2 * Copyright (c) 1993, David Greenman
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD$");
30 #include "opt_capsicum.h"
31 #include "opt_hwpmc_hooks.h"
32 #include "opt_kdtrace.h"
33 #include "opt_ktrace.h"
36 #include <sys/param.h>
37 #include <sys/capability.h>
38 #include <sys/systm.h>
39 #include <sys/capability.h>
40 #include <sys/eventhandler.h>
42 #include <sys/mutex.h>
43 #include <sys/sysproto.h>
44 #include <sys/signalvar.h>
45 #include <sys/kernel.h>
46 #include <sys/mount.h>
47 #include <sys/filedesc.h>
48 #include <sys/fcntl.h>
51 #include <sys/imgact.h>
52 #include <sys/imgact_elf.h>
54 #include <sys/malloc.h>
57 #include <sys/pioctl.h>
58 #include <sys/namei.h>
59 #include <sys/resourcevar.h>
60 #include <sys/sched.h>
62 #include <sys/sf_buf.h>
63 #include <sys/syscallsubr.h>
64 #include <sys/sysent.h>
66 #include <sys/sysctl.h>
67 #include <sys/vnode.h>
70 #include <sys/ktrace.h>
74 #include <vm/vm_param.h>
76 #include <vm/vm_page.h>
77 #include <vm/vm_map.h>
78 #include <vm/vm_kern.h>
79 #include <vm/vm_extern.h>
80 #include <vm/vm_object.h>
81 #include <vm/vm_pager.h>
84 #include <sys/pmckern.h>
87 #include <machine/reg.h>
89 #include <security/audit/audit.h>
90 #include <security/mac/mac_framework.h>
93 #include <sys/dtrace_bsd.h>
94 dtrace_execexit_func_t dtrace_fasttrap_exec;
97 SDT_PROVIDER_DECLARE(proc);
98 SDT_PROBE_DEFINE(proc, kernel, , exec, exec);
99 SDT_PROBE_ARGTYPE(proc, kernel, , exec, 0, "char *");
100 SDT_PROBE_DEFINE(proc, kernel, , exec_failure, exec-failure);
101 SDT_PROBE_ARGTYPE(proc, kernel, , exec_failure, 0, "int");
102 SDT_PROBE_DEFINE(proc, kernel, , exec_success, exec-success);
103 SDT_PROBE_ARGTYPE(proc, kernel, , exec_success, 0, "char *");
105 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
107 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
108 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
109 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
110 static int do_execve(struct thread *td, struct image_args *args,
113 /* XXX This should be vm_size_t. */
114 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD,
115 NULL, 0, sysctl_kern_ps_strings, "LU", "");
117 /* XXX This should be vm_size_t. */
118 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD|
119 CTLFLAG_CAPRD, NULL, 0, sysctl_kern_usrstack, "LU", "");
121 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD,
122 NULL, 0, sysctl_kern_stackprot, "I", "");
124 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
125 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
126 &ps_arg_cache_limit, 0, "");
128 static int map_at_zero = 0;
129 TUNABLE_INT("security.bsd.map_at_zero", &map_at_zero);
130 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RW, &map_at_zero, 0,
131 "Permit processes to map an object at virtual address 0.");
134 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
141 if (req->flags & SCTL_MASK32) {
143 val = (unsigned int)p->p_sysent->sv_psstrings;
144 error = SYSCTL_OUT(req, &val, sizeof(val));
147 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
148 sizeof(p->p_sysent->sv_psstrings));
153 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
160 if (req->flags & SCTL_MASK32) {
162 val = (unsigned int)p->p_sysent->sv_usrstack;
163 error = SYSCTL_OUT(req, &val, sizeof(val));
166 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
167 sizeof(p->p_sysent->sv_usrstack));
172 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
177 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
178 sizeof(p->p_sysent->sv_stackprot)));
182 * Each of the items is a pointer to a `const struct execsw', hence the
183 * double pointer here.
