2 * Copyright (c) 1993, David Greenman
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD$");
30 #include "opt_capsicum.h"
31 #include "opt_hwpmc_hooks.h"
32 #include "opt_kdtrace.h"
33 #include "opt_ktrace.h"
36 #include <sys/param.h>
37 #include <sys/capability.h>
38 #include <sys/systm.h>
39 #include <sys/capability.h>
40 #include <sys/eventhandler.h>
42 #include <sys/mutex.h>
43 #include <sys/sysproto.h>
44 #include <sys/signalvar.h>
45 #include <sys/kernel.h>
46 #include <sys/mount.h>
47 #include <sys/filedesc.h>
48 #include <sys/fcntl.h>
51 #include <sys/imgact.h>
52 #include <sys/imgact_elf.h>
54 #include <sys/malloc.h>
57 #include <sys/pioctl.h>
58 #include <sys/namei.h>
59 #include <sys/resourcevar.h>
60 #include <sys/sched.h>
62 #include <sys/sf_buf.h>
63 #include <sys/syscallsubr.h>
64 #include <sys/sysent.h>
66 #include <sys/sysctl.h>
67 #include <sys/vnode.h>
70 #include <sys/ktrace.h>
74 #include <vm/vm_param.h>
76 #include <vm/vm_page.h>
77 #include <vm/vm_map.h>
78 #include <vm/vm_kern.h>
79 #include <vm/vm_extern.h>
80 #include <vm/vm_object.h>
81 #include <vm/vm_pager.h>
84 #include <sys/pmckern.h>
87 #include <machine/reg.h>
89 #include <security/audit/audit.h>
90 #include <security/mac/mac_framework.h>
93 #include <sys/dtrace_bsd.h>
94 dtrace_execexit_func_t dtrace_fasttrap_exec;
97 SDT_PROVIDER_DECLARE(proc);
98 SDT_PROBE_DEFINE(proc, kernel, , exec, exec);
99 SDT_PROBE_ARGTYPE(proc, kernel, , exec, 0, "char *");
100 SDT_PROBE_DEFINE(proc, kernel, , exec_failure, exec-failure);
101 SDT_PROBE_ARGTYPE(proc, kernel, , exec_failure, 0, "int");
102 SDT_PROBE_DEFINE(proc, kernel, , exec_success, exec-success);
103 SDT_PROBE_ARGTYPE(proc, kernel, , exec_success, 0, "char *");
105 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
107 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
108 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
109 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
110 static int do_execve(struct thread *td, struct image_args *args,
113 /* XXX This should be vm_size_t. */
114 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD,
115 NULL, 0, sysctl_kern_ps_strings, "LU", "");
117 /* XXX This should be vm_size_t. */
118 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD|
119 CTLFLAG_CAPRD, NULL, 0, sysctl_kern_usrstack, "LU", "");
121 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD,
122 NULL, 0, sysctl_kern_stackprot, "I", "");
124 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
125 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
126 &ps_arg_cache_limit, 0, "");
128 static int map_at_zero = 0;
129 TUNABLE_INT("security.bsd.map_at_zero", &map_at_zero);
130 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RW, &map_at_zero, 0,
131 "Permit processes to map an object at virtual address 0.");
134 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
141 if (req->flags & SCTL_MASK32) {
143 val = (unsigned int)p->p_sysent->sv_psstrings;
144 error = SYSCTL_OUT(req, &val, sizeof(val));
147 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
148 sizeof(p->p_sysent->sv_psstrings));
153 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
160 if (req->flags & SCTL_MASK32) {
162 val = (unsigned int)p->p_sysent->sv_usrstack;
163 error = SYSCTL_OUT(req, &val, sizeof(val));
166 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
167 sizeof(p->p_sysent->sv_usrstack));
172 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
177 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
178 sizeof(p->p_sysent->sv_stackprot)));
182 * Each of the items is a pointer to a `const struct execsw', hence the
183 * double pointer here.
