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_ktrace.h"
35 #include <sys/param.h>
36 #include <sys/capability.h>
37 #include <sys/systm.h>
38 #include <sys/capability.h>
39 #include <sys/eventhandler.h>
41 #include <sys/mutex.h>
42 #include <sys/sysproto.h>
43 #include <sys/signalvar.h>
44 #include <sys/kernel.h>
45 #include <sys/mount.h>
46 #include <sys/filedesc.h>
47 #include <sys/fcntl.h>
50 #include <sys/imgact.h>
51 #include <sys/imgact_elf.h>
53 #include <sys/malloc.h>
56 #include <sys/pioctl.h>
57 #include <sys/namei.h>
58 #include <sys/resourcevar.h>
59 #include <sys/rwlock.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_DEFINE1(proc, kernel, , exec, "char *");
99 SDT_PROBE_DEFINE1(proc, kernel, , exec__failure, "int");
100 SDT_PROBE_DEFINE1(proc, kernel, , exec__success, "char *");
102 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
104 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
105 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
106 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
107 static int do_execve(struct thread *td, struct image_args *args,
110 /* XXX This should be vm_size_t. */
111 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD,
112 NULL, 0, sysctl_kern_ps_strings, "LU", "");
114 /* XXX This should be vm_size_t. */
115 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD|
116 CTLFLAG_CAPRD, NULL, 0, sysctl_kern_usrstack, "LU", "");
118 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD,
119 NULL, 0, sysctl_kern_stackprot, "I", "");
121 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
122 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
123 &ps_arg_cache_limit, 0, "");
125 static int disallow_high_osrel;
126 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW,
127 &disallow_high_osrel, 0,
128 "Disallow execution of binaries built for higher version of the world");
130 static int map_at_zero = 0;
131 TUNABLE_INT("security.bsd.map_at_zero", &map_at_zero);
132 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RW, &map_at_zero, 0,
133 "Permit processes to map an object at virtual address 0.");
136 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
143 if (req->flags & SCTL_MASK32) {
145 val = (unsigned int)p->p_sysent->sv_psstrings;
146 error = SYSCTL_OUT(req, &val, sizeof(val));
149 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
150 sizeof(p->p_sysent->sv_psstrings));
155 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
162 if (req->flags & SCTL_MASK32) {
164 val = (unsigned int)p->p_sysent->sv_usrstack;
165 error = SYSCTL_OUT(req, &val, sizeof(val));
168 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
169 sizeof(p->p_sysent->sv_usrstack));
174 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
179 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
180 sizeof(p->p_sysent->sv_stackprot)));
184 * Each of the items is a pointer to a `const struct execsw', hence the
185 * double pointer here.
187 static const struct execsw **execsw;
189 #ifndef _SYS_SYSPROTO_H_
200 struct execve_args /* {
207 struct image_args args;
209 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
210 uap->argv, uap->envv);
212 error = kern_execve(td, &args, NULL);
216 #ifndef _SYS_SYSPROTO_H_
217 struct fexecve_args {
224 sys_fexecve(struct thread *td, struct fexecve_args *uap)
227 struct image_args args;
229 error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
230 uap->argv, uap->envv);
233 error = kern_execve(td, &args, NULL);
238 #ifndef _SYS_SYSPROTO_H_
239 struct __mac_execve_args {
248 sys___mac_execve(td, uap)
250 struct __mac_execve_args /* {
259 struct image_args args;
261 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
262 uap->argv, uap->envv);
264 error = kern_execve(td, &args, uap->mac_p);
272 * XXX: kern_execve has the astonishing property of not always returning to
273 * the caller. If sufficiently bad things happen during the call to
274 * do_execve(), it can end up calling exit1(); as a result, callers must
275 * avoid doing anything which they might need to undo (e.g., allocating
279 kern_execve(td, args, mac_p)
281 struct image_args *args;
284 struct proc *p = td->td_proc;
287 AUDIT_ARG_ARGV(args->begin_argv, args->argc,
288 args->begin_envv - args->begin_argv);
289 AUDIT_ARG_ENVV(args->begin_envv, args->envc,
290 args->endp - args->begin_envv);
291 if (p->p_flag & P_HADTHREADS) {
293 if (thread_single(SINGLE_BOUNDARY)) {
295 exec_free_args(args);
296 return (ERESTART); /* Try again later. */
301 error = do_execve(td, args, mac_p);
303 if (p->p_flag & P_HADTHREADS) {
306 * If success, we upgrade to SINGLE_EXIT state to
307 * force other threads to suicide.
