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_compat.h"
32 #include "opt_hwpmc_hooks.h"
33 #include "opt_ktrace.h"
36 #include <sys/param.h>
37 #include <sys/systm.h>
39 #include <sys/capsicum.h>
40 #include <sys/eventhandler.h>
42 #include <sys/fcntl.h>
43 #include <sys/filedesc.h>
44 #include <sys/imgact.h>
45 #include <sys/imgact_elf.h>
46 #include <sys/kernel.h>
48 #include <sys/malloc.h>
50 #include <sys/mount.h>
51 #include <sys/mutex.h>
52 #include <sys/namei.h>
53 #include <sys/pioctl.h>
56 #include <sys/ptrace.h>
57 #include <sys/resourcevar.h>
58 #include <sys/rwlock.h>
59 #include <sys/sched.h>
61 #include <sys/sf_buf.h>
63 #include <sys/signalvar.h>
66 #include <sys/syscallsubr.h>
67 #include <sys/sysctl.h>
68 #include <sys/sysent.h>
69 #include <sys/sysproto.h>
70 #include <sys/vnode.h>
73 #include <sys/ktrace.h>
77 #include <vm/vm_param.h>
79 #include <vm/vm_page.h>
80 #include <vm/vm_map.h>
81 #include <vm/vm_kern.h>
82 #include <vm/vm_extern.h>
83 #include <vm/vm_object.h>
84 #include <vm/vm_pager.h>
87 #include <sys/pmckern.h>
90 #include <machine/reg.h>
92 #include <security/audit/audit.h>
93 #include <security/mac/mac_framework.h>
96 #include <sys/dtrace_bsd.h>
97 dtrace_execexit_func_t dtrace_fasttrap_exec;
100 SDT_PROVIDER_DECLARE(proc);
101 SDT_PROBE_DEFINE1(proc, , , exec, "char *");
102 SDT_PROBE_DEFINE1(proc, , , exec__failure, "int");
103 SDT_PROBE_DEFINE1(proc, , , exec__success, "char *");
105 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
107 int coredump_pack_fileinfo = 1;
108 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_fileinfo, CTLFLAG_RWTUN,
109 &coredump_pack_fileinfo, 0,
110 "Enable file path packing in 'procstat -f' coredump notes");
112 int coredump_pack_vmmapinfo = 1;
113 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_vmmapinfo, CTLFLAG_RWTUN,
114 &coredump_pack_vmmapinfo, 0,
115 "Enable file path packing in 'procstat -v' coredump notes");
117 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
118 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
119 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
120 static int do_execve(struct thread *td, struct image_args *args,
123 /* XXX This should be vm_size_t. */
124 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD|
125 CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_ps_strings, "LU", "");
127 /* XXX This should be vm_size_t. */
128 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD|
129 CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_usrstack, "LU", "");
131 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD|CTLFLAG_MPSAFE,
132 NULL, 0, sysctl_kern_stackprot, "I", "");
134 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
135 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
136 &ps_arg_cache_limit, 0, "");
138 static int disallow_high_osrel;
139 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW,
140 &disallow_high_osrel, 0,
141 "Disallow execution of binaries built for higher version of the world");
143 static int map_at_zero = 0;
144 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RWTUN, &map_at_zero, 0,
145 "Permit processes to map an object at virtual address 0.");
148 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
155 if (req->flags & SCTL_MASK32) {
157 val = (unsigned int)p->p_sysent->sv_psstrings;
158 error = SYSCTL_OUT(req, &val, sizeof(val));
161 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
162 sizeof(p->p_sysent->sv_psstrings));
167 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
174 if (req->flags & SCTL_MASK32) {
176 val = (unsigned int)p->p_sysent->sv_usrstack;
177 error = SYSCTL_OUT(req, &val, sizeof(val));
180 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
181 sizeof(p->p_sysent->sv_usrstack));
186 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
191 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
192 sizeof(p->p_sysent->sv_stackprot)));
196 * Each of the items is a pointer to a `const struct execsw', hence the
197 * double pointer here.
