2 * Copyright (c) 1993, David Greenman
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD$");
30 #include "opt_capsicum.h"
31 #include "opt_hwpmc_hooks.h"
32 #include "opt_kdtrace.h"
33 #include "opt_ktrace.h"
36 #include <sys/param.h>
37 #include <sys/capability.h>
38 #include <sys/systm.h>
39 #include <sys/capability.h>
40 #include <sys/eventhandler.h>
42 #include <sys/mutex.h>
43 #include <sys/sysproto.h>
44 #include <sys/signalvar.h>
45 #include <sys/kernel.h>
46 #include <sys/mount.h>
47 #include <sys/filedesc.h>
48 #include <sys/fcntl.h>
51 #include <sys/imgact.h>
52 #include <sys/imgact_elf.h>
54 #include <sys/malloc.h>
57 #include <sys/pioctl.h>
58 #include <sys/namei.h>
59 #include <sys/resourcevar.h>
60 #include <sys/rwlock.h>
61 #include <sys/sched.h>
63 #include <sys/sf_buf.h>
64 #include <sys/syscallsubr.h>
65 #include <sys/sysent.h>
67 #include <sys/sysctl.h>
68 #include <sys/vnode.h>
71 #include <sys/ktrace.h>
75 #include <vm/vm_param.h>
77 #include <vm/vm_page.h>
78 #include <vm/vm_map.h>
79 #include <vm/vm_kern.h>
80 #include <vm/vm_extern.h>
81 #include <vm/vm_object.h>
82 #include <vm/vm_pager.h>
85 #include <sys/pmckern.h>
88 #include <machine/reg.h>
90 #include <security/audit/audit.h>
91 #include <security/mac/mac_framework.h>
94 #include <sys/dtrace_bsd.h>
95 dtrace_execexit_func_t dtrace_fasttrap_exec;
98 SDT_PROVIDER_DECLARE(proc);
99 SDT_PROBE_DEFINE(proc, kernel, , exec, exec);
100 SDT_PROBE_ARGTYPE(proc, kernel, , exec, 0, "char *");
101 SDT_PROBE_DEFINE(proc, kernel, , exec_failure, exec-failure);
102 SDT_PROBE_ARGTYPE(proc, kernel, , exec_failure, 0, "int");
103 SDT_PROBE_DEFINE(proc, kernel, , exec_success, exec-success);
104 SDT_PROBE_ARGTYPE(proc, kernel, , exec_success, 0, "char *");
106 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
108 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
109 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
110 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
111 static int do_execve(struct thread *td, struct image_args *args,
114 /* XXX This should be vm_size_t. */
115 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD,
116 NULL, 0, sysctl_kern_ps_strings, "LU", "");
118 /* XXX This should be vm_size_t. */
119 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD|
120 CTLFLAG_CAPRD, NULL, 0, sysctl_kern_usrstack, "LU", "");
122 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD,
123 NULL, 0, sysctl_kern_stackprot, "I", "");
125 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
126 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
127 &ps_arg_cache_limit, 0, "");
129 static int map_at_zero = 0;
130 TUNABLE_INT("security.bsd.map_at_zero", &map_at_zero);
131 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RW, &map_at_zero, 0,
132 "Permit processes to map an object at virtual address 0.");
135 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
142 if (req->flags & SCTL_MASK32) {
144 val = (unsigned int)p->p_sysent->sv_psstrings;
145 error = SYSCTL_OUT(req, &val, sizeof(val));
148 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
149 sizeof(p->p_sysent->sv_psstrings));
154 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
161 if (req->flags & SCTL_MASK32) {
163 val = (unsigned int)p->p_sysent->sv_usrstack;
164 error = SYSCTL_OUT(req, &val, sizeof(val));
167 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
168 sizeof(p->p_sysent->sv_usrstack));
173 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
178 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
179 sizeof(p->p_sysent->sv_stackprot)));
183 * Each of the items is a pointer to a `const struct execsw', hence the
184 * double pointer here.