310 thread_single(SINGLE_EXIT);
320 * In-kernel implementation of execve(). All arguments are assumed to be
321 * userspace pointers from the passed thread.
324 do_execve(td, args, mac_p)
326 struct image_args *args;
329 struct proc *p = td->td_proc;
331 struct ucred *newcred = NULL, *oldcred;
332 struct uidinfo *euip;
333 register_t *stack_base;
335 struct image_params image_params, *imgp;
337 int (*img_first)(struct image_params *);
338 struct pargs *oldargs = NULL, *newargs = NULL;
339 struct sigacts *oldsigacts, *newsigacts;
341 struct vnode *tracevp = NULL;
342 struct ucred *tracecred = NULL;
344 struct vnode *textvp = NULL, *binvp = NULL;
346 int credential_changing;
349 struct label *interpvplabel = NULL;
353 struct pmckern_procexec pe;
355 static const char fexecv_proc_title[] = "(fexecv)";
357 imgp = &image_params;
360 * Lock the process and set the P_INEXEC flag to indicate that
361 * it should be left alone until we're done here. This is
362 * necessary to avoid race conditions - e.g. in ptrace() -
363 * that might allow a local user to illicitly obtain elevated
367 KASSERT((p->p_flag & P_INEXEC) == 0,
368 ("%s(): process already has P_INEXEC flag", __func__));
369 p->p_flag |= P_INEXEC;
373 * Initialize part of the common data
376 imgp->execlabel = NULL;
378 imgp->entry_addr = 0;
379 imgp->reloc_base = 0;
380 imgp->vmspace_destroyed = 0;
381 imgp->interpreted = 0;
383 imgp->interpreter_name = NULL;
384 imgp->auxargs = NULL;
387 imgp->firstpage = NULL;
388 imgp->ps_strings = 0;
389 imgp->auxarg_size = 0;
391 imgp->execpath = imgp->freepath = NULL;
396 imgp->pagesizeslen = 0;
397 imgp->stack_prot = 0;
400 error = mac_execve_enter(imgp, mac_p);
405 imgp->image_header = NULL;
408 * Translate the file name. namei() returns a vnode pointer
409 * in ni_vp amoung other things.
411 * XXXAUDIT: It would be desirable to also audit the name of the
412 * interpreter if this is an interpreted binary.
414 if (args->fname != NULL) {
415 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME
416 | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
419 SDT_PROBE(proc, kernel, , exec, args->fname, 0, 0, 0, 0 );
422 if (args->fname != NULL) {
423 #ifdef CAPABILITY_MODE
425 * While capability mode can't reach this point via direct
426 * path arguments to execve(), we also don't allow
427 * interpreters to be used in capability mode (for now).
428 * Catch indirect lookups and return a permissions error.
430 if (IN_CAPABILITY_MODE(td)) {
442 AUDIT_ARG_FD(args->fd);
444 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
446 error = fgetvp_exec(td, args->fd,
447 cap_rights_init(&rights, CAP_FEXECVE), &binvp);
450 vn_lock(binvp, LK_EXCLUSIVE | LK_RETRY);
451 AUDIT_ARG_VNODE1(binvp);
456 * Check file permissions (also 'opens' file)
458 error = exec_check_permissions(imgp);
460 goto exec_fail_dealloc;
462 imgp->object = imgp->vp->v_object;
463 if (imgp->object != NULL)
464 vm_object_reference(imgp->object);
467 * Set VV_TEXT now so no one can write to the executable while we're
470 * Remember if this was set before and unset it in case this is not
471 * actually an executable image.