199 static const struct execsw **execsw;
201 #ifndef _SYS_SYSPROTO_H_
210 sys_execve(struct thread *td, struct execve_args *uap)
212 struct image_args args;
213 struct vmspace *oldvmspace;
216 error = pre_execve(td, &oldvmspace);
219 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
220 uap->argv, uap->envv);
222 error = kern_execve(td, &args, NULL);
223 post_execve(td, error, oldvmspace);
227 #ifndef _SYS_SYSPROTO_H_
228 struct fexecve_args {
235 sys_fexecve(struct thread *td, struct fexecve_args *uap)
237 struct image_args args;
238 struct vmspace *oldvmspace;
241 error = pre_execve(td, &oldvmspace);
244 error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
245 uap->argv, uap->envv);
248 error = kern_execve(td, &args, NULL);
250 post_execve(td, error, oldvmspace);
254 #ifndef _SYS_SYSPROTO_H_
255 struct __mac_execve_args {
264 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap)
267 struct image_args args;
268 struct vmspace *oldvmspace;
271 error = pre_execve(td, &oldvmspace);
274 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
275 uap->argv, uap->envv);
277 error = kern_execve(td, &args, uap->mac_p);
278 post_execve(td, error, oldvmspace);
286 pre_execve(struct thread *td, struct vmspace **oldvmspace)
291 KASSERT(td == curthread, ("non-current thread %p", td));
294 if ((p->p_flag & P_HADTHREADS) != 0) {
296 if (thread_single(p, SINGLE_BOUNDARY) != 0)
300 KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0,
302 *oldvmspace = p->p_vmspace;
307 post_execve(struct thread *td, int error, struct vmspace *oldvmspace)
311 KASSERT(td == curthread, ("non-current thread %p", td));
313 if ((p->p_flag & P_HADTHREADS) != 0) {
316 * If success, we upgrade to SINGLE_EXIT state to
317 * force other threads to suicide.
320 thread_single(p, SINGLE_EXIT);
322 thread_single_end(p, SINGLE_BOUNDARY);
325 if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
326 KASSERT(p->p_vmspace != oldvmspace,
327 ("oldvmspace still used"));
328 vmspace_free(oldvmspace);
329 td->td_pflags &= ~TDP_EXECVMSPC;
334 * XXX: kern_execve has the astonishing property of not always returning to
335 * the caller. If sufficiently bad things happen during the call to
336 * do_execve(), it can end up calling exit1(); as a result, callers must
337 * avoid doing anything which they might need to undo (e.g., allocating
341 kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p)
344 AUDIT_ARG_ARGV(args->begin_argv, args->argc,
345 args->begin_envv - args->begin_argv);
346 AUDIT_ARG_ENVV(args->begin_envv, args->envc,
347 args->endp - args->begin_envv);
348 return (do_execve(td, args, mac_p));
352 * In-kernel implementation of execve(). All arguments are assumed to be
353 * userspace pointers from the passed thread.
356 do_execve(struct thread *td, struct image_args *args, struct mac *mac_p)
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.
856 * Check if system-wide sampling is in effect or if the
857 * current process is using PMCs. If so, do exec() time
858 * processing. This processing needs to happen AFTER the
859 * P_INEXEC flag is cleared.
861 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
862 VOP_UNLOCK(imgp->vp, 0);
863 pe.pm_credentialschanged = credential_changing;
864 pe.pm_entryaddr = imgp->entry_addr;
866 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
867 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
871 /* Set values passed into the program in registers. */
872 if (p->p_sysent->sv_setregs)
873 (*p->p_sysent->sv_setregs)(td, imgp,
874 (u_long)(uintptr_t)stack_base);
876 exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base);
878 vfs_mark_atime(imgp->vp, td->td_ucred);
880 SDT_PROBE1(proc, , , exec__success, args->fname);
883 if (imgp->firstpage != NULL)
884 exec_unmap_first_page(imgp);
886 if (imgp->vp != NULL) {
888 NDFREE(&nd, NDF_ONLY_PNBUF);
890 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
894 VOP_UNLOCK(imgp->vp, 0);
897 if (imgp->object != NULL)
898 vm_object_deallocate(imgp->object);
900 free(imgp->freepath, M_TEMP);
904 if (p->p_ptevents & PTRACE_EXEC)
905 td->td_dbgflags |= TDB_EXEC;
909 * Stop the process here if its stop event mask has
910 * the S_EXEC bit set.
