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(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(SINGLE_EXIT);
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 = NULL;
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
383 imgp->execlabel = NULL;
385 imgp->entry_addr = 0;
386 imgp->reloc_base = 0;
387 imgp->vmspace_destroyed = 0;
388 imgp->interpreted = 0;
390 imgp->interpreter_name = NULL;
391 imgp->auxargs = NULL;
394 imgp->firstpage = NULL;
395 imgp->ps_strings = 0;
396 imgp->auxarg_size = 0;
398 imgp->execpath = imgp->freepath = NULL;
403 imgp->pagesizeslen = 0;
404 imgp->stack_prot = 0;
407 error = mac_execve_enter(imgp, mac_p);
412 imgp->image_header = NULL;
415 * Translate the file name. namei() returns a vnode pointer
416 * in ni_vp amoung other things.
418 * XXXAUDIT: It would be desirable to also audit the name of the
419 * interpreter if this is an interpreted binary.
421 if (args->fname != NULL) {
422 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME
423 | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
426 SDT_PROBE(proc, kernel, , exec, args->fname, 0, 0, 0, 0 );
429 if (args->fname != NULL) {
430 #ifdef CAPABILITY_MODE
432 * While capability mode can't reach this point via direct
433 * path arguments to execve(), we also don't allow
434 * interpreters to be used in capability mode (for now).
435 * Catch indirect lookups and return a permissions error.
437 if (IN_CAPABILITY_MODE(td)) {
449 AUDIT_ARG_FD(args->fd);
451 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
453 error = fgetvp_exec(td, args->fd,
454 cap_rights_init(&rights, CAP_FEXECVE), &binvp);
457 vn_lock(binvp, LK_EXCLUSIVE | LK_RETRY);
458 AUDIT_ARG_VNODE1(binvp);
463 * Check file permissions (also 'opens' file)
465 error = exec_check_permissions(imgp);
467 goto exec_fail_dealloc;
469 imgp->object = imgp->vp->v_object;
470 if (imgp->object != NULL)
471 vm_object_reference(imgp->object);
474 * Set VV_TEXT now so no one can write to the executable while we're
477 * Remember if this was set before and unset it in case this is not
478 * actually an executable image.
480 textset = VOP_IS_TEXT(imgp->vp);
481 VOP_SET_TEXT(imgp->vp);
483 error = exec_map_first_page(imgp);
485 goto exec_fail_dealloc;
487 imgp->proc->p_osrel = 0;
489 * If the current process has a special image activator it
490 * wants to try first, call it. For example, emulating shell
491 * scripts differently.
494 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
495 error = img_first(imgp);
498 * Loop through the list of image activators, calling each one.
499 * An activator returns -1 if there is no match, 0 on success,
500 * and an error otherwise.
502 for (i = 0; error == -1 && execsw[i]; ++i) {
503 if (execsw[i]->ex_imgact == NULL ||
504 execsw[i]->ex_imgact == img_first) {
507 error = (*execsw[i]->ex_imgact)(imgp);
513 VOP_UNSET_TEXT(imgp->vp);
516 goto exec_fail_dealloc;
520 * Special interpreter operation, cleanup and loop up to try to
521 * activate the interpreter.
523 if (imgp->interpreted) {
524 exec_unmap_first_page(imgp);
526 * VV_TEXT needs to be unset for scripts. There is a short
527 * period before we determine that something is a script where
528 * VV_TEXT will be set. The vnode lock is held over this
529 * entire period so nothing should illegitimately be blocked.
531 VOP_UNSET_TEXT(imgp->vp);
532 /* free name buffer and old vnode */
533 if (args->fname != NULL)
534 NDFREE(&nd, NDF_ONLY_PNBUF);
536 mac_execve_interpreter_enter(binvp, &interpvplabel);
539 VOP_CLOSE(binvp, FREAD, td->td_ucred, td);
543 vm_object_deallocate(imgp->object);
545 /* set new name to that of the interpreter */
546 NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
547 UIO_SYSSPACE, imgp->interpreter_name, td);
548 args->fname = imgp->interpreter_name;
553 * NB: We unlock the vnode here because it is believed that none
554 * of the sv_copyout_strings/sv_fixup operations require the vnode.
556 VOP_UNLOCK(imgp->vp, 0);
559 * Do the best to calculate the full path to the image file.
