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(td, args, mac_p)
358 struct image_args *args;
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;
378 int credential_changing;
381 struct label *interpvplabel = NULL;
385 struct pmckern_procexec pe;
387 static const char fexecv_proc_title[] = "(fexecv)";
389 imgp = &image_params;
392 * Lock the process and set the P_INEXEC flag to indicate that
393 * it should be left alone until we're done here. This is
394 * necessary to avoid race conditions - e.g. in ptrace() -
395 * that might allow a local user to illicitly obtain elevated
399 KASSERT((p->p_flag & P_INEXEC) == 0,
400 ("%s(): process already has P_INEXEC flag", __func__));
401 p->p_flag |= P_INEXEC;
405 * Initialize part of the common data
407 bzero(imgp, sizeof(*imgp));
411 oldcred = p->p_ucred;
414 error = mac_execve_enter(imgp, mac_p);
420 * Translate the file name. namei() returns a vnode pointer
421 * in ni_vp among other things.
423 * XXXAUDIT: It would be desirable to also audit the name of the
424 * interpreter if this is an interpreted binary.
426 if (args->fname != NULL) {
427 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME
428 | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
431 SDT_PROBE1(proc, , , exec, args->fname);
434 if (args->fname != NULL) {
435 #ifdef CAPABILITY_MODE
437 * While capability mode can't reach this point via direct
438 * path arguments to execve(), we also don't allow
439 * interpreters to be used in capability mode (for now).
440 * Catch indirect lookups and return a permissions error.
442 if (IN_CAPABILITY_MODE(td)) {
451 newtextvp = nd.ni_vp;
452 imgp->vp = newtextvp;
454 AUDIT_ARG_FD(args->fd);
456 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
458 error = fgetvp_exec(td, args->fd,
459 cap_rights_init(&rights, CAP_FEXECVE), &newtextvp);
462 vn_lock(newtextvp, LK_EXCLUSIVE | LK_RETRY);
463 AUDIT_ARG_VNODE1(newtextvp);
464 imgp->vp = newtextvp;
468 * Check file permissions (also 'opens' file)
470 error = exec_check_permissions(imgp);
472 goto exec_fail_dealloc;
474 imgp->object = imgp->vp->v_object;
475 if (imgp->object != NULL)
476 vm_object_reference(imgp->object);
479 * Set VV_TEXT now so no one can write to the executable while we're
482 * Remember if this was set before and unset it in case this is not
483 * actually an executable image.
485 textset = VOP_IS_TEXT(imgp->vp);
486 VOP_SET_TEXT(imgp->vp);
488 error = exec_map_first_page(imgp);
490 goto exec_fail_dealloc;
492 imgp->proc->p_osrel = 0;
495 * Implement image setuid/setgid.
497 * Determine new credentials before attempting image activators
498 * so that it can be used by process_exec handlers to determine
499 * credential/setid changes.
501 * Don't honor setuid/setgid if the filesystem prohibits it or if
502 * the process is being traced.
504 * We disable setuid/setgid/etc in capability mode on the basis
505 * that most setugid applications are not written with that
506 * environment in mind, and will therefore almost certainly operate
507 * incorrectly. In principle there's no reason that setugid
508 * applications might not be useful in capability mode, so we may want
509 * to reconsider this conservative design choice in the future.
511 * XXXMAC: For the time being, use NOSUID to also prohibit
512 * transitions on the file system.
514 credential_changing = 0;
515 credential_changing |= (attr.va_mode & S_ISUID) &&
516 oldcred->cr_uid != attr.va_uid;
517 credential_changing |= (attr.va_mode & S_ISGID) &&
518 oldcred->cr_gid != attr.va_gid;
520 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
521 interpvplabel, imgp);
522 credential_changing |= will_transition;
525 if (credential_changing &&
526 #ifdef CAPABILITY_MODE
527 ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
529 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
530 (p->p_flag & P_TRACED) == 0) {
531 imgp->credential_setid = true;
532 VOP_UNLOCK(imgp->vp, 0);
533 imgp->newcred = crdup(oldcred);
534 if (attr.va_mode & S_ISUID) {
535 euip = uifind(attr.va_uid);
536 change_euid(imgp->newcred, euip);
538 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
539 if (attr.va_mode & S_ISGID)
540 change_egid(imgp->newcred, attr.va_gid);
542 * Implement correct POSIX saved-id behavior.