912 STOPEVENT(p, S_EXEC, 0);
915 /* we're done here, clear P_INEXEC */
917 p->p_flag &= ~P_INEXEC;
920 SDT_PROBE1(proc, , , exec__failure, error);
923 if (imgp->newcred != NULL && oldcred != NULL)
924 crfree(imgp->newcred);
927 mac_execve_exit(imgp);
928 mac_execve_interpreter_exit(interpvplabel);
930 exec_free_args(args);
933 * Handle deferred decrement of ref counts.
935 if (oldtextvp != NULL)
940 if (tracecred != NULL)
945 if (oldsigacts != NULL)
946 sigacts_free(oldsigacts);
950 if (error && imgp->vmspace_destroyed) {
951 /* sorry, no more process anymore. exit gracefully */
952 exit1(td, 0, SIGABRT);
965 exec_map_first_page(imgp)
966 struct image_params *imgp;
968 int rv, i, after, initial_pagein;
969 vm_page_t ma[VM_INITIAL_PAGEIN];
972 if (imgp->firstpage != NULL)
973 exec_unmap_first_page(imgp);
975 object = imgp->vp->v_object;
978 VM_OBJECT_WLOCK(object);
979 #if VM_NRESERVLEVEL > 0
980 vm_object_color(object, 0);
982 ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY);
983 if (ma[0]->valid != VM_PAGE_BITS_ALL) {
984 vm_page_xbusy(ma[0]);
985 if (!vm_pager_has_page(object, 0, NULL, &after)) {
988 vm_page_unlock(ma[0]);
989 VM_OBJECT_WUNLOCK(object);
992 initial_pagein = min(after, VM_INITIAL_PAGEIN);
993 KASSERT(initial_pagein <= object->size,
994 ("%s: initial_pagein %d object->size %ju",
995 __func__, initial_pagein, (uintmax_t )object->size));
996 for (i = 1; i < initial_pagein; i++) {
997 if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) {
1000 if (vm_page_tryxbusy(ma[i]))
1003 ma[i] = vm_page_alloc(object, i,
1010 rv = vm_pager_get_pages(object, ma, initial_pagein, NULL, NULL);
1011 if (rv != VM_PAGER_OK) {
1012 for (i = 0; i < initial_pagein; i++) {
1013 vm_page_lock(ma[i]);
1014 vm_page_free(ma[i]);
1015 vm_page_unlock(ma[i]);
1017 VM_OBJECT_WUNLOCK(object);
1020 vm_page_xunbusy(ma[0]);
1021 for (i = 1; i < initial_pagein; i++)
1022 vm_page_readahead_finish(ma[i]);
1024 vm_page_lock(ma[0]);
1025 vm_page_hold(ma[0]);
1026 vm_page_activate(ma[0]);
1027 vm_page_unlock(ma[0]);
1028 VM_OBJECT_WUNLOCK(object);
1030 imgp->firstpage = sf_buf_alloc(ma[0], 0);
1031 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
1037 exec_unmap_first_page(struct image_params *imgp)
1041 if (imgp->firstpage != NULL) {
1042 m = sf_buf_page(imgp->firstpage);
1043 sf_buf_free(imgp->firstpage);
1044 imgp->firstpage = NULL;
1052 * Destroy old address space, and allocate a new stack.
1053 * The new stack is only sgrowsiz large because it is grown
1054 * automatically on a page fault.
