2 * Copyright (c) 1993, David Greenman
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD$");
30 #include "opt_capsicum.h"
31 #include "opt_hwpmc_hooks.h"
32 #include "opt_kdtrace.h"
33 #include "opt_ktrace.h"
36 #include <sys/param.h>
37 #include <sys/capability.h>
38 #include <sys/systm.h>
39 #include <sys/capability.h>
40 #include <sys/eventhandler.h>
42 #include <sys/mutex.h>
43 #include <sys/sysproto.h>
44 #include <sys/signalvar.h>
45 #include <sys/kernel.h>
46 #include <sys/mount.h>
47 #include <sys/filedesc.h>
48 #include <sys/fcntl.h>
51 #include <sys/imgact.h>
52 #include <sys/imgact_elf.h>
54 #include <sys/malloc.h>
57 #include <sys/pioctl.h>
58 #include <sys/namei.h>
59 #include <sys/resourcevar.h>
60 #include <sys/sched.h>
62 #include <sys/sf_buf.h>
63 #include <sys/syscallsubr.h>
64 #include <sys/sysent.h>
66 #include <sys/sysctl.h>
67 #include <sys/vnode.h>
70 #include <sys/ktrace.h>
74 #include <vm/vm_param.h>
76 #include <vm/vm_page.h>
77 #include <vm/vm_map.h>
78 #include <vm/vm_kern.h>
79 #include <vm/vm_extern.h>
80 #include <vm/vm_object.h>
81 #include <vm/vm_pager.h>
84 #include <sys/pmckern.h>
87 #include <machine/reg.h>
89 #include <security/audit/audit.h>
90 #include <security/mac/mac_framework.h>
93 #include <sys/dtrace_bsd.h>
94 dtrace_execexit_func_t dtrace_fasttrap_exec;
97 SDT_PROVIDER_DECLARE(proc);
98 SDT_PROBE_DEFINE(proc, kernel, , exec, exec);
99 SDT_PROBE_ARGTYPE(proc, kernel, , exec, 0, "char *");
100 SDT_PROBE_DEFINE(proc, kernel, , exec_failure, exec-failure);
101 SDT_PROBE_ARGTYPE(proc, kernel, , exec_failure, 0, "int");
102 SDT_PROBE_DEFINE(proc, kernel, , exec_success, exec-success);
103 SDT_PROBE_ARGTYPE(proc, kernel, , exec_success, 0, "char *");
105 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
107 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
108 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
109 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
110 static int do_execve(struct thread *td, struct image_args *args,
113 /* XXX This should be vm_size_t. */
114 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD,
115 NULL, 0, sysctl_kern_ps_strings, "LU", "");
117 /* XXX This should be vm_size_t. */
118 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD|
119 CTLFLAG_CAPRD, NULL, 0, sysctl_kern_usrstack, "LU", "");
121 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD,
122 NULL, 0, sysctl_kern_stackprot, "I", "");
124 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
125 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
126 &ps_arg_cache_limit, 0, "");
128 static int map_at_zero = 0;
129 TUNABLE_INT("security.bsd.map_at_zero", &map_at_zero);
130 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RW, &map_at_zero, 0,
131 "Permit processes to map an object at virtual address 0.");
134 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
141 if (req->flags & SCTL_MASK32) {
143 val = (unsigned int)p->p_sysent->sv_psstrings;
144 error = SYSCTL_OUT(req, &val, sizeof(val));
147 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
148 sizeof(p->p_sysent->sv_psstrings));
153 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
160 if (req->flags & SCTL_MASK32) {
162 val = (unsigned int)p->p_sysent->sv_usrstack;
163 error = SYSCTL_OUT(req, &val, sizeof(val));
166 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
167 sizeof(p->p_sysent->sv_usrstack));
172 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
177 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
178 sizeof(p->p_sysent->sv_stackprot)));
182 * Each of the items is a pointer to a `const struct execsw', hence the
183 * double pointer here.
