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/rwlock.h>
61 #include <sys/sched.h>
63 #include <sys/sf_buf.h>
64 #include <sys/syscallsubr.h>
65 #include <sys/sysent.h>
67 #include <sys/sysctl.h>
68 #include <sys/vnode.h>
71 #include <sys/ktrace.h>
75 #include <vm/vm_param.h>
77 #include <vm/vm_page.h>
78 #include <vm/vm_map.h>
79 #include <vm/vm_kern.h>
80 #include <vm/vm_extern.h>
81 #include <vm/vm_object.h>
82 #include <vm/vm_pager.h>
85 #include <sys/pmckern.h>
88 #include <machine/reg.h>
90 #include <security/audit/audit.h>
91 #include <security/mac/mac_framework.h>
94 #include <sys/dtrace_bsd.h>
95 dtrace_execexit_func_t dtrace_fasttrap_exec;
98 SDT_PROVIDER_DECLARE(proc);
99 SDT_PROBE_DEFINE1(proc, kernel, , exec, "char *");
100 SDT_PROBE_DEFINE1(proc, kernel, , exec__failure, "int");
101 SDT_PROBE_DEFINE1(proc, kernel, , exec__success, "char *");
103 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
105 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
106 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
107 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
108 static int do_execve(struct thread *td, struct image_args *args,
111 /* XXX This should be vm_size_t. */
112 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD,
113 NULL, 0, sysctl_kern_ps_strings, "LU", "");
115 /* XXX This should be vm_size_t. */
116 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD|
117 CTLFLAG_CAPRD, NULL, 0, sysctl_kern_usrstack, "LU", "");
119 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD,
120 NULL, 0, sysctl_kern_stackprot, "I", "");
122 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
123 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
124 &ps_arg_cache_limit, 0, "");
126 static int disallow_high_osrel;
127 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW,
128 &disallow_high_osrel, 0,
129 "Disallow execution of binaries built for higher version of the world");
131 static int map_at_zero = 0;
132 TUNABLE_INT("security.bsd.map_at_zero", &map_at_zero);
133 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RW, &map_at_zero, 0,
134 "Permit processes to map an object at virtual address 0.");
137 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
144 if (req->flags & SCTL_MASK32) {
146 val = (unsigned int)p->p_sysent->sv_psstrings;
147 error = SYSCTL_OUT(req, &val, sizeof(val));
150 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
151 sizeof(p->p_sysent->sv_psstrings));
156 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
163 if (req->flags & SCTL_MASK32) {
165 val = (unsigned int)p->p_sysent->sv_usrstack;
166 error = SYSCTL_OUT(req, &val, sizeof(val));
169 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
170 sizeof(p->p_sysent->sv_usrstack));
175 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
180 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
181 sizeof(p->p_sysent->sv_stackprot)));
185 * Each of the items is a pointer to a `const struct execsw', hence the
186 * double pointer here.
188 static const struct execsw **execsw;
190 #ifndef _SYS_SYSPROTO_H_
201 struct execve_args /* {
208 struct image_args args;
210 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
211 uap->argv, uap->envv);
213 error = kern_execve(td, &args, NULL);
217 #ifndef _SYS_SYSPROTO_H_
218 struct fexecve_args {
225 sys_fexecve(struct thread *td, struct fexecve_args *uap)
228 struct image_args args;
230 error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
231 uap->argv, uap->envv);
234 error = kern_execve(td, &args, NULL);
239 #ifndef _SYS_SYSPROTO_H_
240 struct __mac_execve_args {
249 sys___mac_execve(td, uap)
251 struct __mac_execve_args /* {
260 struct image_args args;
262 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
263 uap->argv, uap->envv);
265 error = kern_execve(td, &args, uap->mac_p);
273 * XXX: kern_execve has the astonishing property of not always returning to
274 * the caller. If sufficiently bad things happen during the call to
275 * do_execve(), it can end up calling exit1(); as a result, callers must
276 * avoid doing anything which they might need to undo (e.g., allocating
280 kern_execve(td, args, mac_p)
282 struct image_args *args;
285 struct proc *p = td->td_proc;
286 struct vmspace *oldvmspace;
289 AUDIT_ARG_ARGV(args->begin_argv, args->argc,
290 args->begin_envv - args->begin_argv);
291 AUDIT_ARG_ENVV(args->begin_envv, args->envc,
292 args->endp - args->begin_envv);
293 if (p->p_flag & P_HADTHREADS) {
295 if (thread_single(SINGLE_BOUNDARY)) {
297 exec_free_args(args);
298 return (ERESTART); /* Try again later. */
303 KASSERT((td->td_pflags & TDP_EXECVMSPC) == 0, ("nested execve"));
304 oldvmspace = td->td_proc->p_vmspace;
305 error = do_execve(td, args, mac_p);
307 if (p->p_flag & P_HADTHREADS) {
310 * If success, we upgrade to SINGLE_EXIT state to
311 * force other threads to suicide.