185 static const struct execsw **execsw;
187 #ifndef _SYS_SYSPROTO_H_
198 struct execve_args /* {
205 struct image_args args;
207 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
208 uap->argv, uap->envv);
210 error = kern_execve(td, &args, NULL);
214 #ifndef _SYS_SYSPROTO_H_
215 struct fexecve_args {
222 sys_fexecve(struct thread *td, struct fexecve_args *uap)
225 struct image_args args;
227 error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
228 uap->argv, uap->envv);
231 error = kern_execve(td, &args, NULL);
236 #ifndef _SYS_SYSPROTO_H_
237 struct __mac_execve_args {
246 sys___mac_execve(td, uap)
248 struct __mac_execve_args /* {
257 struct image_args args;
259 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
260 uap->argv, uap->envv);
262 error = kern_execve(td, &args, uap->mac_p);
270 * XXX: kern_execve has the astonishing property of not always returning to
271 * the caller. If sufficiently bad things happen during the call to
272 * do_execve(), it can end up calling exit1(); as a result, callers must
273 * avoid doing anything which they might need to undo (e.g., allocating
277 kern_execve(td, args, mac_p)
279 struct image_args *args;
282 struct proc *p = td->td_proc;
285 AUDIT_ARG_ARGV(args->begin_argv, args->argc,
286 args->begin_envv - args->begin_argv);
287 AUDIT_ARG_ENVV(args->begin_envv, args->envc,
288 args->endp - args->begin_envv);
289 if (p->p_flag & P_HADTHREADS) {
291 if (thread_single(SINGLE_BOUNDARY)) {
293 exec_free_args(args);
294 return (ERESTART); /* Try again later. */
299 error = do_execve(td, args, mac_p);
301 if (p->p_flag & P_HADTHREADS) {
304 * If success, we upgrade to SINGLE_EXIT state to
305 * force other threads to suicide.
308 thread_single(SINGLE_EXIT);
318 * In-kernel implementation of execve(). All arguments are assumed to be
319 * userspace pointers from the passed thread.
322 do_execve(td, args, mac_p)
324 struct image_args *args;
327 struct proc *p = td->td_proc;
329 struct ucred *newcred = NULL, *oldcred;
330 struct uidinfo *euip;
331 register_t *stack_base;
333 struct image_params image_params, *imgp;
335 int (*img_first)(struct image_params *);
336 struct pargs *oldargs = NULL, *newargs = NULL;
337 struct sigacts *oldsigacts, *newsigacts;
339 struct vnode *tracevp = NULL;
340 struct ucred *tracecred = NULL;
342 struct vnode *textvp = NULL, *binvp = NULL;
343 int credential_changing;
347 struct label *interpvplabel = NULL;
351 struct pmckern_procexec pe;
353 static const char fexecv_proc_title[] = "(fexecv)";
356 imgp = &image_params;
359 * Lock the process and set the P_INEXEC flag to indicate that
360 * it should be left alone until we're done here. This is
361 * necessary to avoid race conditions - e.g. in ptrace() -
362 * that might allow a local user to illicitly obtain elevated
366 KASSERT((p->p_flag & P_INEXEC) == 0,
367 ("%s(): process already has P_INEXEC flag", __func__));
368 p->p_flag |= P_INEXEC;
372 * Initialize part of the common data
375 imgp->execlabel = NULL;
377 imgp->entry_addr = 0;
378 imgp->reloc_base = 0;
379 imgp->vmspace_destroyed = 0;
380 imgp->interpreted = 0;
382 imgp->interpreter_name = NULL;
383 imgp->auxargs = NULL;
386 imgp->firstpage = NULL;
387 imgp->ps_strings = 0;
388 imgp->auxarg_size = 0;
390 imgp->execpath = imgp->freepath = NULL;
395 imgp->pagesizeslen = 0;
396 imgp->stack_prot = 0;
399 error = mac_execve_enter(imgp, mac_p);
404 imgp->image_header = NULL;
407 * Translate the file name. namei() returns a vnode pointer
408 * in ni_vp amoung other things.
410 * XXXAUDIT: It would be desirable to also audit the name of the
411 * interpreter if this is an interpreted binary.
413 if (args->fname != NULL) {
414 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME
415 | MPSAFE | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
418 SDT_PROBE(proc, kernel, , exec, args->fname, 0, 0, 0, 0 );
421 if (args->fname != NULL) {
422 #ifdef CAPABILITY_MODE
424 * While capability mode can't reach this point via direct
425 * path arguments to execve(), we also don't allow
426 * interpreters to be used in capability mode (for now).