185 static const struct execsw **execsw;
187 #ifndef _SYS_SYSPROTO_H_
198 struct execve_args /* {
205 struct image_args args;
207 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
208 uap->argv, uap->envv);
210 error = kern_execve(td, &args, NULL);
214 #ifndef _SYS_SYSPROTO_H_
215 struct fexecve_args {
222 sys_fexecve(struct thread *td, struct fexecve_args *uap)
225 struct image_args args;
227 error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
228 uap->argv, uap->envv);
231 error = kern_execve(td, &args, NULL);
236 #ifndef _SYS_SYSPROTO_H_
237 struct __mac_execve_args {
246 sys___mac_execve(td, uap)
248 struct __mac_execve_args /* {
257 struct image_args args;
259 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
260 uap->argv, uap->envv);
262 error = kern_execve(td, &args, uap->mac_p);
270 * XXX: kern_execve has the astonishing property of not always returning to
271 * the caller. If sufficiently bad things happen during the call to
272 * do_execve(), it can end up calling exit1(); as a result, callers must
273 * avoid doing anything which they might need to undo (e.g., allocating
277 kern_execve(td, args, mac_p)
279 struct image_args *args;
282 struct proc *p = td->td_proc;
285 AUDIT_ARG_ARGV(args->begin_argv, args->argc,
286 args->begin_envv - args->begin_argv);
287 AUDIT_ARG_ENVV(args->begin_envv, args->envc,
288 args->endp - args->begin_envv);
289 if (p->p_flag & P_HADTHREADS) {
291 if (thread_single(SINGLE_BOUNDARY)) {
293 exec_free_args(args);
294 return (ERESTART); /* Try again later. */
299 error = do_execve(td, args, mac_p);
301 if (p->p_flag & P_HADTHREADS) {
304 * If success, we upgrade to SINGLE_EXIT state to
305 * force other threads to suicide.
308 thread_single(SINGLE_EXIT);
318 * In-kernel implementation of execve(). All arguments are assumed to be
319 * userspace pointers from the passed thread.
322 do_execve(td, args, mac_p)
324 struct image_args *args;
327 struct proc *p = td->td_proc;
329 struct ucred *newcred = NULL, *oldcred;
330 struct uidinfo *euip;
331 register_t *stack_base;
333 struct image_params image_params, *imgp;
335 int (*img_first)(struct image_params *);
336 struct pargs *oldargs = NULL, *newargs = NULL;
337 struct sigacts *oldsigacts, *newsigacts;
339 struct vnode *tracevp = NULL;
340 struct ucred *tracecred = NULL;
342 struct vnode *textvp = NULL, *binvp = NULL;
343 int credential_changing;
347 struct label *interpvplabel = NULL;
351 struct pmckern_procexec pe;
353 static const char fexecv_proc_title[] = "(fexecv)";
356 imgp = &image_params;
359 * Lock the process and set the P_INEXEC flag to indicate that
360 * it should be left alone until we're done here. This is
361 * necessary to avoid race conditions - e.g. in ptrace() -
362 * that might allow a local user to illicitly obtain elevated
366 KASSERT((p->p_flag & P_INEXEC) == 0,
367 ("%s(): process already has P_INEXEC flag", __func__));
368 p->p_flag |= P_INEXEC;
372 * Initialize part of the common data
375 imgp->execlabel = NULL;
377 imgp->entry_addr = 0;
378 imgp->reloc_base = 0;
379 imgp->vmspace_destroyed = 0;
380 imgp->interpreted = 0;
382 imgp->interpreter_name = NULL;
383 imgp->auxargs = NULL;
386 imgp->firstpage = NULL;
387 imgp->ps_strings = 0;
388 imgp->auxarg_size = 0;
390 imgp->execpath = imgp->freepath = NULL;
395 imgp->pagesizeslen = 0;
396 imgp->stack_prot = 0;
399 error = mac_execve_enter(imgp, mac_p);
404 imgp->image_header = NULL;
407 * Translate the file name. namei() returns a vnode pointer
408 * in ni_vp amoung other things.
410 * XXXAUDIT: It would be desirable to also audit the name of the
411 * interpreter if this is an interpreted binary.
413 if (args->fname != NULL) {
414 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME
415 | MPSAFE | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
418 SDT_PROBE(proc, kernel, , exec, args->fname, 0, 0, 0, 0 );
421 if (args->fname != NULL) {
422 #ifdef CAPABILITY_MODE
424 * While capability mode can't reach this point via direct
425 * path arguments to execve(), we also don't allow
426 * interpreters to be used in capability mode (for now).