185 static const struct execsw **execsw;
187 #ifndef _SYS_SYSPROTO_H_
198 struct execve_args /* {
205 struct image_args args;
207 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
208 uap->argv, uap->envv);
210 error = kern_execve(td, &args, NULL);
214 #ifndef _SYS_SYSPROTO_H_
215 struct fexecve_args {
222 sys_fexecve(struct thread *td, struct fexecve_args *uap)
225 struct image_args args;
227 error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
228 uap->argv, uap->envv);
231 error = kern_execve(td, &args, NULL);
236 #ifndef _SYS_SYSPROTO_H_
237 struct __mac_execve_args {
246 sys___mac_execve(td, uap)
248 struct __mac_execve_args /* {
257 struct image_args args;
259 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
260 uap->argv, uap->envv);
262 error = kern_execve(td, &args, uap->mac_p);
270 * XXX: kern_execve has the astonishing property of not always returning to
271 * the caller. If sufficiently bad things happen during the call to
272 * do_execve(), it can end up calling exit1(); as a result, callers must
273 * avoid doing anything which they might need to undo (e.g., allocating
277 kern_execve(td, args, mac_p)
279 struct image_args *args;
282 struct proc *p = td->td_proc;
283 struct vmspace *oldvmspace;
286 AUDIT_ARG_ARGV(args->begin_argv, args->argc,
287 args->begin_envv - args->begin_argv);
288 AUDIT_ARG_ENVV(args->begin_envv, args->envc,
289 args->endp - args->begin_envv);
290 if (p->p_flag & P_HADTHREADS) {
292 if (thread_single(SINGLE_BOUNDARY)) {
294 exec_free_args(args);
295 return (ERESTART); /* Try again later. */
300 KASSERT((td->td_pflags & TDP_EXECVMSPC) == 0, ("nested execve"));
301 oldvmspace = td->td_proc->p_vmspace;
302 error = do_execve(td, args, mac_p);
304 if (p->p_flag & P_HADTHREADS) {
307 * If success, we upgrade to SINGLE_EXIT state to
308 * force other threads to suicide.
311 thread_single(SINGLE_EXIT);
316 if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
317 KASSERT(td->td_proc->p_vmspace != oldvmspace,
318 ("oldvmspace still used"));
319 vmspace_free(oldvmspace);
320 td->td_pflags &= ~TDP_EXECVMSPC;
327 * In-kernel implementation of execve(). All arguments are assumed to be
328 * userspace pointers from the passed thread.
331 do_execve(td, args, mac_p)
333 struct image_args *args;
336 struct proc *p = td->td_proc;
338 struct ucred *newcred = NULL, *oldcred;
339 struct uidinfo *euip;
340 register_t *stack_base;
342 struct image_params image_params, *imgp;
344 int (*img_first)(struct image_params *);
345 struct pargs *oldargs = NULL, *newargs = NULL;
346 struct sigacts *oldsigacts, *newsigacts;
348 struct vnode *tracevp = NULL;
349 struct ucred *tracecred = NULL;
351 struct vnode *textvp = NULL, *binvp = NULL;
352 int credential_changing;
356 struct label *interpvplabel = NULL;
360 struct pmckern_procexec pe;
362 static const char fexecv_proc_title[] = "(fexecv)";
365 imgp = &image_params;
368 * Lock the process and set the P_INEXEC flag to indicate that
369 * it should be left alone until we're done here. This is
370 * necessary to avoid race conditions - e.g. in ptrace() -
371 * that might allow a local user to illicitly obtain elevated
375 KASSERT((p->p_flag & P_INEXEC) == 0,
376 ("%s(): process already has P_INEXEC flag", __func__));
377 p->p_flag |= P_INEXEC;
381 * Initialize part of the common data
384 imgp->execlabel = NULL;
386 imgp->entry_addr = 0;
387 imgp->reloc_base = 0;
388 imgp->vmspace_destroyed = 0;
389 imgp->interpreted = 0;
391 imgp->interpreter_name = NULL;
392 imgp->auxargs = NULL;
395 imgp->firstpage = NULL;
396 imgp->ps_strings = 0;
397 imgp->auxarg_size = 0;
399 imgp->execpath = imgp->freepath = NULL;
404 imgp->pagesizeslen = 0;
405 imgp->stack_prot = 0;
408 error = mac_execve_enter(imgp, mac_p);
413 imgp->image_header = NULL;
416 * Translate the file name. namei() returns a vnode pointer
417 * in ni_vp amoung other things.
419 * XXXAUDIT: It would be desirable to also audit the name of the
420 * interpreter if this is an interpreted binary.
422 if (args->fname != NULL) {
423 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME
424 | MPSAFE | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
427 SDT_PROBE(proc, kernel, , exec, args->fname, 0, 0, 0, 0 );
430 if (args->fname != NULL) {
431 #ifdef CAPABILITY_MODE
433 * While capability mode can't reach this point via direct
434 * path arguments to execve(), we also don't allow
435 * interpreters to be used in capability mode (for now).
436 * Catch indirect lookups and return a permissions error.