1057 exec_new_vmspace(struct image_params *imgp, struct sysentvec *sv)
1060 struct proc *p = imgp->proc;
1061 struct vmspace *vmspace = p->p_vmspace;
1063 struct rlimit rlim_stack;
1064 vm_offset_t sv_minuser, stack_addr;
1068 imgp->vmspace_destroyed = 1;
1071 /* May be called with Giant held */
1072 EVENTHANDLER_INVOKE(process_exec, p, imgp);
1075 * Blow away entire process VM, if address space not shared,
1076 * otherwise, create a new VM space so that other threads are
1079 map = &vmspace->vm_map;
1081 sv_minuser = sv->sv_minuser;
1083 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1084 if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser &&
1085 vm_map_max(map) == sv->sv_maxuser) {
1087 pmap_remove_pages(vmspace_pmap(vmspace));
1088 vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1089 /* An exec terminates mlockall(MCL_FUTURE). */
1091 vm_map_modflags(map, 0, MAP_WIREFUTURE);
1094 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1097 vmspace = p->p_vmspace;
1098 map = &vmspace->vm_map;
1101 /* Map a shared page */
1102 obj = sv->sv_shared_page_obj;
1104 vm_object_reference(obj);
1105 error = vm_map_fixed(map, obj, 0,
1106 sv->sv_shared_page_base, sv->sv_shared_page_len,
1107 VM_PROT_READ | VM_PROT_EXECUTE,
1108 VM_PROT_READ | VM_PROT_EXECUTE,
1109 MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1110 if (error != KERN_SUCCESS) {
1111 vm_object_deallocate(obj);
1112 return (vm_mmap_to_errno(error));
1116 /* Allocate a new stack */
1117 if (imgp->stack_sz != 0) {
1118 ssiz = trunc_page(imgp->stack_sz);
1120 lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack);
1122 if (ssiz > rlim_stack.rlim_max)
1123 ssiz = rlim_stack.rlim_max;
1124 if (ssiz > rlim_stack.rlim_cur) {
1125 rlim_stack.rlim_cur = ssiz;
1126 kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack);
1128 } else if (sv->sv_maxssiz != NULL) {
1129 ssiz = *sv->sv_maxssiz;
1133 stack_addr = sv->sv_usrstack - ssiz;
1134 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1135 obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1136 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1137 if (error != KERN_SUCCESS)
1138 return (vm_mmap_to_errno(error));
1141 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they
1142 * are still used to enforce the stack rlimit on the process stack.
1144 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1145 vmspace->vm_maxsaddr = (char *)stack_addr;
1151 * Copy out argument and environment strings from the old process address
1152 * space into the temporary string buffer.
1155 exec_copyin_args(struct image_args *args, char *fname,
1156 enum uio_seg segflg, char **argv, char **envv)
1162 bzero(args, sizeof(*args));
1167 * Allocate demand-paged memory for the file name, argument, and
1168 * environment strings.
1170 error = exec_alloc_args(args);
1175 * Copy the file name.
1177 if (fname != NULL) {
1178 args->fname = args->buf;
1179 error = (segflg == UIO_SYSSPACE) ?