561 if (imgp->auxargs != NULL &&
562 ((args->fname != NULL && args->fname[0] == '/') ||
563 vn_fullpath(td, imgp->vp, &imgp->execpath, &imgp->freepath) != 0))
564 imgp->execpath = args->fname;
566 if (disallow_high_osrel &&
567 P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
569 uprintf("Osrel %d for image %s too high\n", p->p_osrel,
570 imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
571 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
572 goto exec_fail_dealloc;
576 * Copy out strings (args and env) and initialize stack base
578 if (p->p_sysent->sv_copyout_strings)
579 stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
581 stack_base = exec_copyout_strings(imgp);
584 * If custom stack fixup routine present for this process
585 * let it do the stack setup.
586 * Else stuff argument count as first item on stack
588 if (p->p_sysent->sv_fixup != NULL)
589 (*p->p_sysent->sv_fixup)(&stack_base, imgp);
591 suword(--stack_base, imgp->args->argc);
594 * For security and other reasons, the file descriptor table cannot
595 * be shared after an exec.
598 /* close files on exec */
602 * Malloc things before we need locks.
604 i = imgp->args->begin_envv - imgp->args->begin_argv;
605 /* Cache arguments if they fit inside our allowance */
606 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
607 newargs = pargs_alloc(i);
608 bcopy(imgp->args->begin_argv, newargs->ar_args, i);
611 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
613 /* Get a reference to the vnode prior to locking the proc */
617 * For security and other reasons, signal handlers cannot
618 * be shared after an exec. The new process gets a copy of the old
619 * handlers. In execsigs(), the new process will have its signals
622 if (sigacts_shared(p->p_sigacts)) {
623 oldsigacts = p->p_sigacts;
624 newsigacts = sigacts_alloc();
625 sigacts_copy(newsigacts, oldsigacts);
628 newsigacts = NULL; /* satisfy gcc */
633 p->p_sigacts = newsigacts;
634 oldcred = p->p_ucred;
638 /* reset caught signals */
641 /* name this process - nameiexec(p, ndp) */
642 bzero(p->p_comm, sizeof(p->p_comm));
644 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
645 min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
646 else if (vn_commname(binvp, p->p_comm, sizeof(p->p_comm)) != 0)
647 bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
648 bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
650 sched_clear_tdname(td);
654 * mark as execed, wakeup the process that vforked (if any) and tell
655 * it that it now has its own resources back
658 if (p->p_flag & P_PPWAIT) {
659 p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
660 cv_broadcast(&p->p_pwait);
664 * Implement image setuid/setgid.
666 * Don't honor setuid/setgid if the filesystem prohibits it or if
667 * the process is being traced.
669 * We disable setuid/setgid/etc in compatibility mode on the basis
670 * that most setugid applications are not written with that
671 * environment in mind, and will therefore almost certainly operate
672 * incorrectly. In principle there's no reason that setugid
673 * applications might not be useful in capability mode, so we may want
674 * to reconsider this conservative design choice in the future.
676 * XXXMAC: For the time being, use NOSUID to also prohibit
677 * transitions on the file system.
679 credential_changing = 0;
680 credential_changing |= (attr.va_mode & S_ISUID) && oldcred->cr_uid !=
682 credential_changing |= (attr.va_mode & S_ISGID) && oldcred->cr_gid !=
685 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
686 interpvplabel, imgp);
687 credential_changing |= will_transition;
690 if (credential_changing &&
691 #ifdef CAPABILITY_MODE
692 ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
694 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
695 (p->p_flag & P_TRACED) == 0) {
697 * Turn off syscall tracing for set-id programs, except for
698 * root. Record any set-id flags first to make sure that
699 * we do not regain any tracing during a possible block.
704 if (p->p_tracecred != NULL &&
705 priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED, 0))
706 ktrprocexec(p, &tracecred, &tracevp);
709 * Close any file descriptors 0..2 that reference procfs,
710 * then make sure file descriptors 0..2 are in use.
712 * setugidsafety() may call closef() and then pfind()
713 * which may grab the process lock.
714 * fdcheckstd() may call falloc() which may block to
715 * allocate memory, so temporarily drop the process lock.
718 VOP_UNLOCK(imgp->vp, 0);
720 error = fdcheckstd(td);
723 newcred = crdup(oldcred);
724 euip = uifind(attr.va_uid);
725 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
728 * Set the new credentials.