544 * XXXMAC: Note that the current logic will save the
545 * uid and gid if a MAC domain transition occurs, even
546 * though maybe it shouldn't.
548 change_svuid(imgp->newcred, imgp->newcred->cr_uid);
549 change_svgid(imgp->newcred, imgp->newcred->cr_gid);
552 * Implement correct POSIX saved-id behavior.
554 * XXX: It's not clear that the existing behavior is
555 * POSIX-compliant. A number of sources indicate that the
556 * saved uid/gid should only be updated if the new ruid is
557 * not equal to the old ruid, or the new euid is not equal
558 * to the old euid and the new euid is not equal to the old
559 * ruid. The FreeBSD code always updates the saved uid/gid.
560 * Also, this code uses the new (replaced) euid and egid as
561 * the source, which may or may not be the right ones to use.
563 if (oldcred->cr_svuid != oldcred->cr_uid ||
564 oldcred->cr_svgid != oldcred->cr_gid) {
565 VOP_UNLOCK(imgp->vp, 0);
566 imgp->newcred = crdup(oldcred);
567 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
568 change_svuid(imgp->newcred, imgp->newcred->cr_uid);
569 change_svgid(imgp->newcred, imgp->newcred->cr_gid);
572 /* The new credentials are installed into the process later. */
575 * Do the best to calculate the full path to the image file.
577 if (args->fname != NULL && args->fname[0] == '/')
578 imgp->execpath = args->fname;
580 VOP_UNLOCK(imgp->vp, 0);
581 if (vn_fullpath(td, imgp->vp, &imgp->execpath,
582 &imgp->freepath) != 0)
583 imgp->execpath = args->fname;
584 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
588 * If the current process has a special image activator it
589 * wants to try first, call it. For example, emulating shell
590 * scripts differently.
593 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
594 error = img_first(imgp);
597 * Loop through the list of image activators, calling each one.
598 * An activator returns -1 if there is no match, 0 on success,
599 * and an error otherwise.
601 for (i = 0; error == -1 && execsw[i]; ++i) {
602 if (execsw[i]->ex_imgact == NULL ||
603 execsw[i]->ex_imgact == img_first) {
606 error = (*execsw[i]->ex_imgact)(imgp);
612 VOP_UNSET_TEXT(imgp->vp);
615 goto exec_fail_dealloc;
619 * Special interpreter operation, cleanup and loop up to try to
620 * activate the interpreter.
622 if (imgp->interpreted) {
623 exec_unmap_first_page(imgp);
625 * VV_TEXT needs to be unset for scripts. There is a short
626 * period before we determine that something is a script where
627 * VV_TEXT will be set. The vnode lock is held over this
628 * entire period so nothing should illegitimately be blocked.
630 VOP_UNSET_TEXT(imgp->vp);
631 /* free name buffer and old vnode */
632 if (args->fname != NULL)
633 NDFREE(&nd, NDF_ONLY_PNBUF);
635 mac_execve_interpreter_enter(newtextvp, &interpvplabel);
638 VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td);
642 vm_object_deallocate(imgp->object);
644 imgp->credential_setid = false;
645 if (imgp->newcred != NULL) {
646 crfree(imgp->newcred);
647 imgp->newcred = NULL;
649 imgp->execpath = NULL;
650 free(imgp->freepath, M_TEMP);
651 imgp->freepath = NULL;
652 /* set new name to that of the interpreter */
653 NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
654 UIO_SYSSPACE, imgp->interpreter_name, td);
655 args->fname = imgp->interpreter_name;
660 * NB: We unlock the vnode here because it is believed that none
661 * of the sv_copyout_strings/sv_fixup operations require the vnode.