427 * Catch indirect lookups and return a permissions error.
429 if (IN_CAPABILITY_MODE(td)) {
438 vfslocked = NDHASGIANT(&nd);
442 AUDIT_ARG_FD(args->fd);
444 * Some might argue that CAP_READ and/or CAP_MMAP should also
445 * be required here; such arguments will be entertained.
447 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
449 error = fgetvp_exec(td, args->fd, CAP_FEXECVE, &binvp);
452 vfslocked = VFS_LOCK_GIANT(binvp->v_mount);
453 vn_lock(binvp, LK_EXCLUSIVE | LK_RETRY);
454 AUDIT_ARG_VNODE1(binvp);
459 * Check file permissions (also 'opens' file)
461 error = exec_check_permissions(imgp);
463 goto exec_fail_dealloc;
465 imgp->object = imgp->vp->v_object;
466 if (imgp->object != NULL)
467 vm_object_reference(imgp->object);
470 * Set VV_TEXT now so no one can write to the executable while we're
473 * Remember if this was set before and unset it in case this is not
474 * actually an executable image.
476 textset = imgp->vp->v_vflag & VV_TEXT;
477 ASSERT_VOP_ELOCKED(imgp->vp, "vv_text");
478 imgp->vp->v_vflag |= VV_TEXT;
480 error = exec_map_first_page(imgp);
482 goto exec_fail_dealloc;
484 imgp->proc->p_osrel = 0;
486 * If the current process has a special image activator it
487 * wants to try first, call it. For example, emulating shell
488 * scripts differently.
491 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
492 error = img_first(imgp);
495 * Loop through the list of image activators, calling each one.
496 * An activator returns -1 if there is no match, 0 on success,
497 * and an error otherwise.
499 for (i = 0; error == -1 && execsw[i]; ++i) {
500 if (execsw[i]->ex_imgact == NULL ||
501 execsw[i]->ex_imgact == img_first) {
504 error = (*execsw[i]->ex_imgact)(imgp);
510 ASSERT_VOP_ELOCKED(imgp->vp, "vv_text");
511 imgp->vp->v_vflag &= ~VV_TEXT;
515 goto exec_fail_dealloc;
519 * Special interpreter operation, cleanup and loop up to try to
520 * activate the interpreter.
522 if (imgp->interpreted) {
523 exec_unmap_first_page(imgp);
525 * VV_TEXT needs to be unset for scripts. There is a short
526 * period before we determine that something is a script where
527 * VV_TEXT will be set. The vnode lock is held over this
528 * entire period so nothing should illegitimately be blocked.
530 imgp->vp->v_vflag &= ~VV_TEXT;
531 /* free name buffer and old vnode */
532 if (args->fname != NULL)
533 NDFREE(&nd, NDF_ONLY_PNBUF);
535 mac_execve_interpreter_enter(binvp, &interpvplabel);
538 VOP_CLOSE(binvp, FREAD, td->td_ucred, td);
542 vm_object_deallocate(imgp->object);
544 VFS_UNLOCK_GIANT(vfslocked);
546 /* set new name to that of the interpreter */
547 NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME | MPSAFE,
548 UIO_SYSSPACE, imgp->interpreter_name, td);
549 args->fname = imgp->interpreter_name;
554 * NB: We unlock the vnode here because it is believed that none
555 * of the sv_copyout_strings/sv_fixup operations require the vnode.
557 VOP_UNLOCK(imgp->vp, 0);
560 * Do the best to calculate the full path to the image file.
562 if (imgp->auxargs != NULL &&
563 ((args->fname != NULL && args->fname[0] == '/') ||
564 vn_fullpath(td, imgp->vp, &imgp->execpath, &imgp->freepath) != 0))
565 imgp->execpath = args->fname;
568 * Copy out strings (args and env) and initialize stack base
570 if (p->p_sysent->sv_copyout_strings)
571 stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
573 stack_base = exec_copyout_strings(imgp);
576 * If custom stack fixup routine present for this process
577 * let it do the stack setup.