1180 copystr(fname, args->fname, PATH_MAX, &length) :
1181 copyinstr(fname, args->fname, PATH_MAX, &length);
1187 args->begin_argv = args->buf + length;
1188 args->endp = args->begin_argv;
1189 args->stringspace = ARG_MAX;
1192 * extract arguments first
1195 error = fueword(argv++, &argp);
1202 error = copyinstr((void *)(uintptr_t)argp, args->endp,
1203 args->stringspace, &length);
1205 if (error == ENAMETOOLONG)
1209 args->stringspace -= length;
1210 args->endp += length;
1214 args->begin_envv = args->endp;
1217 * extract environment strings
1221 error = fueword(envv++, &envp);
1228 error = copyinstr((void *)(uintptr_t)envp,
1229 args->endp, args->stringspace, &length);
1231 if (error == ENAMETOOLONG)
1235 args->stringspace -= length;
1236 args->endp += length;
1244 exec_free_args(args);
1249 exec_copyin_data_fds(struct thread *td, struct image_args *args,
1250 const void *data, size_t datalen, const int *fds, size_t fdslen)
1252 struct filedesc *ofdp;
1257 memset(args, '\0', sizeof(*args));
1258 ofdp = td->td_proc->p_fd;
1259 if (datalen >= ARG_MAX || fdslen > ofdp->fd_lastfile + 1)
1261 error = exec_alloc_args(args);
1265 args->begin_argv = args->buf;
1266 args->stringspace = ARG_MAX;
1270 * Argument buffer has been provided. Copy it into the
1271 * kernel as a single string and add a terminating null
1274 error = copyin(data, args->begin_argv, datalen);
1277 args->begin_argv[datalen] = '\0';
1278 args->endp = args->begin_argv + datalen + 1;
1279 args->stringspace -= datalen + 1;
1282 * Traditional argument counting. Count the number of
1285 for (p = args->begin_argv; p < args->endp; ++p)
1289 /* No argument buffer provided. */
1290 args->endp = args->begin_argv;
1292 /* There are no environment variables. */
1293 args->begin_envv = args->endp;
1295 /* Create new file descriptor table. */
1296 kfds = malloc(fdslen * sizeof(int), M_TEMP, M_WAITOK);
1297 error = copyin(fds, kfds, fdslen * sizeof(int));
1302 error = fdcopy_remapped(ofdp, kfds, fdslen, &args->fdp);
1309 exec_free_args(args);
1313 struct exec_args_kva {
1316 SLIST_ENTRY(exec_args_kva) next;
1319 static DPCPU_DEFINE(struct exec_args_kva *, exec_args_kva);
1321 static SLIST_HEAD(, exec_args_kva) exec_args_kva_freelist;
1322 static struct mtx exec_args_kva_mtx;
1323 static u_int exec_args_gen;
1326 exec_prealloc_args_kva(void *arg __unused)
1328 struct exec_args_kva *argkva;
1331 SLIST_INIT(&exec_args_kva_freelist);
1332 mtx_init(&exec_args_kva_mtx, "exec args kva", NULL, MTX_DEF);
1333 for (i = 0; i < exec_map_entries; i++) {
1334 argkva = malloc(sizeof(*argkva), M_PARGS, M_WAITOK);
1335 argkva->addr = kmap_alloc_wait(exec_map, exec_map_entry_size);
1336 argkva->gen = exec_args_gen;
1337 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1340 SYSINIT(exec_args_kva, SI_SUB_EXEC, SI_ORDER_ANY, exec_prealloc_args_kva, NULL);
1343 exec_alloc_args_kva(void **cookie)
1345 struct exec_args_kva *argkva;
1347 argkva = (void *)atomic_readandclear_ptr(
1348 (uintptr_t *)DPCPU_PTR(exec_args_kva));
1349 if (argkva == NULL) {
1350 mtx_lock(&exec_args_kva_mtx);
1351 while ((argkva = SLIST_FIRST(&exec_args_kva_freelist)) == NULL)
1352 (void)mtx_sleep(&exec_args_kva_freelist,
1353 &exec_args_kva_mtx, 0, "execkva", 0);
1354 SLIST_REMOVE_HEAD(&exec_args_kva_freelist, next);
1355 mtx_unlock(&exec_args_kva_mtx);
1357 *(struct exec_args_kva **)cookie = argkva;
1358 return (argkva->addr);
1362 exec_release_args_kva(struct exec_args_kva *argkva, u_int gen)
1366 base = argkva->addr;
1367 if (argkva->gen != gen) {
1368 vm_map_madvise(exec_map, base, base + exec_map_entry_size,
1372 if (!atomic_cmpset_ptr((uintptr_t *)DPCPU_PTR(exec_args_kva),
1373 (uintptr_t)NULL, (uintptr_t)argkva)) {
1374 mtx_lock(&exec_args_kva_mtx);
1375 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1376 wakeup_one(&exec_args_kva_freelist);
1377 mtx_unlock(&exec_args_kva_mtx);
1382 exec_free_args_kva(void *cookie)
1385 exec_release_args_kva(cookie, exec_args_gen);
1389 exec_args_kva_lowmem(void *arg __unused)
1391 SLIST_HEAD(, exec_args_kva) head;
1392 struct exec_args_kva *argkva;
1396 gen = atomic_fetchadd_int(&exec_args_gen, 1) + 1;
1399 * Force an madvise of each KVA range. Any currently allocated ranges
1400 * will have MADV_FREE applied once they are freed.
