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/ptrace.h>
57 #include <sys/namei.h>
58 #include <sys/resourcevar.h>
59 #include <sys/rwlock.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_DEFINE1(proc, , , exec, "char *");
99 SDT_PROBE_DEFINE1(proc, , , exec__failure, "int");
100 SDT_PROBE_DEFINE1(proc, , , exec__success, "char *");
102 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
104 int coredump_pack_fileinfo = 1;
105 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_fileinfo, CTLFLAG_RWTUN,
106 &coredump_pack_fileinfo, 0,
107 "Enable file path packing in 'procstat -f' coredump notes");
109 int coredump_pack_vmmapinfo = 1;
110 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_vmmapinfo, CTLFLAG_RWTUN,
111 &coredump_pack_vmmapinfo, 0,
112 "Enable file path packing in 'procstat -v' coredump notes");
114 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
115 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
116 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
117 static int do_execve(struct thread *td, struct image_args *args,
120 /* XXX This should be vm_size_t. */
121 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD,
122 NULL, 0, sysctl_kern_ps_strings, "LU", "");
124 /* XXX This should be vm_size_t. */
125 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD|
126 CTLFLAG_CAPRD, NULL, 0, sysctl_kern_usrstack, "LU", "");
128 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD,
129 NULL, 0, sysctl_kern_stackprot, "I", "");
131 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
132 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
133 &ps_arg_cache_limit, 0, "");
135 static int disallow_high_osrel;
136 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW,
137 &disallow_high_osrel, 0,
138 "Disallow execution of binaries built for higher version of the world");
140 static int map_at_zero = 0;
141 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RWTUN, &map_at_zero, 0,
142 "Permit processes to map an object at virtual address 0.");
145 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
152 if (req->flags & SCTL_MASK32) {
154 val = (unsigned int)p->p_sysent->sv_psstrings;
155 error = SYSCTL_OUT(req, &val, sizeof(val));
158 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
159 sizeof(p->p_sysent->sv_psstrings));
164 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
171 if (req->flags & SCTL_MASK32) {
173 val = (unsigned int)p->p_sysent->sv_usrstack;
174 error = SYSCTL_OUT(req, &val, sizeof(val));
177 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
178 sizeof(p->p_sysent->sv_usrstack));
183 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
188 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
189 sizeof(p->p_sysent->sv_stackprot)));
193 * Each of the items is a pointer to a `const struct execsw', hence the
194 * double pointer here.
196 static const struct execsw **execsw;
198 #ifndef _SYS_SYSPROTO_H_
207 sys_execve(struct thread *td, struct execve_args *uap)
209 struct image_args args;
210 struct vmspace *oldvmspace;
213 error = pre_execve(td, &oldvmspace);
216 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
217 uap->argv, uap->envv);
219 error = kern_execve(td, &args, NULL);
220 post_execve(td, error, oldvmspace);
224 #ifndef _SYS_SYSPROTO_H_
225 struct fexecve_args {
232 sys_fexecve(struct thread *td, struct fexecve_args *uap)
234 struct image_args args;
235 struct vmspace *oldvmspace;
238 error = pre_execve(td, &oldvmspace);
241 error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
242 uap->argv, uap->envv);
245 error = kern_execve(td, &args, NULL);
247 post_execve(td, error, oldvmspace);
251 #ifndef _SYS_SYSPROTO_H_
252 struct __mac_execve_args {
261 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap)
264 struct image_args args;
265 struct vmspace *oldvmspace;
268 error = pre_execve(td, &oldvmspace);
271 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
272 uap->argv, uap->envv);
274 error = kern_execve(td, &args, uap->mac_p);
275 post_execve(td, error, oldvmspace);
283 pre_execve(struct thread *td, struct vmspace **oldvmspace)
288 KASSERT(td == curthread, ("non-current thread %p", td));
291 if ((p->p_flag & P_HADTHREADS) != 0) {
293 if (thread_single(p, SINGLE_BOUNDARY) != 0)
297 KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0,
299 *oldvmspace = p->p_vmspace;
304 post_execve(struct thread *td, int error, struct vmspace *oldvmspace)
308 KASSERT(td == curthread, ("non-current thread %p", td));
310 if ((p->p_flag & P_HADTHREADS) != 0) {
313 * If success, we upgrade to SINGLE_EXIT state to
314 * force other threads to suicide.