578 * Else stuff argument count as first item on stack
580 if (p->p_sysent->sv_fixup != NULL)
581 (*p->p_sysent->sv_fixup)(&stack_base, imgp);
583 suword(--stack_base, imgp->args->argc);
586 * For security and other reasons, the file descriptor table cannot
587 * be shared after an exec.
592 * Malloc things before we need locks.
595 euip = uifind(attr.va_uid);
596 i = imgp->args->begin_envv - imgp->args->begin_argv;
597 /* Cache arguments if they fit inside our allowance */
598 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
599 newargs = pargs_alloc(i);
600 bcopy(imgp->args->begin_argv, newargs->ar_args, i);
603 /* close files on exec */
605 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
607 /* Get a reference to the vnode prior to locking the proc */
611 * For security and other reasons, signal handlers cannot
612 * be shared after an exec. The new process gets a copy of the old
613 * handlers. In execsigs(), the new process will have its signals
617 oldcred = crcopysafe(p, newcred);
618 if (sigacts_shared(p->p_sigacts)) {
619 oldsigacts = p->p_sigacts;
621 newsigacts = sigacts_alloc();
622 sigacts_copy(newsigacts, oldsigacts);
624 p->p_sigacts = newsigacts;
631 /* reset caught signals */
634 /* name this process - nameiexec(p, ndp) */
635 bzero(p->p_comm, sizeof(p->p_comm));
637 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
638 min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
639 else if (vn_commname(binvp, p->p_comm, sizeof(p->p_comm)) != 0)
640 bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
641 bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
643 sched_clear_tdname(td);
647 * mark as execed, wakeup the process that vforked (if any) and tell
648 * it that it now has its own resources back
651 if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
652 p->p_flag &= ~P_PPWAIT;
653 cv_broadcast(&p->p_pwait);
657 * Implement image setuid/setgid.
659 * Don't honor setuid/setgid if the filesystem prohibits it or if
660 * the process is being traced.
662 * We disable setuid/setgid/etc in compatibility mode on the basis
663 * that most setugid applications are not written with that
664 * environment in mind, and will therefore almost certainly operate
665 * incorrectly. In principle there's no reason that setugid
666 * applications might not be useful in capability mode, so we may want
667 * to reconsider this conservative design choice in the future.
669 * XXXMAC: For the time being, use NOSUID to also prohibit
670 * transitions on the file system.
672 credential_changing = 0;
673 credential_changing |= (attr.va_mode & S_ISUID) && oldcred->cr_uid !=
675 credential_changing |= (attr.va_mode & S_ISGID) && oldcred->cr_gid !=
678 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
679 interpvplabel, imgp);
680 credential_changing |= will_transition;
683 if (credential_changing &&
684 #ifdef CAPABILITY_MODE
685 ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
687 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
688 (p->p_flag & P_TRACED) == 0) {
690 * Turn off syscall tracing for set-id programs, except for
691 * root. Record any set-id flags first to make sure that
692 * we do not regain any tracing during a possible block.
697 if (priv_check_cred(oldcred, PRIV_DEBUG_DIFFCRED, 0))
698 ktrprocexec(p, &tracecred, &tracevp);
701 * Close any file descriptors 0..2 that reference procfs,
702 * then make sure file descriptors 0..2 are in use.
704 * setugidsafety() may call closef() and then pfind()
705 * which may grab the process lock.
706 * fdcheckstd() may call falloc() which may block to
707 * allocate memory, so temporarily drop the process lock.
710 VOP_UNLOCK(imgp->vp, 0);
712 error = fdcheckstd(td);
713 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
718 * Set the new credentials.
720 if (attr.va_mode & S_ISUID)
721 change_euid(newcred, euip);
722 if (attr.va_mode & S_ISGID)
723 change_egid(newcred, attr.va_gid);
725 if (will_transition) {
726 mac_vnode_execve_transition(oldcred, newcred, imgp->vp,
727 interpvplabel, imgp);
731 * Implement correct POSIX saved-id behavior.
733 * XXXMAC: Note that the current logic will save the
734 * uid and gid if a MAC domain transition occurs, even
735 * though maybe it shouldn't.
