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/capsicum.h>
37 #include <sys/systm.h>
38 #include <sys/eventhandler.h>
40 #include <sys/mutex.h>
41 #include <sys/sysproto.h>
42 #include <sys/signalvar.h>
43 #include <sys/kernel.h>
44 #include <sys/mount.h>
45 #include <sys/filedesc.h>
46 #include <sys/fcntl.h>
49 #include <sys/imgact.h>
50 #include <sys/imgact_elf.h>
52 #include <sys/malloc.h>
55 #include <sys/pioctl.h>
56 #include <sys/namei.h>
57 #include <sys/resourcevar.h>
58 #include <sys/rwlock.h>
59 #include <sys/sched.h>
61 #include <sys/sf_buf.h>
62 #include <sys/syscallsubr.h>
63 #include <sys/sysent.h>
65 #include <sys/sysctl.h>
66 #include <sys/vnode.h>
69 #include <sys/ktrace.h>
73 #include <vm/vm_param.h>
75 #include <vm/vm_page.h>
76 #include <vm/vm_map.h>
77 #include <vm/vm_kern.h>
78 #include <vm/vm_extern.h>
79 #include <vm/vm_object.h>
80 #include <vm/vm_pager.h>
83 #include <sys/pmckern.h>
86 #include <machine/reg.h>
88 #include <security/audit/audit.h>
89 #include <security/mac/mac_framework.h>
92 #include <sys/dtrace_bsd.h>
93 dtrace_execexit_func_t dtrace_fasttrap_exec;
96 SDT_PROVIDER_DECLARE(proc);
97 SDT_PROBE_DEFINE1(proc, kernel, , exec, "char *");
98 SDT_PROBE_DEFINE1(proc, kernel, , exec__failure, "int");
99 SDT_PROBE_DEFINE1(proc, kernel, , exec__success, "char *");
101 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
103 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
104 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
105 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
106 static int do_execve(struct thread *td, struct image_args *args,
109 /* XXX This should be vm_size_t. */
110 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD,
111 NULL, 0, sysctl_kern_ps_strings, "LU", "");
113 /* XXX This should be vm_size_t. */
114 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD|
115 CTLFLAG_CAPRD, NULL, 0, sysctl_kern_usrstack, "LU", "");
117 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD,
118 NULL, 0, sysctl_kern_stackprot, "I", "");
120 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
121 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
122 &ps_arg_cache_limit, 0, "");
124 static int disallow_high_osrel;
125 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW,
126 &disallow_high_osrel, 0,
127 "Disallow execution of binaries built for higher version of the world");
129 static int map_at_zero = 0;
130 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RWTUN, &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;
283 struct vmspace *oldvmspace;
286 AUDIT_ARG_ARGV(args->begin_argv, args->argc,
287 args->begin_envv - args->begin_argv);
288 AUDIT_ARG_ENVV(args->begin_envv, args->envc,
289 args->endp - args->begin_envv);
290 if (p->p_flag & P_HADTHREADS) {
292 if (thread_single(p, SINGLE_BOUNDARY)) {
294 exec_free_args(args);
295 return (ERESTART); /* Try again later. */
300 KASSERT((td->td_pflags & TDP_EXECVMSPC) == 0, ("nested execve"));
301 oldvmspace = td->td_proc->p_vmspace;
302 error = do_execve(td, args, mac_p);
304 if (p->p_flag & P_HADTHREADS) {
307 * If success, we upgrade to SINGLE_EXIT state to
308 * force other threads to suicide.
311 thread_single(p, SINGLE_EXIT);
313 thread_single_end(p, SINGLE_BOUNDARY);
316 if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
317 KASSERT(td->td_proc->p_vmspace != oldvmspace,
318 ("oldvmspace still used"));
319 vmspace_free(oldvmspace);
320 td->td_pflags &= ~TDP_EXECVMSPC;
327 * In-kernel implementation of execve(). All arguments are assumed to be
328 * userspace pointers from the passed thread.
