcopy(imgp->args->begin_argv, newargs->ar_args, i);
721 * For security and other reasons, signal handlers cannot
722 * be shared after an exec. The new process gets a copy of the old
723 * handlers. In execsigs(), the new process will have its signals
726 if (sigacts_shared(p->p_sigacts)) {
727 oldsigacts = p->p_sigacts;
728 newsigacts = sigacts_alloc();
729 sigacts_copy(newsigacts, oldsigacts);
732 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
736 p->p_sigacts = newsigacts;
740 /* reset caught signals */
743 /* name this process - nameiexec(p, ndp) */
744 bzero(p->p_comm, sizeof(p->p_comm));
746 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
747 min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
748 else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0)
749 bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
750 bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
752 sched_clear_tdname(td);
756 * mark as execed, wakeup the process that vforked (if any) and tell
757 * it that it now has its own resources back
760 if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0)
761 p->p_flag2 &= ~P2_NOTRACE;
762 if (p->p_flag & P_PPWAIT) {
763 p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
764 cv_broadcast(&p->p_pwait);
765 /* STOPs are no longer ignored, arrange for AST */
770 * Implement image setuid/setgid installation.
772 if (imgp->credential_setid) {
774 * Turn off syscall tracing for set-id programs, except for
775 * root. Record any set-id flags first to make sure that
776 * we do not regain any tracing during a possible block.
781 if (p->p_tracecred != NULL &&
782 priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED, 0))
783 ktrprocexec(p, &tracecred, &tracevp);
786 * Close any file descriptors 0..2 that reference procfs,
787 * then make sure file descriptors 0..2 are in use.
789 * Both fdsetugidsafety() and fdcheckstd() may call functions
790 * taking sleepable locks, so temporarily drop our locks.
793 VOP_UNLOCK(imgp->vp, 0);
795 error = fdcheckstd(td);
796 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
798 goto exec_fail_dealloc;
801 if (will_transition) {
802 mac_vnode_execve_transition(oldcred, imgp->newcred,
803 imgp->vp, interpvplabel, imgp);
807 if (oldcred->cr_uid == oldcred->cr_ruid &&
808 oldcred->cr_gid == oldcred->cr_rgid)
809 p->p_flag &= ~P_SUGID;
812 * Set the new credentials.
814 if (imgp->newcred != NULL) {
815 proc_set_cred(p, imgp->newcred);
821 * Store the vp for use in procfs. This vnode was referenced by namei
824 oldtextvp = p->p_textvp;
825 p->p_textvp = newtextvp;
829 * Tell the DTrace fasttrap provider about the exec if it
830 * has declared an interest.
832 if (dtrace_fasttrap_exec)
833 dtrace_fasttrap_exec(p);
837 * Notify others that we exec'd, and clear the P_INEXEC flag
838 * as we're now a bona fide freshly-execed process.
840 KNOTE_LOCKED(p->p_klist, NOTE_EXEC);
841 p->p_flag &= ~P_INEXEC;
843 /* clear "fork but no exec" flag, as we _are_ execing */
844 p->p_acflag &= ~AFORK;
847 * Free any previous argument cache and replace it with
848 * the new argument cache, if any.
858 * Check if system-wide sampling is in effect or if the
859 * current process is using PMCs. If so, do exec() time
860 * processing. This processing needs to happen AFTER the
861 * P_INEXEC flag is cleared.