186 static const struct execsw **execsw;
188 #ifndef _SYS_SYSPROTO_H_
199 struct execve_args /* {
206 struct image_args args;
208 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
209 uap->argv, uap->envv);
211 error = kern_execve(td, &args, NULL);
215 #ifndef _SYS_SYSPROTO_H_
216 struct fexecve_args {
223 sys_fexecve(struct thread *td, struct fexecve_args *uap)
226 struct image_args args;
228 error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
229 uap->argv, uap->envv);
232 error = kern_execve(td, &args, NULL);
237 #ifndef _SYS_SYSPROTO_H_
238 struct __mac_execve_args {
247 sys___mac_execve(td, uap)
249 struct __mac_execve_args /* {
258 struct image_args args;
260 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
261 uap->argv, uap->envv);
263 error = kern_execve(td, &args, uap->mac_p);
271 * XXX: kern_execve has the astonishing property of not always returning to
272 * the caller. If sufficiently bad things happen during the call to
273 * do_execve(), it can end up calling exit1(); as a result, callers must
274 * avoid doing anything which they might need to undo (e.g., allocating
278 kern_execve(td, args, mac_p)
280 struct image_args *args;
283 struct proc *p = td->td_proc;
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 error = do_execve(td, args, mac_p);
302 if (p->p_flag & P_HADTHREADS) {
305 * If success, we upgrade to SINGLE_EXIT state to
306 * force other threads to suicide.
309 thread_single(SINGLE_EXIT);
319 * In-kernel implementation of execve(). All arguments are assumed to be
320 * userspace pointers from the passed thread.
323 do_execve(td, args, mac_p)
325 struct image_args *args;
328 struct proc *p = td->td_proc;
330 struct ucred *newcred = NULL, *oldcred;
331 struct uidinfo *euip;
332 register_t *stack_base;
334 struct image_params image_params, *imgp;
336 int (*img_first)(struct image_params *);
337 struct pargs *oldargs = NULL, *newargs = NULL;
338 struct sigacts *oldsigacts, *newsigacts;
340 struct vnode *tracevp = NULL;
341 struct ucred *tracecred = NULL;
343 struct vnode *textvp = NULL, *binvp = NULL;
344 int credential_changing;
347 struct label *interpvplabel = NULL;
351 struct pmckern_procexec pe;
353 static const char fexecv_proc_title[] = "(fexecv)";
355 imgp = &image_params;
358 * Lock the process and set the P_INEXEC flag to indicate that
359 * it should be left alone until we're done here. This is
360 * necessary to avoid race conditions - e.g. in ptrace() -
361 * that might allow a local user to illicitly obtain elevated
365 KASSERT((p->p_flag & P_INEXEC) == 0,
366 ("%s(): process already has P_INEXEC flag", __func__));
367 p->p_flag |= P_INEXEC;
371 * Initialize part of the common data
374 imgp->execlabel = NULL;
376 imgp->entry_addr = 0;
377 imgp->reloc_base = 0;
378 imgp->vmspace_destroyed = 0;
379 imgp->interpreted = 0;
381 imgp->interpreter_name = NULL;
382 imgp->auxargs = NULL;
385 imgp->firstpage = NULL;
386 imgp->ps_strings = 0;
387 imgp->auxarg_size = 0;
389 imgp->execpath = imgp->freepath = NULL;
394 imgp->pagesizeslen = 0;
395 imgp->stack_prot = 0;
398 error = mac_execve_enter(imgp, mac_p);
403 imgp->image_header = NULL;
406 * Translate the file name. namei() returns a vnode pointer
407 * in ni_vp amoung other things.
409 * XXXAUDIT: It would be desirable to also audit the name of the
410 * interpreter if this is an interpreted binary.
