2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (c) 1993, David Greenman
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
32 #include "opt_capsicum.h"
33 #include "opt_hwpmc_hooks.h"
34 #include "opt_ktrace.h"
37 #include <sys/param.h>
38 #include <sys/systm.h>
40 #include <sys/capsicum.h>
41 #include <sys/eventhandler.h>
43 #include <sys/fcntl.h>
44 #include <sys/filedesc.h>
45 #include <sys/imgact.h>
46 #include <sys/imgact_elf.h>
47 #include <sys/kernel.h>
49 #include <sys/malloc.h>
51 #include <sys/mount.h>
52 #include <sys/mutex.h>
53 #include <sys/namei.h>
54 #include <sys/pioctl.h>
57 #include <sys/ptrace.h>
58 #include <sys/resourcevar.h>
59 #include <sys/rwlock.h>
60 #include <sys/sched.h>
62 #include <sys/sf_buf.h>
64 #include <sys/signalvar.h>
67 #include <sys/syscallsubr.h>
68 #include <sys/sysctl.h>
69 #include <sys/sysent.h>
70 #include <sys/sysproto.h>
71 #include <sys/vnode.h>
74 #include <sys/ktrace.h>
78 #include <vm/vm_param.h>
80 #include <vm/vm_page.h>
81 #include <vm/vm_map.h>
82 #include <vm/vm_kern.h>
83 #include <vm/vm_extern.h>
84 #include <vm/vm_object.h>
85 #include <vm/vm_pager.h>
88 #include <sys/pmckern.h>
91 #include <machine/reg.h>
93 #include <security/audit/audit.h>
94 #include <security/mac/mac_framework.h>
97 #include <sys/dtrace_bsd.h>
98 dtrace_execexit_func_t dtrace_fasttrap_exec;
101 SDT_PROVIDER_DECLARE(proc);
102 SDT_PROBE_DEFINE1(proc, , , exec, "char *");
103 SDT_PROBE_DEFINE1(proc, , , exec__failure, "int");
104 SDT_PROBE_DEFINE1(proc, , , exec__success, "char *");
106 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
108 int coredump_pack_fileinfo = 1;
109 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_fileinfo, CTLFLAG_RWTUN,
110 &coredump_pack_fileinfo, 0,
111 "Enable file path packing in 'procstat -f' coredump notes");
113 int coredump_pack_vmmapinfo = 1;
114 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_vmmapinfo, CTLFLAG_RWTUN,
115 &coredump_pack_vmmapinfo, 0,
116 "Enable file path packing in 'procstat -v' coredump notes");
118 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
119 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
120 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
121 static int do_execve(struct thread *td, struct image_args *args,
124 /* XXX This should be vm_size_t. */
125 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD|
126 CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_ps_strings, "LU", "");
128 /* XXX This should be vm_size_t. */
129 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD|
130 CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_usrstack, "LU", "");
132 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD|CTLFLAG_MPSAFE,
133 NULL, 0, sysctl_kern_stackprot, "I", "");
135 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
136 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
137 &ps_arg_cache_limit, 0, "");
139 static int disallow_high_osrel;
140 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW,
141 &disallow_high_osrel, 0,
142 "Disallow execution of binaries built for higher version of the world");
144 static int map_at_zero = 0;
145 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RWTUN, &map_at_zero, 0,
146 "Permit processes to map an object at virtual address 0.");
148 EVENTHANDLER_LIST_DECLARE(process_exec);
151 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
158 if (req->flags & SCTL_MASK32) {
160 val = (unsigned int)p->p_sysent->sv_psstrings;
161 error = SYSCTL_OUT(req, &val, sizeof(val));
164 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
165 sizeof(p->p_sysent->sv_psstrings));
170 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
177 if (req->flags & SCTL_MASK32) {
179 val = (unsigned int)p->p_sysent->sv_usrstack;
180 error = SYSCTL_OUT(req, &val, sizeof(val));
183 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
184 sizeof(p->p_sysent->sv_usrstack));
189 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
194 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
195 sizeof(p->p_sysent->sv_stackprot)));
199 * Each of the items is a pointer to a `const struct execsw', hence the
200 * double pointer here.
202 static const struct execsw **execsw;
204 #ifndef _SYS_SYSPROTO_H_
213 sys_execve(struct thread *td, struct execve_args *uap)
215 struct image_args args;
216 struct vmspace *oldvmspace;
219 error = pre_execve(td, &oldvmspace);
222 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
223 uap->argv, uap->envv);
225 error = kern_execve(td, &args, NULL);
226 post_execve(td, error, oldvmspace);
230 #ifndef _SYS_SYSPROTO_H_
231 struct fexecve_args {
238 sys_fexecve(struct thread *td, struct fexecve_args *uap)
240 struct image_args args;
241 struct vmspace *oldvmspace;
244 error = pre_execve(td, &oldvmspace);
247 error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
248 uap->argv, uap->envv);
251 error = kern_execve(td, &args, NULL);
253 post_execve(td, error, oldvmspace);
257 #ifndef _SYS_SYSPROTO_H_
258 struct __mac_execve_args {
267 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap)
270 struct image_args args;
271 struct vmspace *oldvmspace;
274 error = pre_execve(td, &oldvmspace);
277 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
278 uap->argv, uap->envv);
280 error = kern_execve(td, &args, uap->mac_p);
281 post_execve(td, error, oldvmspace);
289 pre_execve(struct thread *td, struct vmspace **oldvmspace)
294 KASSERT(td == curthread, ("non-current thread %p", td));
297 if ((p->p_flag & P_HADTHREADS) != 0) {
299 if (thread_single(p, SINGLE_BOUNDARY) != 0)
303 KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0,
305 *oldvmspace = p->p_vmspace;
310 post_execve(struct thread *td, int error, struct vmspace *oldvmspace)
314 KASSERT(td == curthread, ("non-current thread %p", td));
316 if ((p->p_flag & P_HADTHREADS) != 0) {
319 * If success, we upgrade to SINGLE_EXIT state to
320 * force other threads to suicide.