438 if (IN_CAPABILITY_MODE(td)) {
447 vfslocked = NDHASGIANT(&nd);
451 AUDIT_ARG_FD(args->fd);
453 * Some might argue that CAP_READ and/or CAP_MMAP should also
454 * be required here; such arguments will be entertained.
456 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
458 error = fgetvp_exec(td, args->fd, CAP_FEXECVE, &binvp);
461 vfslocked = VFS_LOCK_GIANT(binvp->v_mount);
462 vn_lock(binvp, LK_EXCLUSIVE | LK_RETRY);
463 AUDIT_ARG_VNODE1(binvp);
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 = imgp->vp->v_vflag & VV_TEXT;
486 ASSERT_VOP_ELOCKED(imgp->vp, "vv_text");
487 imgp->vp->v_vflag |= VV_TEXT;
489 error = exec_map_first_page(imgp);
491 goto exec_fail_dealloc;
493 imgp->proc->p_osrel = 0;
495 * If the current process has a special image activator it
496 * wants to try first, call it. For example, emulating shell
497 * scripts differently.
500 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
501 error = img_first(imgp);
504 * Loop through the list of image activators, calling each one.
505 * An activator returns -1 if there is no match, 0 on success,
506 * and an error otherwise.
508 for (i = 0; error == -1 && execsw[i]; ++i) {
509 if (execsw[i]->ex_imgact == NULL ||
510 execsw[i]->ex_imgact == img_first) {
513 error = (*execsw[i]->ex_imgact)(imgp);
519 ASSERT_VOP_ELOCKED(imgp->vp, "vv_text");
520 imgp->vp->v_vflag &= ~VV_TEXT;
524 goto exec_fail_dealloc;
528 * Special interpreter operation, cleanup and loop up to try to
529 * activate the interpreter.
531 if (imgp->interpreted) {
532 exec_unmap_first_page(imgp);
534 * VV_TEXT needs to be unset for scripts. There is a short
535 * period before we determine that something is a script where
536 * VV_TEXT will be set. The vnode lock is held over this
537 * entire period so nothing should illegitimately be blocked.
539 imgp->vp->v_vflag &= ~VV_TEXT;
540 /* free name buffer and old vnode */
541 if (args->fname != NULL)
542 NDFREE(&nd, NDF_ONLY_PNBUF);
544 mac_execve_interpreter_enter(binvp, &interpvplabel);
547 VOP_CLOSE(binvp, FREAD, td->td_ucred, td);
551 vm_object_deallocate(imgp->object);
553 VFS_UNLOCK_GIANT(vfslocked);
555 /* set new name to that of the interpreter */
556 NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME | MPSAFE,
557 UIO_SYSSPACE, imgp->interpreter_name, td);
558 args->fname = imgp->interpreter_name;
563 * NB: We unlock the vnode here because it is believed that none
564 * of the sv_copyout_strings/sv_fixup operations require the vnode.
566 VOP_UNLOCK(imgp->vp, 0);
569 * Do the best to calculate the full path to the image file.
571 if (imgp->auxargs != NULL &&
572 ((args->fname != NULL && args->fname[0] == '/') ||
573 vn_fullpath(td, imgp->vp, &imgp->execpath, &imgp->freepath) != 0))
574 imgp->execpath = args->fname;
577 * Copy out strings (args and env) and initialize stack base
579 if (p->p_sysent->sv_copyout_strings)
580 stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
582 stack_base = exec_copyout_strings(imgp);
585 * If custom stack fixup routine present for this process
586 * let it do the stack setup.
587 * Else stuff argument count as first item on stack
589 if (p->p_sysent->sv_fixup != NULL)
590 (*p->p_sysent->sv_fixup)(&stack_base, imgp);
592 suword(--stack_base, imgp->args->argc);
595 * For security and other reasons, the file descriptor table cannot
596 * be shared after an exec.
601 * Malloc things before we need locks.