473 textset = VOP_IS_TEXT(imgp->vp);
474 VOP_SET_TEXT(imgp->vp);
476 error = exec_map_first_page(imgp);
478 goto exec_fail_dealloc;
480 imgp->proc->p_osrel = 0;
482 * If the current process has a special image activator it
483 * wants to try first, call it. For example, emulating shell
484 * scripts differently.
487 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
488 error = img_first(imgp);
491 * Loop through the list of image activators, calling each one.
492 * An activator returns -1 if there is no match, 0 on success,
493 * and an error otherwise.
495 for (i = 0; error == -1 && execsw[i]; ++i) {
496 if (execsw[i]->ex_imgact == NULL ||
497 execsw[i]->ex_imgact == img_first) {
500 error = (*execsw[i]->ex_imgact)(imgp);
506 VOP_UNSET_TEXT(imgp->vp);
509 goto exec_fail_dealloc;
513 * Special interpreter operation, cleanup and loop up to try to
514 * activate the interpreter.
516 if (imgp->interpreted) {
517 exec_unmap_first_page(imgp);
519 * VV_TEXT needs to be unset for scripts. There is a short
520 * period before we determine that something is a script where
521 * VV_TEXT will be set. The vnode lock is held over this
522 * entire period so nothing should illegitimately be blocked.
524 VOP_UNSET_TEXT(imgp->vp);
525 /* free name buffer and old vnode */
526 if (args->fname != NULL)
527 NDFREE(&nd, NDF_ONLY_PNBUF);
529 mac_execve_interpreter_enter(binvp, &interpvplabel);
532 VOP_CLOSE(binvp, FREAD, td->td_ucred, td);
536 vm_object_deallocate(imgp->object);
538 /* set new name to that of the interpreter */
539 NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
540 UIO_SYSSPACE, imgp->interpreter_name, td);
541 args->fname = imgp->interpreter_name;
546 * NB: We unlock the vnode here because it is believed that none
547 * of the sv_copyout_strings/sv_fixup operations require the vnode.
549 VOP_UNLOCK(imgp->vp, 0);
552 * Do the best to calculate the full path to the image file.
554 if (imgp->auxargs != NULL &&
555 ((args->fname != NULL && args->fname[0] == '/') ||
556 vn_fullpath(td, imgp->vp, &imgp->execpath, &imgp->freepath) != 0))
557 imgp->execpath = args->fname;
559 if (disallow_high_osrel &&
560 P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
562 uprintf("Osrel %d for image %s too high\n", p->p_osrel,
563 imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
564 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
565 goto exec_fail_dealloc;
569 * Copy out strings (args and env) and initialize stack base
571 if (p->p_sysent->sv_copyout_strings)
572 stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
574 stack_base = exec_copyout_strings(imgp);
577 * If custom stack fixup routine present for this process
578 * let it do the stack setup.
579 * Else stuff argument count as first item on stack
581 if (p->p_sysent->sv_fixup != NULL)
582 (*p->p_sysent->sv_fixup)(&stack_base, imgp);
584 suword(--stack_base, imgp->args->argc);
587 * For security and other reasons, the file descriptor table cannot
588 * be shared after an exec.
593 * Malloc things before we need locks.