331 do_execve(td, args, mac_p)
333 struct image_args *args;
336 struct proc *p = td->td_proc;
338 struct ucred *newcred = NULL, *oldcred;
339 struct uidinfo *euip = NULL;
340 register_t *stack_base;
342 struct image_params image_params, *imgp;
344 int (*img_first)(struct image_params *);
345 struct pargs *oldargs = NULL, *newargs = NULL;
346 struct sigacts *oldsigacts, *newsigacts;
348 struct vnode *tracevp = NULL;
349 struct ucred *tracecred = NULL;
351 struct vnode *textvp = NULL, *binvp;
353 int credential_changing;
356 struct label *interpvplabel = NULL;
360 struct pmckern_procexec pe;
362 static const char fexecv_proc_title[] = "(fexecv)";
364 imgp = &image_params;
367 * Lock the process and set the P_INEXEC flag to indicate that
368 * it should be left alone until we're done here. This is
369 * necessary to avoid race conditions - e.g. in ptrace() -
370 * that might allow a local user to illicitly obtain elevated
374 KASSERT((p->p_flag & P_INEXEC) == 0,
375 ("%s(): process already has P_INEXEC flag", __func__));
376 p->p_flag |= P_INEXEC;
380 * Initialize part of the common data
382 bzero(imgp, sizeof(*imgp));
388 error = mac_execve_enter(imgp, mac_p);
394 * Translate the file name. namei() returns a vnode pointer
395 * in ni_vp amoung other things.
397 * XXXAUDIT: It would be desirable to also audit the name of the
398 * interpreter if this is an interpreted binary.
400 if (args->fname != NULL) {
401 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME
402 | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
405 SDT_PROBE(proc, kernel, , exec, args->fname, 0, 0, 0, 0 );
408 if (args->fname != NULL) {
409 #ifdef CAPABILITY_MODE
411 * While capability mode can't reach this point via direct
412 * path arguments to execve(), we also don't allow
413 * interpreters to be used in capability mode (for now).
414 * Catch indirect lookups and return a permissions error.
416 if (IN_CAPABILITY_MODE(td)) {
428 AUDIT_ARG_FD(args->fd);
430 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
432 error = fgetvp_exec(td, args->fd,
433 cap_rights_init(&rights, CAP_FEXECVE), &binvp);
436 vn_lock(binvp, LK_EXCLUSIVE | LK_RETRY);
437 AUDIT_ARG_VNODE1(binvp);
442 * Check file permissions (also 'opens' file)
444 error = exec_check_permissions(imgp);
446 goto exec_fail_dealloc;
448 imgp->object = imgp->vp->v_object;
449 if (imgp->object != NULL)
450 vm_object_reference(imgp->object);
453 * Set VV_TEXT now so no one can write to the executable while we're
456 * Remember if this was set before and unset it in case this is not
457 * actually an executable image.
459 textset = VOP_IS_TEXT(imgp->vp);
460 VOP_SET_TEXT(imgp->vp);
462 error = exec_map_first_page(imgp);
464 goto exec_fail_dealloc;
466 imgp->proc->p_osrel = 0;
468 * If the current process has a special image activator it
469 * wants to try first, call it. For example, emulating shell
470 * scripts differently.
473 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
474 error = img_first(imgp);
477 * Loop through the list of image activators, calling each one.
478 * An activator returns -1 if there is no match, 0 on success,
479 * and an error otherwise.
481 for (i = 0; error == -1 && execsw[i]; ++i) {
482 if (execsw[i]->ex_imgact == NULL ||
483 execsw[i]->ex_imgact == img_first) {
486 error = (*execsw[i]->ex_imgact)(imgp);
492 VOP_UNSET_TEXT(imgp->vp);
495 goto exec_fail_dealloc;
499 * Special interpreter operation, cleanup and loop up to try to
500 * activate the interpreter.
502 if (imgp->interpreted) {
503 exec_unmap_first_page(imgp);
505 * VV_TEXT needs to be unset for scripts. There is a short
506 * period before we determine that something is a script where
507 * VV_TEXT will be set. The vnode lock is held over this
508 * entire period so nothing should illegitimately be blocked.
510 VOP_UNSET_TEXT(imgp->vp);
511 /* free name buffer and old vnode */
512 if (args->fname != NULL)
513 NDFREE(&nd, NDF_ONLY_PNBUF);
515 mac_execve_interpreter_enter(binvp, &interpvplabel);
518 VOP_CLOSE(binvp, FREAD, td->td_ucred, td);
522 vm_object_deallocate(imgp->object);
524 /* set new name to that of the interpreter */
525 NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
526 UIO_SYSSPACE, imgp->interpreter_name, td);
527 args->fname = imgp->interpreter_name;
532 * NB: We unlock the vnode here because it is believed that none
533 * of the sv_copyout_strings/sv_fixup operations require the vnode.