604 euip = uifind(attr.va_uid);
605 i = imgp->args->begin_envv - imgp->args->begin_argv;
606 /* Cache arguments if they fit inside our allowance */
607 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
608 newargs = pargs_alloc(i);
609 bcopy(imgp->args->begin_argv, newargs->ar_args, i);
612 /* close files on exec */
614 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
616 /* Get a reference to the vnode prior to locking the proc */
620 * For security and other reasons, signal handlers cannot
621 * be shared after an exec. The new process gets a copy of the old
622 * handlers. In execsigs(), the new process will have its signals
626 oldcred = crcopysafe(p, newcred);
627 if (sigacts_shared(p->p_sigacts)) {
628 oldsigacts = p->p_sigacts;
630 newsigacts = sigacts_alloc();
631 sigacts_copy(newsigacts, oldsigacts);
633 p->p_sigacts = newsigacts;
640 /* reset caught signals */
643 /* name this process - nameiexec(p, ndp) */
644 bzero(p->p_comm, sizeof(p->p_comm));
646 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
647 min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
648 else if (vn_commname(binvp, p->p_comm, sizeof(p->p_comm)) != 0)
649 bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
650 bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
652 sched_clear_tdname(td);
656 * mark as execed, wakeup the process that vforked (if any) and tell
657 * it that it now has its own resources back
660 if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
661 p->p_flag &= ~P_PPWAIT;
662 cv_broadcast(&p->p_pwait);
666 * Implement image setuid/setgid.
668 * Don't honor setuid/setgid if the filesystem prohibits it or if
669 * the process is being traced.
671 * We disable setuid/setgid/etc in compatibility mode on the basis
672 * that most setugid applications are not written with that
673 * environment in mind, and will therefore almost certainly operate
674 * incorrectly. In principle there's no reason that setugid
675 * applications might not be useful in capability mode, so we may want
676 * to reconsider this conservative design choice in the future.
678 * XXXMAC: For the time being, use NOSUID to also prohibit
679 * transitions on the file system.
681 credential_changing = 0;
682 credential_changing |= (attr.va_mode & S_ISUID) && oldcred->cr_uid !=
684 credential_changing |= (attr.va_mode & S_ISGID) && oldcred->cr_gid !=
687 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
688 interpvplabel, imgp);
689 credential_changing |= will_transition;
692 if (credential_changing &&
693 #ifdef CAPABILITY_MODE
694 ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
696 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
697 (p->p_flag & P_TRACED) == 0) {
699 * Turn off syscall tracing for set-id programs, except for
700 * root. Record any set-id flags first to make sure that
701 * we do not regain any tracing during a possible block.
706 if (priv_check_cred(oldcred, PRIV_DEBUG_DIFFCRED, 0))
707 ktrprocexec(p, &tracecred, &tracevp);
710 * Close any file descriptors 0..2 that reference procfs,
711 * then make sure file descriptors 0..2 are in use.
713 * setugidsafety() may call closef() and then pfind()
714 * which may grab the process lock.
715 * fdcheckstd() may call falloc() which may block to
716 * allocate memory, so temporarily drop the process lock.
719 VOP_UNLOCK(imgp->vp, 0);
721 error = fdcheckstd(td);
722 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
727 * Set the new credentials.
729 if (attr.va_mode & S_ISUID)
730 change_euid(newcred, euip);
731 if (attr.va_mode & S_ISGID)
732 change_egid(newcred, attr.va_gid);
734 if (will_transition) {
735 mac_vnode_execve_transition(oldcred, newcred, imgp->vp,
736 interpvplabel, imgp);
740 * Implement correct POSIX saved-id behavior.
742 * XXXMAC: Note that the current logic will save the
743 * uid and gid if a MAC domain transition occurs, even
744 * though maybe it shouldn't.
746 change_svuid(newcred, newcred->cr_uid);
747 change_svgid(newcred, newcred->cr_gid);
748 p->p_ucred = newcred;
751 if (oldcred->cr_uid == oldcred->cr_ruid &&
752 oldcred->cr_gid == oldcred->cr_rgid)
753 p->p_flag &= ~P_SUGID;
755 * Implement correct POSIX saved-id behavior.
757 * XXX: It's not clear that the existing behavior is
758 * POSIX-compliant. A number of sources indicate that the
759 * saved uid/gid should only be updated if the new ruid is
760 * not equal to the old ruid, or the new euid is not equal
761 * to the old euid and the new euid is not equal to the old
762 * ruid. The FreeBSD code always updates the saved uid/gid.
763 * Also, this code uses the new (replaced) euid and egid as
764 * the source, which may or may not be the right ones to use.
766 if (oldcred->cr_svuid != oldcred->cr_uid ||
767 oldcred->cr_svgid != oldcred->cr_gid) {
768 change_svuid(newcred, newcred->cr_uid);
769 change_svgid(newcred, newcred->cr_gid);
770 p->p_ucred = newcred;
776 * Store the vp for use in procfs. This vnode was referenced prior
777 * to locking the proc lock.