412 if (args->fname != NULL) {
413 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME
414 | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
417 SDT_PROBE(proc, kernel, , exec, args->fname, 0, 0, 0, 0 );
420 if (args->fname != NULL) {
421 #ifdef CAPABILITY_MODE
423 * While capability mode can't reach this point via direct
424 * path arguments to execve(), we also don't allow
425 * interpreters to be used in capability mode (for now).
426 * Catch indirect lookups and return a permissions error.
428 if (IN_CAPABILITY_MODE(td)) {
440 AUDIT_ARG_FD(args->fd);
442 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
444 error = fgetvp_exec(td, args->fd, CAP_FEXECVE, &binvp);
447 vn_lock(binvp, LK_EXCLUSIVE | LK_RETRY);
448 AUDIT_ARG_VNODE1(binvp);
453 * Check file permissions (also 'opens' file)
455 error = exec_check_permissions(imgp);
457 goto exec_fail_dealloc;
459 imgp->object = imgp->vp->v_object;
460 if (imgp->object != NULL)
461 vm_object_reference(imgp->object);
464 * Set VV_TEXT now so no one can write to the executable while we're
467 * Remember if this was set before and unset it in case this is not
468 * actually an executable image.
470 textset = VOP_IS_TEXT(imgp->vp);
471 VOP_SET_TEXT(imgp->vp);
473 error = exec_map_first_page(imgp);
475 goto exec_fail_dealloc;
477 imgp->proc->p_osrel = 0;
479 * If the current process has a special image activator it
480 * wants to try first, call it. For example, emulating shell
481 * scripts differently.
484 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
485 error = img_first(imgp);
488 * Loop through the list of image activators, calling each one.
489 * An activator returns -1 if there is no match, 0 on success,
490 * and an error otherwise.
492 for (i = 0; error == -1 && execsw[i]; ++i) {
493 if (execsw[i]->ex_imgact == NULL ||
494 execsw[i]->ex_imgact == img_first) {
497 error = (*execsw[i]->ex_imgact)(imgp);
503 VOP_UNSET_TEXT(imgp->vp);
506 goto exec_fail_dealloc;
510 * Special interpreter operation, cleanup and loop up to try to
511 * activate the interpreter.
513 if (imgp->interpreted) {
514 exec_unmap_first_page(imgp);
516 * VV_TEXT needs to be unset for scripts. There is a short
517 * period before we determine that something is a script where
518 * VV_TEXT will be set. The vnode lock is held over this
519 * entire period so nothing should illegitimately be blocked.
521 VOP_UNSET_TEXT(imgp->vp);
522 /* free name buffer and old vnode */
523 if (args->fname != NULL)
524 NDFREE(&nd, NDF_ONLY_PNBUF);
526 mac_execve_interpreter_enter(binvp, &interpvplabel);
529 VOP_CLOSE(binvp, FREAD, td->td_ucred, td);
533 vm_object_deallocate(imgp->object);
535 /* set new name to that of the interpreter */
536 NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
537 UIO_SYSSPACE, imgp->interpreter_name, td);
538 args->fname = imgp->interpreter_name;
543 * NB: We unlock the vnode here because it is believed that none
544 * of the sv_copyout_strings/sv_fixup operations require the vnode.
546 VOP_UNLOCK(imgp->vp, 0);
549 * Do the best to calculate the full path to the image file.
551 if (imgp->auxargs != NULL &&
552 ((args->fname != NULL && args->fname[0] == '/') ||
553 vn_fullpath(td, imgp->vp, &imgp->execpath, &imgp->freepath) != 0))
554 imgp->execpath = args->fname;
557 * Copy out strings (args and env) and initialize stack base
559 if (p->p_sysent->sv_copyout_strings)
560 stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
562 stack_base = exec_copyout_strings(imgp);
565 * If custom stack fixup routine present for this process
566 * let it do the stack setup.