596 euip = uifind(attr.va_uid);
597 i = imgp->args->begin_envv - imgp->args->begin_argv;
598 /* Cache arguments if they fit inside our allowance */
599 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
600 newargs = pargs_alloc(i);
601 bcopy(imgp->args->begin_argv, newargs->ar_args, i);
604 /* close files on exec */
606 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
608 /* Get a reference to the vnode prior to locking the proc */
612 * For security and other reasons, signal handlers cannot
613 * be shared after an exec. The new process gets a copy of the old
614 * handlers. In execsigs(), the new process will have its signals
618 oldcred = crcopysafe(p, newcred);
619 if (sigacts_shared(p->p_sigacts)) {
620 oldsigacts = p->p_sigacts;
622 newsigacts = sigacts_alloc();
623 sigacts_copy(newsigacts, oldsigacts);
625 p->p_sigacts = newsigacts;
632 /* reset caught signals */
635 /* name this process - nameiexec(p, ndp) */
636 bzero(p->p_comm, sizeof(p->p_comm));
638 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
639 min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
640 else if (vn_commname(binvp, p->p_comm, sizeof(p->p_comm)) != 0)
641 bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
642 bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
644 sched_clear_tdname(td);
648 * mark as execed, wakeup the process that vforked (if any) and tell
649 * it that it now has its own resources back
652 if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
653 p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
654 cv_broadcast(&p->p_pwait);
658 * Implement image setuid/setgid.
660 * Don't honor setuid/setgid if the filesystem prohibits it or if
661 * the process is being traced.
663 * We disable setuid/setgid/etc in compatibility mode on the basis
664 * that most setugid applications are not written with that
665 * environment in mind, and will therefore almost certainly operate
666 * incorrectly. In principle there's no reason that setugid
667 * applications might not be useful in capability mode, so we may want
668 * to reconsider this conservative design choice in the future.
670 * XXXMAC: For the time being, use NOSUID to also prohibit
671 * transitions on the file system.
673 credential_changing = 0;
674 credential_changing |= (attr.va_mode & S_ISUID) && oldcred->cr_uid !=
676 credential_changing |= (attr.va_mode & S_ISGID) && oldcred->cr_gid !=
679 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
680 interpvplabel, imgp);
681 credential_changing |= will_transition;
684 if (credential_changing &&
685 #ifdef CAPABILITY_MODE
686 ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
688 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
689 (p->p_flag & P_TRACED) == 0) {
691 * Turn off syscall tracing for set-id programs, except for
692 * root. Record any set-id flags first to make sure that
693 * we do not regain any tracing during a possible block.
698 if (p->p_tracecred != NULL &&
699 priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED, 0))
700 ktrprocexec(p, &tracecred, &tracevp);
703 * Close any file descriptors 0..2 that reference procfs,
704 * then make sure file descriptors 0..2 are in use.
706 * setugidsafety() may call closef() and then pfind()
707 * which may grab the process lock.
708 * fdcheckstd() may call falloc() which may block to
709 * allocate memory, so temporarily drop the process lock.
712 VOP_UNLOCK(imgp->vp, 0);
714 error = fdcheckstd(td);
715 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
720 * Set the new credentials.
722 if (attr.va_mode & S_ISUID)
723 change_euid(newcred, euip);
724 if (attr.va_mode & S_ISGID)
725 change_egid(newcred, attr.va_gid);
727 if (will_transition) {
728 mac_vnode_execve_transition(oldcred, newcred, imgp->vp,
729 interpvplabel, imgp);
733 * Implement correct POSIX saved-id behavior.
735 * XXXMAC: Note that the current logic will save the
736 * uid and gid if a MAC domain transition occurs, even
737 * though maybe it shouldn't.
739 change_svuid(newcred, newcred->cr_uid);
740 change_svgid(newcred, newcred->cr_gid);
741 p->p_ucred = newcred;
744 if (oldcred->cr_uid == oldcred->cr_ruid &&
745 oldcred->cr_gid == oldcred->cr_rgid)
746 p->p_flag &= ~P_SUGID;
748 * Implement correct POSIX saved-id behavior.
750 * XXX: It's not clear that the existing behavior is
751 * POSIX-compliant. A number of sources indicate that the
752 * saved uid/gid should only be updated if the new ruid is
753 * not equal to the old ruid, or the new euid is not equal
754 * to the old euid and the new euid is not equal to the old
755 * ruid. The FreeBSD code always updates the saved uid/gid.
756 * Also, this code uses the new (replaced) euid and egid as
757 * the source, which may or may not be the right ones to use.