427 * Catch indirect lookups and return a permissions error.
429 if (IN_CAPABILITY_MODE(td)) {
438 vfslocked = NDHASGIANT(&nd);
442 AUDIT_ARG_FD(args->fd);
444 * Some might argue that CAP_READ and/or CAP_MMAP should also
445 * be required here; such arguments will be entertained.
447 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
449 error = fgetvp_exec(td, args->fd, CAP_FEXECVE, &binvp);
452 vfslocked = VFS_LOCK_GIANT(binvp->v_mount);
453 vn_lock(binvp, LK_EXCLUSIVE | LK_RETRY);
454 AUDIT_ARG_VNODE1(binvp);
459 * Check file permissions (also 'opens' file)
461 error = exec_check_permissions(imgp);
463 goto exec_fail_dealloc;
465 imgp->object = imgp->vp->v_object;
466 if (imgp->object != NULL)
467 vm_object_reference(imgp->object);
470 * Set VV_TEXT now so no one can write to the executable while we're
473 * Remember if this was set before and unset it in case this is not
474 * actually an executable image.
476 textset = VOP_IS_TEXT(imgp->vp);
477 VOP_SET_TEXT(imgp->vp);
479 error = exec_map_first_page(imgp);
481 goto exec_fail_dealloc;
483 imgp->proc->p_osrel = 0;
485 * If the current process has a special image activator it
486 * wants to try first, call it. For example, emulating shell
487 * scripts differently.
490 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
491 error = img_first(imgp);
494 * Loop through the list of image activators, calling each one.
495 * An activator returns -1 if there is no match, 0 on success,
496 * and an error otherwise.
498 for (i = 0; error == -1 && execsw[i]; ++i) {
499 if (execsw[i]->ex_imgact == NULL ||
500 execsw[i]->ex_imgact == img_first) {
503 error = (*execsw[i]->ex_imgact)(imgp);
509 VOP_UNSET_TEXT(imgp->vp);
512 goto exec_fail_dealloc;
516 * Special interpreter operation, cleanup and loop up to try to
517 * activate the interpreter.
519 if (imgp->interpreted) {
520 exec_unmap_first_page(imgp);
522 * VV_TEXT needs to be unset for scripts. There is a short
523 * period before we determine that something is a script where
524 * VV_TEXT will be set. The vnode lock is held over this
525 * entire period so nothing should illegitimately be blocked.
527 VOP_UNSET_TEXT(imgp->vp);
528 /* free name buffer and old vnode */
529 if (args->fname != NULL)
530 NDFREE(&nd, NDF_ONLY_PNBUF);
532 mac_execve_interpreter_enter(binvp, &interpvplabel);
535 VOP_CLOSE(binvp, FREAD, td->td_ucred, td);
539 vm_object_deallocate(imgp->object);
541 VFS_UNLOCK_GIANT(vfslocked);
543 /* set new name to that of the interpreter */
544 NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME | MPSAFE,
545 UIO_SYSSPACE, imgp->interpreter_name, td);
546 args->fname = imgp->interpreter_name;
551 * NB: We unlock the vnode here because it is believed that none
552 * of the sv_copyout_strings/sv_fixup operations require the vnode.
554 VOP_UNLOCK(imgp->vp, 0);
557 * Do the best to calculate the full path to the image file.
559 if (imgp->auxargs != NULL &&
560 ((args->fname != NULL && args->fname[0] == '/') ||
561 vn_fullpath(td, imgp->vp, &imgp->execpath, &imgp->freepath) != 0))
562 imgp->execpath = args->fname;
565 * Copy out strings (args and env) and initialize stack base
567 if (p->p_sysent->sv_copyout_strings)
568 stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
570 stack_base = exec_copyout_strings(imgp);
573 * If custom stack fixup routine present for this process
574 * let it do the stack setup.
575 * Else stuff argument count as first item on stack
577 if (p->p_sysent->sv_fixup != NULL)
578 (*p->p_sysent->sv_fixup)(&stack_base, imgp);
580 suword(--stack_base, imgp->args->argc);
583 * For security and other reasons, the file descriptor table cannot
584 * be shared after an exec.