863 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
864 VOP_UNLOCK(imgp->vp, 0);
865 pe.pm_credentialschanged = credential_changing;
866 pe.pm_entryaddr = imgp->entry_addr;
868 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
869 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
873 /* Set values passed into the program in registers. */
874 if (p->p_sysent->sv_setregs)
875 (*p->p_sysent->sv_setregs)(td, imgp,
876 (u_long)(uintptr_t)stack_base);
878 exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base);
880 vfs_mark_atime(imgp->vp, td->td_ucred);
882 SDT_PROBE1(proc, , , exec__success, args->fname);
885 if (imgp->firstpage != NULL)
886 exec_unmap_first_page(imgp);
888 if (imgp->vp != NULL) {
890 NDFREE(&nd, NDF_ONLY_PNBUF);
892 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
896 VOP_UNLOCK(imgp->vp, 0);
899 if (imgp->object != NULL)
900 vm_object_deallocate(imgp->object);
902 free(imgp->freepath, M_TEMP);
905 if (p->p_ptevents & PTRACE_EXEC) {
907 if (p->p_ptevents & PTRACE_EXEC)
908 td->td_dbgflags |= TDB_EXEC;
913 * Stop the process here if its stop event mask has
914 * the S_EXEC bit set.
916 STOPEVENT(p, S_EXEC, 0);
919 /* we're done here, clear P_INEXEC */
921 p->p_flag &= ~P_INEXEC;
924 SDT_PROBE1(proc, , , exec__failure, error);
927 if (imgp->newcred != NULL && oldcred != NULL)
928 crfree(imgp->newcred);
931 mac_execve_exit(imgp);
932 mac_execve_interpreter_exit(interpvplabel);
934 exec_free_args(args);
937 * Handle deferred decrement of ref counts.
939 if (oldtextvp != NULL)
944 if (tracecred != NULL)
949 if (oldsigacts != NULL)
950 sigacts_free(oldsigacts);
954 if (error && imgp->vmspace_destroyed) {
955 /* sorry, no more process anymore. exit gracefully */
956 exit1(td, 0, SIGABRT);
969 exec_map_first_page(imgp)
970 struct image_params *imgp;
972 int rv, i, after, initial_pagein;
973 vm_page_t ma[VM_INITIAL_PAGEIN];
976 if (imgp->firstpage != NULL)
977 exec_unmap_first_page(imgp);
979 object = imgp->vp->v_object;
982 VM_OBJECT_WLOCK(object);
983 #if VM_NRESERVLEVEL > 0
984 vm_object_color(object, 0);
986 ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY);
987 if (ma[0]->valid != VM_PAGE_BITS_ALL) {
988 vm_page_xbusy(ma[0]);
989 if (!vm_pager_has_page(object, 0, NULL, &after)) {
992 vm_page_unlock(ma[0]);
993 VM_OBJECT_WUNLOCK(object);
996 initial_pagein = min(after, VM_INITIAL_PAGEIN);
997 KASSERT(initial_pagein <= object->size,
998 ("%s: initial_pagein %d object->size %ju",
999 __func__, initial_pagein, (uintmax_t )object->size));
1000 for (i = 1; i < initial_pagein; i++) {
1001 if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) {
1004 if (vm_page_tryxbusy(ma[i]))
1007 ma[i] = vm_page_alloc(object, i,
1014 rv = vm_pager_get_pages(object, ma, initial_pagein, NULL, NULL);
1015 if (rv != VM_PAGER_OK) {
1016 for (i = 0; i < initial_pagein; i++) {
1017 vm_page_lock(ma[i]);
1018 vm_page_free(ma[i]);
1019 vm_page_unlock(ma[i]);
1021 VM_OBJECT_WUNLOCK(object);
1024 vm_page_xunbusy(ma[0]);
1025 for (i = 1; i < initial_pagein; i++)
1026 vm_page_readahead_finish(ma[i]);
1028 vm_page_lock(ma[0]);
1029 vm_page_hold(ma[0]);
1030 vm_page_activate(ma[0]);
1031 vm_page_unlock(ma[0]);
1032 VM_OBJECT_WUNLOCK(object);
1034 imgp->firstpage = sf_buf_alloc(ma[0], 0);
1035 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
1041 exec_unmap_first_page(struct image_params *imgp)
1045 if (imgp->firstpage != NULL) {
1046 m = sf_buf_page(imgp->firstpage);
1047 sf_buf_free(imgp->firstpage);
1048 imgp->firstpage = NULL;
1056 * Destroy old address space, and allocate a new stack.