663 VOP_UNLOCK(imgp->vp, 0);
665 if (disallow_high_osrel &&
666 P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
668 uprintf("Osrel %d for image %s too high\n", p->p_osrel,
669 imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
670 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
671 goto exec_fail_dealloc;
674 /* ABI enforces the use of Capsicum. Switch into capabilities mode. */
675 if (SV_PROC_FLAG(p, SV_CAPSICUM))
676 sys_cap_enter(td, NULL);
679 * Copy out strings (args and env) and initialize stack base
681 if (p->p_sysent->sv_copyout_strings)
682 stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
684 stack_base = exec_copyout_strings(imgp);
687 * If custom stack fixup routine present for this process
688 * let it do the stack setup.
689 * Else stuff argument count as first item on stack
691 if (p->p_sysent->sv_fixup != NULL)
692 (*p->p_sysent->sv_fixup)(&stack_base, imgp);
694 suword(--stack_base, imgp->args->argc);
696 if (args->fdp != NULL) {
697 /* Install a brand new file descriptor table. */
698 fdinstall_remapped(td, args->fdp);
702 * Keep on using the existing file descriptor table. For
703 * security and other reasons, the file descriptor table
704 * cannot be shared after an exec.
707 /* close files on exec */
712 * Malloc things before we need locks.
714 i = imgp->args->begin_envv - imgp->args->begin_argv;
715 /* Cache arguments if they fit inside our allowance */
716 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
717 newargs = pargs_alloc(i);
718 bcopy(imgp->args->begin_argv, newargs->ar_args, i);
722 * For security and other reasons, signal handlers cannot
723 * be shared after an exec. The new process gets a copy of the old
724 * handlers. In execsigs(), the new process will have its signals
727 if (sigacts_shared(p->p_sigacts)) {
728 oldsigacts = p->p_sigacts;
729 newsigacts = sigacts_alloc();
730 sigacts_copy(newsigacts, oldsigacts);
733 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
737 p->p_sigacts = newsigacts;
741 /* reset caught signals */
744 /* name this process - nameiexec(p, ndp) */
745 bzero(p->p_comm, sizeof(p->p_comm));
747 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
748 min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
749 else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0)
750 bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
751 bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
753 sched_clear_tdname(td);
757 * mark as execed, wakeup the process that vforked (if any) and tell
758 * it that it now has its own resources back
761 if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0)
762 p->p_flag2 &= ~P2_NOTRACE;
763 if (p->p_flag & P_PPWAIT) {
764 p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
765 cv_broadcast(&p->p_pwait);
766 /* STOPs are no longer ignored, arrange for AST */
771 * Implement image setuid/setgid installation.
773 if (imgp->credential_setid) {
775 * Turn off syscall tracing for set-id programs, except for
776 * root. Record any set-id flags first to make sure that
777 * we do not regain any tracing during a possible block.
782 if (p->p_tracecred != NULL &&
783 priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED, 0))
784 ktrprocexec(p, &tracecred, &tracevp);
787 * Close any file descriptors 0..2 that reference procfs,
788 * then make sure file descriptors 0..2 are in use.
790 * Both fdsetugidsafety() and fdcheckstd() may call functions
791 * taking sleepable locks, so temporarily drop our locks.
794 VOP_UNLOCK(imgp->vp, 0);
796 error = fdcheckstd(td);
797 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
799 goto exec_fail_dealloc;
802 if (will_transition) {
803 mac_vnode_execve_transition(oldcred, imgp->newcred,
804 imgp->vp, interpvplabel, imgp);
808 if (oldcred->cr_uid == oldcred->cr_ruid &&
809 oldcred->cr_gid == oldcred->cr_rgid)
810 p->p_flag &= ~P_SUGID;
813 * Set the new credentials.
815 if (imgp->newcred != NULL) {
816 proc_set_cred(p, imgp->newcred);
822 * Store the vp for use in procfs. This vnode was referenced by namei
825 oldtextvp = p->p_textvp;
826 p->p_textvp = newtextvp;
830 * Tell the DTrace fasttrap provider about the exec if it
831 * has declared an interest.