317 thread_single(p, SINGLE_EXIT);
319 thread_single_end(p, SINGLE_BOUNDARY);
322 if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
323 KASSERT(p->p_vmspace != oldvmspace,
324 ("oldvmspace still used"));
325 vmspace_free(oldvmspace);
326 td->td_pflags &= ~TDP_EXECVMSPC;
331 * XXX: kern_execve has the astonishing property of not always returning to
332 * the caller. If sufficiently bad things happen during the call to
333 * do_execve(), it can end up calling exit1(); as a result, callers must
334 * avoid doing anything which they might need to undo (e.g., allocating
338 kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p)
341 AUDIT_ARG_ARGV(args->begin_argv, args->argc,
342 args->begin_envv - args->begin_argv);
343 AUDIT_ARG_ENVV(args->begin_envv, args->envc,
344 args->endp - args->begin_envv);
345 return (do_execve(td, args, mac_p));
349 * In-kernel implementation of execve(). All arguments are assumed to be
350 * userspace pointers from the passed thread.
353 do_execve(td, args, mac_p)
355 struct image_args *args;
358 struct proc *p = td->td_proc;
360 struct ucred *oldcred;
361 struct uidinfo *euip = NULL;
362 register_t *stack_base;
364 struct image_params image_params, *imgp;
366 int (*img_first)(struct image_params *);
367 struct pargs *oldargs = NULL, *newargs = NULL;
368 struct sigacts *oldsigacts = NULL, *newsigacts = NULL;
370 struct vnode *tracevp = NULL;
371 struct ucred *tracecred = NULL;
373 struct vnode *oldtextvp = NULL, *newtextvp;
375 int credential_changing;
378 struct label *interpvplabel = NULL;
382 struct pmckern_procexec pe;
384 static const char fexecv_proc_title[] = "(fexecv)";
386 imgp = &image_params;
389 * Lock the process and set the P_INEXEC flag to indicate that
390 * it should be left alone until we're done here. This is
391 * necessary to avoid race conditions - e.g. in ptrace() -
392 * that might allow a local user to illicitly obtain elevated
396 KASSERT((p->p_flag & P_INEXEC) == 0,
397 ("%s(): process already has P_INEXEC flag", __func__));
398 p->p_flag |= P_INEXEC;
402 * Initialize part of the common data
404 bzero(imgp, sizeof(*imgp));
408 oldcred = p->p_ucred;
411 error = mac_execve_enter(imgp, mac_p);
417 * Translate the file name. namei() returns a vnode pointer
418 * in ni_vp among other things.
420 * XXXAUDIT: It would be desirable to also audit the name of the
421 * interpreter if this is an interpreted binary.
423 if (args->fname != NULL) {
424 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME
425 | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
428 SDT_PROBE1(proc, , , exec, args->fname);
431 if (args->fname != NULL) {
432 #ifdef CAPABILITY_MODE
434 * While capability mode can't reach this point via direct
435 * path arguments to execve(), we also don't allow
436 * interpreters to be used in capability mode (for now).
437 * Catch indirect lookups and return a permissions error.
439 if (IN_CAPABILITY_MODE(td)) {
448 newtextvp = nd.ni_vp;
449 imgp->vp = newtextvp;
451 AUDIT_ARG_FD(args->fd);
453 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
455 error = fgetvp_exec(td, args->fd,
456 cap_rights_init(&rights, CAP_FEXECVE), &newtextvp);
459 vn_lock(newtextvp, LK_EXCLUSIVE | LK_RETRY);
460 AUDIT_ARG_VNODE1(newtextvp);
461 imgp->vp = newtextvp;
465 * Check file permissions (also 'opens' file)
467 error = exec_check_permissions(imgp);
469 goto exec_fail_dealloc;
471 imgp->object = imgp->vp->v_object;
472 if (imgp->object != NULL)
473 vm_object_reference(imgp->object);
476 * Set VV_TEXT now so no one can write to the executable while we're
479 * Remember if this was set before and unset it in case this is not
480 * actually an executable image.
482 textset = VOP_IS_TEXT(imgp->vp);
483 VOP_SET_TEXT(imgp->vp);
485 error = exec_map_first_page(imgp);
487 goto exec_fail_dealloc;
489 imgp->proc->p_osrel = 0;
492 * Implement image setuid/setgid.
494 * Determine new credentials before attempting image activators
495 * so that it can be used by process_exec handlers to determine
496 * credential/setid changes.
498 * Don't honor setuid/setgid if the filesystem prohibits it or if
499 * the process is being traced.