779 textvp = p->p_textvp;
784 * Tell the DTrace fasttrap provider about the exec if it
785 * has declared an interest.
787 if (dtrace_fasttrap_exec)
788 dtrace_fasttrap_exec(p);
792 * Notify others that we exec'd, and clear the P_INEXEC flag
793 * as we're now a bona fide freshly-execed process.
795 KNOTE_LOCKED(&p->p_klist, NOTE_EXEC);
796 p->p_flag &= ~P_INEXEC;
798 /* clear "fork but no exec" flag, as we _are_ execing */
799 p->p_acflag &= ~AFORK;
802 * Free any previous argument cache and replace it with
803 * the new argument cache, if any.
811 * Check if system-wide sampling is in effect or if the
812 * current process is using PMCs. If so, do exec() time
813 * processing. This processing needs to happen AFTER the
814 * P_INEXEC flag is cleared.
816 * The proc lock needs to be released before taking the PMC
819 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
821 VOP_UNLOCK(imgp->vp, 0);
822 pe.pm_credentialschanged = credential_changing;
823 pe.pm_entryaddr = imgp->entry_addr;
825 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
826 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
829 #else /* !HWPMC_HOOKS */
833 /* Set values passed into the program in registers. */
834 if (p->p_sysent->sv_setregs)
835 (*p->p_sysent->sv_setregs)(td, imgp,
836 (u_long)(uintptr_t)stack_base);
838 exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base);
840 vfs_mark_atime(imgp->vp, td->td_ucred);
842 SDT_PROBE(proc, kernel, , exec_success, args->fname, 0, 0, 0, 0);
846 * Free any resources malloc'd earlier that we didn't use.
853 VOP_UNLOCK(imgp->vp, 0);
856 * Handle deferred decrement of ref counts.
858 if (textvp != NULL) {
861 tvfslocked = VFS_LOCK_GIANT(textvp->v_mount);
863 VFS_UNLOCK_GIANT(tvfslocked);
865 if (binvp && error != 0)
868 if (tracevp != NULL) {
871 tvfslocked = VFS_LOCK_GIANT(tracevp->v_mount);
873 VFS_UNLOCK_GIANT(tvfslocked);
875 if (tracecred != NULL)
878 vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
881 if (oldsigacts != NULL)
882 sigacts_free(oldsigacts);
887 * free various allocated resources
889 if (imgp->firstpage != NULL)
890 exec_unmap_first_page(imgp);
892 if (imgp->vp != NULL) {
894 NDFREE(&nd, NDF_ONLY_PNBUF);
896 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
900 if (imgp->object != NULL)
901 vm_object_deallocate(imgp->object);
903 free(imgp->freepath, M_TEMP);
907 td->td_dbgflags |= TDB_EXEC;
911 * Stop the process here if its stop event mask has
912 * the S_EXEC bit set.
914 STOPEVENT(p, S_EXEC, 0);
919 /* we're done here, clear P_INEXEC */
921 p->p_flag &= ~P_INEXEC;
924 SDT_PROBE(proc, kernel, , exec_failure, error, 0, 0, 0, 0);
928 mac_execve_exit(imgp);
929 mac_execve_interpreter_exit(interpvplabel);
931 VFS_UNLOCK_GIANT(vfslocked);
932 exec_free_args(args);
934 if (error && imgp->vmspace_destroyed) {
935 /* sorry, no more process anymore. exit gracefully */
936 exit1(td, W_EXITCODE(0, SIGABRT));
949 exec_map_first_page(imgp)
950 struct image_params *imgp;
954 vm_page_t ma[VM_INITIAL_PAGEIN];
957 if (imgp->firstpage != NULL)
958 exec_unmap_first_page(imgp);
960 object = imgp->vp->v_object;
963 VM_OBJECT_LOCK(object);
964 #if VM_NRESERVLEVEL > 0
965 if ((object->flags & OBJ_COLORED) == 0) {
966 object->flags |= OBJ_COLORED;
967 object->pg_color = 0;
970 ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
971 if (ma[0]->valid != VM_PAGE_BITS_ALL) {
972 initial_pagein = VM_INITIAL_PAGEIN;
973 if (initial_pagein > object->size)
974 initial_pagein = object->size;
975 for (i = 1; i < initial_pagein; i++) {
976 if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) {
979 if ((ma[i]->oflags & VPO_BUSY) || ma[i]->busy)
983 ma[i] = vm_page_alloc(object, i,
984 VM_ALLOC_NORMAL | VM_ALLOC_IFNOTCACHED);
990 rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
991 ma[0] = vm_page_lookup(object, 0);
992 if ((rv != VM_PAGER_OK) || (ma[0] == NULL)) {
996 vm_page_unlock(ma[0]);
998 VM_OBJECT_UNLOCK(object);
1002 vm_page_lock(ma[0]);
1003 vm_page_hold(ma[0]);
1004 vm_page_unlock(ma[0]);
1005 vm_page_wakeup(ma[0]);
1006 VM_OBJECT_UNLOCK(object);
1008 imgp->firstpage = sf_buf_alloc(ma[0], 0);
1009 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
1015 exec_unmap_first_page(imgp)
1016 struct image_params *imgp;
1020 if (imgp->firstpage != NULL) {
1021 m = sf_buf_page(imgp->firstpage);
1022 sf_buf_free(imgp->firstpage);
1023 imgp->firstpage = NULL;
1031 * Destroy old address space, and allocate a new stack
1032 * The new stack is only SGROWSIZ large because it is grown
1033 * automatically in trap.c.