833 if (dtrace_fasttrap_exec)
834 dtrace_fasttrap_exec(p);
838 * Notify others that we exec'd, and clear the P_INEXEC flag
839 * as we're now a bona fide freshly-execed process.
841 KNOTE_LOCKED(p->p_klist, NOTE_EXEC);
842 p->p_flag &= ~P_INEXEC;
844 /* clear "fork but no exec" flag, as we _are_ execing */
845 p->p_acflag &= ~AFORK;
848 * Free any previous argument cache and replace it with
849 * the new argument cache, if any.
857 * Check if system-wide sampling is in effect or if the
858 * current process is using PMCs. If so, do exec() time
859 * processing. This processing needs to happen AFTER the
860 * P_INEXEC flag is cleared.
862 * The proc lock needs to be released before taking the PMC
865 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
867 VOP_UNLOCK(imgp->vp, 0);
868 pe.pm_credentialschanged = credential_changing;
869 pe.pm_entryaddr = imgp->entry_addr;
871 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
872 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
875 #else /* !HWPMC_HOOKS */
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);
912 if (p->p_ptevents & PTRACE_EXEC)
913 td->td_dbgflags |= TDB_EXEC;
917 * Stop the process here if its stop event mask has
918 * the S_EXEC bit set.
920 STOPEVENT(p, S_EXEC, 0);
923 /* we're done here, clear P_INEXEC */
925 p->p_flag &= ~P_INEXEC;
928 SDT_PROBE1(proc, , , exec__failure, error);
931 if (imgp->newcred != NULL && oldcred != NULL)
932 crfree(imgp->newcred);
935 mac_execve_exit(imgp);
936 mac_execve_interpreter_exit(interpvplabel);
938 exec_free_args(args);
941 * Handle deferred decrement of ref counts.
943 if (oldtextvp != NULL)
948 if (tracecred != NULL)
953 if (oldsigacts != NULL)
954 sigacts_free(oldsigacts);
958 if (error && imgp->vmspace_destroyed) {
959 /* sorry, no more process anymore. exit gracefully */
960 exit1(td, 0, SIGABRT);
973 exec_map_first_page(imgp)
974 struct image_params *imgp;
976 int rv, i, after, initial_pagein;
977 vm_page_t ma[VM_INITIAL_PAGEIN];
980 if (imgp->firstpage != NULL)
981 exec_unmap_first_page(imgp);
983 object = imgp->vp->v_object;
986 VM_OBJECT_WLOCK(object);
987 #if VM_NRESERVLEVEL > 0
988 vm_object_color(object, 0);
990 ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY);
991 if (ma[0]->valid != VM_PAGE_BITS_ALL) {
992 vm_page_xbusy(ma[0]);
993 if (!vm_pager_has_page(object, 0, NULL, &after)) {
996 vm_page_unlock(ma[0]);
997 VM_OBJECT_WUNLOCK(object);
1000 initial_pagein = min(after, VM_INITIAL_PAGEIN);
1001 KASSERT(initial_pagein <= object->size,
1002 ("%s: initial_pagein %d object->size %ju",
1003 __func__, initial_pagein, (uintmax_t )object->size));
1004 for (i = 1; i < initial_pagein; i++) {
1005 if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) {
1008 if (vm_page_tryxbusy(ma[i]))
1011 ma[i] = vm_page_alloc(object, i,
1018 rv = vm_pager_get_pages(object, ma, initial_pagein, NULL, NULL);
1019 if (rv != VM_PAGER_OK) {
1020 for (i = 0; i < initial_pagein; i++) {
1021 vm_page_lock(ma[i]);
1022 vm_page_free(ma[i]);
1023 vm_page_unlock(ma[i]);
1025 VM_OBJECT_WUNLOCK(object);
1028 vm_page_xunbusy(ma[0]);
1029 for (i = 1; i < initial_pagein; i++)
1030 vm_page_readahead_finish(ma[i]);
1032 vm_page_lock(ma[0]);
1033 vm_page_hold(ma[0]);
1034 vm_page_activate(ma[0]);
1035 vm_page_unlock(ma[0]);
1036 VM_OBJECT_WUNLOCK(object);
1038 imgp->firstpage = sf_buf_alloc(ma[0], 0);
1039 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
1045 exec_unmap_first_page(imgp)
1046 struct image_params *imgp;
1050 if (imgp->firstpage != NULL) {
1051 m = sf_buf_page(imgp->firstpage);
1052 sf_buf_free(imgp->firstpage);
1053 imgp->firstpage = NULL;
1061 * Destroy old address space, and allocate a new stack
1062 * The new stack is only SGROWSIZ large because it is grown
1063 * automatically in trap.c.