501 * We disable setuid/setgid/etc in capability mode on the basis
502 * that most setugid applications are not written with that
503 * environment in mind, and will therefore almost certainly operate
504 * incorrectly. In principle there's no reason that setugid
505 * applications might not be useful in capability mode, so we may want
506 * to reconsider this conservative design choice in the future.
508 * XXXMAC: For the time being, use NOSUID to also prohibit
509 * transitions on the file system.
511 credential_changing = 0;
512 credential_changing |= (attr.va_mode & S_ISUID) &&
513 oldcred->cr_uid != attr.va_uid;
514 credential_changing |= (attr.va_mode & S_ISGID) &&
515 oldcred->cr_gid != attr.va_gid;
517 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
518 interpvplabel, imgp);
519 credential_changing |= will_transition;
522 if (credential_changing &&
523 #ifdef CAPABILITY_MODE
524 ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
526 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
527 (p->p_flag & P_TRACED) == 0) {
528 imgp->credential_setid = true;
529 VOP_UNLOCK(imgp->vp, 0);
530 imgp->newcred = crdup(oldcred);
531 if (attr.va_mode & S_ISUID) {
532 euip = uifind(attr.va_uid);
533 change_euid(imgp->newcred, euip);
535 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
536 if (attr.va_mode & S_ISGID)
537 change_egid(imgp->newcred, attr.va_gid);
539 * Implement correct POSIX saved-id behavior.
541 * XXXMAC: Note that the current logic will save the
542 * uid and gid if a MAC domain transition occurs, even
543 * though maybe it shouldn't.
545 change_svuid(imgp->newcred, imgp->newcred->cr_uid);
546 change_svgid(imgp->newcred, imgp->newcred->cr_gid);
549 * Implement correct POSIX saved-id behavior.
551 * XXX: It's not clear that the existing behavior is
552 * POSIX-compliant. A number of sources indicate that the
553 * saved uid/gid should only be updated if the new ruid is
554 * not equal to the old ruid, or the new euid is not equal
555 * to the old euid and the new euid is not equal to the old
556 * ruid. The FreeBSD code always updates the saved uid/gid.
557 * Also, this code uses the new (replaced) euid and egid as
558 * the source, which may or may not be the right ones to use.
560 if (oldcred->cr_svuid != oldcred->cr_uid ||
561 oldcred->cr_svgid != oldcred->cr_gid) {
562 VOP_UNLOCK(imgp->vp, 0);
563 imgp->newcred = crdup(oldcred);
564 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
565 change_svuid(imgp->newcred, imgp->newcred->cr_uid);
566 change_svgid(imgp->newcred, imgp->newcred->cr_gid);
569 /* The new credentials are installed into the process later. */
572 * Do the best to calculate the full path to the image file.
574 if (args->fname != NULL && args->fname[0] == '/')
575 imgp->execpath = args->fname;
577 VOP_UNLOCK(imgp->vp, 0);
578 if (vn_fullpath(td, imgp->vp, &imgp->execpath,
579 &imgp->freepath) != 0)
580 imgp->execpath = args->fname;
581 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
585 * If the current process has a special image activator it
586 * wants to try first, call it. For example, emulating shell
587 * scripts differently.
590 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
591 error = img_first(imgp);
594 * Loop through the list of image activators, calling each one.
595 * An activator returns -1 if there is no match, 0 on success,
596 * and an error otherwise.
598 for (i = 0; error == -1 && execsw[i]; ++i) {
599 if (execsw[i]->ex_imgact == NULL ||
600 execsw[i]->ex_imgact == img_first) {
603 error = (*execsw[i]->ex_imgact)(imgp);
609 VOP_UNSET_TEXT(imgp->vp);
612 goto exec_fail_dealloc;
616 * Special interpreter operation, cleanup and loop up to try to
617 * activate the interpreter.
619 if (imgp->interpreted) {
620 exec_unmap_first_page(imgp);
622 * VV_TEXT needs to be unset for scripts. There is a short
623 * period before we determine that something is a script where
624 * VV_TEXT will be set. The vnode lock is held over this
625 * entire period so nothing should illegitimately be blocked.