314 thread_single(SINGLE_EXIT);
319 if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
320 KASSERT(td->td_proc->p_vmspace != oldvmspace,
321 ("oldvmspace still used"));
322 vmspace_free(oldvmspace);
323 td->td_pflags &= ~TDP_EXECVMSPC;
330 * In-kernel implementation of execve(). All arguments are assumed to be
331 * userspace pointers from the passed thread.
334 do_execve(td, args, mac_p)
336 struct image_args *args;
339 struct proc *p = td->td_proc;
341 struct ucred *newcred = NULL, *oldcred;
342 struct uidinfo *euip = NULL;
343 register_t *stack_base;
345 struct image_params image_params, *imgp;
347 int (*img_first)(struct image_params *);
348 struct pargs *oldargs = NULL, *newargs = NULL;
349 struct sigacts *oldsigacts, *newsigacts;
351 struct vnode *tracevp = NULL;
352 struct ucred *tracecred = NULL;
354 struct vnode *textvp = NULL, *binvp = NULL;
356 int credential_changing;
359 struct label *interpvplabel = NULL;
363 struct pmckern_procexec pe;
365 static const char fexecv_proc_title[] = "(fexecv)";
367 imgp = &image_params;
370 * Lock the process and set the P_INEXEC flag to indicate that
371 * it should be left alone until we're done here. This is
372 * necessary to avoid race conditions - e.g. in ptrace() -
373 * that might allow a local user to illicitly obtain elevated
377 KASSERT((p->p_flag & P_INEXEC) == 0,
378 ("%s(): process already has P_INEXEC flag", __func__));
379 p->p_flag |= P_INEXEC;
383 * Initialize part of the common data
386 imgp->execlabel = NULL;
388 imgp->entry_addr = 0;
389 imgp->reloc_base = 0;
390 imgp->vmspace_destroyed = 0;
391 imgp->interpreted = 0;
393 imgp->interpreter_name = NULL;
394 imgp->auxargs = NULL;
397 imgp->firstpage = NULL;
398 imgp->ps_strings = 0;
399 imgp->auxarg_size = 0;
401 imgp->execpath = imgp->freepath = NULL;
406 imgp->pagesizeslen = 0;
407 imgp->stack_prot = 0;
410 error = mac_execve_enter(imgp, mac_p);
415 imgp->image_header = NULL;
418 * Translate the file name. namei() returns a vnode pointer
419 * in ni_vp amoung other things.
421 * XXXAUDIT: It would be desirable to also audit the name of the
422 * interpreter if this is an interpreted binary.
424 if (args->fname != NULL) {
425 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME
426 | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
429 SDT_PROBE(proc, kernel, , exec, args->fname, 0, 0, 0, 0 );
432 if (args->fname != NULL) {
433 #ifdef CAPABILITY_MODE
435 * While capability mode can't reach this point via direct
436 * path arguments to execve(), we also don't allow
437 * interpreters to be used in capability mode (for now).
438 * Catch indirect lookups and return a permissions error.