737 change_svuid(newcred, newcred->cr_uid);
738 change_svgid(newcred, newcred->cr_gid);
739 p->p_ucred = newcred;
742 if (oldcred->cr_uid == oldcred->cr_ruid &&
743 oldcred->cr_gid == oldcred->cr_rgid)
744 p->p_flag &= ~P_SUGID;
746 * Implement correct POSIX saved-id behavior.
748 * XXX: It's not clear that the existing behavior is
749 * POSIX-compliant. A number of sources indicate that the
750 * saved uid/gid should only be updated if the new ruid is
751 * not equal to the old ruid, or the new euid is not equal
752 * to the old euid and the new euid is not equal to the old
753 * ruid. The FreeBSD code always updates the saved uid/gid.
754 * Also, this code uses the new (replaced) euid and egid as
755 * the source, which may or may not be the right ones to use.
757 if (oldcred->cr_svuid != oldcred->cr_uid ||
758 oldcred->cr_svgid != oldcred->cr_gid) {
759 change_svuid(newcred, newcred->cr_uid);
760 change_svgid(newcred, newcred->cr_gid);
761 p->p_ucred = newcred;
767 * Store the vp for use in procfs. This vnode was referenced prior
768 * to locking the proc lock.
770 textvp = p->p_textvp;
775 * Tell the DTrace fasttrap provider about the exec if it
776 * has declared an interest.
778 if (dtrace_fasttrap_exec)
779 dtrace_fasttrap_exec(p);
783 * Notify others that we exec'd, and clear the P_INEXEC flag
784 * as we're now a bona fide freshly-execed process.
786 KNOTE_LOCKED(&p->p_klist, NOTE_EXEC);
787 p->p_flag &= ~P_INEXEC;
789 /* clear "fork but no exec" flag, as we _are_ execing */
790 p->p_acflag &= ~AFORK;
793 * Free any previous argument cache and replace it with
794 * the new argument cache, if any.
802 * Check if system-wide sampling is in effect or if the
803 * current process is using PMCs. If so, do exec() time
804 * processing. This processing needs to happen AFTER the
805 * P_INEXEC flag is cleared.
807 * The proc lock needs to be released before taking the PMC
810 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
812 VOP_UNLOCK(imgp->vp, 0);
813 pe.pm_credentialschanged = credential_changing;
814 pe.pm_entryaddr = imgp->entry_addr;
816 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
817 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
820 #else /* !HWPMC_HOOKS */
824 /* Set values passed into the program in registers. */
825 if (p->p_sysent->sv_setregs)
826 (*p->p_sysent->sv_setregs)(td, imgp,
827 (u_long)(uintptr_t)stack_base);
829 exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base);
831 vfs_mark_atime(imgp->vp, td->td_ucred);
833 SDT_PROBE(proc, kernel, , exec_success, args->fname, 0, 0, 0, 0);
837 * Free any resources malloc'd earlier that we didn't use.
844 VOP_UNLOCK(imgp->vp, 0);
847 * Handle deferred decrement of ref counts.
849 if (textvp != NULL) {
852 tvfslocked = VFS_LOCK_GIANT(textvp->v_mount);
854 VFS_UNLOCK_GIANT(tvfslocked);
856 if (binvp && error != 0)
859 if (tracevp != NULL) {
862 tvfslocked = VFS_LOCK_GIANT(tracevp->v_mount);
864 VFS_UNLOCK_GIANT(tvfslocked);
866 if (tracecred != NULL)
869 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
872 if (oldsigacts != NULL)
873 sigacts_free(oldsigacts);
878 * free various allocated resources
880 if (imgp->firstpage != NULL)
881 exec_unmap_first_page(imgp);
883 if (imgp->vp != NULL) {
885 NDFREE(&nd, NDF_ONLY_PNBUF);
887 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
891 if (imgp->object != NULL)
892 vm_object_deallocate(imgp->object);
894 free(imgp->freepath, M_TEMP);
898 td->td_dbgflags |= TDB_EXEC;
902 * Stop the process here if its stop event mask has
903 * the S_EXEC bit set.