759 if (oldcred->cr_svuid != oldcred->cr_uid ||
760 oldcred->cr_svgid != oldcred->cr_gid) {
761 change_svuid(newcred, newcred->cr_uid);
762 change_svgid(newcred, newcred->cr_gid);
763 p->p_ucred = newcred;
769 * Store the vp for use in procfs. This vnode was referenced prior
770 * to locking the proc lock.
772 textvp = p->p_textvp;
777 * Tell the DTrace fasttrap provider about the exec if it
778 * has declared an interest.
780 if (dtrace_fasttrap_exec)
781 dtrace_fasttrap_exec(p);
785 * Notify others that we exec'd, and clear the P_INEXEC flag
786 * as we're now a bona fide freshly-execed process.
788 KNOTE_LOCKED(&p->p_klist, NOTE_EXEC);
789 p->p_flag &= ~P_INEXEC;
791 /* clear "fork but no exec" flag, as we _are_ execing */
792 p->p_acflag &= ~AFORK;
795 * Free any previous argument cache and replace it with
796 * the new argument cache, if any.
804 * Check if system-wide sampling is in effect or if the
805 * current process is using PMCs. If so, do exec() time
806 * processing. This processing needs to happen AFTER the
807 * P_INEXEC flag is cleared.
809 * The proc lock needs to be released before taking the PMC
812 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
814 VOP_UNLOCK(imgp->vp, 0);
815 pe.pm_credentialschanged = credential_changing;
816 pe.pm_entryaddr = imgp->entry_addr;
818 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
819 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
822 #else /* !HWPMC_HOOKS */
826 /* Set values passed into the program in registers. */
827 if (p->p_sysent->sv_setregs)
828 (*p->p_sysent->sv_setregs)(td, imgp,
829 (u_long)(uintptr_t)stack_base);
831 exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base);
833 vfs_mark_atime(imgp->vp, td->td_ucred);
835 SDT_PROBE(proc, kernel, , exec__success, args->fname, 0, 0, 0, 0);
839 * Free any resources malloc'd earlier that we didn't use.
846 VOP_UNLOCK(imgp->vp, 0);
849 * Handle deferred decrement of ref counts.
853 if (binvp && error != 0)
858 if (tracecred != NULL)
861 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
864 if (oldsigacts != NULL)
865 sigacts_free(oldsigacts);
870 * free various allocated resources
872 if (imgp->firstpage != NULL)
873 exec_unmap_first_page(imgp);
875 if (imgp->vp != NULL) {
877 NDFREE(&nd, NDF_ONLY_PNBUF);
879 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
883 if (imgp->object != NULL)
884 vm_object_deallocate(imgp->object);
886 free(imgp->freepath, M_TEMP);
890 td->td_dbgflags |= TDB_EXEC;
894 * Stop the process here if its stop event mask has
895 * the S_EXEC bit set.