1403 mtx_lock(&exec_args_kva_mtx);
1404 SLIST_SWAP(&head, &exec_args_kva_freelist, exec_args_kva);
1405 mtx_unlock(&exec_args_kva_mtx);
1406 while ((argkva = SLIST_FIRST(&head)) != NULL) {
1407 SLIST_REMOVE_HEAD(&head, next);
1408 exec_release_args_kva(argkva, gen);
1412 argkva = (void *)atomic_readandclear_ptr(
1413 (uintptr_t *)DPCPU_ID_PTR(i, exec_args_kva));
1415 exec_release_args_kva(argkva, gen);
1418 EVENTHANDLER_DEFINE(vm_lowmem, exec_args_kva_lowmem, NULL,
1419 EVENTHANDLER_PRI_ANY);
1422 * Allocate temporary demand-paged, zero-filled memory for the file name,
1423 * argument, and environment strings.
1426 exec_alloc_args(struct image_args *args)
1429 args->buf = (char *)exec_alloc_args_kva(&args->bufkva);
1434 exec_free_args(struct image_args *args)
1437 if (args->buf != NULL) {
1438 exec_free_args_kva(args->bufkva);
1441 if (args->fname_buf != NULL) {
1442 free(args->fname_buf, M_TEMP);
1443 args->fname_buf = NULL;
1445 if (args->fdp != NULL)
1446 fdescfree_remapped(args->fdp);
1450 * Copy strings out to the new process address space, constructing new arg
1451 * and env vector tables. Return a pointer to the base so that it can be used
1452 * as the initial stack pointer.
1455 exec_copyout_strings(struct image_params *imgp)
1461 register_t *stack_base;
1462 struct ps_strings *arginfo;
1464 size_t execpath_len;
1465 int szsigcode, szps;
1466 char canary[sizeof(long) * 8];
1468 szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1470 * Calculate string base and vector table pointers.
1471 * Also deal with signal trampoline code for this exec type.
1473 if (imgp->execpath != NULL && imgp->auxargs != NULL)
1474 execpath_len = strlen(imgp->execpath) + 1;
1479 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1480 if (p->p_sysent->sv_sigcode_base == 0) {
1481 if (p->p_sysent->sv_szsigcode != NULL)
1482 szsigcode = *(p->p_sysent->sv_szsigcode);
1484 destp = (uintptr_t)arginfo;
1489 if (szsigcode != 0) {
1491 destp = rounddown2(destp, sizeof(void *));
1492 copyout(p->p_sysent->sv_sigcode, (void *)destp, szsigcode);
1496 * Copy the image path for the rtld.
1498 if (execpath_len != 0) {
1499 destp -= execpath_len;
1500 imgp->execpathp = destp;
1501 copyout(imgp->execpath, (void *)destp, execpath_len);
1505 * Prepare the canary for SSP.
1507 arc4rand(canary, sizeof(canary), 0);
1508 destp -= sizeof(canary);
1509 imgp->canary = destp;
1510 copyout(canary, (void *)destp, sizeof(canary));
1511 imgp->canarylen = sizeof(canary);
1514 * Prepare the pagesizes array.