322 if (error == EJUSTRETURN)
323 thread_single(p, SINGLE_EXIT);
325 thread_single_end(p, SINGLE_BOUNDARY);
328 if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
329 KASSERT(p->p_vmspace != oldvmspace,
330 ("oldvmspace still used"));
331 vmspace_free(oldvmspace);
332 td->td_pflags &= ~TDP_EXECVMSPC;
337 * XXX: kern_execve has the astonishing property of not always returning to
338 * the caller. If sufficiently bad things happen during the call to
339 * do_execve(), it can end up calling exit1(); as a result, callers must
340 * avoid doing anything which they might need to undo (e.g., allocating
344 kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p)
347 AUDIT_ARG_ARGV(args->begin_argv, args->argc,
348 args->begin_envv - args->begin_argv);
349 AUDIT_ARG_ENVV(args->begin_envv, args->envc,
350 args->endp - args->begin_envv);
351 return (do_execve(td, args, mac_p));
355 * In-kernel implementation of execve(). All arguments are assumed to be
356 * userspace pointers from the passed thread.
359 do_execve(struct thread *td, struct image_args *args, struct mac *mac_p)
361 struct proc *p = td->td_proc;
363 struct ucred *oldcred;
364 struct uidinfo *euip = NULL;
365 register_t *stack_base;
367 struct image_params image_params, *imgp;
369 int (*img_first)(struct image_params *);
370 struct pargs *oldargs = NULL, *newargs = NULL;
371 struct sigacts *oldsigacts = NULL, *newsigacts = NULL;
373 struct vnode *tracevp = NULL;
374 struct ucred *tracecred = NULL;
376 struct vnode *oldtextvp = NULL, *newtextvp;
377 int credential_changing;
380 struct label *interpvplabel = NULL;
384 struct pmckern_procexec pe;
386 static const char fexecv_proc_title[] = "(fexecv)";
388 imgp = &image_params;
391 * Lock the process and set the P_INEXEC flag to indicate that
392 * it should be left alone until we're done here. This is
393 * necessary to avoid race conditions - e.g. in ptrace() -
394 * that might allow a local user to illicitly obtain elevated
398 KASSERT((p->p_flag & P_INEXEC) == 0,
399 ("%s(): process already has P_INEXEC flag", __func__));
400 p->p_flag |= P_INEXEC;
404 * Initialize part of the common data
406 bzero(imgp, sizeof(*imgp));
410 oldcred = p->p_ucred;
413 error = mac_execve_enter(imgp, mac_p);
419 * Translate the file name. namei() returns a vnode pointer
420 * in ni_vp among other things.
422 * XXXAUDIT: It would be desirable to also audit the name of the
423 * interpreter if this is an interpreted binary.
425 if (args->fname != NULL) {
426 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME
427 | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
430 SDT_PROBE1(proc, , , exec, args->fname);
433 if (args->fname != NULL) {
434 #ifdef CAPABILITY_MODE
436 * While capability mode can't reach this point via direct
437 * path arguments to execve(), we also don't allow
438 * interpreters to be used in capability mode (for now).
439 * Catch indirect lookups and return a permissions error.
441 if (IN_CAPABILITY_MODE(td)) {
450 newtextvp = nd.ni_vp;
451 imgp->vp = newtextvp;
453 AUDIT_ARG_FD(args->fd);
455 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
457 error = fgetvp_exec(td, args->fd, &cap_fexecve_rights, &newtextvp);
460 vn_lock(newtextvp, LK_EXCLUSIVE | LK_RETRY);
461 AUDIT_ARG_VNODE1(newtextvp);
462 imgp->vp = newtextvp;
466 * Check file permissions (also 'opens' file)
468 error = exec_check_permissions(imgp);
470 goto exec_fail_dealloc;
472 imgp->object = imgp->vp->v_object;
473 if (imgp->object != NULL)
474 vm_object_reference(imgp->object);
477 * Set VV_TEXT now so no one can write to the executable while we're
480 * Remember if this was set before and unset it in case this is not
481 * actually an executable image.
483 textset = VOP_IS_TEXT(imgp->vp);
484 VOP_SET_TEXT(imgp->vp);
486 error = exec_map_first_page(imgp);
488 goto exec_fail_dealloc;
490 imgp->proc->p_osrel = 0;
493 * Implement image setuid/setgid.