589 * Malloc things before we need locks.
592 euip = uifind(attr.va_uid);
593 i = imgp->args->begin_envv - imgp->args->begin_argv;
594 /* Cache arguments if they fit inside our allowance */
595 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
596 newargs = pargs_alloc(i);
597 bcopy(imgp->args->begin_argv, newargs->ar_args, i);
600 /* close files on exec */
602 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
604 /* Get a reference to the vnode prior to locking the proc */
608 * For security and other reasons, signal handlers cannot
609 * be shared after an exec. The new process gets a copy of the old
610 * handlers. In execsigs(), the new process will have its signals
614 oldcred = crcopysafe(p, newcred);
615 if (sigacts_shared(p->p_sigacts)) {
616 oldsigacts = p->p_sigacts;
618 newsigacts = sigacts_alloc();
619 sigacts_copy(newsigacts, oldsigacts);
621 p->p_sigacts = newsigacts;
628 /* reset caught signals */
631 /* name this process - nameiexec(p, ndp) */
632 bzero(p->p_comm, sizeof(p->p_comm));
634 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
635 min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
636 else if (vn_commname(binvp, p->p_comm, sizeof(p->p_comm)) != 0)
637 bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
638 bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
640 sched_clear_tdname(td);
644 * mark as execed, wakeup the process that vforked (if any) and tell
645 * it that it now has its own resources back
648 if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
649 p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
650 cv_broadcast(&p->p_pwait);
654 * Implement image setuid/setgid.
656 * Don't honor setuid/setgid if the filesystem prohibits it or if
657 * the process is being traced.
659 * We disable setuid/setgid/etc in compatibility mode on the basis
660 * that most setugid applications are not written with that
661 * environment in mind, and will therefore almost certainly operate
662 * incorrectly. In principle there's no reason that setugid
663 * applications might not be useful in capability mode, so we may want
664 * to reconsider this conservative design choice in the future.
666 * XXXMAC: For the time being, use NOSUID to also prohibit
667 * transitions on the file system.
669 credential_changing = 0;
670 credential_changing |= (attr.va_mode & S_ISUID) && oldcred->cr_uid !=
672 credential_changing |= (attr.va_mode & S_ISGID) && oldcred->cr_gid !=
675 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
676 interpvplabel, imgp);
677 credential_changing |= will_transition;
680 if (credential_changing &&
681 #ifdef CAPABILITY_MODE
682 ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
684 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
685 (p->p_flag & P_TRACED) == 0) {
687 * Turn off syscall tracing for set-id programs, except for
688 * root. Record any set-id flags first to make sure that
689 * we do not regain any tracing during a possible block.
694 if (p->p_tracecred != NULL &&
695 priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED, 0))
696 ktrprocexec(p, &tracecred, &tracevp);
699 * Close any file descriptors 0..2 that reference procfs,
700 * then make sure file descriptors 0..2 are in use.
702 * setugidsafety() may call closef() and then pfind()
703 * which may grab the process lock.
704 * fdcheckstd() may call falloc() which may block to
705 * allocate memory, so temporarily drop the process lock.
708 VOP_UNLOCK(imgp->vp, 0);
710 error = fdcheckstd(td);
711 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
716 * Set the new credentials.
718 if (attr.va_mode & S_ISUID)
719 change_euid(newcred, euip);
720 if (attr.va_mode & S_ISGID)
721 change_egid(newcred, attr.va_gid);
723 if (will_transition) {
724 mac_vnode_execve_transition(oldcred, newcred, imgp->vp,
725 interpvplabel, imgp);
729 * Implement correct POSIX saved-id behavior.
731 * XXXMAC: Note that the current logic will save the
732 * uid and gid if a MAC domain transition occurs, even
733 * though maybe it shouldn't.
735 change_svuid(newcred, newcred->cr_uid);
736 change_svgid(newcred, newcred->cr_gid);
737 p->p_ucred = newcred;
740 if (oldcred->cr_uid == oldcred->cr_ruid &&
741 oldcred->cr_gid == oldcred->cr_rgid)
742 p->p_flag &= ~P_SUGID;
744 * Implement correct POSIX saved-id behavior.