730 if (attr.va_mode & S_ISUID)
731 change_euid(newcred, euip);
732 if (attr.va_mode & S_ISGID)
733 change_egid(newcred, attr.va_gid);
735 if (will_transition) {
736 mac_vnode_execve_transition(oldcred, newcred, imgp->vp,
737 interpvplabel, imgp);
741 * Implement correct POSIX saved-id behavior.
743 * XXXMAC: Note that the current logic will save the
744 * uid and gid if a MAC domain transition occurs, even
745 * though maybe it shouldn't.
747 change_svuid(newcred, newcred->cr_uid);
748 change_svgid(newcred, newcred->cr_gid);
749 p->p_ucred = newcred;
751 if (oldcred->cr_uid == oldcred->cr_ruid &&
752 oldcred->cr_gid == oldcred->cr_rgid)
753 p->p_flag &= ~P_SUGID;
755 * Implement correct POSIX saved-id behavior.
757 * XXX: It's not clear that the existing behavior is
758 * POSIX-compliant. A number of sources indicate that the
759 * saved uid/gid should only be updated if the new ruid is
760 * not equal to the old ruid, or the new euid is not equal
761 * to the old euid and the new euid is not equal to the old
762 * ruid. The FreeBSD code always updates the saved uid/gid.
763 * Also, this code uses the new (replaced) euid and egid as
764 * the source, which may or may not be the right ones to use.
766 if (oldcred->cr_svuid != oldcred->cr_uid ||
767 oldcred->cr_svgid != oldcred->cr_gid) {
769 VOP_UNLOCK(imgp->vp, 0);
770 newcred = crdup(oldcred);
771 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
773 change_svuid(newcred, newcred->cr_uid);
774 change_svgid(newcred, newcred->cr_gid);
775 p->p_ucred = newcred;
780 * Store the vp for use in procfs. This vnode was referenced prior
781 * to locking the proc lock.
783 textvp = p->p_textvp;
788 * Tell the DTrace fasttrap provider about the exec if it
789 * has declared an interest.
791 if (dtrace_fasttrap_exec)
792 dtrace_fasttrap_exec(p);
796 * Notify others that we exec'd, and clear the P_INEXEC flag
797 * as we're now a bona fide freshly-execed process.
799 KNOTE_LOCKED(&p->p_klist, NOTE_EXEC);
800 p->p_flag &= ~P_INEXEC;
802 /* clear "fork but no exec" flag, as we _are_ execing */
803 p->p_acflag &= ~AFORK;
806 * Free any previous argument cache and replace it with
807 * the new argument cache, if any.
815 * Check if system-wide sampling is in effect or if the
816 * current process is using PMCs. If so, do exec() time
817 * processing. This processing needs to happen AFTER the
818 * P_INEXEC flag is cleared.
820 * The proc lock needs to be released before taking the PMC
823 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
825 VOP_UNLOCK(imgp->vp, 0);
826 pe.pm_credentialschanged = credential_changing;
827 pe.pm_entryaddr = imgp->entry_addr;
829 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
830 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
833 #else /* !HWPMC_HOOKS */
837 /* Set values passed into the program in registers. */
838 if (p->p_sysent->sv_setregs)
839 (*p->p_sysent->sv_setregs)(td, imgp,
840 (u_long)(uintptr_t)stack_base);
842 exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base);
844 vfs_mark_atime(imgp->vp, td->td_ucred);
846 SDT_PROBE(proc, kernel, , exec__success, args->fname, 0, 0, 0, 0);
848 VOP_UNLOCK(imgp->vp, 0);
851 * Free any resources malloc'd earlier that we didn't use.
859 * Handle deferred decrement of ref counts.
863 if (binvp && error != 0)
868 if (tracecred != NULL)
871 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
874 if (oldsigacts != NULL)
875 sigacts_free(oldsigacts);
880 * free various allocated resources
882 if (imgp->firstpage != NULL)
883 exec_unmap_first_page(imgp);
885 if (imgp->vp != NULL) {
887 NDFREE(&nd, NDF_ONLY_PNBUF);
889 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
893 if (imgp->object != NULL)
894 vm_object_deallocate(imgp->object);
896 free(imgp->freepath, M_TEMP);
900 td->td_dbgflags |= TDB_EXEC;
904 * Stop the process here if its stop event mask has
905 * the S_EXEC bit set.