495 * Determine new credentials before attempting image activators
496 * so that it can be used by process_exec handlers to determine
497 * credential/setid changes.
499 * Don't honor setuid/setgid if the filesystem prohibits it or if
500 * the process is being traced.
502 * We disable setuid/setgid/etc in capability mode on the basis
503 * that most setugid applications are not written with that
504 * environment in mind, and will therefore almost certainly operate
505 * incorrectly. In principle there's no reason that setugid
506 * applications might not be useful in capability mode, so we may want
507 * to reconsider this conservative design choice in the future.
509 * XXXMAC: For the time being, use NOSUID to also prohibit
510 * transitions on the file system.
512 credential_changing = 0;
513 credential_changing |= (attr.va_mode & S_ISUID) &&
514 oldcred->cr_uid != attr.va_uid;
515 credential_changing |= (attr.va_mode & S_ISGID) &&
516 oldcred->cr_gid != attr.va_gid;
518 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
519 interpvplabel, imgp);
520 credential_changing |= will_transition;
523 /* Don't inherit PROC_PDEATHSIG_CTL value if setuid/setgid. */
524 if (credential_changing)
525 imgp->proc->p_pdeathsig = 0;
527 if (credential_changing &&
528 #ifdef CAPABILITY_MODE
529 ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
531 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
532 (p->p_flag & P_TRACED) == 0) {
533 imgp->credential_setid = true;
534 VOP_UNLOCK(imgp->vp, 0);
535 imgp->newcred = crdup(oldcred);
536 if (attr.va_mode & S_ISUID) {
537 euip = uifind(attr.va_uid);
538 change_euid(imgp->newcred, euip);
540 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
541 if (attr.va_mode & S_ISGID)
542 change_egid(imgp->newcred, attr.va_gid);
544 * Implement correct POSIX saved-id behavior.
546 * XXXMAC: Note that the current logic will save the
547 * uid and gid if a MAC domain transition occurs, even
548 * though maybe it shouldn't.
550 change_svuid(imgp->newcred, imgp->newcred->cr_uid);
551 change_svgid(imgp->newcred, imgp->newcred->cr_gid);
554 * Implement correct POSIX saved-id behavior.
556 * XXX: It's not clear that the existing behavior is
557 * POSIX-compliant. A number of sources indicate that the
558 * saved uid/gid should only be updated if the new ruid is
559 * not equal to the old ruid, or the new euid is not equal
560 * to the old euid and the new euid is not equal to the old
561 * ruid. The FreeBSD code always updates the saved uid/gid.
562 * Also, this code uses the new (replaced) euid and egid as
563 * the source, which may or may not be the right ones to use.
565 if (oldcred->cr_svuid != oldcred->cr_uid ||
566 oldcred->cr_svgid != oldcred->cr_gid) {
567 VOP_UNLOCK(imgp->vp, 0);
568 imgp->newcred = crdup(oldcred);
569 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
570 change_svuid(imgp->newcred, imgp->newcred->cr_uid);
571 change_svgid(imgp->newcred, imgp->newcred->cr_gid);
574 /* The new credentials are installed into the process later. */
577 * Do the best to calculate the full path to the image file.
579 if (args->fname != NULL && args->fname[0] == '/')
580 imgp->execpath = args->fname;
582 VOP_UNLOCK(imgp->vp, 0);
583 if (vn_fullpath(td, imgp->vp, &imgp->execpath,
584 &imgp->freepath) != 0)
585 imgp->execpath = args->fname;
586 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
590 * If the current process has a special image activator it
591 * wants to try first, call it. For example, emulating shell
592 * scripts differently.
595 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
596 error = img_first(imgp);
599 * Loop through the list of image activators, calling each one.
600 * An activator returns -1 if there is no match, 0 on success,
601 * and an error otherwise.
603 for (i = 0; error == -1 && execsw[i]; ++i) {
604 if (execsw[i]->ex_imgact == NULL ||
605 execsw[i]->ex_imgact == img_first) {
608 error = (*execsw[i]->ex_imgact)(imgp);
614 VOP_UNSET_TEXT(imgp->vp);
617 goto exec_fail_dealloc;
621 * Special interpreter operation, cleanup and loop up to try to
622 * activate the interpreter.
624 if (imgp->interpreted) {
625 exec_unmap_first_page(imgp);
627 * VV_TEXT needs to be unset for scripts. There is a short
628 * period before we determine that something is a script where
629 * VV_TEXT will be set. The vnode lock is held over this
630 * entire period so nothing should illegitimately be blocked.
632 VOP_UNSET_TEXT(imgp->vp);
633 /* free name buffer and old vnode */
634 if (args->fname != NULL)
635 NDFREE(&nd, NDF_ONLY_PNBUF);
637 mac_execve_interpreter_enter(newtextvp, &interpvplabel);
640 VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td);
644 vm_object_deallocate(imgp->object);
646 imgp->credential_setid = false;
647 if (imgp->newcred != NULL) {
648 crfree(imgp->newcred);
649 imgp->newcred = NULL;
651 imgp->execpath = NULL;
652 free(imgp->freepath, M_TEMP);
653 imgp->freepath = NULL;
654 /* set new name to that of the interpreter */
655 NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
656 UIO_SYSSPACE, imgp->interpreter_name, td);
657 args->fname = imgp->interpreter_name;
662 * NB: We unlock the vnode here because it is believed that none
663 * of the sv_copyout_strings/sv_fixup operations require the vnode.