535 VOP_UNLOCK(imgp->vp, 0);
538 * Do the best to calculate the full path to the image file.
540 if (imgp->auxargs != NULL &&
541 ((args->fname != NULL && args->fname[0] == '/') ||
542 vn_fullpath(td, imgp->vp, &imgp->execpath, &imgp->freepath) != 0))
543 imgp->execpath = args->fname;
545 if (disallow_high_osrel &&
546 P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
548 uprintf("Osrel %d for image %s too high\n", p->p_osrel,
549 imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
550 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
551 goto exec_fail_dealloc;
555 * Copy out strings (args and env) and initialize stack base
557 if (p->p_sysent->sv_copyout_strings)
558 stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
560 stack_base = exec_copyout_strings(imgp);
563 * If custom stack fixup routine present for this process
564 * let it do the stack setup.
565 * Else stuff argument count as first item on stack
567 if (p->p_sysent->sv_fixup != NULL)
568 (*p->p_sysent->sv_fixup)(&stack_base, imgp);
570 suword(--stack_base, imgp->args->argc);
573 * For security and other reasons, the file descriptor table cannot
574 * be shared after an exec.
577 /* close files on exec */
581 * Malloc things before we need locks.
583 i = imgp->args->begin_envv - imgp->args->begin_argv;
584 /* Cache arguments if they fit inside our allowance */
585 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
586 newargs = pargs_alloc(i);
587 bcopy(imgp->args->begin_argv, newargs->ar_args, i);
590 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
592 /* Get a reference to the vnode prior to locking the proc */
596 * For security and other reasons, signal handlers cannot
597 * be shared after an exec. The new process gets a copy of the old
598 * handlers. In execsigs(), the new process will have its signals
601 if (sigacts_shared(p->p_sigacts)) {
602 oldsigacts = p->p_sigacts;
603 newsigacts = sigacts_alloc();
604 sigacts_copy(newsigacts, oldsigacts);
607 newsigacts = NULL; /* satisfy gcc */
612 p->p_sigacts = newsigacts;
613 oldcred = p->p_ucred;
617 /* reset caught signals */
620 /* name this process - nameiexec(p, ndp) */
621 bzero(p->p_comm, sizeof(p->p_comm));
623 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
624 min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
625 else if (vn_commname(binvp, p->p_comm, sizeof(p->p_comm)) != 0)
626 bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
627 bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
629 sched_clear_tdname(td);
633 * mark as execed, wakeup the process that vforked (if any) and tell
634 * it that it now has its own resources back
637 if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0)
638 p->p_flag2 &= ~P2_NOTRACE;
639 if (p->p_flag & P_PPWAIT) {
640 p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
641 cv_broadcast(&p->p_pwait);
645 * Implement image setuid/setgid.
647 * Don't honor setuid/setgid if the filesystem prohibits it or if
648 * the process is being traced.
650 * We disable setuid/setgid/etc in compatibility mode on the basis
651 * that most setugid applications are not written with that
652 * environment in mind, and will therefore almost certainly operate
653 * incorrectly. In principle there's no reason that setugid
654 * applications might not be useful in capability mode, so we may want
655 * to reconsider this conservative design choice in the future.
657 * XXXMAC: For the time being, use NOSUID to also prohibit
658 * transitions on the file system.
660 credential_changing = 0;
661 credential_changing |= (attr.va_mode & S_ISUID) && oldcred->cr_uid !=
663 credential_changing |= (attr.va_mode & S_ISGID) && oldcred->cr_gid !=
666 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
667 interpvplabel, imgp);
668 credential_changing |= will_transition;
671 if (credential_changing &&
672 #ifdef CAPABILITY_MODE
673 ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
675 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
676 (p->p_flag & P_TRACED) == 0) {
678 * Turn off syscall tracing for set-id programs, except for
679 * root. Record any set-id flags first to make sure that
680 * we do not regain any tracing during a possible block.