907 STOPEVENT(p, S_EXEC, 0);
912 /* we're done here, clear P_INEXEC */
914 p->p_flag &= ~P_INEXEC;
917 SDT_PROBE(proc, kernel, , exec__failure, error, 0, 0, 0, 0);
921 mac_execve_exit(imgp);
922 mac_execve_interpreter_exit(interpvplabel);
924 exec_free_args(args);
926 if (error && imgp->vmspace_destroyed) {
927 /* sorry, no more process anymore. exit gracefully */
928 exit1(td, W_EXITCODE(0, SIGABRT));
941 exec_map_first_page(imgp)
942 struct image_params *imgp;
946 vm_page_t ma[VM_INITIAL_PAGEIN];
949 if (imgp->firstpage != NULL)
950 exec_unmap_first_page(imgp);
952 object = imgp->vp->v_object;
955 VM_OBJECT_WLOCK(object);
956 #if VM_NRESERVLEVEL > 0
957 if ((object->flags & OBJ_COLORED) == 0) {
958 object->flags |= OBJ_COLORED;
959 object->pg_color = 0;
962 ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL);
963 if (ma[0]->valid != VM_PAGE_BITS_ALL) {
964 initial_pagein = VM_INITIAL_PAGEIN;
965 if (initial_pagein > object->size)
966 initial_pagein = object->size;
967 for (i = 1; i < initial_pagein; i++) {
968 if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) {
971 if (vm_page_tryxbusy(ma[i]))
974 ma[i] = vm_page_alloc(object, i,
975 VM_ALLOC_NORMAL | VM_ALLOC_IFNOTCACHED);
981 rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
982 ma[0] = vm_page_lookup(object, 0);
983 if ((rv != VM_PAGER_OK) || (ma[0] == NULL)) {
987 vm_page_unlock(ma[0]);
989 VM_OBJECT_WUNLOCK(object);
993 vm_page_xunbusy(ma[0]);
996 vm_page_unlock(ma[0]);
997 VM_OBJECT_WUNLOCK(object);
999 imgp->firstpage = sf_buf_alloc(ma[0], 0);
1000 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
1006 exec_unmap_first_page(imgp)
1007 struct image_params *imgp;
1011 if (imgp->firstpage != NULL) {
1012 m = sf_buf_page(imgp->firstpage);
1013 sf_buf_free(imgp->firstpage);
1014 imgp->firstpage = NULL;
1022 * Destroy old address space, and allocate a new stack
1023 * The new stack is only SGROWSIZ large because it is grown
1024 * automatically in trap.c.
1027 exec_new_vmspace(imgp, sv)
1028 struct image_params *imgp;
1029 struct sysentvec *sv;
1032 struct proc *p = imgp->proc;
1033 struct vmspace *vmspace = p->p_vmspace;
1035 vm_offset_t sv_minuser, stack_addr;
1039 imgp->vmspace_destroyed = 1;
1042 /* May be called with Giant held */
1043 EVENTHANDLER_INVOKE(process_exec, p, imgp);
1046 * Blow away entire process VM, if address space not shared,
1047 * otherwise, create a new VM space so that other threads are
1050 map = &vmspace->vm_map;
1052 sv_minuser = sv->sv_minuser;
1054 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1055 if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser &&
1056 vm_map_max(map) == sv->sv_maxuser) {
1058 pmap_remove_pages(vmspace_pmap(vmspace));
1059 vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1061 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1064 vmspace = p->p_vmspace;
1065 map = &vmspace->vm_map;
1068 /* Map a shared page */
1069 obj = sv->sv_shared_page_obj;
1071 vm_object_reference(obj);
1072 error = vm_map_fixed(map, obj, 0,
1073 sv->sv_shared_page_base, sv->sv_shared_page_len,
1074 VM_PROT_READ | VM_PROT_EXECUTE,
1075 VM_PROT_READ | VM_PROT_EXECUTE,
1076 MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1078 vm_object_deallocate(obj);
1083 /* Allocate a new stack */
1084 if (sv->sv_maxssiz != NULL)
1085 ssiz = *sv->sv_maxssiz;
1088 stack_addr = sv->sv_usrstack - ssiz;
1089 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1090 obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1092 VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1097 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they
1098 * are still used to enforce the stack rlimit on the process stack.