665 VOP_UNLOCK(imgp->vp, 0);
667 if (disallow_high_osrel &&
668 P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
670 uprintf("Osrel %d for image %s too high\n", p->p_osrel,
671 imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
672 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
673 goto exec_fail_dealloc;
676 /* ABI enforces the use of Capsicum. Switch into capabilities mode. */
677 if (SV_PROC_FLAG(p, SV_CAPSICUM))
678 sys_cap_enter(td, NULL);
681 * Copy out strings (args and env) and initialize stack base
683 if (p->p_sysent->sv_copyout_strings)
684 stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
686 stack_base = exec_copyout_strings(imgp);
689 * If custom stack fixup routine present for this process
690 * let it do the stack setup.
691 * Else stuff argument count as first item on stack
693 if (p->p_sysent->sv_fixup != NULL)
694 error = (*p->p_sysent->sv_fixup)(&stack_base, imgp);
696 error = suword(--stack_base, imgp->args->argc) == 0 ?
699 goto exec_fail_dealloc;
701 if (args->fdp != NULL) {
702 /* Install a brand new file descriptor table. */
703 fdinstall_remapped(td, args->fdp);
707 * Keep on using the existing file descriptor table. For
708 * security and other reasons, the file descriptor table
709 * cannot be shared after an exec.
712 /* close files on exec */
717 * Malloc things before we need locks.
719 i = imgp->args->begin_envv - imgp->args->begin_argv;
720 /* Cache arguments if they fit inside our allowance */
721 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
722 newargs = pargs_alloc(i);
723 bcopy(imgp->args->begin_argv, newargs->ar_args, i);
727 * For security and other reasons, signal handlers cannot
728 * be shared after an exec. The new process gets a copy of the old
729 * handlers. In execsigs(), the new process will have its signals
732 if (sigacts_shared(p->p_sigacts)) {
733 oldsigacts = p->p_sigacts;
734 newsigacts = sigacts_alloc();
735 sigacts_copy(newsigacts, oldsigacts);
738 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
742 p->p_sigacts = newsigacts;
746 /* reset caught signals */
749 /* name this process - nameiexec(p, ndp) */
750 bzero(p->p_comm, sizeof(p->p_comm));
752 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
753 min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
754 else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0)
755 bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
756 bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
758 sched_clear_tdname(td);
762 * mark as execed, wakeup the process that vforked (if any) and tell
763 * it that it now has its own resources back
766 if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0)
767 p->p_flag2 &= ~P2_NOTRACE;
768 if (p->p_flag & P_PPWAIT) {
769 p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
770 cv_broadcast(&p->p_pwait);
771 /* STOPs are no longer ignored, arrange for AST */
776 * Implement image setuid/setgid installation.
778 if (imgp->credential_setid) {
780 * Turn off syscall tracing for set-id programs, except for
781 * root. Record any set-id flags first to make sure that
782 * we do not regain any tracing during a possible block.
787 if (p->p_tracecred != NULL &&
788 priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED, 0))
789 ktrprocexec(p, &tracecred, &tracevp);
792 * Close any file descriptors 0..2 that reference procfs,
793 * then make sure file descriptors 0..2 are in use.
795 * Both fdsetugidsafety() and fdcheckstd() may call functions
796 * taking sleepable locks, so temporarily drop our locks.
799 VOP_UNLOCK(imgp->vp, 0);
801 error = fdcheckstd(td);
802 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
804 goto exec_fail_dealloc;
807 if (will_transition) {
808 mac_vnode_execve_transition(oldcred, imgp->newcred,
809 imgp->vp, interpvplabel, imgp);
813 if (oldcred->cr_uid == oldcred->cr_ruid &&
814 oldcred->cr_gid == oldcred->cr_rgid)
815 p->p_flag &= ~P_SUGID;
818 * Set the new credentials.
820 if (imgp->newcred != NULL) {
821 proc_set_cred(p, imgp->newcred);
827 * Store the vp for use in procfs. This vnode was referenced by namei
830 oldtextvp = p->p_textvp;
831 p->p_textvp = newtextvp;
835 * Tell the DTrace fasttrap provider about the exec if it
836 * has declared an interest.
838 if (dtrace_fasttrap_exec)
839 dtrace_fasttrap_exec(p);
843 * Notify others that we exec'd, and clear the P_INEXEC flag
844 * as we're now a bona fide freshly-execed process.
846 KNOTE_LOCKED(p->p_klist, NOTE_EXEC);
847 p->p_flag &= ~P_INEXEC;
849 /* clear "fork but no exec" flag, as we _are_ execing */
850 p->p_acflag &= ~AFORK;
853 * Free any previous argument cache and replace it with
854 * the new argument cache, if any.
864 * Check if system-wide sampling is in effect or if the
865 * current process is using PMCs. If so, do exec() time
866 * processing. This processing needs to happen AFTER the
867 * P_INEXEC flag is cleared.