567 * Else stuff argument count as first item on stack
569 if (p->p_sysent->sv_fixup != NULL)
570 (*p->p_sysent->sv_fixup)(&stack_base, imgp);
572 suword(--stack_base, imgp->args->argc);
575 * For security and other reasons, the file descriptor table cannot
576 * be shared after an exec.
581 * Malloc things before we need locks.
584 euip = uifind(attr.va_uid);
585 i = imgp->args->begin_envv - imgp->args->begin_argv;
586 /* Cache arguments if they fit inside our allowance */
587 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
588 newargs = pargs_alloc(i);
589 bcopy(imgp->args->begin_argv, newargs->ar_args, i);
592 /* close files on exec */
594 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
596 /* Get a reference to the vnode prior to locking the proc */
600 * For security and other reasons, signal handlers cannot
601 * be shared after an exec. The new process gets a copy of the old
602 * handlers. In execsigs(), the new process will have its signals
606 oldcred = crcopysafe(p, newcred);
607 if (sigacts_shared(p->p_sigacts)) {
608 oldsigacts = p->p_sigacts;
610 newsigacts = sigacts_alloc();
611 sigacts_copy(newsigacts, oldsigacts);
613 p->p_sigacts = newsigacts;
620 /* reset caught signals */
623 /* name this process - nameiexec(p, ndp) */
624 bzero(p->p_comm, sizeof(p->p_comm));
626 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
627 min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
628 else if (vn_commname(binvp, p->p_comm, sizeof(p->p_comm)) != 0)
629 bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
630 bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
632 sched_clear_tdname(td);
636 * mark as execed, wakeup the process that vforked (if any) and tell
637 * it that it now has its own resources back
640 if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
641 p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
642 cv_broadcast(&p->p_pwait);
646 * Implement image setuid/setgid.
648 * Don't honor setuid/setgid if the filesystem prohibits it or if
649 * the process is being traced.
651 * We disable setuid/setgid/etc in compatibility mode on the basis
652 * that most setugid applications are not written with that
653 * environment in mind, and will therefore almost certainly operate
654 * incorrectly. In principle there's no reason that setugid
655 * applications might not be useful in capability mode, so we may want
656 * to reconsider this conservative design choice in the future.
658 * XXXMAC: For the time being, use NOSUID to also prohibit
659 * transitions on the file system.
661 credential_changing = 0;
662 credential_changing |= (attr.va_mode & S_ISUID) && oldcred->cr_uid !=
664 credential_changing |= (attr.va_mode & S_ISGID) && oldcred->cr_gid !=
667 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
668 interpvplabel, imgp);
669 credential_changing |= will_transition;
672 if (credential_changing &&
673 #ifdef CAPABILITY_MODE
674 ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
676 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
677 (p->p_flag & P_TRACED) == 0) {
679 * Turn off syscall tracing for set-id programs, except for
680 * root. Record any set-id flags first to make sure that
681 * we do not regain any tracing during a possible block.
686 if (p->p_tracecred != NULL &&
687 priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED, 0))
688 ktrprocexec(p, &tracecred, &tracevp);
691 * Close any file descriptors 0..2 that reference procfs,
692 * then make sure file descriptors 0..2 are in use.
694 * setugidsafety() may call closef() and then pfind()
695 * which may grab the process lock.
696 * fdcheckstd() may call falloc() which may block to
697 * allocate memory, so temporarily drop the process lock.
700 VOP_UNLOCK(imgp->vp, 0);
702 error = fdcheckstd(td);
703 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
708 * Set the new credentials.
710 if (attr.va_mode & S_ISUID)
711 change_euid(newcred, euip);
712 if (attr.va_mode & S_ISGID)
713 change_egid(newcred, attr.va_gid);
715 if (will_transition) {
716 mac_vnode_execve_transition(oldcred, newcred, imgp->vp,
717 interpvplabel, imgp);
721 * Implement correct POSIX saved-id behavior.
723 * XXXMAC: Note that the current logic will save the
724 * uid and gid if a MAC domain transition occurs, even
725 * though maybe it shouldn't.