685 if (p->p_tracecred != NULL &&
686 priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED, 0))
687 ktrprocexec(p, &tracecred, &tracevp);
690 * Close any file descriptors 0..2 that reference procfs,
691 * then make sure file descriptors 0..2 are in use.
693 * Both fdsetugidsafety() and fdcheckstd() may call functions
694 * taking sleepable locks, so temporarily drop our locks.
697 VOP_UNLOCK(imgp->vp, 0);
699 error = fdcheckstd(td);
702 newcred = crdup(oldcred);
703 euip = uifind(attr.va_uid);
704 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
707 * Set the new credentials.
709 if (attr.va_mode & S_ISUID)
710 change_euid(newcred, euip);
711 if (attr.va_mode & S_ISGID)
712 change_egid(newcred, attr.va_gid);
714 if (will_transition) {
715 mac_vnode_execve_transition(oldcred, newcred, imgp->vp,
716 interpvplabel, imgp);
720 * Implement correct POSIX saved-id behavior.
722 * XXXMAC: Note that the current logic will save the
723 * uid and gid if a MAC domain transition occurs, even
724 * though maybe it shouldn't.
726 change_svuid(newcred, newcred->cr_uid);
727 change_svgid(newcred, newcred->cr_gid);
728 proc_set_cred(p, newcred);
730 if (oldcred->cr_uid == oldcred->cr_ruid &&
731 oldcred->cr_gid == oldcred->cr_rgid)
732 p->p_flag &= ~P_SUGID;
734 * Implement correct POSIX saved-id behavior.
736 * XXX: It's not clear that the existing behavior is
737 * POSIX-compliant. A number of sources indicate that the
738 * saved uid/gid should only be updated if the new ruid is
739 * not equal to the old ruid, or the new euid is not equal
740 * to the old euid and the new euid is not equal to the old
741 * ruid. The FreeBSD code always updates the saved uid/gid.
742 * Also, this code uses the new (replaced) euid and egid as
743 * the source, which may or may not be the right ones to use.
745 if (oldcred->cr_svuid != oldcred->cr_uid ||
746 oldcred->cr_svgid != oldcred->cr_gid) {
748 VOP_UNLOCK(imgp->vp, 0);
749 newcred = crdup(oldcred);
750 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
752 change_svuid(newcred, newcred->cr_uid);
753 change_svgid(newcred, newcred->cr_gid);
754 proc_set_cred(p, newcred);
759 * Store the vp for use in procfs. This vnode was referenced prior
760 * to locking the proc lock.
762 textvp = p->p_textvp;
767 * Tell the DTrace fasttrap provider about the exec if it
768 * has declared an interest.
770 if (dtrace_fasttrap_exec)
771 dtrace_fasttrap_exec(p);
775 * Notify others that we exec'd, and clear the P_INEXEC flag
776 * as we're now a bona fide freshly-execed process.
778 KNOTE_LOCKED(&p->p_klist, NOTE_EXEC);
779 p->p_flag &= ~P_INEXEC;
781 /* clear "fork but no exec" flag, as we _are_ execing */
782 p->p_acflag &= ~AFORK;
785 * Free any previous argument cache and replace it with
786 * the new argument cache, if any.
794 * Check if system-wide sampling is in effect or if the
795 * current process is using PMCs. If so, do exec() time
796 * processing. This processing needs to happen AFTER the
797 * P_INEXEC flag is cleared.
799 * The proc lock needs to be released before taking the PMC
802 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
804 VOP_UNLOCK(imgp->vp, 0);
805 pe.pm_credentialschanged = credential_changing;
806 pe.pm_entryaddr = imgp->entry_addr;
808 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
809 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
812 #else /* !HWPMC_HOOKS */
816 /* Set values passed into the program in registers. */
817 if (p->p_sysent->sv_setregs)
818 (*p->p_sysent->sv_setregs)(td, imgp,
819 (u_long)(uintptr_t)stack_base);
821 exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base);
823 vfs_mark_atime(imgp->vp, td->td_ucred);
825 SDT_PROBE(proc, kernel, , exec__success, args->fname, 0, 0, 0, 0);
827 VOP_UNLOCK(imgp->vp, 0);
830 * Free any resources malloc'd earlier that we didn't use.
838 * Handle deferred decrement of ref counts.