1100 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1101 vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - ssiz;
1107 * Copy out argument and environment strings from the old process address
1108 * space into the temporary string buffer.
1111 exec_copyin_args(struct image_args *args, char *fname,
1112 enum uio_seg segflg, char **argv, char **envv)
1118 bzero(args, sizeof(*args));
1123 * Allocate demand-paged memory for the file name, argument, and
1124 * environment strings.
1126 error = exec_alloc_args(args);
1131 * Copy the file name.
1133 if (fname != NULL) {
1134 args->fname = args->buf;
1135 error = (segflg == UIO_SYSSPACE) ?
1136 copystr(fname, args->fname, PATH_MAX, &length) :
1137 copyinstr(fname, args->fname, PATH_MAX, &length);
1143 args->begin_argv = args->buf + length;
1144 args->endp = args->begin_argv;
1145 args->stringspace = ARG_MAX;
1148 * extract arguments first
1150 while ((argp = (caddr_t) (intptr_t) fuword(argv++))) {
1151 if (argp == (caddr_t) -1) {
1155 if ((error = copyinstr(argp, args->endp,
1156 args->stringspace, &length))) {
1157 if (error == ENAMETOOLONG)
1161 args->stringspace -= length;
1162 args->endp += length;
1166 args->begin_envv = args->endp;
1169 * extract environment strings
1172 while ((envp = (caddr_t)(intptr_t)fuword(envv++))) {
1173 if (envp == (caddr_t)-1) {
1177 if ((error = copyinstr(envp, args->endp,
1178 args->stringspace, &length))) {
1179 if (error == ENAMETOOLONG)
1183 args->stringspace -= length;
1184 args->endp += length;
1192 exec_free_args(args);
1197 * Allocate temporary demand-paged, zero-filled memory for the file name,
1198 * argument, and environment strings. Returns zero if the allocation succeeds
1199 * and ENOMEM otherwise.
1202 exec_alloc_args(struct image_args *args)
1205 args->buf = (char *)kmap_alloc_wait(exec_map, PATH_MAX + ARG_MAX);
1206 return (args->buf != NULL ? 0 : ENOMEM);
1210 exec_free_args(struct image_args *args)
1213 if (args->buf != NULL) {
1214 kmap_free_wakeup(exec_map, (vm_offset_t)args->buf,
1215 PATH_MAX + ARG_MAX);
1218 if (args->fname_buf != NULL) {
1219 free(args->fname_buf, M_TEMP);
1220 args->fname_buf = NULL;
1225 * Copy strings out to the new process address space, constructing new arg
1226 * and env vector tables. Return a pointer to the base so that it can be used
1227 * as the initial stack pointer.
1230 exec_copyout_strings(imgp)
1231 struct image_params *imgp;
1237 register_t *stack_base;
1238 struct ps_strings *arginfo;
1240 size_t execpath_len;
1241 int szsigcode, szps;
1242 char canary[sizeof(long) * 8];
1244 szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1246 * Calculate string base and vector table pointers.
1247 * Also deal with signal trampoline code for this exec type.
1249 if (imgp->execpath != NULL && imgp->auxargs != NULL)
1250 execpath_len = strlen(imgp->execpath) + 1;
1255 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1256 if (p->p_sysent->sv_sigcode_base == 0) {
1257 if (p->p_sysent->sv_szsigcode != NULL)
1258 szsigcode = *(p->p_sysent->sv_szsigcode);
1260 destp = (uintptr_t)arginfo;
1265 if (szsigcode != 0) {
1267 destp = rounddown2(destp, sizeof(void *));
1268 copyout(p->p_sysent->sv_sigcode, (void *)destp, szsigcode);
1272 * Copy the image path for the rtld.
1274 if (execpath_len != 0) {
1275 destp -= execpath_len;
1276 imgp->execpathp = destp;
1277 copyout(imgp->execpath, (void *)destp, execpath_len);
1281 * Prepare the canary for SSP.
1283 arc4rand(canary, sizeof(canary), 0);
1284 destp -= sizeof(canary);
1285 imgp->canary = destp;
1286 copyout(canary, (void *)destp, sizeof(canary));
1287 imgp->canarylen = sizeof(canary);
1290 * Prepare the pagesizes array.