746 * XXX: It's not clear that the existing behavior is
747 * POSIX-compliant. A number of sources indicate that the
748 * saved uid/gid should only be updated if the new ruid is
749 * not equal to the old ruid, or the new euid is not equal
750 * to the old euid and the new euid is not equal to the old
751 * ruid. The FreeBSD code always updates the saved uid/gid.
752 * Also, this code uses the new (replaced) euid and egid as
753 * the source, which may or may not be the right ones to use.
755 if (oldcred->cr_svuid != oldcred->cr_uid ||
756 oldcred->cr_svgid != oldcred->cr_gid) {
757 change_svuid(newcred, newcred->cr_uid);
758 change_svgid(newcred, newcred->cr_gid);
759 p->p_ucred = newcred;
765 * Store the vp for use in procfs. This vnode was referenced prior
766 * to locking the proc lock.
768 textvp = p->p_textvp;
773 * Tell the DTrace fasttrap provider about the exec if it
774 * has declared an interest.
776 if (dtrace_fasttrap_exec)
777 dtrace_fasttrap_exec(p);
781 * Notify others that we exec'd, and clear the P_INEXEC flag
782 * as we're now a bona fide freshly-execed process.
784 KNOTE_LOCKED(&p->p_klist, NOTE_EXEC);
785 p->p_flag &= ~P_INEXEC;
787 /* clear "fork but no exec" flag, as we _are_ execing */
788 p->p_acflag &= ~AFORK;
791 * Free any previous argument cache and replace it with
792 * the new argument cache, if any.
800 * Check if system-wide sampling is in effect or if the
801 * current process is using PMCs. If so, do exec() time
802 * processing. This processing needs to happen AFTER the
803 * P_INEXEC flag is cleared.
805 * The proc lock needs to be released before taking the PMC
808 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
810 VOP_UNLOCK(imgp->vp, 0);
811 pe.pm_credentialschanged = credential_changing;
812 pe.pm_entryaddr = imgp->entry_addr;
814 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
815 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
818 #else /* !HWPMC_HOOKS */
822 /* Set values passed into the program in registers. */
823 if (p->p_sysent->sv_setregs)
824 (*p->p_sysent->sv_setregs)(td, imgp,
825 (u_long)(uintptr_t)stack_base);
827 exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base);
829 vfs_mark_atime(imgp->vp, td->td_ucred);
831 SDT_PROBE(proc, kernel, , exec_success, args->fname, 0, 0, 0, 0);
835 * Free any resources malloc'd earlier that we didn't use.
842 VOP_UNLOCK(imgp->vp, 0);
845 * Handle deferred decrement of ref counts.
847 if (textvp != NULL) {
850 tvfslocked = VFS_LOCK_GIANT(textvp->v_mount);
852 VFS_UNLOCK_GIANT(tvfslocked);
854 if (binvp && error != 0)
857 if (tracevp != NULL) {
860 tvfslocked = VFS_LOCK_GIANT(tracevp->v_mount);
862 VFS_UNLOCK_GIANT(tvfslocked);
864 if (tracecred != NULL)
867 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
870 if (oldsigacts != NULL)
871 sigacts_free(oldsigacts);
876 * free various allocated resources
878 if (imgp->firstpage != NULL)
879 exec_unmap_first_page(imgp);
881 if (imgp->vp != NULL) {
883 NDFREE(&nd, NDF_ONLY_PNBUF);
885 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
889 if (imgp->object != NULL)
890 vm_object_deallocate(imgp->object);
892 free(imgp->freepath, M_TEMP);
896 td->td_dbgflags |= TDB_EXEC;
900 * Stop the process here if its stop event mask has
901 * the S_EXEC bit set.