627 VOP_UNSET_TEXT(imgp->vp);
628 /* free name buffer and old vnode */
629 if (args->fname != NULL)
630 NDFREE(&nd, NDF_ONLY_PNBUF);
632 mac_execve_interpreter_enter(newtextvp, &interpvplabel);
635 VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td);
639 vm_object_deallocate(imgp->object);
641 imgp->credential_setid = false;
642 if (imgp->newcred != NULL) {
643 crfree(imgp->newcred);
644 imgp->newcred = NULL;
646 imgp->execpath = NULL;
647 free(imgp->freepath, M_TEMP);
648 imgp->freepath = NULL;
649 /* set new name to that of the interpreter */
650 NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
651 UIO_SYSSPACE, imgp->interpreter_name, td);
652 args->fname = imgp->interpreter_name;
657 * NB: We unlock the vnode here because it is believed that none
658 * of the sv_copyout_strings/sv_fixup operations require the vnode.
660 VOP_UNLOCK(imgp->vp, 0);
662 if (disallow_high_osrel &&
663 P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
665 uprintf("Osrel %d for image %s too high\n", p->p_osrel,
666 imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
667 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
668 goto exec_fail_dealloc;
671 /* ABI enforces the use of Capsicum. Switch into capabilities mode. */
672 if (SV_PROC_FLAG(p, SV_CAPSICUM))
673 sys_cap_enter(td, NULL);
676 * Copy out strings (args and env) and initialize stack base
678 if (p->p_sysent->sv_copyout_strings)
679 stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
681 stack_base = exec_copyout_strings(imgp);
684 * If custom stack fixup routine present for this process
685 * let it do the stack setup.
686 * Else stuff argument count as first item on stack
688 if (p->p_sysent->sv_fixup != NULL)
689 (*p->p_sysent->sv_fixup)(&stack_base, imgp);
691 suword(--stack_base, imgp->args->argc);
693 if (args->fdp != NULL) {
694 /* Install a brand new file descriptor table. */
695 fdinstall_remapped(td, args->fdp);
699 * Keep on using the existing file descriptor table. For
700 * security and other reasons, the file descriptor table
701 * cannot be shared after an exec.
704 /* close files on exec */
709 * Malloc things before we need locks.
711 i = imgp->args->begin_envv - imgp->args->begin_argv;
712 /* Cache arguments if they fit inside our allowance */
713 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
714 newargs = pargs_alloc(i);
715 bcopy(imgp->args->begin_argv, newargs->ar_args, i);
719 * For security and other reasons, signal handlers cannot
720 * be shared after an exec. The new process gets a copy of the old
721 * handlers. In execsigs(), the new process will have its signals
724 if (sigacts_shared(p->p_sigacts)) {
725 oldsigacts = p->p_sigacts;
726 newsigacts = sigacts_alloc();
727 sigacts_copy(newsigacts, oldsigacts);
730 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
734 p->p_sigacts = newsigacts;
738 /* reset caught signals */
741 /* name this process - nameiexec(p, ndp) */
742 bzero(p->p_comm, sizeof(p->p_comm));
744 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
745 min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
746 else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0)
747 bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
748 bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
750 sched_clear_tdname(td);
754 * mark as execed, wakeup the process that vforked (if any) and tell
755 * it that it now has its own resources back
758 if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0)
759 p->p_flag2 &= ~P2_NOTRACE;
760 if (p->p_flag & P_PPWAIT) {
761 p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
762 cv_broadcast(&p->p_pwait);
763 /* STOPs are no longer ignored, arrange for AST */
768 * Implement image setuid/setgid installation.
770 if (imgp->credential_setid) {
772 * Turn off syscall tracing for set-id programs, except for
773 * root. Record any set-id flags first to make sure that
774 * we do not regain any tracing during a possible block.
779 if (p->p_tracecred != NULL &&
780 priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED, 0))
781 ktrprocexec(p, &tracecred, &tracevp);
784 * Close any file descriptors 0..2 that reference procfs,
785 * then make sure file descriptors 0..2 are in use.
787 * Both fdsetugidsafety() and fdcheckstd() may call functions
788 * taking sleepable locks, so temporarily drop our locks.
791 VOP_UNLOCK(imgp->vp, 0);
793 error = fdcheckstd(td);
794 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
796 goto exec_fail_dealloc;
799 if (will_transition) {
800 mac_vnode_execve_transition(oldcred, imgp->newcred,
801 imgp->vp, interpvplabel, imgp);
805 if (oldcred->cr_uid == oldcred->cr_ruid &&
806 oldcred->cr_gid == oldcred->cr_rgid)
807 p->p_flag &= ~P_SUGID;
810 * Set the new credentials.