897 STOPEVENT(p, S_EXEC, 0);
902 /* we're done here, clear P_INEXEC */
904 p->p_flag &= ~P_INEXEC;
907 SDT_PROBE(proc, kernel, , exec__failure, error, 0, 0, 0, 0);
911 mac_execve_exit(imgp);
912 mac_execve_interpreter_exit(interpvplabel);
914 exec_free_args(args);
916 if (error && imgp->vmspace_destroyed) {
917 /* sorry, no more process anymore. exit gracefully */
918 exit1(td, W_EXITCODE(0, SIGABRT));
931 exec_map_first_page(imgp)
932 struct image_params *imgp;
936 vm_page_t ma[VM_INITIAL_PAGEIN];
939 if (imgp->firstpage != NULL)
940 exec_unmap_first_page(imgp);
942 object = imgp->vp->v_object;
945 VM_OBJECT_WLOCK(object);
946 #if VM_NRESERVLEVEL > 0
947 if ((object->flags & OBJ_COLORED) == 0) {
948 object->flags |= OBJ_COLORED;
949 object->pg_color = 0;
952 ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL);
953 if (ma[0]->valid != VM_PAGE_BITS_ALL) {
954 initial_pagein = VM_INITIAL_PAGEIN;
955 if (initial_pagein > object->size)
956 initial_pagein = object->size;
957 for (i = 1; i < initial_pagein; i++) {
958 if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) {
961 if (vm_page_tryxbusy(ma[i]))
964 ma[i] = vm_page_alloc(object, i,
965 VM_ALLOC_NORMAL | VM_ALLOC_IFNOTCACHED);
971 rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
972 ma[0] = vm_page_lookup(object, 0);
973 if ((rv != VM_PAGER_OK) || (ma[0] == NULL)) {
977 vm_page_unlock(ma[0]);
979 VM_OBJECT_WUNLOCK(object);
983 vm_page_xunbusy(ma[0]);
986 vm_page_unlock(ma[0]);
987 VM_OBJECT_WUNLOCK(object);
989 imgp->firstpage = sf_buf_alloc(ma[0], 0);
990 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
996 exec_unmap_first_page(imgp)
997 struct image_params *imgp;
1001 if (imgp->firstpage != NULL) {
1002 m = sf_buf_page(imgp->firstpage);
1003 sf_buf_free(imgp->firstpage);
1004 imgp->firstpage = NULL;
1012 * Destroy old address space, and allocate a new stack
1013 * The new stack is only SGROWSIZ large because it is grown
1014 * automatically in trap.c.
1017 exec_new_vmspace(imgp, sv)
1018 struct image_params *imgp;
1019 struct sysentvec *sv;
1022 struct proc *p = imgp->proc;
1023 struct vmspace *vmspace = p->p_vmspace;
1025 vm_offset_t sv_minuser, stack_addr;
1029 imgp->vmspace_destroyed = 1;
1032 /* May be called with Giant held */
1033 EVENTHANDLER_INVOKE(process_exec, p, imgp);
1036 * Blow away entire process VM, if address space not shared,
1037 * otherwise, create a new VM space so that other threads are
1040 map = &vmspace->vm_map;
1042 sv_minuser = sv->sv_minuser;
1044 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1045 if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser &&
1046 vm_map_max(map) == sv->sv_maxuser) {
1048 pmap_remove_pages(vmspace_pmap(vmspace));
1049 vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1051 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1054 vmspace = p->p_vmspace;
1055 map = &vmspace->vm_map;
1058 /* Map a shared page */
1059 obj = sv->sv_shared_page_obj;
1061 vm_object_reference(obj);
1062 error = vm_map_fixed(map, obj, 0,
1063 sv->sv_shared_page_base, sv->sv_shared_page_len,
1064 VM_PROT_READ | VM_PROT_EXECUTE,
1065 VM_PROT_READ | VM_PROT_EXECUTE,
1066 MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1068 vm_object_deallocate(obj);
1073 /* Allocate a new stack */
1074 if (sv->sv_maxssiz != NULL)
1075 ssiz = *sv->sv_maxssiz;
1078 stack_addr = sv->sv_usrstack - ssiz;
1079 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1080 obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1082 VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1087 /* Allocate a new register stack */
1088 stack_addr = IA64_BACKINGSTORE;
1089 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1090 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_UP);
1095 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
1096 * VM_STACK case, but they are still used to monitor the size of the
1097 * process stack so we can check the stack rlimit.
1099 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1100 vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - ssiz;
1106 * Copy out argument and environment strings from the old process address
1107 * space into the temporary string buffer.
1110 exec_copyin_args(struct image_args *args, char *fname,
1111 enum uio_seg segflg, char **argv, char **envv)
1117 bzero(args, sizeof(*args));
1122 * Allocate demand-paged memory for the file name, argument, and
1123 * environment strings.
1125 error = exec_alloc_args(args);
1130 * Copy the file name.
1132 if (fname != NULL) {
1133 args->fname = args->buf;
1134 error = (segflg == UIO_SYSSPACE) ?