903 STOPEVENT(p, S_EXEC, 0);
908 /* we're done here, clear P_INEXEC */
910 p->p_flag &= ~P_INEXEC;
913 SDT_PROBE(proc, kernel, , exec_failure, error, 0, 0, 0, 0);
917 mac_execve_exit(imgp);
918 mac_execve_interpreter_exit(interpvplabel);
920 VFS_UNLOCK_GIANT(vfslocked);
921 exec_free_args(args);
923 if (error && imgp->vmspace_destroyed) {
924 /* sorry, no more process anymore. exit gracefully */
925 exit1(td, W_EXITCODE(0, SIGABRT));
938 exec_map_first_page(imgp)
939 struct image_params *imgp;
943 vm_page_t ma[VM_INITIAL_PAGEIN];
946 if (imgp->firstpage != NULL)
947 exec_unmap_first_page(imgp);
949 object = imgp->vp->v_object;
952 VM_OBJECT_LOCK(object);
953 #if VM_NRESERVLEVEL > 0
954 if ((object->flags & OBJ_COLORED) == 0) {
955 object->flags |= OBJ_COLORED;
956 object->pg_color = 0;
959 ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
960 if (ma[0]->valid != VM_PAGE_BITS_ALL) {
961 initial_pagein = VM_INITIAL_PAGEIN;
962 if (initial_pagein > object->size)
963 initial_pagein = object->size;
964 for (i = 1; i < initial_pagein; i++) {
965 if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) {
968 if ((ma[i]->oflags & VPO_BUSY) || ma[i]->busy)
972 ma[i] = vm_page_alloc(object, i,
973 VM_ALLOC_NORMAL | VM_ALLOC_IFNOTCACHED);
979 rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
980 ma[0] = vm_page_lookup(object, 0);
981 if ((rv != VM_PAGER_OK) || (ma[0] == NULL)) {
985 vm_page_unlock(ma[0]);
987 VM_OBJECT_UNLOCK(object);
993 vm_page_unlock(ma[0]);
994 vm_page_wakeup(ma[0]);
995 VM_OBJECT_UNLOCK(object);
997 imgp->firstpage = sf_buf_alloc(ma[0], 0);
998 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
1004 exec_unmap_first_page(imgp)
1005 struct image_params *imgp;
1009 if (imgp->firstpage != NULL) {
1010 m = sf_buf_page(imgp->firstpage);
1011 sf_buf_free(imgp->firstpage);
1012 imgp->firstpage = NULL;
1020 * Destroy old address space, and allocate a new stack
1021 * The new stack is only SGROWSIZ large because it is grown
1022 * automatically in trap.c.
1025 exec_new_vmspace(imgp, sv)
1026 struct image_params *imgp;
1027 struct sysentvec *sv;
1030 struct proc *p = imgp->proc;
1031 struct vmspace *vmspace = p->p_vmspace;
1033 vm_offset_t sv_minuser, stack_addr;
1037 imgp->vmspace_destroyed = 1;
1040 /* May be called with Giant held */
1041 EVENTHANDLER_INVOKE(process_exec, p, imgp);
1044 * Blow away entire process VM, if address space not shared,
1045 * otherwise, create a new VM space so that other threads are
1048 map = &vmspace->vm_map;
1050 sv_minuser = sv->sv_minuser;
1052 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1053 if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser &&
1054 vm_map_max(map) == sv->sv_maxuser) {
1056 pmap_remove_pages(vmspace_pmap(vmspace));
1057 vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1059 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1062 vmspace = p->p_vmspace;
1063 map = &vmspace->vm_map;
1066 /* Map a shared page */
1067 obj = sv->sv_shared_page_obj;
1069 vm_object_reference(obj);
1070 error = vm_map_fixed(map, obj, 0,
1071 sv->sv_shared_page_base, sv->sv_shared_page_len,
1072 VM_PROT_READ | VM_PROT_EXECUTE,
1073 VM_PROT_READ | VM_PROT_EXECUTE,
1074 MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1076 vm_object_deallocate(obj);
1081 /* Allocate a new stack */
1082 if (sv->sv_maxssiz != NULL)
1083 ssiz = *sv->sv_maxssiz;
1086 stack_addr = sv->sv_usrstack - ssiz;
1087 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1088 obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1090 VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1095 /* Allocate a new register stack */
1096 stack_addr = IA64_BACKINGSTORE;
1097 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1098 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_UP);
1103 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
1104 * VM_STACK case, but they are still used to monitor the size of the
1105 * process stack so we can check the stack rlimit.
1107 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1108 vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - ssiz;
1114 * Copy out argument and environment strings from the old process address
1115 * space into the temporary string buffer.