1066 exec_new_vmspace(imgp, sv)
1067 struct image_params *imgp;
1068 struct sysentvec *sv;
1071 struct proc *p = imgp->proc;
1072 struct vmspace *vmspace = p->p_vmspace;
1074 struct rlimit rlim_stack;
1075 vm_offset_t sv_minuser, stack_addr;
1079 imgp->vmspace_destroyed = 1;
1082 /* May be called with Giant held */
1083 EVENTHANDLER_INVOKE(process_exec, p, imgp);
1086 * Blow away entire process VM, if address space not shared,
1087 * otherwise, create a new VM space so that other threads are
1090 map = &vmspace->vm_map;
1092 sv_minuser = sv->sv_minuser;
1094 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1095 if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser &&
1096 vm_map_max(map) == sv->sv_maxuser) {
1098 pmap_remove_pages(vmspace_pmap(vmspace));
1099 vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1101 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1104 vmspace = p->p_vmspace;
1105 map = &vmspace->vm_map;
1108 /* Map a shared page */
1109 obj = sv->sv_shared_page_obj;
1111 vm_object_reference(obj);
1112 error = vm_map_fixed(map, obj, 0,
1113 sv->sv_shared_page_base, sv->sv_shared_page_len,
1114 VM_PROT_READ | VM_PROT_EXECUTE,
1115 VM_PROT_READ | VM_PROT_EXECUTE,
1116 MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1118 vm_object_deallocate(obj);
1123 /* Allocate a new stack */
1124 if (imgp->stack_sz != 0) {
1125 ssiz = trunc_page(imgp->stack_sz);
1127 lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack);
1129 if (ssiz > rlim_stack.rlim_max)
1130 ssiz = rlim_stack.rlim_max;
1131 if (ssiz > rlim_stack.rlim_cur) {
1132 rlim_stack.rlim_cur = ssiz;
1133 kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack);
1135 } else if (sv->sv_maxssiz != NULL) {
1136 ssiz = *sv->sv_maxssiz;
1140 stack_addr = sv->sv_usrstack - ssiz;
1141 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1142 obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1144 VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1149 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they
1150 * are still used to enforce the stack rlimit on the process stack.
1152 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1153 vmspace->vm_maxsaddr = (char *)stack_addr;
1159 * Copy out argument and environment strings from the old process address
1160 * space into the temporary string buffer.
1163 exec_copyin_args(struct image_args *args, char *fname,
1164 enum uio_seg segflg, char **argv, char **envv)
1170 bzero(args, sizeof(*args));
1175 * Allocate demand-paged memory for the file name, argument, and
1176 * environment strings.
1178 error = exec_alloc_args(args);
1183 * Copy the file name.
1185 if (fname != NULL) {
1186 args->fname = args->buf;
1187 error = (segflg == UIO_SYSSPACE) ?