440 if (IN_CAPABILITY_MODE(td)) {
452 AUDIT_ARG_FD(args->fd);
454 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
456 error = fgetvp_exec(td, args->fd,
457 cap_rights_init(&rights, CAP_FEXECVE), &binvp);
460 vn_lock(binvp, LK_EXCLUSIVE | LK_RETRY);
461 AUDIT_ARG_VNODE1(binvp);
466 * Check file permissions (also 'opens' file)
468 error = exec_check_permissions(imgp);
470 goto exec_fail_dealloc;
472 imgp->object = imgp->vp->v_object;
473 if (imgp->object != NULL)
474 vm_object_reference(imgp->object);
477 * Set VV_TEXT now so no one can write to the executable while we're
480 * Remember if this was set before and unset it in case this is not
481 * actually an executable image.
483 textset = VOP_IS_TEXT(imgp->vp);
484 VOP_SET_TEXT(imgp->vp);
486 error = exec_map_first_page(imgp);
488 goto exec_fail_dealloc;
490 imgp->proc->p_osrel = 0;
492 * If the current process has a special image activator it
493 * wants to try first, call it. For example, emulating shell
494 * scripts differently.
497 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
498 error = img_first(imgp);
501 * Loop through the list of image activators, calling each one.
502 * An activator returns -1 if there is no match, 0 on success,
503 * and an error otherwise.
505 for (i = 0; error == -1 && execsw[i]; ++i) {
506 if (execsw[i]->ex_imgact == NULL ||
507 execsw[i]->ex_imgact == img_first) {
510 error = (*execsw[i]->ex_imgact)(imgp);
516 VOP_UNSET_TEXT(imgp->vp);
519 goto exec_fail_dealloc;
523 * Special interpreter operation, cleanup and loop up to try to
524 * activate the interpreter.
526 if (imgp->interpreted) {
527 exec_unmap_first_page(imgp);
529 * VV_TEXT needs to be unset for scripts. There is a short
530 * period before we determine that something is a script where
531 * VV_TEXT will be set. The vnode lock is held over this
532 * entire period so nothing should illegitimately be blocked.
534 VOP_UNSET_TEXT(imgp->vp);
535 /* free name buffer and old vnode */
536 if (args->fname != NULL)
537 NDFREE(&nd, NDF_ONLY_PNBUF);
539 mac_execve_interpreter_enter(binvp, &interpvplabel);
542 VOP_CLOSE(binvp, FREAD, td->td_ucred, td);
546 vm_object_deallocate(imgp->object);
548 /* set new name to that of the interpreter */
549 NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
550 UIO_SYSSPACE, imgp->interpreter_name, td);
551 args->fname = imgp->interpreter_name;
556 * NB: We unlock the vnode here because it is believed that none
557 * of the sv_copyout_strings/sv_fixup operations require the vnode.
559 VOP_UNLOCK(imgp->vp, 0);
562 * Do the best to calculate the full path to the image file.
564 if (imgp->auxargs != NULL &&
565 ((args->fname != NULL && args->fname[0] == '/') ||
566 vn_fullpath(td, imgp->vp, &imgp->execpath, &imgp->freepath) != 0))
567 imgp->execpath = args->fname;
569 if (disallow_high_osrel &&
570 P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
572 uprintf("Osrel %d for image %s too high\n", p->p_osrel,
573 imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
574 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
575 goto exec_fail_dealloc;
579 * Copy out strings (args and env) and initialize stack base
581 if (p->p_sysent->sv_copyout_strings)
582 stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
584 stack_base = exec_copyout_strings(imgp);
587 * If custom stack fixup routine present for this process
588 * let it do the stack setup.
589 * Else stuff argument count as first item on stack
591 if (p->p_sysent->sv_fixup != NULL)
592 (*p->p_sysent->sv_fixup)(&stack_base, imgp);
594 suword(--stack_base, imgp->args->argc);
597 * For security and other reasons, the file descriptor table cannot
598 * be shared after an exec.
601 /* close files on exec */
605 * Malloc things before we need locks.