847 if (tracecred != NULL)
850 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
853 if (oldsigacts != NULL)
854 sigacts_free(oldsigacts);
859 * free various allocated resources
861 if (imgp->firstpage != NULL)
862 exec_unmap_first_page(imgp);
864 if (imgp->vp != NULL) {
866 NDFREE(&nd, NDF_ONLY_PNBUF);
868 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
872 if (imgp->object != NULL)
873 vm_object_deallocate(imgp->object);
875 free(imgp->freepath, M_TEMP);
879 td->td_dbgflags |= TDB_EXEC;
883 * Stop the process here if its stop event mask has
884 * the S_EXEC bit set.
886 STOPEVENT(p, S_EXEC, 0);
891 /* we're done here, clear P_INEXEC */
893 p->p_flag &= ~P_INEXEC;
896 SDT_PROBE(proc, kernel, , exec__failure, error, 0, 0, 0, 0);
900 mac_execve_exit(imgp);
901 mac_execve_interpreter_exit(interpvplabel);
903 exec_free_args(args);
905 if (error && imgp->vmspace_destroyed) {
906 /* sorry, no more process anymore. exit gracefully */
907 exit1(td, W_EXITCODE(0, SIGABRT));
920 exec_map_first_page(imgp)
921 struct image_params *imgp;
925 vm_page_t ma[VM_INITIAL_PAGEIN];
928 if (imgp->firstpage != NULL)
929 exec_unmap_first_page(imgp);
931 object = imgp->vp->v_object;
934 VM_OBJECT_WLOCK(object);
935 #if VM_NRESERVLEVEL > 0
936 vm_object_color(object, 0);
938 ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL);
939 if (ma[0]->valid != VM_PAGE_BITS_ALL) {
940 initial_pagein = VM_INITIAL_PAGEIN;
941 if (initial_pagein > object->size)
942 initial_pagein = object->size;
943 for (i = 1; i < initial_pagein; i++) {
944 if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) {
947 if (vm_page_tryxbusy(ma[i]))
950 ma[i] = vm_page_alloc(object, i,
951 VM_ALLOC_NORMAL | VM_ALLOC_IFNOTCACHED);
957 rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
958 ma[0] = vm_page_lookup(object, 0);
959 if ((rv != VM_PAGER_OK) || (ma[0] == NULL)) {
963 vm_page_unlock(ma[0]);
965 VM_OBJECT_WUNLOCK(object);
969 vm_page_xunbusy(ma[0]);
972 vm_page_activate(ma[0]);
973 vm_page_unlock(ma[0]);
974 VM_OBJECT_WUNLOCK(object);
976 imgp->firstpage = sf_buf_alloc(ma[0], 0);
977 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
983 exec_unmap_first_page(imgp)
984 struct image_params *imgp;
988 if (imgp->firstpage != NULL) {
989 m = sf_buf_page(imgp->firstpage);
990 sf_buf_free(imgp->firstpage);
991 imgp->firstpage = NULL;
999 * Destroy old address space, and allocate a new stack
1000 * The new stack is only SGROWSIZ large because it is grown
1001 * automatically in trap.c.
1004 exec_new_vmspace(imgp, sv)
1005 struct image_params *imgp;
1006 struct sysentvec *sv;
1009 struct proc *p = imgp->proc;
1010 struct vmspace *vmspace = p->p_vmspace;
1012 struct rlimit rlim_stack;
1013 vm_offset_t sv_minuser, stack_addr;
1017 imgp->vmspace_destroyed = 1;
1020 /* May be called with Giant held */
1021 EVENTHANDLER_INVOKE(process_exec, p, imgp);
1024 * Blow away entire process VM, if address space not shared,
1025 * otherwise, create a new VM space so that other threads are
1028 map = &vmspace->vm_map;
1030 sv_minuser = sv->sv_minuser;
1032 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1033 if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser &&
1034 vm_map_max(map) == sv->sv_maxuser) {
1036 pmap_remove_pages(vmspace_pmap(vmspace));
1037 vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1039 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1042 vmspace = p->p_vmspace;
1043 map = &vmspace->vm_map;
1046 /* Map a shared page */
1047 obj = sv->sv_shared_page_obj;
1049 vm_object_reference(obj);
1050 error = vm_map_fixed(map, obj, 0,
1051 sv->sv_shared_page_base, sv->sv_shared_page_len,
1052 VM_PROT_READ | VM_PROT_EXECUTE,
1053 VM_PROT_READ | VM_PROT_EXECUTE,
1054 MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1056 vm_object_deallocate(obj);
1061 /* Allocate a new stack */
1062 if (imgp->stack_sz != 0) {
1063 ssiz = trunc_page(imgp->stack_sz);
1065 lim_rlimit(p, RLIMIT_STACK, &rlim_stack);
1067 if (ssiz > rlim_stack.rlim_max)
1068 ssiz = rlim_stack.rlim_max;
1069 if (ssiz > rlim_stack.rlim_cur) {
1070 rlim_stack.rlim_cur = ssiz;
1071 kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack);
1073 } else if (sv->sv_maxssiz != NULL) {
1074 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 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they
1088 * are still used to enforce the stack rlimit on the process stack.