812 if (imgp->newcred != NULL) {
813 proc_set_cred(p, imgp->newcred);
819 * Store the vp for use in procfs. This vnode was referenced by namei
822 oldtextvp = p->p_textvp;
823 p->p_textvp = newtextvp;
827 * Tell the DTrace fasttrap provider about the exec if it
828 * has declared an interest.
830 if (dtrace_fasttrap_exec)
831 dtrace_fasttrap_exec(p);
835 * Notify others that we exec'd, and clear the P_INEXEC flag
836 * as we're now a bona fide freshly-execed process.
838 KNOTE_LOCKED(p->p_klist, NOTE_EXEC);
839 p->p_flag &= ~P_INEXEC;
841 /* clear "fork but no exec" flag, as we _are_ execing */
842 p->p_acflag &= ~AFORK;
845 * Free any previous argument cache and replace it with
846 * the new argument cache, if any.
854 * Check if system-wide sampling is in effect or if the
855 * current process is using PMCs. If so, do exec() time
856 * processing. This processing needs to happen AFTER the
857 * P_INEXEC flag is cleared.
859 * The proc lock needs to be released before taking the PMC
862 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
864 VOP_UNLOCK(imgp->vp, 0);
865 pe.pm_credentialschanged = credential_changing;
866 pe.pm_entryaddr = imgp->entry_addr;
868 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
869 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
872 #else /* !HWPMC_HOOKS */
876 /* Set values passed into the program in registers. */
877 if (p->p_sysent->sv_setregs)
878 (*p->p_sysent->sv_setregs)(td, imgp,
879 (u_long)(uintptr_t)stack_base);
881 exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base);
883 vfs_mark_atime(imgp->vp, td->td_ucred);
885 SDT_PROBE1(proc, , , exec__success, args->fname);
888 if (imgp->firstpage != NULL)
889 exec_unmap_first_page(imgp);
891 if (imgp->vp != NULL) {
893 NDFREE(&nd, NDF_ONLY_PNBUF);
895 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
899 VOP_UNLOCK(imgp->vp, 0);
902 if (imgp->object != NULL)
903 vm_object_deallocate(imgp->object);
905 free(imgp->freepath, M_TEMP);
909 if (p->p_ptevents & PTRACE_EXEC)
910 td->td_dbgflags |= TDB_EXEC;
914 * Stop the process here if its stop event mask has
915 * the S_EXEC bit set.
917 STOPEVENT(p, S_EXEC, 0);
920 /* we're done here, clear P_INEXEC */
922 p->p_flag &= ~P_INEXEC;
925 SDT_PROBE1(proc, , , exec__failure, error);
928 if (imgp->newcred != NULL && oldcred != NULL)
929 crfree(imgp->newcred);
932 mac_execve_exit(imgp);
933 mac_execve_interpreter_exit(interpvplabel);
935 exec_free_args(args);
938 * Handle deferred decrement of ref counts.
940 if (oldtextvp != NULL)
945 if (tracecred != NULL)
950 if (oldsigacts != NULL)
951 sigacts_free(oldsigacts);
955 if (error && imgp->vmspace_destroyed) {
956 /* sorry, no more process anymore. exit gracefully */
957 exit1(td, 0, SIGABRT);
970 exec_map_first_page(imgp)
971 struct image_params *imgp;
973 int rv, i, after, initial_pagein;
974 vm_page_t ma[VM_INITIAL_PAGEIN];
977 if (imgp->firstpage != NULL)
978 exec_unmap_first_page(imgp);
980 object = imgp->vp->v_object;
983 VM_OBJECT_WLOCK(object);
984 #if VM_NRESERVLEVEL > 0
985 vm_object_color(object, 0);
987 ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY);
988 if (ma[0]->valid != VM_PAGE_BITS_ALL) {
989 vm_page_xbusy(ma[0]);
990 if (!vm_pager_has_page(object, 0, NULL, &after)) {
993 vm_page_unlock(ma[0]);
994 VM_OBJECT_WUNLOCK(object);
997 initial_pagein = min(after, VM_INITIAL_PAGEIN);
998 KASSERT(initial_pagein <= object->size,
999 ("%s: initial_pagein %d object->size %ju",
1000 __func__, initial_pagein, (uintmax_t )object->size));
1001 for (i = 1; i < initial_pagein; i++) {
1002 if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) {
1005 if (vm_page_tryxbusy(ma[i]))
1008 ma[i] = vm_page_alloc(object, i,
1009 VM_ALLOC_NORMAL | VM_ALLOC_IFNOTCACHED);
1015 rv = vm_pager_get_pages(object, ma, initial_pagein, NULL, NULL);
1016 if (rv != VM_PAGER_OK) {
1017 for (i = 0; i < initial_pagein; i++) {
1018 vm_page_lock(ma[i]);
1019 vm_page_free(ma[i]);
1020 vm_page_unlock(ma[i]);
1022 VM_OBJECT_WUNLOCK(object);
1025 vm_page_xunbusy(ma[0]);
1026 for (i = 1; i < initial_pagein; i++)
1027 vm_page_readahead_finish(ma[i]);
1029 vm_page_lock(ma[0]);
1030 vm_page_hold(ma[0]);
1031 vm_page_activate(ma[0]);
1032 vm_page_unlock(ma[0]);
1033 VM_OBJECT_WUNLOCK(object);
1035 imgp->firstpage = sf_buf_alloc(ma[0], 0);
1036 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
1042 exec_unmap_first_page(imgp)
1043 struct image_params *imgp;
1047 if (imgp->firstpage != NULL) {
1048 m = sf_buf_page(imgp->firstpage);
1049 sf_buf_free(imgp->firstpage);
1050 imgp->firstpage = NULL;
1058 * Destroy old address space, and allocate a new stack
1059 * The new stack is only SGROWSIZ large because it is grown
1060 * automatically in trap.c.