607 i = imgp->args->begin_envv - imgp->args->begin_argv;
608 /* Cache arguments if they fit inside our allowance */
609 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
610 newargs = pargs_alloc(i);
611 bcopy(imgp->args->begin_argv, newargs->ar_args, i);
614 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
616 /* Get a reference to the vnode prior to locking the proc */
620 * For security and other reasons, signal handlers cannot
621 * be shared after an exec. The new process gets a copy of the old
622 * handlers. In execsigs(), the new process will have its signals
625 if (sigacts_shared(p->p_sigacts)) {
626 oldsigacts = p->p_sigacts;
627 newsigacts = sigacts_alloc();
628 sigacts_copy(newsigacts, oldsigacts);
631 newsigacts = NULL; /* satisfy gcc */
636 p->p_sigacts = newsigacts;
637 oldcred = p->p_ucred;
641 /* reset caught signals */
644 /* name this process - nameiexec(p, ndp) */
645 bzero(p->p_comm, sizeof(p->p_comm));
647 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
648 min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
649 else if (vn_commname(binvp, p->p_comm, sizeof(p->p_comm)) != 0)
650 bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
651 bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
653 sched_clear_tdname(td);
657 * mark as execed, wakeup the process that vforked (if any) and tell
658 * it that it now has its own resources back
661 if (p->p_flag & P_PPWAIT) {
662 p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
663 cv_broadcast(&p->p_pwait);
667 * Implement image setuid/setgid.
669 * Don't honor setuid/setgid if the filesystem prohibits it or if
670 * the process is being traced.
672 * We disable setuid/setgid/etc in compatibility mode on the basis
673 * that most setugid applications are not written with that
674 * environment in mind, and will therefore almost certainly operate
675 * incorrectly. In principle there's no reason that setugid
676 * applications might not be useful in capability mode, so we may want
677 * to reconsider this conservative design choice in the future.
679 * XXXMAC: For the time being, use NOSUID to also prohibit
680 * transitions on the file system.
682 credential_changing = 0;
683 credential_changing |= (attr.va_mode & S_ISUID) && oldcred->cr_uid !=
685 credential_changing |= (attr.va_mode & S_ISGID) && oldcred->cr_gid !=
688 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
689 interpvplabel, imgp);
690 credential_changing |= will_transition;
693 if (credential_changing &&
694 #ifdef CAPABILITY_MODE
695 ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
697 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
698 (p->p_flag & P_TRACED) == 0) {
700 * Turn off syscall tracing for set-id programs, except for
701 * root. Record any set-id flags first to make sure that
702 * we do not regain any tracing during a possible block.
707 if (p->p_tracecred != NULL &&
708 priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED, 0))
709 ktrprocexec(p, &tracecred, &tracevp);
712 * Close any file descriptors 0..2 that reference procfs,
713 * then make sure file descriptors 0..2 are in use.
715 * setugidsafety() may call closef() and then pfind()
716 * which may grab the process lock.
717 * fdcheckstd() may call falloc() which may block to
718 * allocate memory, so temporarily drop the process lock.
721 VOP_UNLOCK(imgp->vp, 0);
723 error = fdcheckstd(td);
726 newcred = crdup(oldcred);
727 euip = uifind(attr.va_uid);
728 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
731 * Set the new credentials.
733 if (attr.va_mode & S_ISUID)
734 change_euid(newcred, euip);
735 if (attr.va_mode & S_ISGID)
736 change_egid(newcred, attr.va_gid);
738 if (will_transition) {
739 mac_vnode_execve_transition(oldcred, newcred, imgp->vp,
740 interpvplabel, imgp);
744 * Implement correct POSIX saved-id behavior.
746 * XXXMAC: Note that the current logic will save the
747 * uid and gid if a MAC domain transition occurs, even
748 * though maybe it shouldn't.
750 change_svuid(newcred, newcred->cr_uid);
751 change_svgid(newcred, newcred->cr_gid);
752 p->p_ucred = newcred;
754 if (oldcred->cr_uid == oldcred->cr_ruid &&
755 oldcred->cr_gid == oldcred->cr_rgid)
756 p->p_flag &= ~P_SUGID;
758 * Implement correct POSIX saved-id behavior.