1057 * The new stack is only sgrowsiz large because it is grown
1058 * automatically on a page fault.
1061 exec_new_vmspace(struct image_params *imgp, struct sysentvec *sv)
1064 struct proc *p = imgp->proc;
1065 struct vmspace *vmspace = p->p_vmspace;
1067 struct rlimit rlim_stack;
1068 vm_offset_t sv_minuser, stack_addr;
1072 imgp->vmspace_destroyed = 1;
1075 /* May be called with Giant held */
1076 EVENTHANDLER_DIRECT_INVOKE(process_exec, p, imgp);
1079 * Blow away entire process VM, if address space not shared,
1080 * otherwise, create a new VM space so that other threads are
1083 map = &vmspace->vm_map;
1085 sv_minuser = sv->sv_minuser;
1087 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1088 if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser &&
1089 vm_map_max(map) == sv->sv_maxuser) {
1091 pmap_remove_pages(vmspace_pmap(vmspace));
1092 vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1093 /* An exec terminates mlockall(MCL_FUTURE). */
1095 vm_map_modflags(map, 0, MAP_WIREFUTURE);
1098 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1101 vmspace = p->p_vmspace;
1102 map = &vmspace->vm_map;
1105 /* Map a shared page */
1106 obj = sv->sv_shared_page_obj;
1108 vm_object_reference(obj);
1109 error = vm_map_fixed(map, obj, 0,
1110 sv->sv_shared_page_base, sv->sv_shared_page_len,
1111 VM_PROT_READ | VM_PROT_EXECUTE,
1112 VM_PROT_READ | VM_PROT_EXECUTE,
1113 MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1114 if (error != KERN_SUCCESS) {
1115 vm_object_deallocate(obj);
1116 return (vm_mmap_to_errno(error));
1120 /* Allocate a new stack */
1121 if (imgp->stack_sz != 0) {
1122 ssiz = trunc_page(imgp->stack_sz);
1124 lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack);
1126 if (ssiz > rlim_stack.rlim_max)
1127 ssiz = rlim_stack.rlim_max;
1128 if (ssiz > rlim_stack.rlim_cur) {
1129 rlim_stack.rlim_cur = ssiz;
1130 kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack);
1132 } else if (sv->sv_maxssiz != NULL) {
1133 ssiz = *sv->sv_maxssiz;
1137 stack_addr = sv->sv_usrstack - ssiz;
1138 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1139 obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1140 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1141 if (error != KERN_SUCCESS)
1142 return (vm_mmap_to_errno(error));
1145 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they
1146 * are still used to enforce the stack rlimit on the process stack.
1148 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1149 vmspace->vm_maxsaddr = (char *)stack_addr;
1155 * Copy out argument and environment strings from the old process address
1156 * space into the temporary string buffer.
1159 exec_copyin_args(struct image_args *args, char *fname,
1160 enum uio_seg segflg, char **argv, char **envv)
1166 bzero(args, sizeof(*args));
1171 * Allocate demand-paged memory for the file name, argument, and
1172 * environment strings.
1174 error = exec_alloc_args(args);
1179 * Copy the file name.
1181 if (fname != NULL) {
1182 args->fname = args->buf;
1183 error = (segflg == UIO_SYSSPACE) ?