727 change_svuid(newcred, newcred->cr_uid);
728 change_svgid(newcred, newcred->cr_gid);
729 p->p_ucred = newcred;
732 if (oldcred->cr_uid == oldcred->cr_ruid &&
733 oldcred->cr_gid == oldcred->cr_rgid)
734 p->p_flag &= ~P_SUGID;
736 * Implement correct POSIX saved-id behavior.
738 * XXX: It's not clear that the existing behavior is
739 * POSIX-compliant. A number of sources indicate that the
740 * saved uid/gid should only be updated if the new ruid is
741 * not equal to the old ruid, or the new euid is not equal
742 * to the old euid and the new euid is not equal to the old
743 * ruid. The FreeBSD code always updates the saved uid/gid.
744 * Also, this code uses the new (replaced) euid and egid as
745 * the source, which may or may not be the right ones to use.
747 if (oldcred->cr_svuid != oldcred->cr_uid ||
748 oldcred->cr_svgid != oldcred->cr_gid) {
749 change_svuid(newcred, newcred->cr_uid);
750 change_svgid(newcred, newcred->cr_gid);
751 p->p_ucred = newcred;
757 * Store the vp for use in procfs. This vnode was referenced prior
758 * to locking the proc lock.
760 textvp = p->p_textvp;
765 * Tell the DTrace fasttrap provider about the exec if it
766 * has declared an interest.
768 if (dtrace_fasttrap_exec)
769 dtrace_fasttrap_exec(p);
773 * Notify others that we exec'd, and clear the P_INEXEC flag
774 * as we're now a bona fide freshly-execed process.
776 KNOTE_LOCKED(&p->p_klist, NOTE_EXEC);
777 p->p_flag &= ~P_INEXEC;
779 /* clear "fork but no exec" flag, as we _are_ execing */
780 p->p_acflag &= ~AFORK;
783 * Free any previous argument cache and replace it with
784 * the new argument cache, if any.
792 * Check if system-wide sampling is in effect or if the
793 * current process is using PMCs. If so, do exec() time
794 * processing. This processing needs to happen AFTER the
795 * P_INEXEC flag is cleared.
797 * The proc lock needs to be released before taking the PMC
800 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
802 VOP_UNLOCK(imgp->vp, 0);
803 pe.pm_credentialschanged = credential_changing;
804 pe.pm_entryaddr = imgp->entry_addr;
806 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
807 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
810 #else /* !HWPMC_HOOKS */
814 /* Set values passed into the program in registers. */
815 if (p->p_sysent->sv_setregs)
816 (*p->p_sysent->sv_setregs)(td, imgp,
817 (u_long)(uintptr_t)stack_base);
819 exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base);
821 vfs_mark_atime(imgp->vp, td->td_ucred);
823 SDT_PROBE(proc, kernel, , exec_success, args->fname, 0, 0, 0, 0);
827 * Free any resources malloc'd earlier that we didn't use.
834 VOP_UNLOCK(imgp->vp, 0);
837 * Handle deferred decrement of ref counts.
841 if (binvp && error != 0)
846 if (tracecred != NULL)
849 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
852 if (oldsigacts != NULL)
853 sigacts_free(oldsigacts);
858 * free various allocated resources
860 if (imgp->firstpage != NULL)
861 exec_unmap_first_page(imgp);
863 if (imgp->vp != NULL) {
865 NDFREE(&nd, NDF_ONLY_PNBUF);
867 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
871 if (imgp->object != NULL)
872 vm_object_deallocate(imgp->object);
874 free(imgp->freepath, M_TEMP);
878 td->td_dbgflags |= TDB_EXEC;
882 * Stop the process here if its stop event mask has
883 * the S_EXEC bit set.