869 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
870 VOP_UNLOCK(imgp->vp, 0);
871 pe.pm_credentialschanged = credential_changing;
872 pe.pm_entryaddr = imgp->entry_addr;
874 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
875 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
879 /* Set values passed into the program in registers. */
880 if (p->p_sysent->sv_setregs)
881 (*p->p_sysent->sv_setregs)(td, imgp,
882 (u_long)(uintptr_t)stack_base);
884 exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base);
886 vfs_mark_atime(imgp->vp, td->td_ucred);
888 SDT_PROBE1(proc, , , exec__success, args->fname);
891 if (imgp->firstpage != NULL)
892 exec_unmap_first_page(imgp);
894 if (imgp->vp != NULL) {
896 NDFREE(&nd, NDF_ONLY_PNBUF);
898 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
902 VOP_UNLOCK(imgp->vp, 0);
905 if (imgp->object != NULL)
906 vm_object_deallocate(imgp->object);
908 free(imgp->freepath, M_TEMP);
911 if (p->p_ptevents & PTRACE_EXEC) {
913 if (p->p_ptevents & PTRACE_EXEC)
914 td->td_dbgflags |= TDB_EXEC;
919 * Stop the process here if its stop event mask has
920 * the S_EXEC bit set.
922 STOPEVENT(p, S_EXEC, 0);
925 /* we're done here, clear P_INEXEC */
927 p->p_flag &= ~P_INEXEC;
930 SDT_PROBE1(proc, , , exec__failure, error);
933 if (imgp->newcred != NULL && oldcred != NULL)
934 crfree(imgp->newcred);
937 mac_execve_exit(imgp);
938 mac_execve_interpreter_exit(interpvplabel);
940 exec_free_args(args);
943 * Handle deferred decrement of ref counts.
945 if (oldtextvp != NULL)
950 if (tracecred != NULL)
955 if (oldsigacts != NULL)
956 sigacts_free(oldsigacts);
960 if (error && imgp->vmspace_destroyed) {
961 /* sorry, no more process anymore. exit gracefully */
962 exit1(td, 0, SIGABRT);
972 * We don't want cpu_set_syscall_retval() to overwrite any of
973 * the register values put in place by exec_setregs().
974 * Implementations of cpu_set_syscall_retval() will leave
975 * registers unmodified when returning EJUSTRETURN.
977 return (error == 0 ? EJUSTRETURN : error);
981 exec_map_first_page(struct image_params *imgp)
983 int rv, i, after, initial_pagein;
984 vm_page_t ma[VM_INITIAL_PAGEIN];
987 if (imgp->firstpage != NULL)
988 exec_unmap_first_page(imgp);
990 object = imgp->vp->v_object;
993 VM_OBJECT_WLOCK(object);
994 #if VM_NRESERVLEVEL > 0
995 vm_object_color(object, 0);
997 ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY);
998 if (ma[0]->valid != VM_PAGE_BITS_ALL) {
999 vm_page_xbusy(ma[0]);
1000 if (!vm_pager_has_page(object, 0, NULL, &after)) {
1001 vm_page_lock(ma[0]);
1002 vm_page_free(ma[0]);
1003 vm_page_unlock(ma[0]);
1004 VM_OBJECT_WUNLOCK(object);
1007 initial_pagein = min(after, VM_INITIAL_PAGEIN);
1008 KASSERT(initial_pagein <= object->size,
1009 ("%s: initial_pagein %d object->size %ju",
1010 __func__, initial_pagein, (uintmax_t )object->size));
1011 for (i = 1; i < initial_pagein; i++) {
1012 if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) {
1015 if (!vm_page_tryxbusy(ma[i]))
1018 ma[i] = vm_page_alloc(object, i,
1025 rv = vm_pager_get_pages(object, ma, initial_pagein, NULL, NULL);
1026 if (rv != VM_PAGER_OK) {
1027 for (i = 0; i < initial_pagein; i++) {
1028 vm_page_lock(ma[i]);
1029 vm_page_free(ma[i]);
1030 vm_page_unlock(ma[i]);
1032 VM_OBJECT_WUNLOCK(object);
1035 vm_page_xunbusy(ma[0]);
1036 for (i = 1; i < initial_pagein; i++)
1037 vm_page_readahead_finish(ma[i]);
1039 vm_page_lock(ma[0]);
1040 vm_page_hold(ma[0]);
1041 vm_page_activate(ma[0]);
1042 vm_page_unlock(ma[0]);
1043 VM_OBJECT_WUNLOCK(object);
1045 imgp->firstpage = sf_buf_alloc(ma[0], 0);
1046 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
1052 exec_unmap_first_page(struct image_params *imgp)
1056 if (imgp->firstpage != NULL) {
1057 m = sf_buf_page(imgp->firstpage);
1058 sf_buf_free(imgp->firstpage);
1059 imgp->firstpage = NULL;
1067 * Destroy old address space, and allocate a new stack.
1068 * The new stack is only sgrowsiz large because it is grown
1069 * automatically on a page fault.