1135 copystr(fname, args->fname, PATH_MAX, &length) :
1136 copyinstr(fname, args->fname, PATH_MAX, &length);
1142 args->begin_argv = args->buf + length;
1143 args->endp = args->begin_argv;
1144 args->stringspace = ARG_MAX;
1147 * extract arguments first
1149 while ((argp = (caddr_t) (intptr_t) fuword(argv++))) {
1150 if (argp == (caddr_t) -1) {
1154 if ((error = copyinstr(argp, args->endp,
1155 args->stringspace, &length))) {
1156 if (error == ENAMETOOLONG)
1160 args->stringspace -= length;
1161 args->endp += length;
1165 args->begin_envv = args->endp;
1168 * extract environment strings
1171 while ((envp = (caddr_t)(intptr_t)fuword(envv++))) {
1172 if (envp == (caddr_t)-1) {
1176 if ((error = copyinstr(envp, args->endp,
1177 args->stringspace, &length))) {
1178 if (error == ENAMETOOLONG)
1182 args->stringspace -= length;
1183 args->endp += length;
1191 exec_free_args(args);
1196 * Allocate temporary demand-paged, zero-filled memory for the file name,
1197 * argument, and environment strings. Returns zero if the allocation succeeds
1198 * and ENOMEM otherwise.
1201 exec_alloc_args(struct image_args *args)
1204 args->buf = (char *)kmap_alloc_wait(exec_map, PATH_MAX + ARG_MAX);
1205 return (args->buf != NULL ? 0 : ENOMEM);
1209 exec_free_args(struct image_args *args)
1212 if (args->buf != NULL) {
1213 kmap_free_wakeup(exec_map, (vm_offset_t)args->buf,
1214 PATH_MAX + ARG_MAX);
1217 if (args->fname_buf != NULL) {
1218 free(args->fname_buf, M_TEMP);
1219 args->fname_buf = NULL;
1224 * Copy strings out to the new process address space, constructing new arg
1225 * and env vector tables. Return a pointer to the base so that it can be used
1226 * as the initial stack pointer.
1229 exec_copyout_strings(imgp)
1230 struct image_params *imgp;
1234 char *stringp, *destp;
1235 register_t *stack_base;
1236 struct ps_strings *arginfo;
1238 size_t execpath_len;
1239 int szsigcode, szps;
1240 char canary[sizeof(long) * 8];
1242 szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1244 * Calculate string base and vector table pointers.
1245 * Also deal with signal trampoline code for this exec type.
1247 if (imgp->execpath != NULL && imgp->auxargs != NULL)
1248 execpath_len = strlen(imgp->execpath) + 1;
1253 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1254 if (p->p_sysent->sv_sigcode_base == 0) {
1255 if (p->p_sysent->sv_szsigcode != NULL)
1256 szsigcode = *(p->p_sysent->sv_szsigcode);
1258 destp = (caddr_t)arginfo - szsigcode - SPARE_USRSPACE -
1259 roundup(execpath_len, sizeof(char *)) -
1260 roundup(sizeof(canary), sizeof(char *)) -
1261 roundup(szps, sizeof(char *)) -
1262 roundup((ARG_MAX - imgp->args->stringspace), sizeof(char *));
1268 copyout(p->p_sysent->sv_sigcode, ((caddr_t)arginfo -
1269 szsigcode), szsigcode);
1272 * Copy the image path for the rtld.
1274 if (execpath_len != 0) {
1275 imgp->execpathp = (uintptr_t)arginfo - szsigcode - execpath_len;
1276 copyout(imgp->execpath, (void *)imgp->execpathp,
1281 * Prepare the canary for SSP.
1283 arc4rand(canary, sizeof(canary), 0);
1284 imgp->canary = (uintptr_t)arginfo - szsigcode - execpath_len -
1286 copyout(canary, (void *)imgp->canary, sizeof(canary));
1287 imgp->canarylen = sizeof(canary);
1290 * Prepare the pagesizes array.