1063 exec_new_vmspace(imgp, sv)
1064 struct image_params *imgp;
1065 struct sysentvec *sv;
1068 struct proc *p = imgp->proc;
1069 struct vmspace *vmspace = p->p_vmspace;
1071 struct rlimit rlim_stack;
1072 vm_offset_t sv_minuser, stack_addr;
1076 imgp->vmspace_destroyed = 1;
1079 /* May be called with Giant held */
1080 EVENTHANDLER_INVOKE(process_exec, p, imgp);
1083 * Blow away entire process VM, if address space not shared,
1084 * otherwise, create a new VM space so that other threads are
1087 map = &vmspace->vm_map;
1089 sv_minuser = sv->sv_minuser;
1091 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1092 if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser &&
1093 vm_map_max(map) == sv->sv_maxuser) {
1095 pmap_remove_pages(vmspace_pmap(vmspace));
1096 vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1098 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1101 vmspace = p->p_vmspace;
1102 map = &vmspace->vm_map;
1105 /* Map a shared page */
1106 obj = sv->sv_shared_page_obj;
1108 vm_object_reference(obj);
1109 error = vm_map_fixed(map, obj, 0,
1110 sv->sv_shared_page_base, sv->sv_shared_page_len,
1111 VM_PROT_READ | VM_PROT_EXECUTE,
1112 VM_PROT_READ | VM_PROT_EXECUTE,
1113 MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1115 vm_object_deallocate(obj);
1120 /* Allocate a new stack */
1121 if (imgp->stack_sz != 0) {
1122 ssiz = trunc_page(imgp->stack_sz);
1124 lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack);
1126 if (ssiz > rlim_stack.rlim_max)
1127 ssiz = rlim_stack.rlim_max;
1128 if (ssiz > rlim_stack.rlim_cur) {
1129 rlim_stack.rlim_cur = ssiz;
1130 kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack);
1132 } else if (sv->sv_maxssiz != NULL) {
1133 ssiz = *sv->sv_maxssiz;
1137 stack_addr = sv->sv_usrstack - ssiz;
1138 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1139 obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1141 VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1146 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they
1147 * are still used to enforce the stack rlimit on the process stack.
1149 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1150 vmspace->vm_maxsaddr = (char *)stack_addr;
1156 * Copy out argument and environment strings from the old process address
1157 * space into the temporary string buffer.
1160 exec_copyin_args(struct image_args *args, char *fname,
1161 enum uio_seg segflg, char **argv, char **envv)
1167 bzero(args, sizeof(*args));
1172 * Allocate demand-paged memory for the file name, argument, and
1173 * environment strings.
1175 error = exec_alloc_args(args);
1180 * Copy the file name.
1182 if (fname != NULL) {
1183 args->fname = args->buf;
1184 error = (segflg == UIO_SYSSPACE) ?