1072 exec_new_vmspace(struct image_params *imgp, struct sysentvec *sv)
1075 struct proc *p = imgp->proc;
1076 struct vmspace *vmspace = p->p_vmspace;
1078 struct rlimit rlim_stack;
1079 vm_offset_t sv_minuser, stack_addr;
1083 imgp->vmspace_destroyed = 1;
1086 /* May be called with Giant held */
1087 EVENTHANDLER_DIRECT_INVOKE(process_exec, p, imgp);
1090 * Blow away entire process VM, if address space not shared,
1091 * otherwise, create a new VM space so that other threads are
1094 map = &vmspace->vm_map;
1096 sv_minuser = sv->sv_minuser;
1098 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1099 if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser &&
1100 vm_map_max(map) == sv->sv_maxuser) {
1102 pmap_remove_pages(vmspace_pmap(vmspace));
1103 vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1104 /* An exec terminates mlockall(MCL_FUTURE). */
1106 vm_map_modflags(map, 0, MAP_WIREFUTURE);
1109 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1112 vmspace = p->p_vmspace;
1113 map = &vmspace->vm_map;
1116 /* Map a shared page */
1117 obj = sv->sv_shared_page_obj;
1119 vm_object_reference(obj);
1120 error = vm_map_fixed(map, obj, 0,
1121 sv->sv_shared_page_base, sv->sv_shared_page_len,
1122 VM_PROT_READ | VM_PROT_EXECUTE,
1123 VM_PROT_READ | VM_PROT_EXECUTE,
1124 MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1125 if (error != KERN_SUCCESS) {
1126 vm_object_deallocate(obj);
1127 return (vm_mmap_to_errno(error));
1131 /* Allocate a new stack */
1132 if (imgp->stack_sz != 0) {
1133 ssiz = trunc_page(imgp->stack_sz);
1135 lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack);
1137 if (ssiz > rlim_stack.rlim_max)
1138 ssiz = rlim_stack.rlim_max;
1139 if (ssiz > rlim_stack.rlim_cur) {
1140 rlim_stack.rlim_cur = ssiz;
1141 kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack);
1143 } else if (sv->sv_maxssiz != NULL) {
1144 ssiz = *sv->sv_maxssiz;
1148 stack_addr = sv->sv_usrstack - ssiz;
1149 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1150 obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1151 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1152 if (error != KERN_SUCCESS)
1153 return (vm_mmap_to_errno(error));
1156 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they
1157 * are still used to enforce the stack rlimit on the process stack.
1159 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1160 vmspace->vm_maxsaddr = (char *)stack_addr;
1166 * Copy out argument and environment strings from the old process address
1167 * space into the temporary string buffer.
1170 exec_copyin_args(struct image_args *args, char *fname,
1171 enum uio_seg segflg, char **argv, char **envv)
1177 bzero(args, sizeof(*args));
1182 * Allocate demand-paged memory for the file name, argument, and
1183 * environment strings.
1185 error = exec_alloc_args(args);
1190 * Copy the file name.
1192 if (fname != NULL) {
1193 args->fname = args->buf;
1194 error = (segflg == UIO_SYSSPACE) ?
1195 copystr(fname, args->fname, PATH_MAX, &length) :
1196 copyinstr(fname, args->fname, PATH_MAX, &length);
1202 args->begin_argv = args->buf + length;
1203 args->endp = args->begin_argv;
1204 args->stringspace = ARG_MAX;
1207 * extract arguments first
1210 error = fueword(argv++, &argp);
1217 error = copyinstr((void *)(uintptr_t)argp, args->endp,
1218 args->stringspace, &length);
1220 if (error == ENAMETOOLONG)
1224 args->stringspace -= length;
1225 args->endp += length;
1229 args->begin_envv = args->endp;
1232 * extract environment strings
1236 error = fueword(envv++, &envp);
1243 error = copyinstr((void *)(uintptr_t)envp,
1244 args->endp, args->stringspace, &length);
1246 if (error == ENAMETOOLONG)
1250 args->stringspace -= length;
1251 args->endp += length;
1259 exec_free_args(args);
1264 exec_copyin_data_fds(struct thread *td, struct image_args *args,
1265 const void *data, size_t datalen, const int *fds, size_t fdslen)
1267 struct filedesc *ofdp;
1272 memset(args, '\0', sizeof(*args));
1273 ofdp = td->td_proc->p_fd;
1274 if (datalen >= ARG_MAX || fdslen > ofdp->fd_lastfile + 1)
1276 error = exec_alloc_args(args);
1280 args->begin_argv = args->buf;
1281 args->stringspace = ARG_MAX;
1285 * Argument buffer has been provided. Copy it into the
1286 * kernel as a single string and add a terminating null
1289 error = copyin(data, args->begin_argv, datalen);
1292 args->begin_argv[datalen] = '\0';