1184 copystr(fname, args->fname, PATH_MAX, &length) :
1185 copyinstr(fname, args->fname, PATH_MAX, &length);
1191 args->begin_argv = args->buf + length;
1192 args->endp = args->begin_argv;
1193 args->stringspace = ARG_MAX;
1196 * extract arguments first
1199 error = fueword(argv++, &argp);
1206 error = copyinstr((void *)(uintptr_t)argp, args->endp,
1207 args->stringspace, &length);
1209 if (error == ENAMETOOLONG)
1213 args->stringspace -= length;
1214 args->endp += length;
1218 args->begin_envv = args->endp;
1221 * extract environment strings
1225 error = fueword(envv++, &envp);
1232 error = copyinstr((void *)(uintptr_t)envp,
1233 args->endp, args->stringspace, &length);
1235 if (error == ENAMETOOLONG)
1239 args->stringspace -= length;
1240 args->endp += length;
1248 exec_free_args(args);
1253 exec_copyin_data_fds(struct thread *td, struct image_args *args,
1254 const void *data, size_t datalen, const int *fds, size_t fdslen)
1256 struct filedesc *ofdp;
1261 memset(args, '\0', sizeof(*args));
1262 ofdp = td->td_proc->p_fd;
1263 if (datalen >= ARG_MAX || fdslen > ofdp->fd_lastfile + 1)
1265 error = exec_alloc_args(args);
1269 args->begin_argv = args->buf;
1270 args->stringspace = ARG_MAX;
1274 * Argument buffer has been provided. Copy it into the
1275 * kernel as a single string and add a terminating null
1278 error = copyin(data, args->begin_argv, datalen);
1281 args->begin_argv[datalen] = '\0';
1282 args->endp = args->begin_argv + datalen + 1;
1283 args->stringspace -= datalen + 1;
1286 * Traditional argument counting. Count the number of
1289 for (p = args->begin_argv; p < args->endp; ++p)
1293 /* No argument buffer provided. */
1294 args->endp = args->begin_argv;
1296 /* There are no environment variables. */
1297 args->begin_envv = args->endp;
1299 /* Create new file descriptor table. */
1300 kfds = malloc(fdslen * sizeof(int), M_TEMP, M_WAITOK);
1301 error = copyin(fds, kfds, fdslen * sizeof(int));
1306 error = fdcopy_remapped(ofdp, kfds, fdslen, &args->fdp);
1313 exec_free_args(args);
1317 struct exec_args_kva {
1320 SLIST_ENTRY(exec_args_kva) next;
1323 static DPCPU_DEFINE(struct exec_args_kva *, exec_args_kva);
1325 static SLIST_HEAD(, exec_args_kva) exec_args_kva_freelist;
1326 static struct mtx exec_args_kva_mtx;
1327 static u_int exec_args_gen;
1330 exec_prealloc_args_kva(void *arg __unused)
1332 struct exec_args_kva *argkva;
1335 SLIST_INIT(&exec_args_kva_freelist);
1336 mtx_init(&exec_args_kva_mtx, "exec args kva", NULL, MTX_DEF);
1337 for (i = 0; i < exec_map_entries; i++) {
1338 argkva = malloc(sizeof(*argkva), M_PARGS, M_WAITOK);
1339 argkva->addr = kmap_alloc_wait(exec_map, exec_map_entry_size);
1340 argkva->gen = exec_args_gen;
1341 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1344 SYSINIT(exec_args_kva, SI_SUB_EXEC, SI_ORDER_ANY, exec_prealloc_args_kva, NULL);
1347 exec_alloc_args_kva(void **cookie)
1349 struct exec_args_kva *argkva;
1351 argkva = (void *)atomic_readandclear_ptr(
1352 (uintptr_t *)DPCPU_PTR(exec_args_kva));
1353 if (argkva == NULL) {
1354 mtx_lock(&exec_args_kva_mtx);
1355 while ((argkva = SLIST_FIRST(&exec_args_kva_freelist)) == NULL)
1356 (void)mtx_sleep(&exec_args_kva_freelist,
1357 &exec_args_kva_mtx, 0, "execkva", 0);
1358 SLIST_REMOVE_HEAD(&exec_args_kva_freelist, next);
1359 mtx_unlock(&exec_args_kva_mtx);
1361 *(struct exec_args_kva **)cookie = argkva;
1362 return (argkva->addr);
1366 exec_release_args_kva(struct exec_args_kva *argkva, u_int gen)
1370 base = argkva->addr;
1371 if (argkva->gen != gen) {
1372 vm_map_madvise(exec_map, base, base + exec_map_entry_size,
1376 if (!atomic_cmpset_ptr((uintptr_t *)DPCPU_PTR(exec_args_kva),
1377 (uintptr_t)NULL, (uintptr_t)argkva)) {
1378 mtx_lock(&exec_args_kva_mtx);
1379 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1380 wakeup_one(&exec_args_kva_freelist);
1381 mtx_unlock(&exec_args_kva_mtx);
1386 exec_free_args_kva(void *cookie)
1389 exec_release_args_kva(cookie, exec_args_gen);
1393 exec_args_kva_lowmem(void *arg __unused)
1395 SLIST_HEAD(, exec_args_kva) head;
1396 struct exec_args_kva *argkva;
1400 gen = atomic_fetchadd_int(&exec_args_gen, 1) + 1;
1403 * Force an madvise of each KVA range. Any currently allocated ranges
1404 * will have MADV_FREE applied once they are freed.