905 STOPEVENT(p, S_EXEC, 0);
910 /* we're done here, clear P_INEXEC */
912 p->p_flag &= ~P_INEXEC;
915 SDT_PROBE(proc, kernel, , exec_failure, error, 0, 0, 0, 0);
919 mac_execve_exit(imgp);
920 mac_execve_interpreter_exit(interpvplabel);
922 VFS_UNLOCK_GIANT(vfslocked);
923 exec_free_args(args);
925 if (error && imgp->vmspace_destroyed) {
926 /* sorry, no more process anymore. exit gracefully */
927 exit1(td, W_EXITCODE(0, SIGABRT));
940 exec_map_first_page(imgp)
941 struct image_params *imgp;
945 vm_page_t ma[VM_INITIAL_PAGEIN];
948 if (imgp->firstpage != NULL)
949 exec_unmap_first_page(imgp);
951 object = imgp->vp->v_object;
954 VM_OBJECT_LOCK(object);
955 #if VM_NRESERVLEVEL > 0
956 if ((object->flags & OBJ_COLORED) == 0) {
957 object->flags |= OBJ_COLORED;
958 object->pg_color = 0;
961 ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
962 if (ma[0]->valid != VM_PAGE_BITS_ALL) {
963 initial_pagein = VM_INITIAL_PAGEIN;
964 if (initial_pagein > object->size)
965 initial_pagein = object->size;
966 for (i = 1; i < initial_pagein; i++) {
967 if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) {
970 if ((ma[i]->oflags & VPO_BUSY) || ma[i]->busy)
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_UNLOCK(object);
995 vm_page_unlock(ma[0]);
996 vm_page_wakeup(ma[0]);
997 VM_OBJECT_UNLOCK(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, VM_PROT_ALL,
1075 MAP_COPY_ON_WRITE | MAP_ACC_NO_CHARGE);
1077 vm_object_deallocate(obj);
1082 /* Allocate a new stack */
1083 if (sv->sv_maxssiz != NULL)
1084 ssiz = *sv->sv_maxssiz;
1087 stack_addr = sv->sv_usrstack - ssiz;
1088 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1089 obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1091 VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1096 /* Allocate a new register stack */
1097 stack_addr = IA64_BACKINGSTORE;
1098 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1099 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_UP);
1104 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
1105 * VM_STACK case, but they are still used to monitor the size of the
1106 * process stack so we can check the stack rlimit.
1108 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1109 vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - ssiz;
1115 * Copy out argument and environment strings from the old process address
1116 * space into the temporary string buffer.
1119 exec_copyin_args(struct image_args *args, char *fname,
1120 enum uio_seg segflg, char **argv, char **envv)
1126 bzero(args, sizeof(*args));
1131 * Allocate demand-paged memory for the file name, argument, and
1132 * environment strings.
1134 error = exec_alloc_args(args);
1139 * Copy the file name.
1141 if (fname != NULL) {
1142 args->fname = args->buf;
1143 error = (segflg == UIO_SYSSPACE) ?
1144 copystr(fname, args->fname, PATH_MAX, &length) :
1145 copyinstr(fname, args->fname, PATH_MAX, &length);
1151 args->begin_argv = args->buf + length;
1152 args->endp = args->begin_argv;
1153 args->stringspace = ARG_MAX;
1156 * extract arguments first
1158 while ((argp = (caddr_t) (intptr_t) fuword(argv++))) {
1159 if (argp == (caddr_t) -1) {
1163 if ((error = copyinstr(argp, args->endp,
1164 args->stringspace, &length))) {
1165 if (error == ENAMETOOLONG)
1169 args->stringspace -= length;
1170 args->endp += length;
1174 args->begin_envv = args->endp;
1177 * extract environment strings
1180 while ((envp = (caddr_t)(intptr_t)fuword(envv++))) {
1181 if (envp == (caddr_t)-1) {
1185 if ((error = copyinstr(envp, args->endp,
1186 args->stringspace, &length))) {
1187 if (error == ENAMETOOLONG)
1191 args->stringspace -= length;
1192 args->endp += length;
1200 exec_free_args(args);
1205 * Allocate temporary demand-paged, zero-filled memory for the file name,
1206 * argument, and environment strings. Returns zero if the allocation succeeds
1207 * and ENOMEM otherwise.