1293 args->endp = args->begin_argv + datalen + 1;
1294 args->stringspace -= datalen + 1;
1297 * Traditional argument counting. Count the number of
1300 for (p = args->begin_argv; p < args->endp; ++p)
1304 /* No argument buffer provided. */
1305 args->endp = args->begin_argv;
1307 /* There are no environment variables. */
1308 args->begin_envv = args->endp;
1310 /* Create new file descriptor table. */
1311 kfds = malloc(fdslen * sizeof(int), M_TEMP, M_WAITOK);
1312 error = copyin(fds, kfds, fdslen * sizeof(int));
1317 error = fdcopy_remapped(ofdp, kfds, fdslen, &args->fdp);
1324 exec_free_args(args);
1328 struct exec_args_kva {
1331 SLIST_ENTRY(exec_args_kva) next;
1334 DPCPU_DEFINE_STATIC(struct exec_args_kva *, exec_args_kva);
1336 static SLIST_HEAD(, exec_args_kva) exec_args_kva_freelist;
1337 static struct mtx exec_args_kva_mtx;
1338 static u_int exec_args_gen;
1341 exec_prealloc_args_kva(void *arg __unused)
1343 struct exec_args_kva *argkva;
1346 SLIST_INIT(&exec_args_kva_freelist);
1347 mtx_init(&exec_args_kva_mtx, "exec args kva", NULL, MTX_DEF);
1348 for (i = 0; i < exec_map_entries; i++) {
1349 argkva = malloc(sizeof(*argkva), M_PARGS, M_WAITOK);
1350 argkva->addr = kmap_alloc_wait(exec_map, exec_map_entry_size);
1351 argkva->gen = exec_args_gen;
1352 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1355 SYSINIT(exec_args_kva, SI_SUB_EXEC, SI_ORDER_ANY, exec_prealloc_args_kva, NULL);
1358 exec_alloc_args_kva(void **cookie)
1360 struct exec_args_kva *argkva;
1362 argkva = (void *)atomic_readandclear_ptr(
1363 (uintptr_t *)DPCPU_PTR(exec_args_kva));
1364 if (argkva == NULL) {
1365 mtx_lock(&exec_args_kva_mtx);
1366 while ((argkva = SLIST_FIRST(&exec_args_kva_freelist)) == NULL)
1367 (void)mtx_sleep(&exec_args_kva_freelist,
1368 &exec_args_kva_mtx, 0, "execkva", 0);
1369 SLIST_REMOVE_HEAD(&exec_args_kva_freelist, next);
1370 mtx_unlock(&exec_args_kva_mtx);
1372 *(struct exec_args_kva **)cookie = argkva;
1373 return (argkva->addr);
1377 exec_release_args_kva(struct exec_args_kva *argkva, u_int gen)
1381 base = argkva->addr;
1382 if (argkva->gen != gen) {
1383 (void)vm_map_madvise(exec_map, base, base + exec_map_entry_size,
1387 if (!atomic_cmpset_ptr((uintptr_t *)DPCPU_PTR(exec_args_kva),
1388 (uintptr_t)NULL, (uintptr_t)argkva)) {
1389 mtx_lock(&exec_args_kva_mtx);
1390 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1391 wakeup_one(&exec_args_kva_freelist);
1392 mtx_unlock(&exec_args_kva_mtx);
1397 exec_free_args_kva(void *cookie)
1400 exec_release_args_kva(cookie, exec_args_gen);
1404 exec_args_kva_lowmem(void *arg __unused)
1406 SLIST_HEAD(, exec_args_kva) head;
1407 struct exec_args_kva *argkva;
1411 gen = atomic_fetchadd_int(&exec_args_gen, 1) + 1;
1414 * Force an madvise of each KVA range. Any currently allocated ranges
1415 * will have MADV_FREE applied once they are freed.
1418 mtx_lock(&exec_args_kva_mtx);
1419 SLIST_SWAP(&head, &exec_args_kva_freelist, exec_args_kva);
1420 mtx_unlock(&exec_args_kva_mtx);
1421 while ((argkva = SLIST_FIRST(&head)) != NULL) {
1422 SLIST_REMOVE_HEAD(&head, next);
1423 exec_release_args_kva(argkva, gen);
1427 argkva = (void *)atomic_readandclear_ptr(
1428 (uintptr_t *)DPCPU_ID_PTR(i, exec_args_kva));
1430 exec_release_args_kva(argkva, gen);
1433 EVENTHANDLER_DEFINE(vm_lowmem, exec_args_kva_lowmem, NULL,
1434 EVENTHANDLER_PRI_ANY);
1437 * Allocate temporary demand-paged, zero-filled memory for the file name,
1438 * argument, and environment strings.
1441 exec_alloc_args(struct image_args *args)
1444 args->buf = (char *)exec_alloc_args_kva(&args->bufkva);
1449 exec_free_args(struct image_args *args)
1452 if (args->buf != NULL) {
1453 exec_free_args_kva(args->bufkva);
1456 if (args->fname_buf != NULL) {
1457 free(args->fname_buf, M_TEMP);
1458 args->fname_buf = NULL;
1460 if (args->fdp != NULL)
1461 fdescfree_remapped(args->fdp);
1465 * Copy strings out to the new process address space, constructing new arg
1466 * and env vector tables. Return a pointer to the base so that it can be used
1467 * as the initial stack pointer.
1470 exec_copyout_strings(struct image_params *imgp)
1476 register_t *stack_base;
1477 struct ps_strings *arginfo;
1479 size_t execpath_len;
1480 int szsigcode, szps;
1481 char canary[sizeof(long) * 8];
1483 szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1485 * Calculate string base and vector table pointers.