1407 mtx_lock(&exec_args_kva_mtx);
1408 SLIST_SWAP(&head, &exec_args_kva_freelist, exec_args_kva);
1409 mtx_unlock(&exec_args_kva_mtx);
1410 while ((argkva = SLIST_FIRST(&head)) != NULL) {
1411 SLIST_REMOVE_HEAD(&head, next);
1412 exec_release_args_kva(argkva, gen);
1416 argkva = (void *)atomic_readandclear_ptr(
1417 (uintptr_t *)DPCPU_ID_PTR(i, exec_args_kva));
1419 exec_release_args_kva(argkva, gen);
1422 EVENTHANDLER_DEFINE(vm_lowmem, exec_args_kva_lowmem, NULL,
1423 EVENTHANDLER_PRI_ANY);
1426 * Allocate temporary demand-paged, zero-filled memory for the file name,
1427 * argument, and environment strings.
1430 exec_alloc_args(struct image_args *args)
1433 args->buf = (char *)exec_alloc_args_kva(&args->bufkva);
1438 exec_free_args(struct image_args *args)
1441 if (args->buf != NULL) {
1442 exec_free_args_kva(args->bufkva);
1445 if (args->fname_buf != NULL) {
1446 free(args->fname_buf, M_TEMP);
1447 args->fname_buf = NULL;
1449 if (args->fdp != NULL)
1450 fdescfree_remapped(args->fdp);
1454 * Copy strings out to the new process address space, constructing new arg
1455 * and env vector tables. Return a pointer to the base so that it can be used
1456 * as the initial stack pointer.
1459 exec_copyout_strings(struct image_params *imgp)
1465 register_t *stack_base;
1466 struct ps_strings *arginfo;
1468 size_t execpath_len;
1469 int szsigcode, szps;
1470 char canary[sizeof(long) * 8];
1472 szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1474 * Calculate string base and vector table pointers.
1475 * Also deal with signal trampoline code for this exec type.
1477 if (imgp->execpath != NULL && imgp->auxargs != NULL)
1478 execpath_len = strlen(imgp->execpath) + 1;
1483 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1484 if (p->p_sysent->sv_sigcode_base == 0) {
1485 if (p->p_sysent->sv_szsigcode != NULL)
1486 szsigcode = *(p->p_sysent->sv_szsigcode);
1488 destp = (uintptr_t)arginfo;
1493 if (szsigcode != 0) {
1495 destp = rounddown2(destp, sizeof(void *));
1496 copyout(p->p_sysent->sv_sigcode, (void *)destp, szsigcode);
1500 * Copy the image path for the rtld.
1502 if (execpath_len != 0) {
1503 destp -= execpath_len;
1504 imgp->execpathp = destp;
1505 copyout(imgp->execpath, (void *)destp, execpath_len);
1509 * Prepare the canary for SSP.
1511 arc4rand(canary, sizeof(canary), 0);
1512 destp -= sizeof(canary);
1513 imgp->canary = destp;
1514 copyout(canary, (void *)destp, sizeof(canary));
1515 imgp->canarylen = sizeof(canary);
1518 * Prepare the pagesizes array.