1036 exec_new_vmspace(imgp, sv)
1037 struct image_params *imgp;
1038 struct sysentvec *sv;
1041 struct proc *p = imgp->proc;
1042 struct vmspace *vmspace = p->p_vmspace;
1044 vm_offset_t sv_minuser, stack_addr;
1048 imgp->vmspace_destroyed = 1;
1051 /* May be called with Giant held */
1052 EVENTHANDLER_INVOKE(process_exec, p, imgp);
1055 * Blow away entire process VM, if address space not shared,
1056 * otherwise, create a new VM space so that other threads are
1059 map = &vmspace->vm_map;
1061 sv_minuser = sv->sv_minuser;
1063 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1064 if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser &&
1065 vm_map_max(map) == sv->sv_maxuser) {
1067 pmap_remove_pages(vmspace_pmap(vmspace));
1068 vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1070 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1073 vmspace = p->p_vmspace;
1074 map = &vmspace->vm_map;
1077 /* Map a shared page */
1078 obj = sv->sv_shared_page_obj;
1080 vm_object_reference(obj);
1081 error = vm_map_fixed(map, obj, 0,
1082 sv->sv_shared_page_base, sv->sv_shared_page_len,
1083 VM_PROT_READ | VM_PROT_EXECUTE, VM_PROT_ALL,
1084 MAP_COPY_ON_WRITE | MAP_ACC_NO_CHARGE);
1086 vm_object_deallocate(obj);
1091 /* Allocate a new stack */
1092 if (sv->sv_maxssiz != NULL)
1093 ssiz = *sv->sv_maxssiz;
1096 stack_addr = sv->sv_usrstack - ssiz;
1097 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1098 obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1100 VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1105 /* Allocate a new register stack */
1106 stack_addr = IA64_BACKINGSTORE;
1107 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1108 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_UP);
1113 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
1114 * VM_STACK case, but they are still used to monitor the size of the
1115 * process stack so we can check the stack rlimit.
1117 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1118 vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - ssiz;
1124 * Copy out argument and environment strings from the old process address
1125 * space into the temporary string buffer.
1128 exec_copyin_args(struct image_args *args, char *fname,
1129 enum uio_seg segflg, char **argv, char **envv)
1135 bzero(args, sizeof(*args));
1140 * Allocate demand-paged memory for the file name, argument, and
1141 * environment strings.
1143 error = exec_alloc_args(args);
1148 * Copy the file name.
1150 if (fname != NULL) {
1151 args->fname = args->buf;
1152 error = (segflg == UIO_SYSSPACE) ?
1153 copystr(fname, args->fname, PATH_MAX, &length) :
1154 copyinstr(fname, args->fname, PATH_MAX, &length);
1160 args->begin_argv = args->buf + length;
1161 args->endp = args->begin_argv;
1162 args->stringspace = ARG_MAX;
1165 * extract arguments first
1167 while ((argp = (caddr_t) (intptr_t) fuword(argv++))) {
1168 if (argp == (caddr_t) -1) {
1172 if ((error = copyinstr(argp, args->endp,
1173 args->stringspace, &length))) {
1174 if (error == ENAMETOOLONG)
1178 args->stringspace -= length;
1179 args->endp += length;
1183 args->begin_envv = args->endp;
1186 * extract environment strings
1189 while ((envp = (caddr_t)(intptr_t)fuword(envv++))) {
1190 if (envp == (caddr_t)-1) {
1194 if ((error = copyinstr(envp, args->endp,
1195 args->stringspace, &length))) {
1196 if (error == ENAMETOOLONG)
1200 args->stringspace -= length;
1201 args->endp += length;
1209 exec_free_args(args);
1214 * Allocate temporary demand-paged, zero-filled memory for the file name,
1215 * argument, and environment strings. Returns zero if the allocation succeeds
1216 * and ENOMEM otherwise.