760 * XXX: It's not clear that the existing behavior is
761 * POSIX-compliant. A number of sources indicate that the
762 * saved uid/gid should only be updated if the new ruid is
763 * not equal to the old ruid, or the new euid is not equal
764 * to the old euid and the new euid is not equal to the old
765 * ruid. The FreeBSD code always updates the saved uid/gid.
766 * Also, this code uses the new (replaced) euid and egid as
767 * the source, which may or may not be the right ones to use.
769 if (oldcred->cr_svuid != oldcred->cr_uid ||
770 oldcred->cr_svgid != oldcred->cr_gid) {
772 VOP_UNLOCK(imgp->vp, 0);
773 newcred = crdup(oldcred);
774 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
776 change_svuid(newcred, newcred->cr_uid);
777 change_svgid(newcred, newcred->cr_gid);
778 p->p_ucred = newcred;
783 * Store the vp for use in procfs. This vnode was referenced prior
784 * to locking the proc lock.
786 textvp = p->p_textvp;
791 * Tell the DTrace fasttrap provider about the exec if it
792 * has declared an interest.
794 if (dtrace_fasttrap_exec)
795 dtrace_fasttrap_exec(p);
799 * Notify others that we exec'd, and clear the P_INEXEC flag
800 * as we're now a bona fide freshly-execed process.
802 KNOTE_LOCKED(&p->p_klist, NOTE_EXEC);
803 p->p_flag &= ~P_INEXEC;
805 /* clear "fork but no exec" flag, as we _are_ execing */
806 p->p_acflag &= ~AFORK;
809 * Free any previous argument cache and replace it with
810 * the new argument cache, if any.
818 * Check if system-wide sampling is in effect or if the
819 * current process is using PMCs. If so, do exec() time
820 * processing. This processing needs to happen AFTER the
821 * P_INEXEC flag is cleared.
823 * The proc lock needs to be released before taking the PMC
826 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
828 VOP_UNLOCK(imgp->vp, 0);
829 pe.pm_credentialschanged = credential_changing;
830 pe.pm_entryaddr = imgp->entry_addr;
832 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
833 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
836 #else /* !HWPMC_HOOKS */
840 /* Set values passed into the program in registers. */
841 if (p->p_sysent->sv_setregs)
842 (*p->p_sysent->sv_setregs)(td, imgp,
843 (u_long)(uintptr_t)stack_base);
845 exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base);
847 vfs_mark_atime(imgp->vp, td->td_ucred);
849 SDT_PROBE(proc, kernel, , exec__success, args->fname, 0, 0, 0, 0);
851 VOP_UNLOCK(imgp->vp, 0);
854 * Free any resources malloc'd earlier that we didn't use.
862 * Handle deferred decrement of ref counts.
866 if (binvp && error != 0)
871 if (tracecred != NULL)
874 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
877 if (oldsigacts != NULL)
878 sigacts_free(oldsigacts);
883 * free various allocated resources
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 if (imgp->object != NULL)
897 vm_object_deallocate(imgp->object);
899 free(imgp->freepath, M_TEMP);
903 td->td_dbgflags |= TDB_EXEC;
907 * Stop the process here if its stop event mask has
908 * the S_EXEC bit set.