1292 imgp->pagesizes = (uintptr_t)arginfo - szsigcode - execpath_len -
1293 roundup(sizeof(canary), sizeof(char *)) - szps;
1294 copyout(pagesizes, (void *)imgp->pagesizes, szps);
1295 imgp->pagesizeslen = szps;
1298 * If we have a valid auxargs ptr, prepare some room
1301 if (imgp->auxargs) {
1303 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1304 * lower compatibility.
1306 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1309 * The '+ 2' is for the null pointers at the end of each of
1310 * the arg and env vector sets,and imgp->auxarg_size is room
1311 * for argument of Runtime loader.
1313 vectp = (char **)(destp - (imgp->args->argc +
1314 imgp->args->envc + 2 + imgp->auxarg_size)
1318 * The '+ 2' is for the null pointers at the end of each of
1319 * the arg and env vector sets
1321 vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc + 2) *
1326 * vectp also becomes our initial stack base
1328 stack_base = (register_t *)vectp;
1330 stringp = imgp->args->begin_argv;
1331 argc = imgp->args->argc;
1332 envc = imgp->args->envc;
1335 * Copy out strings - arguments and environment.
1337 copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
1340 * Fill in "ps_strings" struct for ps, w, etc.
1342 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1343 suword32(&arginfo->ps_nargvstr, argc);
1346 * Fill in argument portion of vector table.
1348 for (; argc > 0; --argc) {
1349 suword(vectp++, (long)(intptr_t)destp);
1350 while (*stringp++ != 0)
1355 /* a null vector table pointer separates the argp's from the envp's */
1358 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1359 suword32(&arginfo->ps_nenvstr, envc);
1362 * Fill in environment portion of vector table.
1364 for (; envc > 0; --envc) {
1365 suword(vectp++, (long)(intptr_t)destp);
1366 while (*stringp++ != 0)
1371 /* end of vector table is a null pointer */
1374 return (stack_base);
1378 * Check permissions of file to execute.
1379 * Called with imgp->vp locked.
1380 * Return 0 for success or error code on failure.
1383 exec_check_permissions(imgp)
1384 struct image_params *imgp;
1386 struct vnode *vp = imgp->vp;
1387 struct vattr *attr = imgp->attr;
1389 int error, writecount;
1393 /* Get file attributes */
1394 error = VOP_GETATTR(vp, attr, td->td_ucred);
1399 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1405 * 1) Check if file execution is disabled for the filesystem that
1406 * this file resides on.
1407 * 2) Ensure that at least one execute bit is on. Otherwise, a
1408 * privileged user will always succeed, and we don't want this
1409 * to happen unless the file really is executable.
1410 * 3) Ensure that the file is a regular file.
1412 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1413 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1414 (attr->va_type != VREG))
1418 * Zero length files can't be exec'd
1420 if (attr->va_size == 0)
1424 * Check for execute permission to file based on current credentials.
1426 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1431 * Check number of open-for-writes on the file and deny execution
1434 error = VOP_GET_WRITECOUNT(vp, &writecount);
1437 if (writecount != 0)
1441 * Call filesystem specific open routine (which does nothing in the
1444 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1451 * Exec handler registration
1454 exec_register(execsw_arg)
1455 const struct execsw *execsw_arg;
1457 const struct execsw **es, **xs, **newexecsw;
1458 int count = 2; /* New slot and trailing NULL */
1461 for (es = execsw; *es; es++)
1463 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1464 if (newexecsw == NULL)
1468 for (es = execsw; *es; es++)
1473 free(execsw, M_TEMP);
1479 exec_unregister(execsw_arg)
1480 const struct execsw *execsw_arg;
1482 const struct execsw **es, **xs, **newexecsw;
1486 panic("unregister with no handlers left?\n");
1488 for (es = execsw; *es; es++) {
1489 if (*es == execsw_arg)
1494 for (es = execsw; *es; es++)
1495 if (*es != execsw_arg)
1497 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1498 if (newexecsw == NULL)
1501 for (es = execsw; *es; es++)
1502 if (*es != execsw_arg)
1506 free(execsw, M_TEMP);