1517 destp = rounddown2(destp, sizeof(void *));
1518 imgp->pagesizes = destp;
1519 copyout(pagesizes, (void *)destp, szps);
1520 imgp->pagesizeslen = szps;
1522 destp -= ARG_MAX - imgp->args->stringspace;
1523 destp = rounddown2(destp, sizeof(void *));
1526 * If we have a valid auxargs ptr, prepare some room
1529 if (imgp->auxargs) {
1531 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1532 * lower compatibility.
1534 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1537 * The '+ 2' is for the null pointers at the end of each of
1538 * the arg and env vector sets,and imgp->auxarg_size is room
1539 * for argument of Runtime loader.
1541 vectp = (char **)(destp - (imgp->args->argc +
1542 imgp->args->envc + 2 + imgp->auxarg_size)
1546 * The '+ 2' is for the null pointers at the end of each of
1547 * the arg and env vector sets
1549 vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc
1550 + 2) * sizeof(char *));
1554 * vectp also becomes our initial stack base
1556 stack_base = (register_t *)vectp;
1558 stringp = imgp->args->begin_argv;
1559 argc = imgp->args->argc;
1560 envc = imgp->args->envc;
1563 * Copy out strings - arguments and environment.
1565 copyout(stringp, (void *)destp, ARG_MAX - imgp->args->stringspace);
1568 * Fill in "ps_strings" struct for ps, w, etc.
1570 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1571 suword32(&arginfo->ps_nargvstr, argc);
1574 * Fill in argument portion of vector table.
1576 for (; argc > 0; --argc) {
1577 suword(vectp++, (long)(intptr_t)destp);
1578 while (*stringp++ != 0)
1583 /* a null vector table pointer separates the argp's from the envp's */
1586 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1587 suword32(&arginfo->ps_nenvstr, envc);
1590 * Fill in environment portion of vector table.
1592 for (; envc > 0; --envc) {
1593 suword(vectp++, (long)(intptr_t)destp);
1594 while (*stringp++ != 0)
1599 /* end of vector table is a null pointer */
1602 return (stack_base);
1606 * Check permissions of file to execute.
1607 * Called with imgp->vp locked.
1608 * Return 0 for success or error code on failure.
1611 exec_check_permissions(struct image_params *imgp)
1613 struct vnode *vp = imgp->vp;
1614 struct vattr *attr = imgp->attr;
1616 int error, writecount;
1620 /* Get file attributes */
1621 error = VOP_GETATTR(vp, attr, td->td_ucred);
1626 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1632 * 1) Check if file execution is disabled for the filesystem that
1633 * this file resides on.
1634 * 2) Ensure that at least one execute bit is on. Otherwise, a
1635 * privileged user will always succeed, and we don't want this
1636 * to happen unless the file really is executable.
1637 * 3) Ensure that the file is a regular file.
1639 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1640 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1641 (attr->va_type != VREG))
1645 * Zero length files can't be exec'd
1647 if (attr->va_size == 0)
1651 * Check for execute permission to file based on current credentials.
1653 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1658 * Check number of open-for-writes on the file and deny execution
1661 error = VOP_GET_WRITECOUNT(vp, &writecount);
1664 if (writecount != 0)
1668 * Call filesystem specific open routine (which does nothing in the
1671 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1678 * Exec handler registration
1681 exec_register(const struct execsw *execsw_arg)
1683 const struct execsw **es, **xs, **newexecsw;
1684 int count = 2; /* New slot and trailing NULL */
1687 for (es = execsw; *es; es++)
1689 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1692 for (es = execsw; *es; es++)
1697 free(execsw, M_TEMP);
1703 exec_unregister(const struct execsw *execsw_arg)
1705 const struct execsw **es, **xs, **newexecsw;
1709 panic("unregister with no handlers left?\n");
1711 for (es = execsw; *es; es++) {
1712 if (*es == execsw_arg)
1717 for (es = execsw; *es; es++)
1718 if (*es != execsw_arg)
1720 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1722 for (es = execsw; *es; es++)
1723 if (*es != execsw_arg)
1727 free(execsw, M_TEMP);