910 STOPEVENT(p, S_EXEC, 0);
915 /* we're done here, clear P_INEXEC */
917 p->p_flag &= ~P_INEXEC;
920 SDT_PROBE(proc, kernel, , exec__failure, error, 0, 0, 0, 0);
924 mac_execve_exit(imgp);
925 mac_execve_interpreter_exit(interpvplabel);
927 exec_free_args(args);
929 if (error && imgp->vmspace_destroyed) {
930 /* sorry, no more process anymore. exit gracefully */
931 exit1(td, W_EXITCODE(0, SIGABRT));
944 exec_map_first_page(imgp)
945 struct image_params *imgp;
949 vm_page_t ma[VM_INITIAL_PAGEIN];
952 if (imgp->firstpage != NULL)
953 exec_unmap_first_page(imgp);
955 object = imgp->vp->v_object;
958 VM_OBJECT_WLOCK(object);
959 #if VM_NRESERVLEVEL > 0
960 if ((object->flags & OBJ_COLORED) == 0) {
961 object->flags |= OBJ_COLORED;
962 object->pg_color = 0;
965 ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL);
966 if (ma[0]->valid != VM_PAGE_BITS_ALL) {
967 initial_pagein = VM_INITIAL_PAGEIN;
968 if (initial_pagein > object->size)
969 initial_pagein = object->size;
970 for (i = 1; i < initial_pagein; i++) {
971 if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) {
974 if (vm_page_tryxbusy(ma[i]))
977 ma[i] = vm_page_alloc(object, i,
978 VM_ALLOC_NORMAL | VM_ALLOC_IFNOTCACHED);
984 rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
985 ma[0] = vm_page_lookup(object, 0);
986 if ((rv != VM_PAGER_OK) || (ma[0] == NULL)) {
990 vm_page_unlock(ma[0]);
992 VM_OBJECT_WUNLOCK(object);
996 vm_page_xunbusy(ma[0]);
999 vm_page_activate(ma[0]);
1000 vm_page_unlock(ma[0]);
1001 VM_OBJECT_WUNLOCK(object);
1003 imgp->firstpage = sf_buf_alloc(ma[0], 0);
1004 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
1010 exec_unmap_first_page(imgp)
1011 struct image_params *imgp;
1015 if (imgp->firstpage != NULL) {
1016 m = sf_buf_page(imgp->firstpage);
1017 sf_buf_free(imgp->firstpage);
1018 imgp->firstpage = NULL;
1026 * Destroy old address space, and allocate a new stack
1027 * The new stack is only SGROWSIZ large because it is grown
1028 * automatically in trap.c.
1031 exec_new_vmspace(imgp, sv)
1032 struct image_params *imgp;
1033 struct sysentvec *sv;
1036 struct proc *p = imgp->proc;
1037 struct vmspace *vmspace = p->p_vmspace;
1039 vm_offset_t sv_minuser, stack_addr;
1043 imgp->vmspace_destroyed = 1;
1046 /* May be called with Giant held */
1047 EVENTHANDLER_INVOKE(process_exec, p, imgp);
1050 * Blow away entire process VM, if address space not shared,
1051 * otherwise, create a new VM space so that other threads are
1054 map = &vmspace->vm_map;
1056 sv_minuser = sv->sv_minuser;
1058 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1059 if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser &&
1060 vm_map_max(map) == sv->sv_maxuser) {
1062 pmap_remove_pages(vmspace_pmap(vmspace));
1063 vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1065 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1068 vmspace = p->p_vmspace;
1069 map = &vmspace->vm_map;
1072 /* Map a shared page */
1073 obj = sv->sv_shared_page_obj;
1075 vm_object_reference(obj);
1076 error = vm_map_fixed(map, obj, 0,
1077 sv->sv_shared_page_base, sv->sv_shared_page_len,
1078 VM_PROT_READ | VM_PROT_EXECUTE,
1079 VM_PROT_READ | VM_PROT_EXECUTE,
1080 MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1082 vm_object_deallocate(obj);
1087 /* Allocate a new stack */
1088 if (sv->sv_maxssiz != NULL)
1089 ssiz = *sv->sv_maxssiz;
1092 stack_addr = sv->sv_usrstack - ssiz;
1093 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1094 obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1096 VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1101 /* Allocate a new register stack */
1102 stack_addr = IA64_BACKINGSTORE;
1103 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1104 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_UP);
1109 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
1110 * VM_STACK case, but they are still used to monitor the size of the
1111 * process stack so we can check the stack rlimit.
1113 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1114 vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - ssiz;
1120 * Copy out argument and environment strings from the old process address
1121 * space into the temporary string buffer.
1124 exec_copyin_args(struct image_args *args, char *fname,
1125 enum uio_seg segflg, char **argv, char **envv)
1131 bzero(args, sizeof(*args));
1136 * Allocate demand-paged memory for the file name, argument, and
1137 * environment strings.