1210 exec_alloc_args(struct image_args *args)
1213 args->buf = (char *)kmem_alloc_wait(exec_map, PATH_MAX + ARG_MAX);
1214 return (args->buf != NULL ? 0 : ENOMEM);
1218 exec_free_args(struct image_args *args)
1221 if (args->buf != NULL) {
1222 kmem_free_wakeup(exec_map, (vm_offset_t)args->buf,
1223 PATH_MAX + ARG_MAX);
1226 if (args->fname_buf != NULL) {
1227 free(args->fname_buf, M_TEMP);
1228 args->fname_buf = NULL;
1233 * Copy strings out to the new process address space, constructing new arg
1234 * and env vector tables. Return a pointer to the base so that it can be used
1235 * as the initial stack pointer.
1238 exec_copyout_strings(imgp)
1239 struct image_params *imgp;
1243 char *stringp, *destp;
1244 register_t *stack_base;
1245 struct ps_strings *arginfo;
1247 size_t execpath_len;
1248 int szsigcode, szps;
1249 char canary[sizeof(long) * 8];
1251 szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1253 * Calculate string base and vector table pointers.
1254 * Also deal with signal trampoline code for this exec type.
1256 if (imgp->execpath != NULL && imgp->auxargs != NULL)
1257 execpath_len = strlen(imgp->execpath) + 1;
1262 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1263 if (p->p_sysent->sv_sigcode_base == 0) {
1264 if (p->p_sysent->sv_szsigcode != NULL)
1265 szsigcode = *(p->p_sysent->sv_szsigcode);
1267 destp = (caddr_t)arginfo - szsigcode - SPARE_USRSPACE -
1268 roundup(execpath_len, sizeof(char *)) -
1269 roundup(sizeof(canary), sizeof(char *)) -
1270 roundup(szps, sizeof(char *)) -
1271 roundup((ARG_MAX - imgp->args->stringspace), sizeof(char *));
1277 copyout(p->p_sysent->sv_sigcode, ((caddr_t)arginfo -
1278 szsigcode), szsigcode);
1281 * Copy the image path for the rtld.
1283 if (execpath_len != 0) {
1284 imgp->execpathp = (uintptr_t)arginfo - szsigcode - execpath_len;
1285 copyout(imgp->execpath, (void *)imgp->execpathp,
1290 * Prepare the canary for SSP.
1292 arc4rand(canary, sizeof(canary), 0);
1293 imgp->canary = (uintptr_t)arginfo - szsigcode - execpath_len -
1295 copyout(canary, (void *)imgp->canary, sizeof(canary));
1296 imgp->canarylen = sizeof(canary);
1299 * Prepare the pagesizes array.
1301 imgp->pagesizes = (uintptr_t)arginfo - szsigcode - execpath_len -
1302 roundup(sizeof(canary), sizeof(char *)) - szps;
1303 copyout(pagesizes, (void *)imgp->pagesizes, szps);
1304 imgp->pagesizeslen = szps;
1307 * If we have a valid auxargs ptr, prepare some room
1310 if (imgp->auxargs) {
1312 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1313 * lower compatibility.
1315 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1318 * The '+ 2' is for the null pointers at the end of each of
1319 * the arg and env vector sets,and imgp->auxarg_size is room
1320 * for argument of Runtime loader.
1322 vectp = (char **)(destp - (imgp->args->argc +
1323 imgp->args->envc + 2 + imgp->auxarg_size)
1327 * The '+ 2' is for the null pointers at the end of each of
1328 * the arg and env vector sets
1330 vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc + 2) *
1335 * vectp also becomes our initial stack base
1337 stack_base = (register_t *)vectp;
1339 stringp = imgp->args->begin_argv;
1340 argc = imgp->args->argc;
1341 envc = imgp->args->envc;
1344 * Copy out strings - arguments and environment.
1346 copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
1349 * Fill in "ps_strings" struct for ps, w, etc.
1351 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1352 suword32(&arginfo->ps_nargvstr, argc);
1355 * Fill in argument portion of vector table.
1357 for (; argc > 0; --argc) {
1358 suword(vectp++, (long)(intptr_t)destp);
1359 while (*stringp++ != 0)
1364 /* a null vector table pointer separates the argp's from the envp's */
1367 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1368 suword32(&arginfo->ps_nenvstr, envc);
1371 * Fill in environment portion of vector table.