1188 copystr(fname, args->fname, PATH_MAX, &length) :
1189 copyinstr(fname, args->fname, PATH_MAX, &length);
1195 args->begin_argv = args->buf + length;
1196 args->endp = args->begin_argv;
1197 args->stringspace = ARG_MAX;
1200 * extract arguments first
1203 error = fueword(argv++, &argp);
1210 error = copyinstr((void *)(uintptr_t)argp, args->endp,
1211 args->stringspace, &length);
1213 if (error == ENAMETOOLONG)
1217 args->stringspace -= length;
1218 args->endp += length;
1222 args->begin_envv = args->endp;
1225 * extract environment strings
1229 error = fueword(envv++, &envp);
1236 error = copyinstr((void *)(uintptr_t)envp,
1237 args->endp, args->stringspace, &length);
1239 if (error == ENAMETOOLONG)
1243 args->stringspace -= length;
1244 args->endp += length;
1252 exec_free_args(args);
1257 exec_copyin_data_fds(struct thread *td, struct image_args *args,
1258 const void *data, size_t datalen, const int *fds, size_t fdslen)
1260 struct filedesc *ofdp;
1265 memset(args, '\0', sizeof(*args));
1266 ofdp = td->td_proc->p_fd;
1267 if (datalen >= ARG_MAX || fdslen > ofdp->fd_lastfile + 1)
1269 error = exec_alloc_args(args);
1273 args->begin_argv = args->buf;
1274 args->stringspace = ARG_MAX;
1278 * Argument buffer has been provided. Copy it into the
1279 * kernel as a single string and add a terminating null
1282 error = copyin(data, args->begin_argv, datalen);
1285 args->begin_argv[datalen] = '\0';
1286 args->endp = args->begin_argv + datalen + 1;
1287 args->stringspace -= datalen + 1;
1290 * Traditional argument counting. Count the number of
1293 for (p = args->begin_argv; p < args->endp; ++p)
1297 /* No argument buffer provided. */
1298 args->endp = args->begin_argv;
1300 /* There are no environment variables. */
1301 args->begin_envv = args->endp;
1303 /* Create new file descriptor table. */
1304 kfds = malloc(fdslen * sizeof(int), M_TEMP, M_WAITOK);
1305 error = copyin(fds, kfds, fdslen * sizeof(int));
1310 error = fdcopy_remapped(ofdp, kfds, fdslen, &args->fdp);
1317 exec_free_args(args);
1321 struct exec_args_kva {
1324 SLIST_ENTRY(exec_args_kva) next;
1327 static DPCPU_DEFINE(struct exec_args_kva *, exec_args_kva);
1329 static SLIST_HEAD(, exec_args_kva) exec_args_kva_freelist;
1330 static struct mtx exec_args_kva_mtx;
1331 static u_int exec_args_gen;
1334 exec_prealloc_args_kva(void *arg __unused)
1336 struct exec_args_kva *argkva;
1339 SLIST_INIT(&exec_args_kva_freelist);
1340 mtx_init(&exec_args_kva_mtx, "exec args kva", NULL, MTX_DEF);
1341 for (i = 0; i < exec_map_entries; i++) {
1342 argkva = malloc(sizeof(*argkva), M_PARGS, M_WAITOK);
1343 argkva->addr = kmap_alloc_wait(exec_map, exec_map_entry_size);
1344 argkva->gen = exec_args_gen;
1345 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1348 SYSINIT(exec_args_kva, SI_SUB_EXEC, SI_ORDER_ANY, exec_prealloc_args_kva, NULL);
1351 exec_alloc_args_kva(void **cookie)
1353 struct exec_args_kva *argkva;
1355 argkva = (void *)atomic_readandclear_ptr(
1356 (uintptr_t *)DPCPU_PTR(exec_args_kva));
1357 if (argkva == NULL) {
1358 mtx_lock(&exec_args_kva_mtx);
1359 while ((argkva = SLIST_FIRST(&exec_args_kva_freelist)) == NULL)
1360 (void)mtx_sleep(&exec_args_kva_freelist,
1361 &exec_args_kva_mtx, 0, "execkva", 0);
1362 SLIST_REMOVE_HEAD(&exec_args_kva_freelist, next);
1363 mtx_unlock(&exec_args_kva_mtx);
1365 *(struct exec_args_kva **)cookie = argkva;
1366 return (argkva->addr);
1370 exec_release_args_kva(struct exec_args_kva *argkva, u_int gen)
1374 base = argkva->addr;
1375 if (argkva->gen != gen) {
1376 vm_map_madvise(exec_map, base, base + exec_map_entry_size,
1380 if (!atomic_cmpset_ptr((uintptr_t *)DPCPU_PTR(exec_args_kva),
1381 (uintptr_t)NULL, (uintptr_t)argkva)) {
1382 mtx_lock(&exec_args_kva_mtx);
1383 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1384 wakeup_one(&exec_args_kva_freelist);
1385 mtx_unlock(&exec_args_kva_mtx);
1390 exec_free_args_kva(void *cookie)
1393 exec_release_args_kva(cookie, exec_args_gen);
1397 exec_args_kva_lowmem(void *arg __unused)
1399 SLIST_HEAD(, exec_args_kva) head;
1400 struct exec_args_kva *argkva;
1404 gen = atomic_fetchadd_int(&exec_args_gen, 1) + 1;
1407 * Force an madvise of each KVA range. Any currently allocated ranges
1408 * will have MADV_FREE applied once they are freed.