1139 error = exec_alloc_args(args);
1144 * Copy the file name.
1146 if (fname != NULL) {
1147 args->fname = args->buf;
1148 error = (segflg == UIO_SYSSPACE) ?
1149 copystr(fname, args->fname, PATH_MAX, &length) :
1150 copyinstr(fname, args->fname, PATH_MAX, &length);
1156 args->begin_argv = args->buf + length;
1157 args->endp = args->begin_argv;
1158 args->stringspace = ARG_MAX;
1161 * extract arguments first
1163 while ((argp = (caddr_t) (intptr_t) fuword(argv++))) {
1164 if (argp == (caddr_t) -1) {
1168 if ((error = copyinstr(argp, args->endp,
1169 args->stringspace, &length))) {
1170 if (error == ENAMETOOLONG)
1174 args->stringspace -= length;
1175 args->endp += length;
1179 args->begin_envv = args->endp;
1182 * extract environment strings
1185 while ((envp = (caddr_t)(intptr_t)fuword(envv++))) {
1186 if (envp == (caddr_t)-1) {
1190 if ((error = copyinstr(envp, args->endp,
1191 args->stringspace, &length))) {
1192 if (error == ENAMETOOLONG)
1196 args->stringspace -= length;
1197 args->endp += length;
1205 exec_free_args(args);
1210 * Allocate temporary demand-paged, zero-filled memory for the file name,
1211 * argument, and environment strings. Returns zero if the allocation succeeds
1212 * and ENOMEM otherwise.
1215 exec_alloc_args(struct image_args *args)
1218 args->buf = (char *)kmap_alloc_wait(exec_map, PATH_MAX + ARG_MAX);
1219 return (args->buf != NULL ? 0 : ENOMEM);
1223 exec_free_args(struct image_args *args)
1226 if (args->buf != NULL) {
1227 kmap_free_wakeup(exec_map, (vm_offset_t)args->buf,
1228 PATH_MAX + ARG_MAX);
1231 if (args->fname_buf != NULL) {
1232 free(args->fname_buf, M_TEMP);
1233 args->fname_buf = NULL;
1238 * Copy strings out to the new process address space, constructing new arg
1239 * and env vector tables. Return a pointer to the base so that it can be used
1240 * as the initial stack pointer.
1243 exec_copyout_strings(imgp)
1244 struct image_params *imgp;
1250 register_t *stack_base;
1251 struct ps_strings *arginfo;
1253 size_t execpath_len;
1254 int szsigcode, szps;
1255 char canary[sizeof(long) * 8];
1257 szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1259 * Calculate string base and vector table pointers.
1260 * Also deal with signal trampoline code for this exec type.
1262 if (imgp->execpath != NULL && imgp->auxargs != NULL)
1263 execpath_len = strlen(imgp->execpath) + 1;
1268 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1269 if (p->p_sysent->sv_sigcode_base == 0) {
1270 if (p->p_sysent->sv_szsigcode != NULL)
1271 szsigcode = *(p->p_sysent->sv_szsigcode);
1273 destp = (uintptr_t)arginfo;
1278 if (szsigcode != 0) {
1280 destp = rounddown2(destp, sizeof(void *));
1281 copyout(p->p_sysent->sv_sigcode, (void *)destp, szsigcode);
1285 * Copy the image path for the rtld.
1287 if (execpath_len != 0) {
1288 destp -= execpath_len;
1289 imgp->execpathp = destp;
1290 copyout(imgp->execpath, (void *)destp, execpath_len);
1294 * Prepare the canary for SSP.
1296 arc4rand(canary, sizeof(canary), 0);
1297 destp -= sizeof(canary);
1298 imgp->canary = destp;
1299 copyout(canary, (void *)destp, sizeof(canary));
1300 imgp->canarylen = sizeof(canary);
1303 * Prepare the pagesizes array.