1118 exec_copyin_args(struct image_args *args, char *fname,
1119 enum uio_seg segflg, char **argv, char **envv)
1125 bzero(args, sizeof(*args));
1130 * Allocate demand-paged memory for the file name, argument, and
1131 * environment strings.
1133 error = exec_alloc_args(args);
1138 * Copy the file name.
1140 if (fname != NULL) {
1141 args->fname = args->buf;
1142 error = (segflg == UIO_SYSSPACE) ?
1143 copystr(fname, args->fname, PATH_MAX, &length) :
1144 copyinstr(fname, args->fname, PATH_MAX, &length);
1150 args->begin_argv = args->buf + length;
1151 args->endp = args->begin_argv;
1152 args->stringspace = ARG_MAX;
1155 * extract arguments first
1157 while ((argp = (caddr_t) (intptr_t) fuword(argv++))) {
1158 if (argp == (caddr_t) -1) {
1162 if ((error = copyinstr(argp, args->endp,
1163 args->stringspace, &length))) {
1164 if (error == ENAMETOOLONG)
1168 args->stringspace -= length;
1169 args->endp += length;
1173 args->begin_envv = args->endp;
1176 * extract environment strings
1179 while ((envp = (caddr_t)(intptr_t)fuword(envv++))) {
1180 if (envp == (caddr_t)-1) {
1184 if ((error = copyinstr(envp, args->endp,
1185 args->stringspace, &length))) {
1186 if (error == ENAMETOOLONG)
1190 args->stringspace -= length;
1191 args->endp += length;
1199 exec_free_args(args);
1204 * Allocate temporary demand-paged, zero-filled memory for the file name,
1205 * argument, and environment strings. Returns zero if the allocation succeeds
1206 * and ENOMEM otherwise.
1209 exec_alloc_args(struct image_args *args)
1212 args->buf = (char *)kmem_alloc_wait(exec_map, PATH_MAX + ARG_MAX);
1213 return (args->buf != NULL ? 0 : ENOMEM);
1217 exec_free_args(struct image_args *args)
1220 if (args->buf != NULL) {
1221 kmem_free_wakeup(exec_map, (vm_offset_t)args->buf,
1222 PATH_MAX + ARG_MAX);
1225 if (args->fname_buf != NULL) {
1226 free(args->fname_buf, M_TEMP);
1227 args->fname_buf = NULL;
1232 * Copy strings out to the new process address space, constructing new arg
1233 * and env vector tables. Return a pointer to the base so that it can be used
1234 * as the initial stack pointer.
1237 exec_copyout_strings(imgp)
1238 struct image_params *imgp;
1242 char *stringp, *destp;
1243 register_t *stack_base;
1244 struct ps_strings *arginfo;
1246 size_t execpath_len;
1247 int szsigcode, szps;
1248 char canary[sizeof(long) * 8];
1250 szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1252 * Calculate string base and vector table pointers.
1253 * Also deal with signal trampoline code for this exec type.
1255 if (imgp->execpath != NULL && imgp->auxargs != NULL)
1256 execpath_len = strlen(imgp->execpath) + 1;
1261 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1262 if (p->p_sysent->sv_sigcode_base == 0) {
1263 if (p->p_sysent->sv_szsigcode != NULL)
1264 szsigcode = *(p->p_sysent->sv_szsigcode);
1266 destp = (caddr_t)arginfo - szsigcode - SPARE_USRSPACE -
1267 roundup(execpath_len, sizeof(char *)) -
1268 roundup(sizeof(canary), sizeof(char *)) -
1269 roundup(szps, sizeof(char *)) -
1270 roundup((ARG_MAX - imgp->args->stringspace), sizeof(char *));
1276 copyout(p->p_sysent->sv_sigcode, ((caddr_t)arginfo -
1277 szsigcode), szsigcode);
1280 * Copy the image path for the rtld.
1282 if (execpath_len != 0) {
1283 imgp->execpathp = (uintptr_t)arginfo - szsigcode - execpath_len;
1284 copyout(imgp->execpath, (void *)imgp->execpathp,
1289 * Prepare the canary for SSP.
1291 arc4rand(canary, sizeof(canary), 0);
1292 imgp->canary = (uintptr_t)arginfo - szsigcode - execpath_len -
1294 copyout(canary, (void *)imgp->canary, sizeof(canary));
1295 imgp->canarylen = sizeof(canary);
1298 * Prepare the pagesizes array.