1411 mtx_lock(&exec_args_kva_mtx);
1412 SLIST_SWAP(&head, &exec_args_kva_freelist, exec_args_kva);
1413 mtx_unlock(&exec_args_kva_mtx);
1414 while ((argkva = SLIST_FIRST(&head)) != NULL) {
1415 SLIST_REMOVE_HEAD(&head, next);
1416 exec_release_args_kva(argkva, gen);
1420 argkva = (void *)atomic_readandclear_ptr(
1421 (uintptr_t *)DPCPU_ID_PTR(i, exec_args_kva));
1423 exec_release_args_kva(argkva, gen);
1426 EVENTHANDLER_DEFINE(vm_lowmem, exec_args_kva_lowmem, NULL,
1427 EVENTHANDLER_PRI_ANY);
1430 * Allocate temporary demand-paged, zero-filled memory for the file name,
1431 * argument, and environment strings.
1434 exec_alloc_args(struct image_args *args)
1437 args->buf = (char *)exec_alloc_args_kva(&args->bufkva);
1442 exec_free_args(struct image_args *args)
1445 if (args->buf != NULL) {
1446 exec_free_args_kva(args->bufkva);
1449 if (args->fname_buf != NULL) {
1450 free(args->fname_buf, M_TEMP);
1451 args->fname_buf = NULL;
1453 if (args->fdp != NULL)
1454 fdescfree_remapped(args->fdp);
1458 * Copy strings out to the new process address space, constructing new arg
1459 * and env vector tables. Return a pointer to the base so that it can be used
1460 * as the initial stack pointer.
1463 exec_copyout_strings(imgp)
1464 struct image_params *imgp;
1470 register_t *stack_base;
1471 struct ps_strings *arginfo;
1473 size_t execpath_len;
1474 int szsigcode, szps;
1475 char canary[sizeof(long) * 8];
1477 szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1479 * Calculate string base and vector table pointers.
1480 * Also deal with signal trampoline code for this exec type.
1482 if (imgp->execpath != NULL && imgp->auxargs != NULL)
1483 execpath_len = strlen(imgp->execpath) + 1;
1488 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1489 if (p->p_sysent->sv_sigcode_base == 0) {
1490 if (p->p_sysent->sv_szsigcode != NULL)
1491 szsigcode = *(p->p_sysent->sv_szsigcode);
1493 destp = (uintptr_t)arginfo;
1498 if (szsigcode != 0) {
1500 destp = rounddown2(destp, sizeof(void *));
1501 copyout(p->p_sysent->sv_sigcode, (void *)destp, szsigcode);
1505 * Copy the image path for the rtld.
1507 if (execpath_len != 0) {
1508 destp -= execpath_len;
1509 imgp->execpathp = destp;
1510 copyout(imgp->execpath, (void *)destp, execpath_len);
1514 * Prepare the canary for SSP.
1516 arc4rand(canary, sizeof(canary), 0);
1517 destp -= sizeof(canary);
1518 imgp->canary = destp;
1519 copyout(canary, (void *)destp, sizeof(canary));
1520 imgp->canarylen = sizeof(canary);
1523 * Prepare the pagesizes array.