1219 exec_alloc_args(struct image_args *args)
1222 args->buf = (char *)kmem_alloc_wait(exec_map, PATH_MAX + ARG_MAX);
1223 return (args->buf != NULL ? 0 : ENOMEM);
1227 exec_free_args(struct image_args *args)
1230 if (args->buf != NULL) {
1231 kmem_free_wakeup(exec_map, (vm_offset_t)args->buf,
1232 PATH_MAX + ARG_MAX);
1235 if (args->fname_buf != NULL) {
1236 free(args->fname_buf, M_TEMP);
1237 args->fname_buf = NULL;
1242 * Copy strings out to the new process address space, constructing new arg
1243 * and env vector tables. Return a pointer to the base so that it can be used
1244 * as the initial stack pointer.
1247 exec_copyout_strings(imgp)
1248 struct image_params *imgp;
1252 char *stringp, *destp;
1253 register_t *stack_base;
1254 struct ps_strings *arginfo;
1256 size_t execpath_len;
1257 int szsigcode, szps;
1258 char canary[sizeof(long) * 8];
1260 szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1262 * Calculate string base and vector table pointers.
1263 * Also deal with signal trampoline code for this exec type.
1265 if (imgp->execpath != NULL && imgp->auxargs != NULL)
1266 execpath_len = strlen(imgp->execpath) + 1;
1271 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1272 if (p->p_sysent->sv_sigcode_base == 0) {
1273 if (p->p_sysent->sv_szsigcode != NULL)
1274 szsigcode = *(p->p_sysent->sv_szsigcode);
1276 destp = (caddr_t)arginfo - szsigcode - SPARE_USRSPACE -
1277 roundup(execpath_len, sizeof(char *)) -
1278 roundup(sizeof(canary), sizeof(char *)) -
1279 roundup(szps, sizeof(char *)) -
1280 roundup((ARG_MAX - imgp->args->stringspace), sizeof(char *));
1286 copyout(p->p_sysent->sv_sigcode, ((caddr_t)arginfo -
1287 szsigcode), szsigcode);
1290 * Copy the image path for the rtld.
1292 if (execpath_len != 0) {
1293 imgp->execpathp = (uintptr_t)arginfo - szsigcode - execpath_len;
1294 copyout(imgp->execpath, (void *)imgp->execpathp,
1299 * Prepare the canary for SSP.
1301 arc4rand(canary, sizeof(canary), 0);
1302 imgp->canary = (uintptr_t)arginfo - szsigcode - execpath_len -
1304 copyout(canary, (void *)imgp->canary, sizeof(canary));
1305 imgp->canarylen = sizeof(canary);
1308 * Prepare the pagesizes array.
1310 imgp->pagesizes = (uintptr_t)arginfo - szsigcode - execpath_len -
1311 roundup(sizeof(canary), sizeof(char *)) - szps;
1312 copyout(pagesizes, (void *)imgp->pagesizes, szps);
1313 imgp->pagesizeslen = szps;
1316 * If we have a valid auxargs ptr, prepare some room
1319 if (imgp->auxargs) {
1321 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1322 * lower compatibility.
1324 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1327 * The '+ 2' is for the null pointers at the end of each of
1328 * the arg and env vector sets,and imgp->auxarg_size is room
1329 * for argument of Runtime loader.
1331 vectp = (char **)(destp - (imgp->args->argc +
1332 imgp->args->envc + 2 + imgp->auxarg_size)
1336 * The '+ 2' is for the null pointers at the end of each of
1337 * the arg and env vector sets
1339 vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc + 2) *
1344 * vectp also becomes our initial stack base
1346 stack_base = (register_t *)vectp;
1348 stringp = imgp->args->begin_argv;
1349 argc = imgp->args->argc;
1350 envc = imgp->args->envc;
1353 * Copy out strings - arguments and environment.
1355 copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
1358 * Fill in "ps_strings" struct for ps, w, etc.
1360 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1361 suword32(&arginfo->ps_nargvstr, argc);
1364 * Fill in argument portion of vector table.