1185 copystr(fname, args->fname, PATH_MAX, &length) :
1186 copyinstr(fname, args->fname, PATH_MAX, &length);
1192 args->begin_argv = args->buf + length;
1193 args->endp = args->begin_argv;
1194 args->stringspace = ARG_MAX;
1197 * extract arguments first
1200 error = fueword(argv++, &argp);
1207 error = copyinstr((void *)(uintptr_t)argp, args->endp,
1208 args->stringspace, &length);
1210 if (error == ENAMETOOLONG)
1214 args->stringspace -= length;
1215 args->endp += length;
1219 args->begin_envv = args->endp;
1222 * extract environment strings
1226 error = fueword(envv++, &envp);
1233 error = copyinstr((void *)(uintptr_t)envp,
1234 args->endp, args->stringspace, &length);
1236 if (error == ENAMETOOLONG)
1240 args->stringspace -= length;
1241 args->endp += length;
1249 exec_free_args(args);
1254 exec_copyin_data_fds(struct thread *td, struct image_args *args,
1255 const void *data, size_t datalen, const int *fds, size_t fdslen)
1257 struct filedesc *ofdp;
1262 memset(args, '\0', sizeof(*args));
1263 ofdp = td->td_proc->p_fd;
1264 if (datalen >= ARG_MAX || fdslen > ofdp->fd_lastfile + 1)
1266 error = exec_alloc_args(args);
1270 args->begin_argv = args->buf;
1271 args->stringspace = ARG_MAX;
1275 * Argument buffer has been provided. Copy it into the
1276 * kernel as a single string and add a terminating null
1279 error = copyin(data, args->begin_argv, datalen);
1282 args->begin_argv[datalen] = '\0';
1283 args->endp = args->begin_argv + datalen + 1;
1284 args->stringspace -= datalen + 1;
1287 * Traditional argument counting. Count the number of
1290 for (p = args->begin_argv; p < args->endp; ++p)
1294 /* No argument buffer provided. */
1295 args->endp = args->begin_argv;
1297 /* There are no environment variables. */
1298 args->begin_envv = args->endp;
1300 /* Create new file descriptor table. */
1301 kfds = malloc(fdslen * sizeof(int), M_TEMP, M_WAITOK);
1302 error = copyin(fds, kfds, fdslen * sizeof(int));
1307 error = fdcopy_remapped(ofdp, kfds, fdslen, &args->fdp);
1314 exec_free_args(args);
1319 * Allocate temporary demand-paged, zero-filled memory for the file name,
1320 * argument, and environment strings. Returns zero if the allocation succeeds
1321 * and ENOMEM otherwise.
1324 exec_alloc_args(struct image_args *args)
1327 args->buf = (char *)kmap_alloc_wait(exec_map, PATH_MAX + ARG_MAX);
1328 return (args->buf != NULL ? 0 : ENOMEM);
1332 exec_free_args(struct image_args *args)
1335 if (args->buf != NULL) {
1336 kmap_free_wakeup(exec_map, (vm_offset_t)args->buf,
1337 PATH_MAX + ARG_MAX);
1340 if (args->fname_buf != NULL) {
1341 free(args->fname_buf, M_TEMP);
1342 args->fname_buf = NULL;
1344 if (args->fdp != NULL)
1345 fdescfree_remapped(args->fdp);
1349 * Copy strings out to the new process address space, constructing new arg
1350 * and env vector tables. Return a pointer to the base so that it can be used
1351 * as the initial stack pointer.
1354 exec_copyout_strings(imgp)
1355 struct image_params *imgp;
1361 register_t *stack_base;
1362 struct ps_strings *arginfo;
1364 size_t execpath_len;
1365 int szsigcode, szps;
1366 char canary[sizeof(long) * 8];
1368 szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1370 * Calculate string base and vector table pointers.
1371 * Also deal with signal trampoline code for this exec type.
1373 if (imgp->execpath != NULL && imgp->auxargs != NULL)
1374 execpath_len = strlen(imgp->execpath) + 1;
1379 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1380 if (p->p_sysent->sv_sigcode_base == 0) {
1381 if (p->p_sysent->sv_szsigcode != NULL)
1382 szsigcode = *(p->p_sysent->sv_szsigcode);
1384 destp = (uintptr_t)arginfo;
1389 if (szsigcode != 0) {
1391 destp = rounddown2(destp, sizeof(void *));
1392 copyout(p->p_sysent->sv_sigcode, (void *)destp, szsigcode);
1396 * Copy the image path for the rtld.
1398 if (execpath_len != 0) {
1399 destp -= execpath_len;
1400 imgp->execpathp = destp;
1401 copyout(imgp->execpath, (void *)destp, execpath_len);
1405 * Prepare the canary for SSP.