1526 destp = rounddown2(destp, sizeof(void *));
1527 imgp->pagesizes = destp;
1528 copyout(pagesizes, (void *)destp, szps);
1529 imgp->pagesizeslen = szps;
1531 destp -= ARG_MAX - imgp->args->stringspace;
1532 destp = rounddown2(destp, sizeof(void *));
1535 * If we have a valid auxargs ptr, prepare some room
1538 if (imgp->auxargs) {
1540 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1541 * lower compatibility.
1543 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1546 * The '+ 2' is for the null pointers at the end of each of
1547 * the arg and env vector sets,and imgp->auxarg_size is room
1548 * for argument of Runtime loader.
1550 vectp = (char **)(destp - (imgp->args->argc +
1551 imgp->args->envc + 2 + imgp->auxarg_size)
1555 * The '+ 2' is for the null pointers at the end of each of
1556 * the arg and env vector sets
1558 vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc
1559 + 2) * sizeof(char *));
1563 * vectp also becomes our initial stack base
1565 stack_base = (register_t *)vectp;
1567 stringp = imgp->args->begin_argv;
1568 argc = imgp->args->argc;
1569 envc = imgp->args->envc;
1572 * Copy out strings - arguments and environment.
1574 copyout(stringp, (void *)destp, ARG_MAX - imgp->args->stringspace);
1577 * Fill in "ps_strings" struct for ps, w, etc.
1579 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1580 suword32(&arginfo->ps_nargvstr, argc);
1583 * Fill in argument portion of vector table.
1585 for (; argc > 0; --argc) {
1586 suword(vectp++, (long)(intptr_t)destp);
1587 while (*stringp++ != 0)
1592 /* a null vector table pointer separates the argp's from the envp's */
1595 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1596 suword32(&arginfo->ps_nenvstr, envc);
1599 * Fill in environment portion of vector table.
1601 for (; envc > 0; --envc) {
1602 suword(vectp++, (long)(intptr_t)destp);
1603 while (*stringp++ != 0)
1608 /* end of vector table is a null pointer */
1611 return (stack_base);
1615 * Check permissions of file to execute.
1616 * Called with imgp->vp locked.
1617 * Return 0 for success or error code on failure.
1620 exec_check_permissions(imgp)
1621 struct image_params *imgp;
1623 struct vnode *vp = imgp->vp;
1624 struct vattr *attr = imgp->attr;
1626 int error, writecount;
1630 /* Get file attributes */
1631 error = VOP_GETATTR(vp, attr, td->td_ucred);
1636 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1642 * 1) Check if file execution is disabled for the filesystem that
1643 * this file resides on.
1644 * 2) Ensure that at least one execute bit is on. Otherwise, a
1645 * privileged user will always succeed, and we don't want this
1646 * to happen unless the file really is executable.
1647 * 3) Ensure that the file is a regular file.
1649 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1650 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1651 (attr->va_type != VREG))
1655 * Zero length files can't be exec'd
1657 if (attr->va_size == 0)
1661 * Check for execute permission to file based on current credentials.
1663 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1668 * Check number of open-for-writes on the file and deny execution
1671 error = VOP_GET_WRITECOUNT(vp, &writecount);
1674 if (writecount != 0)
1678 * Call filesystem specific open routine (which does nothing in the
1681 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1688 * Exec handler registration
1691 exec_register(execsw_arg)
1692 const struct execsw *execsw_arg;
1694 const struct execsw **es, **xs, **newexecsw;
1695 int count = 2; /* New slot and trailing NULL */
1698 for (es = execsw; *es; es++)
1700 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1703 for (es = execsw; *es; es++)
1708 free(execsw, M_TEMP);
1714 exec_unregister(execsw_arg)
1715 const struct execsw *execsw_arg;
1717 const struct execsw **es, **xs, **newexecsw;
1721 panic("unregister with no handlers left?\n");
1723 for (es = execsw; *es; es++) {
1724 if (*es == execsw_arg)
1729 for (es = execsw; *es; es++)
1730 if (*es != execsw_arg)
1732 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1734 for (es = execsw; *es; es++)
1735 if (*es != execsw_arg)
1739 free(execsw, M_TEMP);