1366 for (; argc > 0; --argc) {
1367 suword(vectp++, (long)(intptr_t)destp);
1368 while (*stringp++ != 0)
1373 /* a null vector table pointer separates the argp's from the envp's */
1376 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1377 suword32(&arginfo->ps_nenvstr, envc);
1380 * Fill in environment portion of vector table.
1382 for (; envc > 0; --envc) {
1383 suword(vectp++, (long)(intptr_t)destp);
1384 while (*stringp++ != 0)
1389 /* end of vector table is a null pointer */
1392 return (stack_base);
1396 * Check permissions of file to execute.
1397 * Called with imgp->vp locked.
1398 * Return 0 for success or error code on failure.
1401 exec_check_permissions(imgp)
1402 struct image_params *imgp;
1404 struct vnode *vp = imgp->vp;
1405 struct vattr *attr = imgp->attr;
1411 /* Get file attributes */
1412 error = VOP_GETATTR(vp, attr, td->td_ucred);
1417 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1423 * 1) Check if file execution is disabled for the filesystem that
1424 * this file resides on.
1425 * 2) Ensure that at least one execute bit is on. Otherwise, a
1426 * privileged user will always succeed, and we don't want this
1427 * to happen unless the file really is executable.
1428 * 3) Ensure that the file is a regular file.
1430 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1431 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1432 (attr->va_type != VREG))
1436 * Zero length files can't be exec'd
1438 if (attr->va_size == 0)
1442 * Check for execute permission to file based on current credentials.
1444 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1449 * Check number of open-for-writes on the file and deny execution
1452 if (vp->v_writecount)
1456 * Call filesystem specific open routine (which does nothing in the
1459 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1466 * Exec handler registration
1469 exec_register(execsw_arg)
1470 const struct execsw *execsw_arg;
1472 const struct execsw **es, **xs, **newexecsw;
1473 int count = 2; /* New slot and trailing NULL */
1476 for (es = execsw; *es; es++)
1478 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1479 if (newexecsw == NULL)
1483 for (es = execsw; *es; es++)
1488 free(execsw, M_TEMP);
1494 exec_unregister(execsw_arg)
1495 const struct execsw *execsw_arg;
1497 const struct execsw **es, **xs, **newexecsw;
1501 panic("unregister with no handlers left?\n");
1503 for (es = execsw; *es; es++) {
1504 if (*es == execsw_arg)
1509 for (es = execsw; *es; es++)
1510 if (*es != execsw_arg)
1512 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1513 if (newexecsw == NULL)
1516 for (es = execsw; *es; es++)
1517 if (*es != execsw_arg)
1521 free(execsw, M_TEMP);
1526 static vm_object_t shared_page_obj;
1527 static int shared_page_free;
1530 shared_page_fill(int size, int align, const char *data)
1537 VM_OBJECT_LOCK(shared_page_obj);
1538 m = vm_page_grab(shared_page_obj, 0, VM_ALLOC_RETRY);
1539 res = roundup(shared_page_free, align);
1540 if (res + size >= IDX_TO_OFF(shared_page_obj->size))
1543 VM_OBJECT_UNLOCK(shared_page_obj);
1544 s = sf_buf_alloc(m, SFB_DEFAULT);
1546 bcopy(data, (void *)(sk + res), size);
1547 shared_page_free = res + size;
1549 VM_OBJECT_LOCK(shared_page_obj);
1552 VM_OBJECT_UNLOCK(shared_page_obj);
1557 shared_page_init(void *dummy __unused)
1561 shared_page_obj = vm_pager_allocate(OBJT_PHYS, 0, PAGE_SIZE,
1562 VM_PROT_DEFAULT, 0, NULL);
1563 VM_OBJECT_LOCK(shared_page_obj);
1564 m = vm_page_grab(shared_page_obj, 0, VM_ALLOC_RETRY | VM_ALLOC_NOBUSY |
1566 m->valid = VM_PAGE_BITS_ALL;
1567 VM_OBJECT_UNLOCK(shared_page_obj);
1570 SYSINIT(shp, SI_SUB_EXEC, SI_ORDER_FIRST, (sysinit_cfunc_t)shared_page_init,
1574 exec_sysvec_init(void *param)
1576 struct sysentvec *sv;
1578 sv = (struct sysentvec *)param;
1580 if ((sv->sv_flags & SV_SHP) == 0)
1582 sv->sv_shared_page_obj = shared_page_obj;
1583 sv->sv_sigcode_base = sv->sv_shared_page_base +
1584 shared_page_fill(*(sv->sv_szsigcode), 16, sv->sv_sigcode);