1090 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1091 vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - ssiz;
1097 * Copy out argument and environment strings from the old process address
1098 * space into the temporary string buffer.
1101 exec_copyin_args(struct image_args *args, char *fname,
1102 enum uio_seg segflg, char **argv, char **envv)
1108 bzero(args, sizeof(*args));
1113 * Allocate demand-paged memory for the file name, argument, and
1114 * environment strings.
1116 error = exec_alloc_args(args);
1121 * Copy the file name.
1123 if (fname != NULL) {
1124 args->fname = args->buf;
1125 error = (segflg == UIO_SYSSPACE) ?
1126 copystr(fname, args->fname, PATH_MAX, &length) :
1127 copyinstr(fname, args->fname, PATH_MAX, &length);
1133 args->begin_argv = args->buf + length;
1134 args->endp = args->begin_argv;
1135 args->stringspace = ARG_MAX;
1138 * extract arguments first
1141 error = fueword(argv++, &argp);
1148 error = copyinstr((void *)(uintptr_t)argp, args->endp,
1149 args->stringspace, &length);
1151 if (error == ENAMETOOLONG)
1155 args->stringspace -= length;
1156 args->endp += length;
1160 args->begin_envv = args->endp;
1163 * extract environment strings
1167 error = fueword(envv++, &envp);
1174 error = copyinstr((void *)(uintptr_t)envp,
1175 args->endp, args->stringspace, &length);
1177 if (error == ENAMETOOLONG)
1181 args->stringspace -= length;
1182 args->endp += length;
1190 exec_free_args(args);
1195 * Allocate temporary demand-paged, zero-filled memory for the file name,
1196 * argument, and environment strings. Returns zero if the allocation succeeds
1197 * and ENOMEM otherwise.
1200 exec_alloc_args(struct image_args *args)
1203 args->buf = (char *)kmap_alloc_wait(exec_map, PATH_MAX + ARG_MAX);
1204 return (args->buf != NULL ? 0 : ENOMEM);
1208 exec_free_args(struct image_args *args)
1211 if (args->buf != NULL) {
1212 kmap_free_wakeup(exec_map, (vm_offset_t)args->buf,
1213 PATH_MAX + ARG_MAX);
1216 if (args->fname_buf != NULL) {
1217 free(args->fname_buf, M_TEMP);
1218 args->fname_buf = NULL;
1223 * Copy strings out to the new process address space, constructing new arg
1224 * and env vector tables. Return a pointer to the base so that it can be used
1225 * as the initial stack pointer.
1228 exec_copyout_strings(imgp)
1229 struct image_params *imgp;
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 = (uintptr_t)arginfo;
1263 if (szsigcode != 0) {
1265 destp = rounddown2(destp, sizeof(void *));
1266 copyout(p->p_sysent->sv_sigcode, (void *)destp, szsigcode);
1270 * Copy the image path for the rtld.
1272 if (execpath_len != 0) {
1273 destp -= execpath_len;
1274 imgp->execpathp = destp;
1275 copyout(imgp->execpath, (void *)destp, execpath_len);
1279 * Prepare the canary for SSP.
1281 arc4rand(canary, sizeof(canary), 0);
1282 destp -= sizeof(canary);
1283 imgp->canary = destp;
1284 copyout(canary, (void *)destp, sizeof(canary));
1285 imgp->canarylen = sizeof(canary);
1288 * Prepare the pagesizes array.