1407 arc4rand(canary, sizeof(canary), 0);
1408 destp -= sizeof(canary);
1409 imgp->canary = destp;
1410 copyout(canary, (void *)destp, sizeof(canary));
1411 imgp->canarylen = sizeof(canary);
1414 * Prepare the pagesizes array.
1417 destp = rounddown2(destp, sizeof(void *));
1418 imgp->pagesizes = destp;
1419 copyout(pagesizes, (void *)destp, szps);
1420 imgp->pagesizeslen = szps;
1422 destp -= ARG_MAX - imgp->args->stringspace;
1423 destp = rounddown2(destp, sizeof(void *));
1426 * If we have a valid auxargs ptr, prepare some room
1429 if (imgp->auxargs) {
1431 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1432 * lower compatibility.
1434 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1437 * The '+ 2' is for the null pointers at the end of each of
1438 * the arg and env vector sets,and imgp->auxarg_size is room
1439 * for argument of Runtime loader.
1441 vectp = (char **)(destp - (imgp->args->argc +
1442 imgp->args->envc + 2 + imgp->auxarg_size)
1446 * The '+ 2' is for the null pointers at the end of each of
1447 * the arg and env vector sets
1449 vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc
1450 + 2) * sizeof(char *));
1454 * vectp also becomes our initial stack base
1456 stack_base = (register_t *)vectp;
1458 stringp = imgp->args->begin_argv;
1459 argc = imgp->args->argc;
1460 envc = imgp->args->envc;
1463 * Copy out strings - arguments and environment.
1465 copyout(stringp, (void *)destp, ARG_MAX - imgp->args->stringspace);
1468 * Fill in "ps_strings" struct for ps, w, etc.
1470 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1471 suword32(&arginfo->ps_nargvstr, argc);
1474 * Fill in argument portion of vector table.
1476 for (; argc > 0; --argc) {
1477 suword(vectp++, (long)(intptr_t)destp);
1478 while (*stringp++ != 0)
1483 /* a null vector table pointer separates the argp's from the envp's */
1486 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1487 suword32(&arginfo->ps_nenvstr, envc);
1490 * Fill in environment portion of vector table.
1492 for (; envc > 0; --envc) {
1493 suword(vectp++, (long)(intptr_t)destp);
1494 while (*stringp++ != 0)
1499 /* end of vector table is a null pointer */
1502 return (stack_base);
1506 * Check permissions of file to execute.
1507 * Called with imgp->vp locked.
1508 * Return 0 for success or error code on failure.
1511 exec_check_permissions(imgp)
1512 struct image_params *imgp;
1514 struct vnode *vp = imgp->vp;
1515 struct vattr *attr = imgp->attr;
1517 int error, writecount;
1521 /* Get file attributes */
1522 error = VOP_GETATTR(vp, attr, td->td_ucred);
1527 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1533 * 1) Check if file execution is disabled for the filesystem that
1534 * this file resides on.
1535 * 2) Ensure that at least one execute bit is on. Otherwise, a
1536 * privileged user will always succeed, and we don't want this
1537 * to happen unless the file really is executable.
1538 * 3) Ensure that the file is a regular file.
1540 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1541 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1542 (attr->va_type != VREG))
1546 * Zero length files can't be exec'd
1548 if (attr->va_size == 0)
1552 * Check for execute permission to file based on current credentials.
1554 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1559 * Check number of open-for-writes on the file and deny execution
1562 error = VOP_GET_WRITECOUNT(vp, &writecount);
1565 if (writecount != 0)
1569 * Call filesystem specific open routine (which does nothing in the
1572 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1579 * Exec handler registration
1582 exec_register(execsw_arg)
1583 const struct execsw *execsw_arg;
1585 const struct execsw **es, **xs, **newexecsw;
1586 int count = 2; /* New slot and trailing NULL */
1589 for (es = execsw; *es; es++)
1591 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1594 for (es = execsw; *es; es++)
1599 free(execsw, M_TEMP);
1605 exec_unregister(execsw_arg)
1606 const struct execsw *execsw_arg;
1608 const struct execsw **es, **xs, **newexecsw;
1612 panic("unregister with no handlers left?\n");
1614 for (es = execsw; *es; es++) {
1615 if (*es == execsw_arg)
1620 for (es = execsw; *es; es++)
1621 if (*es != execsw_arg)
1623 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1625 for (es = execsw; *es; es++)
1626 if (*es != execsw_arg)
1630 free(execsw, M_TEMP);