885 STOPEVENT(p, S_EXEC, 0);
890 /* we're done here, clear P_INEXEC */
892 p->p_flag &= ~P_INEXEC;
895 SDT_PROBE(proc, kernel, , exec_failure, error, 0, 0, 0, 0);
899 mac_execve_exit(imgp);
900 mac_execve_interpreter_exit(interpvplabel);
902 exec_free_args(args);
904 if (error && imgp->vmspace_destroyed) {
905 /* sorry, no more process anymore. exit gracefully */
906 exit1(td, W_EXITCODE(0, SIGABRT));
919 exec_map_first_page(imgp)
920 struct image_params *imgp;
924 vm_page_t ma[VM_INITIAL_PAGEIN];
927 if (imgp->firstpage != NULL)
928 exec_unmap_first_page(imgp);
930 object = imgp->vp->v_object;
933 VM_OBJECT_WLOCK(object);
934 #if VM_NRESERVLEVEL > 0
935 if ((object->flags & OBJ_COLORED) == 0) {
936 object->flags |= OBJ_COLORED;
937 object->pg_color = 0;
940 ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
941 if (ma[0]->valid != VM_PAGE_BITS_ALL) {
942 initial_pagein = VM_INITIAL_PAGEIN;
943 if (initial_pagein > object->size)
944 initial_pagein = object->size;
945 for (i = 1; i < initial_pagein; i++) {
946 if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) {
949 if (vm_page_tryxbusy(ma[i]))
952 ma[i] = vm_page_alloc(object, i,
953 VM_ALLOC_NORMAL | VM_ALLOC_IFNOTCACHED);
959 rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
960 ma[0] = vm_page_lookup(object, 0);
961 if ((rv != VM_PAGER_OK) || (ma[0] == NULL)) {
965 vm_page_unlock(ma[0]);
967 VM_OBJECT_WUNLOCK(object);
971 vm_page_xunbusy(ma[0]);
974 vm_page_unlock(ma[0]);
975 VM_OBJECT_WUNLOCK(object);
977 imgp->firstpage = sf_buf_alloc(ma[0], 0);
978 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
984 exec_unmap_first_page(imgp)
985 struct image_params *imgp;
989 if (imgp->firstpage != NULL) {
990 m = sf_buf_page(imgp->firstpage);
991 sf_buf_free(imgp->firstpage);
992 imgp->firstpage = NULL;
1000 * Destroy old address space, and allocate a new stack
1001 * The new stack is only SGROWSIZ large because it is grown
1002 * automatically in trap.c.
1005 exec_new_vmspace(imgp, sv)
1006 struct image_params *imgp;
1007 struct sysentvec *sv;
1010 struct proc *p = imgp->proc;
1011 struct vmspace *vmspace = p->p_vmspace;
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 (sv->sv_maxssiz != NULL)
1063 ssiz = *sv->sv_maxssiz;
1066 stack_addr = sv->sv_usrstack - ssiz;
1067 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1068 obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1070 VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1075 /* Allocate a new register stack */
1076 stack_addr = IA64_BACKINGSTORE;
1077 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1078 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_UP);
1083 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
1084 * VM_STACK case, but they are still used to monitor the size of the
1085 * process stack so we can check the stack rlimit.
1087 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1088 vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - ssiz;
1094 * Copy out argument and environment strings from the old process address
1095 * space into the temporary string buffer.
1098 exec_copyin_args(struct image_args *args, char *fname,
1099 enum uio_seg segflg, char **argv, char **envv)
1105 bzero(args, sizeof(*args));
1110 * Allocate demand-paged memory for the file name, argument, and
1111 * environment strings.
1113 error = exec_alloc_args(args);
1118 * Copy the file name.
1120 if (fname != NULL) {
1121 args->fname = args->buf;
1122 error = (segflg == UIO_SYSSPACE) ?