1300 imgp->pagesizes = (uintptr_t)arginfo - szsigcode - execpath_len -
1301 roundup(sizeof(canary), sizeof(char *)) - szps;
1302 copyout(pagesizes, (void *)imgp->pagesizes, szps);
1303 imgp->pagesizeslen = szps;
1306 * If we have a valid auxargs ptr, prepare some room
1309 if (imgp->auxargs) {
1311 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1312 * lower compatibility.
1314 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1317 * The '+ 2' is for the null pointers at the end of each of
1318 * the arg and env vector sets,and imgp->auxarg_size is room
1319 * for argument of Runtime loader.
1321 vectp = (char **)(destp - (imgp->args->argc +
1322 imgp->args->envc + 2 + imgp->auxarg_size)
1326 * The '+ 2' is for the null pointers at the end of each of
1327 * the arg and env vector sets
1329 vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc + 2) *
1334 * vectp also becomes our initial stack base
1336 stack_base = (register_t *)vectp;
1338 stringp = imgp->args->begin_argv;
1339 argc = imgp->args->argc;
1340 envc = imgp->args->envc;
1343 * Copy out strings - arguments and environment.
1345 copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
1348 * Fill in "ps_strings" struct for ps, w, etc.
1350 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1351 suword32(&arginfo->ps_nargvstr, argc);
1354 * Fill in argument portion of vector table.
1356 for (; argc > 0; --argc) {
1357 suword(vectp++, (long)(intptr_t)destp);
1358 while (*stringp++ != 0)
1363 /* a null vector table pointer separates the argp's from the envp's */
1366 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1367 suword32(&arginfo->ps_nenvstr, envc);
1370 * Fill in environment portion of vector table.
1372 for (; envc > 0; --envc) {
1373 suword(vectp++, (long)(intptr_t)destp);
1374 while (*stringp++ != 0)
1379 /* end of vector table is a null pointer */
1382 return (stack_base);
1386 * Check permissions of file to execute.
1387 * Called with imgp->vp locked.
1388 * Return 0 for success or error code on failure.
1391 exec_check_permissions(imgp)
1392 struct image_params *imgp;
1394 struct vnode *vp = imgp->vp;
1395 struct vattr *attr = imgp->attr;
1397 int error, writecount;
1401 /* Get file attributes */
1402 error = VOP_GETATTR(vp, attr, td->td_ucred);
1407 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1413 * 1) Check if file execution is disabled for the filesystem that
1414 * this file resides on.
1415 * 2) Ensure that at least one execute bit is on. Otherwise, a
1416 * privileged user will always succeed, and we don't want this
1417 * to happen unless the file really is executable.
1418 * 3) Ensure that the file is a regular file.
1420 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1421 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1422 (attr->va_type != VREG))
1426 * Zero length files can't be exec'd
1428 if (attr->va_size == 0)
1432 * Check for execute permission to file based on current credentials.
1434 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1439 * Check number of open-for-writes on the file and deny execution
1442 error = VOP_GET_WRITECOUNT(vp, &writecount);
1445 if (writecount != 0)
1449 * Call filesystem specific open routine (which does nothing in the
1452 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1459 * Exec handler registration
1462 exec_register(execsw_arg)
1463 const struct execsw *execsw_arg;
1465 const struct execsw **es, **xs, **newexecsw;
1466 int count = 2; /* New slot and trailing NULL */
1469 for (es = execsw; *es; es++)
1471 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1472 if (newexecsw == NULL)
1476 for (es = execsw; *es; es++)
1481 free(execsw, M_TEMP);
1487 exec_unregister(execsw_arg)
1488 const struct execsw *execsw_arg;
1490 const struct execsw **es, **xs, **newexecsw;
1494 panic("unregister with no handlers left?\n");
1496 for (es = execsw; *es; es++) {
1497 if (*es == execsw_arg)
1502 for (es = execsw; *es; es++)
1503 if (*es != execsw_arg)
1505 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1506 if (newexecsw == NULL)
1509 for (es = execsw; *es; es++)
1510 if (*es != execsw_arg)
1514 free(execsw, M_TEMP);