1306 destp = rounddown2(destp, sizeof(void *));
1307 imgp->pagesizes = destp;
1308 copyout(pagesizes, (void *)destp, szps);
1309 imgp->pagesizeslen = szps;
1311 destp -= ARG_MAX - imgp->args->stringspace;
1312 destp = rounddown2(destp, sizeof(void *));
1315 * If we have a valid auxargs ptr, prepare some room
1318 if (imgp->auxargs) {
1320 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1321 * lower compatibility.
1323 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1326 * The '+ 2' is for the null pointers at the end of each of
1327 * the arg and env vector sets,and imgp->auxarg_size is room
1328 * for argument of Runtime loader.
1330 vectp = (char **)(destp - (imgp->args->argc +
1331 imgp->args->envc + 2 + imgp->auxarg_size)
1335 * The '+ 2' is for the null pointers at the end of each of
1336 * the arg and env vector sets
1338 vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc
1339 + 2) * sizeof(char *));
1343 * vectp also becomes our initial stack base
1345 stack_base = (register_t *)vectp;
1347 stringp = imgp->args->begin_argv;
1348 argc = imgp->args->argc;
1349 envc = imgp->args->envc;
1352 * Copy out strings - arguments and environment.
1354 copyout(stringp, (void *)destp, ARG_MAX - imgp->args->stringspace);
1357 * Fill in "ps_strings" struct for ps, w, etc.
1359 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1360 suword32(&arginfo->ps_nargvstr, argc);
1363 * Fill in argument portion of vector table.
1365 for (; argc > 0; --argc) {
1366 suword(vectp++, (long)(intptr_t)destp);
1367 while (*stringp++ != 0)
1372 /* a null vector table pointer separates the argp's from the envp's */
1375 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1376 suword32(&arginfo->ps_nenvstr, envc);
1379 * Fill in environment portion of vector table.
1381 for (; envc > 0; --envc) {
1382 suword(vectp++, (long)(intptr_t)destp);
1383 while (*stringp++ != 0)
1388 /* end of vector table is a null pointer */
1391 return (stack_base);
1395 * Check permissions of file to execute.
1396 * Called with imgp->vp locked.
1397 * Return 0 for success or error code on failure.
1400 exec_check_permissions(imgp)
1401 struct image_params *imgp;
1403 struct vnode *vp = imgp->vp;
1404 struct vattr *attr = imgp->attr;
1406 int error, writecount;
1410 /* Get file attributes */
1411 error = VOP_GETATTR(vp, attr, td->td_ucred);
1416 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1422 * 1) Check if file execution is disabled for the filesystem that
1423 * this file resides on.
1424 * 2) Ensure that at least one execute bit is on. Otherwise, a
1425 * privileged user will always succeed, and we don't want this
1426 * to happen unless the file really is executable.
1427 * 3) Ensure that the file is a regular file.
1429 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1430 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1431 (attr->va_type != VREG))
1435 * Zero length files can't be exec'd
1437 if (attr->va_size == 0)
1441 * Check for execute permission to file based on current credentials.
1443 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1448 * Check number of open-for-writes on the file and deny execution
1451 error = VOP_GET_WRITECOUNT(vp, &writecount);
1454 if (writecount != 0)
1458 * Call filesystem specific open routine (which does nothing in the
1461 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1468 * Exec handler registration
1471 exec_register(execsw_arg)
1472 const struct execsw *execsw_arg;
1474 const struct execsw **es, **xs, **newexecsw;
1475 int count = 2; /* New slot and trailing NULL */
1478 for (es = execsw; *es; es++)
1480 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1481 if (newexecsw == NULL)
1485 for (es = execsw; *es; es++)
1490 free(execsw, M_TEMP);
1496 exec_unregister(execsw_arg)
1497 const struct execsw *execsw_arg;
1499 const struct execsw **es, **xs, **newexecsw;
1503 panic("unregister with no handlers left?\n");
1505 for (es = execsw; *es; es++) {
1506 if (*es == execsw_arg)
1511 for (es = execsw; *es; es++)
1512 if (*es != execsw_arg)
1514 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1515 if (newexecsw == NULL)
1518 for (es = execsw; *es; es++)
1519 if (*es != execsw_arg)
1523 free(execsw, M_TEMP);