1123 copystr(fname, args->fname, PATH_MAX, &length) :
1124 copyinstr(fname, args->fname, PATH_MAX, &length);
1130 args->begin_argv = args->buf + length;
1131 args->endp = args->begin_argv;
1132 args->stringspace = ARG_MAX;
1135 * extract arguments first
1137 while ((argp = (caddr_t) (intptr_t) fuword(argv++))) {
1138 if (argp == (caddr_t) -1) {
1142 if ((error = copyinstr(argp, args->endp,
1143 args->stringspace, &length))) {
1144 if (error == ENAMETOOLONG)
1148 args->stringspace -= length;
1149 args->endp += length;
1153 args->begin_envv = args->endp;
1156 * extract environment strings
1159 while ((envp = (caddr_t)(intptr_t)fuword(envv++))) {
1160 if (envp == (caddr_t)-1) {
1164 if ((error = copyinstr(envp, args->endp,
1165 args->stringspace, &length))) {
1166 if (error == ENAMETOOLONG)
1170 args->stringspace -= length;
1171 args->endp += length;
1179 exec_free_args(args);
1184 * Allocate temporary demand-paged, zero-filled memory for the file name,
1185 * argument, and environment strings. Returns zero if the allocation succeeds
1186 * and ENOMEM otherwise.
1189 exec_alloc_args(struct image_args *args)
1192 args->buf = (char *)kmap_alloc_wait(exec_map, PATH_MAX + ARG_MAX);
1193 return (args->buf != NULL ? 0 : ENOMEM);
1197 exec_free_args(struct image_args *args)
1200 if (args->buf != NULL) {
1201 kmap_free_wakeup(exec_map, (vm_offset_t)args->buf,
1202 PATH_MAX + ARG_MAX);
1205 if (args->fname_buf != NULL) {
1206 free(args->fname_buf, M_TEMP);
1207 args->fname_buf = NULL;
1212 * Copy strings out to the new process address space, constructing new arg
1213 * and env vector tables. Return a pointer to the base so that it can be used
1214 * as the initial stack pointer.
1217 exec_copyout_strings(imgp)
1218 struct image_params *imgp;
1222 char *stringp, *destp;
1223 register_t *stack_base;
1224 struct ps_strings *arginfo;
1226 size_t execpath_len;
1227 int szsigcode, szps;
1228 char canary[sizeof(long) * 8];
1230 szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1232 * Calculate string base and vector table pointers.
1233 * Also deal with signal trampoline code for this exec type.
1235 if (imgp->execpath != NULL && imgp->auxargs != NULL)
1236 execpath_len = strlen(imgp->execpath) + 1;
1241 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1242 if (p->p_sysent->sv_sigcode_base == 0) {
1243 if (p->p_sysent->sv_szsigcode != NULL)
1244 szsigcode = *(p->p_sysent->sv_szsigcode);
1246 destp = (caddr_t)arginfo - szsigcode - SPARE_USRSPACE -
1247 roundup(execpath_len, sizeof(char *)) -
1248 roundup(sizeof(canary), sizeof(char *)) -
1249 roundup(szps, sizeof(char *)) -
1250 roundup((ARG_MAX - imgp->args->stringspace), sizeof(char *));
1256 copyout(p->p_sysent->sv_sigcode, ((caddr_t)arginfo -
1257 szsigcode), szsigcode);
1260 * Copy the image path for the rtld.
1262 if (execpath_len != 0) {
1263 imgp->execpathp = (uintptr_t)arginfo - szsigcode - execpath_len;
1264 copyout(imgp->execpath, (void *)imgp->execpathp,
1269 * Prepare the canary for SSP.
1271 arc4rand(canary, sizeof(canary), 0);
1272 imgp->canary = (uintptr_t)arginfo - szsigcode - execpath_len -
1274 copyout(canary, (void *)imgp->canary, sizeof(canary));
1275 imgp->canarylen = sizeof(canary);
1278 * Prepare the pagesizes array.