1486 * Also deal with signal trampoline code for this exec type.
1488 if (imgp->execpath != NULL && imgp->auxargs != NULL)
1489 execpath_len = strlen(imgp->execpath) + 1;
1494 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1495 if (p->p_sysent->sv_sigcode_base == 0) {
1496 if (p->p_sysent->sv_szsigcode != NULL)
1497 szsigcode = *(p->p_sysent->sv_szsigcode);
1499 destp = (uintptr_t)arginfo;
1504 if (szsigcode != 0) {
1506 destp = rounddown2(destp, sizeof(void *));
1507 copyout(p->p_sysent->sv_sigcode, (void *)destp, szsigcode);
1511 * Copy the image path for the rtld.
1513 if (execpath_len != 0) {
1514 destp -= execpath_len;
1515 destp = rounddown2(destp, sizeof(void *));
1516 imgp->execpathp = destp;
1517 copyout(imgp->execpath, (void *)destp, execpath_len);
1521 * Prepare the canary for SSP.
1523 arc4rand(canary, sizeof(canary), 0);
1524 destp -= sizeof(canary);
1525 imgp->canary = destp;
1526 copyout(canary, (void *)destp, sizeof(canary));
1527 imgp->canarylen = sizeof(canary);
1530 * Prepare the pagesizes array.
1533 destp = rounddown2(destp, sizeof(void *));
1534 imgp->pagesizes = destp;
1535 copyout(pagesizes, (void *)destp, szps);
1536 imgp->pagesizeslen = szps;
1538 destp -= ARG_MAX - imgp->args->stringspace;
1539 destp = rounddown2(destp, sizeof(void *));
1541 vectp = (char **)destp;
1542 if (imgp->auxargs) {
1544 * Allocate room on the stack for the ELF auxargs
1545 * array. It has up to AT_COUNT entries.
1547 vectp -= howmany(AT_COUNT * sizeof(Elf_Auxinfo),
1552 * Allocate room for the argv[] and env vectors including the
1553 * terminating NULL pointers.
1555 vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
1558 * vectp also becomes our initial stack base
1560 stack_base = (register_t *)vectp;
1562 stringp = imgp->args->begin_argv;
1563 argc = imgp->args->argc;
1564 envc = imgp->args->envc;
1567 * Copy out strings - arguments and environment.
1569 copyout(stringp, (void *)destp, ARG_MAX - imgp->args->stringspace);
1572 * Fill in "ps_strings" struct for ps, w, etc.
1574 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1575 suword32(&arginfo->ps_nargvstr, argc);
1578 * Fill in argument portion of vector table.
1580 for (; argc > 0; --argc) {
1581 suword(vectp++, (long)(intptr_t)destp);
1582 while (*stringp++ != 0)
1587 /* a null vector table pointer separates the argp's from the envp's */
1590 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1591 suword32(&arginfo->ps_nenvstr, envc);
1594 * Fill in environment portion of vector table.
1596 for (; envc > 0; --envc) {
1597 suword(vectp++, (long)(intptr_t)destp);
1598 while (*stringp++ != 0)
1603 /* end of vector table is a null pointer */
1606 return (stack_base);
1610 * Check permissions of file to execute.
1611 * Called with imgp->vp locked.
1612 * Return 0 for success or error code on failure.
1615 exec_check_permissions(struct image_params *imgp)
1617 struct vnode *vp = imgp->vp;
1618 struct vattr *attr = imgp->attr;
1620 int error, writecount;
1624 /* Get file attributes */
1625 error = VOP_GETATTR(vp, attr, td->td_ucred);
1630 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1636 * 1) Check if file execution is disabled for the filesystem that
1637 * this file resides on.
1638 * 2) Ensure that at least one execute bit is on. Otherwise, a
1639 * privileged user will always succeed, and we don't want this
1640 * to happen unless the file really is executable.
1641 * 3) Ensure that the file is a regular file.
1643 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1644 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1645 (attr->va_type != VREG))
1649 * Zero length files can't be exec'd
1651 if (attr->va_size == 0)
1655 * Check for execute permission to file based on current credentials.
1657 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1662 * Check number of open-for-writes on the file and deny execution
1665 error = VOP_GET_WRITECOUNT(vp, &writecount);
1668 if (writecount != 0)
1672 * Call filesystem specific open routine (which does nothing in the
1675 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1682 * Exec handler registration
1685 exec_register(const struct execsw *execsw_arg)
1687 const struct execsw **es, **xs, **newexecsw;
1688 u_int count = 2; /* New slot and trailing NULL */
1691 for (es = execsw; *es; es++)
1693 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1696 for (es = execsw; *es; es++)
1701 free(execsw, M_TEMP);
1707 exec_unregister(const struct execsw *execsw_arg)
1709 const struct execsw **es, **xs, **newexecsw;
1713 panic("unregister with no handlers left?\n");
1715 for (es = execsw; *es; es++) {
1716 if (*es == execsw_arg)
1721 for (es = execsw; *es; es++)
1722 if (*es != execsw_arg)
1724 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1726 for (es = execsw; *es; es++)
1727 if (*es != execsw_arg)
1731 free(execsw, M_TEMP);