1521 destp = rounddown2(destp, sizeof(void *));
1522 imgp->pagesizes = destp;
1523 copyout(pagesizes, (void *)destp, szps);
1524 imgp->pagesizeslen = szps;
1526 destp -= ARG_MAX - imgp->args->stringspace;
1527 destp = rounddown2(destp, sizeof(void *));
1530 * If we have a valid auxargs ptr, prepare some room
1533 if (imgp->auxargs) {
1535 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1536 * lower compatibility.
1538 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1541 * The '+ 2' is for the null pointers at the end of each of
1542 * the arg and env vector sets,and imgp->auxarg_size is room
1543 * for argument of Runtime loader.
1545 vectp = (char **)(destp - (imgp->args->argc +
1546 imgp->args->envc + 2 + imgp->auxarg_size)
1550 * The '+ 2' is for the null pointers at the end of each of
1551 * the arg and env vector sets
1553 vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc
1554 + 2) * sizeof(char *));
1558 * vectp also becomes our initial stack base
1560 stack_base = (register_t *)vectp;
1562 stringp = imgp->args->begin_argv;
1563 argc = imgp->args->argc;
1564 envc = imgp->args->envc;
1567 * Copy out strings - arguments and environment.
1569 copyout(stringp, (void *)destp, ARG_MAX - imgp->args->stringspace);
1572 * Fill in "ps_strings" struct for ps, w, etc.
1574 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1575 suword32(&arginfo->ps_nargvstr, argc);
1578 * Fill in argument portion of vector table.
1580 for (; argc > 0; --argc) {
1581 suword(vectp++, (long)(intptr_t)destp);
1582 while (*stringp++ != 0)
1587 /* a null vector table pointer separates the argp's from the envp's */
1590 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1591 suword32(&arginfo->ps_nenvstr, envc);
1594 * Fill in environment portion of vector table.
1596 for (; envc > 0; --envc) {
1597 suword(vectp++, (long)(intptr_t)destp);
1598 while (*stringp++ != 0)
1603 /* end of vector table is a null pointer */
1606 return (stack_base);
1610 * Check permissions of file to execute.
1611 * Called with imgp->vp locked.
1612 * Return 0 for success or error code on failure.
1615 exec_check_permissions(struct image_params *imgp)
1617 struct vnode *vp = imgp->vp;
1618 struct vattr *attr = imgp->attr;
1620 int error, writecount;
1624 /* Get file attributes */
1625 error = VOP_GETATTR(vp, attr, td->td_ucred);
1630 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1636 * 1) Check if file execution is disabled for the filesystem that
1637 * this file resides on.
1638 * 2) Ensure that at least one execute bit is on. Otherwise, a
1639 * privileged user will always succeed, and we don't want this
1640 * to happen unless the file really is executable.
1641 * 3) Ensure that the file is a regular file.
1643 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1644 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1645 (attr->va_type != VREG))
1649 * Zero length files can't be exec'd
1651 if (attr->va_size == 0)
1655 * Check for execute permission to file based on current credentials.
1657 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1662 * Check number of open-for-writes on the file and deny execution
1665 error = VOP_GET_WRITECOUNT(vp, &writecount);
1668 if (writecount != 0)
1672 * Call filesystem specific open routine (which does nothing in the
1675 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1682 * Exec handler registration
1685 exec_register(const struct execsw *execsw_arg)
1687 const struct execsw **es, **xs, **newexecsw;
1688 int count = 2; /* New slot and trailing NULL */
1691 for (es = execsw; *es; es++)
1693 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1696 for (es = execsw; *es; es++)
1701 free(execsw, M_TEMP);
1707 exec_unregister(const struct execsw *execsw_arg)
1709 const struct execsw **es, **xs, **newexecsw;
1713 panic("unregister with no handlers left?\n");
1715 for (es = execsw; *es; es++) {
1716 if (*es == execsw_arg)
1721 for (es = execsw; *es; es++)
1722 if (*es != execsw_arg)
1724 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1726 for (es = execsw; *es; es++)
1727 if (*es != execsw_arg)
1731 free(execsw, M_TEMP);