1280 imgp->pagesizes = (uintptr_t)arginfo - szsigcode - execpath_len -
1281 roundup(sizeof(canary), sizeof(char *)) - szps;
1282 copyout(pagesizes, (void *)imgp->pagesizes, szps);
1283 imgp->pagesizeslen = szps;
1286 * If we have a valid auxargs ptr, prepare some room
1289 if (imgp->auxargs) {
1291 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1292 * lower compatibility.
1294 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1297 * The '+ 2' is for the null pointers at the end of each of
1298 * the arg and env vector sets,and imgp->auxarg_size is room
1299 * for argument of Runtime loader.
1301 vectp = (char **)(destp - (imgp->args->argc +
1302 imgp->args->envc + 2 + imgp->auxarg_size)
1306 * The '+ 2' is for the null pointers at the end of each of
1307 * the arg and env vector sets
1309 vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc + 2) *
1314 * vectp also becomes our initial stack base
1316 stack_base = (register_t *)vectp;
1318 stringp = imgp->args->begin_argv;
1319 argc = imgp->args->argc;
1320 envc = imgp->args->envc;
1323 * Copy out strings - arguments and environment.
1325 copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
1328 * Fill in "ps_strings" struct for ps, w, etc.
1330 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1331 suword32(&arginfo->ps_nargvstr, argc);
1334 * Fill in argument portion of vector table.
1336 for (; argc > 0; --argc) {
1337 suword(vectp++, (long)(intptr_t)destp);
1338 while (*stringp++ != 0)
1343 /* a null vector table pointer separates the argp's from the envp's */
1346 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1347 suword32(&arginfo->ps_nenvstr, envc);
1350 * Fill in environment portion of vector table.
1352 for (; envc > 0; --envc) {
1353 suword(vectp++, (long)(intptr_t)destp);
1354 while (*stringp++ != 0)
1359 /* end of vector table is a null pointer */
1362 return (stack_base);
1366 * Check permissions of file to execute.
1367 * Called with imgp->vp locked.
1368 * Return 0 for success or error code on failure.
1371 exec_check_permissions(imgp)
1372 struct image_params *imgp;
1374 struct vnode *vp = imgp->vp;
1375 struct vattr *attr = imgp->attr;
1377 int error, writecount;
1381 /* Get file attributes */
1382 error = VOP_GETATTR(vp, attr, td->td_ucred);
1387 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1393 * 1) Check if file execution is disabled for the filesystem that
1394 * this file resides on.
1395 * 2) Ensure that at least one execute bit is on. Otherwise, a
1396 * privileged user will always succeed, and we don't want this
1397 * to happen unless the file really is executable.
1398 * 3) Ensure that the file is a regular file.
1400 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1401 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1402 (attr->va_type != VREG))
1406 * Zero length files can't be exec'd
1408 if (attr->va_size == 0)
1412 * Check for execute permission to file based on current credentials.
1414 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1419 * Check number of open-for-writes on the file and deny execution
1422 error = VOP_GET_WRITECOUNT(vp, &writecount);
1425 if (writecount != 0)
1429 * Call filesystem specific open routine (which does nothing in the
1432 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1439 * Exec handler registration
1442 exec_register(execsw_arg)
1443 const struct execsw *execsw_arg;
1445 const struct execsw **es, **xs, **newexecsw;
1446 int count = 2; /* New slot and trailing NULL */
1449 for (es = execsw; *es; es++)
1451 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1452 if (newexecsw == NULL)
1456 for (es = execsw; *es; es++)
1461 free(execsw, M_TEMP);
1467 exec_unregister(execsw_arg)
1468 const struct execsw *execsw_arg;
1470 const struct execsw **es, **xs, **newexecsw;
1474 panic("unregister with no handlers left?\n");
1476 for (es = execsw; *es; es++) {
1477 if (*es == execsw_arg)
1482 for (es = execsw; *es; es++)
1483 if (*es != execsw_arg)
1485 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1486 if (newexecsw == NULL)
1489 for (es = execsw; *es; es++)
1490 if (*es != execsw_arg)
1494 free(execsw, M_TEMP);