1293 destp = rounddown2(destp, sizeof(void *));
1294 imgp->pagesizes = destp;
1295 copyout(pagesizes, (void *)destp, szps);
1296 imgp->pagesizeslen = szps;
1298 destp -= ARG_MAX - imgp->args->stringspace;
1299 destp = rounddown2(destp, sizeof(void *));
1302 * If we have a valid auxargs ptr, prepare some room
1305 if (imgp->auxargs) {
1307 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1308 * lower compatibility.
1310 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1313 * The '+ 2' is for the null pointers at the end of each of
1314 * the arg and env vector sets,and imgp->auxarg_size is room
1315 * for argument of Runtime loader.
1317 vectp = (char **)(destp - (imgp->args->argc +
1318 imgp->args->envc + 2 + imgp->auxarg_size)
1322 * The '+ 2' is for the null pointers at the end of each of
1323 * the arg and env vector sets
1325 vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc
1326 + 2) * sizeof(char *));
1330 * vectp also becomes our initial stack base
1332 stack_base = (register_t *)vectp;
1334 stringp = imgp->args->begin_argv;
1335 argc = imgp->args->argc;
1336 envc = imgp->args->envc;
1339 * Copy out strings - arguments and environment.
1341 copyout(stringp, (void *)destp, ARG_MAX - imgp->args->stringspace);
1344 * Fill in "ps_strings" struct for ps, w, etc.
1346 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1347 suword32(&arginfo->ps_nargvstr, argc);
1350 * Fill in argument portion of vector table.
1352 for (; argc > 0; --argc) {
1353 suword(vectp++, (long)(intptr_t)destp);
1354 while (*stringp++ != 0)
1359 /* a null vector table pointer separates the argp's from the envp's */
1362 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1363 suword32(&arginfo->ps_nenvstr, envc);
1366 * Fill in environment portion of vector table.
1368 for (; envc > 0; --envc) {
1369 suword(vectp++, (long)(intptr_t)destp);
1370 while (*stringp++ != 0)
1375 /* end of vector table is a null pointer */
1378 return (stack_base);
1382 * Check permissions of file to execute.
1383 * Called with imgp->vp locked.
1384 * Return 0 for success or error code on failure.
1387 exec_check_permissions(imgp)
1388 struct image_params *imgp;
1390 struct vnode *vp = imgp->vp;
1391 struct vattr *attr = imgp->attr;
1393 int error, writecount;
1397 /* Get file attributes */
1398 error = VOP_GETATTR(vp, attr, td->td_ucred);
1403 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1409 * 1) Check if file execution is disabled for the filesystem that
1410 * this file resides on.
1411 * 2) Ensure that at least one execute bit is on. Otherwise, a
1412 * privileged user will always succeed, and we don't want this
1413 * to happen unless the file really is executable.
1414 * 3) Ensure that the file is a regular file.
1416 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1417 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1418 (attr->va_type != VREG))
1422 * Zero length files can't be exec'd
1424 if (attr->va_size == 0)
1428 * Check for execute permission to file based on current credentials.
1430 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1435 * Check number of open-for-writes on the file and deny execution
1438 error = VOP_GET_WRITECOUNT(vp, &writecount);
1441 if (writecount != 0)
1445 * Call filesystem specific open routine (which does nothing in the
1448 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1455 * Exec handler registration
1458 exec_register(execsw_arg)
1459 const struct execsw *execsw_arg;
1461 const struct execsw **es, **xs, **newexecsw;
1462 int count = 2; /* New slot and trailing NULL */
1465 for (es = execsw; *es; es++)
1467 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1468 if (newexecsw == NULL)
1472 for (es = execsw; *es; es++)
1477 free(execsw, M_TEMP);
1483 exec_unregister(execsw_arg)
1484 const struct execsw *execsw_arg;
1486 const struct execsw **es, **xs, **newexecsw;
1490 panic("unregister with no handlers left?\n");
1492 for (es = execsw; *es; es++) {
1493 if (*es == execsw_arg)
1498 for (es = execsw; *es; es++)
1499 if (*es != execsw_arg)
1501 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1502 if (newexecsw == NULL)
1505 for (es = execsw; *es; es++)
1506 if (*es != execsw_arg)
1510 free(execsw, M_TEMP);