1373 for (; envc > 0; --envc) {
1374 suword(vectp++, (long)(intptr_t)destp);
1375 while (*stringp++ != 0)
1380 /* end of vector table is a null pointer */
1383 return (stack_base);
1387 * Check permissions of file to execute.
1388 * Called with imgp->vp locked.
1389 * Return 0 for success or error code on failure.
1392 exec_check_permissions(imgp)
1393 struct image_params *imgp;
1395 struct vnode *vp = imgp->vp;
1396 struct vattr *attr = imgp->attr;
1402 /* Get file attributes */
1403 error = VOP_GETATTR(vp, attr, td->td_ucred);
1408 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1414 * 1) Check if file execution is disabled for the filesystem that
1415 * this file resides on.
1416 * 2) Ensure that at least one execute bit is on. Otherwise, a
1417 * privileged user will always succeed, and we don't want this
1418 * to happen unless the file really is executable.
1419 * 3) Ensure that the file is a regular file.
1421 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1422 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1423 (attr->va_type != VREG))
1427 * Zero length files can't be exec'd
1429 if (attr->va_size == 0)
1433 * Check for execute permission to file based on current credentials.
1435 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1440 * Check number of open-for-writes on the file and deny execution
1443 if (vp->v_writecount)
1447 * Call filesystem specific open routine (which does nothing in the
1450 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1457 * Exec handler registration
1460 exec_register(execsw_arg)
1461 const struct execsw *execsw_arg;
1463 const struct execsw **es, **xs, **newexecsw;
1464 int count = 2; /* New slot and trailing NULL */
1467 for (es = execsw; *es; es++)
1469 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1470 if (newexecsw == NULL)
1474 for (es = execsw; *es; es++)
1479 free(execsw, M_TEMP);
1485 exec_unregister(execsw_arg)
1486 const struct execsw *execsw_arg;
1488 const struct execsw **es, **xs, **newexecsw;
1492 panic("unregister with no handlers left?\n");
1494 for (es = execsw; *es; es++) {
1495 if (*es == execsw_arg)
1500 for (es = execsw; *es; es++)
1501 if (*es != execsw_arg)
1503 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1504 if (newexecsw == NULL)
1507 for (es = execsw; *es; es++)
1508 if (*es != execsw_arg)
1512 free(execsw, M_TEMP);
1517 static vm_object_t shared_page_obj;
1518 static int shared_page_free;
1521 shared_page_fill(int size, int align, const char *data)
1528 VM_OBJECT_LOCK(shared_page_obj);
1529 m = vm_page_grab(shared_page_obj, 0, VM_ALLOC_RETRY);
1530 res = roundup(shared_page_free, align);
1531 if (res + size >= IDX_TO_OFF(shared_page_obj->size))
1534 VM_OBJECT_UNLOCK(shared_page_obj);
1535 s = sf_buf_alloc(m, SFB_DEFAULT);
1537 bcopy(data, (void *)(sk + res), size);
1538 shared_page_free = res + size;
1540 VM_OBJECT_LOCK(shared_page_obj);
1543 VM_OBJECT_UNLOCK(shared_page_obj);
1548 shared_page_init(void *dummy __unused)
1552 shared_page_obj = vm_pager_allocate(OBJT_PHYS, 0, PAGE_SIZE,
1553 VM_PROT_DEFAULT, 0, NULL);
1554 VM_OBJECT_LOCK(shared_page_obj);
1555 m = vm_page_grab(shared_page_obj, 0, VM_ALLOC_RETRY | VM_ALLOC_NOBUSY |
1557 m->valid = VM_PAGE_BITS_ALL;
1558 VM_OBJECT_UNLOCK(shared_page_obj);
1561 SYSINIT(shp, SI_SUB_EXEC, SI_ORDER_FIRST, (sysinit_cfunc_t)shared_page_init,
1565 exec_sysvec_init(void *param)
1567 struct sysentvec *sv;
1569 sv = (struct sysentvec *)param;
1571 if ((sv->sv_flags & SV_SHP) == 0)
1573 sv->sv_shared_page_obj = shared_page_obj;
1574 sv->sv_sigcode_base = sv->sv_shared_page_base +
1575 shared_page_fill(*(sv->sv_szsigcode), 16, sv->sv_sigcode);