1291 destp = rounddown2(destp, sizeof(void *));
1292 imgp->pagesizes = destp;
1293 copyout(pagesizes, (void *)destp, szps);
1294 imgp->pagesizeslen = szps;
1296 destp -= ARG_MAX - imgp->args->stringspace;
1297 destp = rounddown2(destp, sizeof(void *));
1300 * If we have a valid auxargs ptr, prepare some room
1303 if (imgp->auxargs) {
1305 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1306 * lower compatibility.
1308 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1311 * The '+ 2' is for the null pointers at the end of each of
1312 * the arg and env vector sets,and imgp->auxarg_size is room
1313 * for argument of Runtime loader.
1315 vectp = (char **)(destp - (imgp->args->argc +
1316 imgp->args->envc + 2 + imgp->auxarg_size)
1320 * The '+ 2' is for the null pointers at the end of each of
1321 * the arg and env vector sets
1323 vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc
1324 + 2) * sizeof(char *));
1328 * vectp also becomes our initial stack base
1330 stack_base = (register_t *)vectp;
1332 stringp = imgp->args->begin_argv;
1333 argc = imgp->args->argc;
1334 envc = imgp->args->envc;
1337 * Copy out strings - arguments and environment.
1339 copyout(stringp, (void *)destp, ARG_MAX - imgp->args->stringspace);
1342 * Fill in "ps_strings" struct for ps, w, etc.
1344 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1345 suword32(&arginfo->ps_nargvstr, argc);
1348 * Fill in argument portion of vector table.
1350 for (; argc > 0; --argc) {
1351 suword(vectp++, (long)(intptr_t)destp);
1352 while (*stringp++ != 0)
1357 /* a null vector table pointer separates the argp's from the envp's */
1360 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1361 suword32(&arginfo->ps_nenvstr, envc);
1364 * Fill in environment portion of vector table.
1366 for (; envc > 0; --envc) {
1367 suword(vectp++, (long)(intptr_t)destp);
1368 while (*stringp++ != 0)
1373 /* end of vector table is a null pointer */
1376 return (stack_base);
1380 * Check permissions of file to execute.
1381 * Called with imgp->vp locked.
1382 * Return 0 for success or error code on failure.
1385 exec_check_permissions(imgp)
1386 struct image_params *imgp;
1388 struct vnode *vp = imgp->vp;
1389 struct vattr *attr = imgp->attr;
1391 int error, writecount;
1395 /* Get file attributes */
1396 error = VOP_GETATTR(vp, attr, td->td_ucred);
1401 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1407 * 1) Check if file execution is disabled for the filesystem that
1408 * this file resides on.
1409 * 2) Ensure that at least one execute bit is on. Otherwise, a
1410 * privileged user will always succeed, and we don't want this
1411 * to happen unless the file really is executable.
1412 * 3) Ensure that the file is a regular file.
1414 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1415 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1416 (attr->va_type != VREG))
1420 * Zero length files can't be exec'd
1422 if (attr->va_size == 0)
1426 * Check for execute permission to file based on current credentials.
1428 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1433 * Check number of open-for-writes on the file and deny execution
1436 error = VOP_GET_WRITECOUNT(vp, &writecount);
1439 if (writecount != 0)
1443 * Call filesystem specific open routine (which does nothing in the
1446 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1453 * Exec handler registration
1456 exec_register(execsw_arg)
1457 const struct execsw *execsw_arg;
1459 const struct execsw **es, **xs, **newexecsw;
1460 int count = 2; /* New slot and trailing NULL */
1463 for (es = execsw; *es; es++)
1465 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1466 if (newexecsw == NULL)
1470 for (es = execsw; *es; es++)
1475 free(execsw, M_TEMP);
1481 exec_unregister(execsw_arg)
1482 const struct execsw *execsw_arg;
1484 const struct execsw **es, **xs, **newexecsw;
1488 panic("unregister with no handlers left?\n");
1490 for (es = execsw; *es; es++) {
1491 if (*es == execsw_arg)
1496 for (es = execsw; *es; es++)
1497 if (*es != execsw_arg)
1499 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1500 if (newexecsw == NULL)
1503 for (es = execsw; *es; es++)
1504 if (*es != execsw_arg)
1508 free(execsw, M_TEMP);