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>
61 #include <sys/sf_buf.h>
62 #include <sys/syscallsubr.h>
63 #include <sys/sysent.h>
65 #include <sys/sysctl.h>
66 #include <sys/vnode.h>
69 #include <sys/ktrace.h>
73 #include <vm/vm_param.h>
75 #include <vm/vm_page.h>
76 #include <vm/vm_map.h>
77 #include <vm/vm_kern.h>
78 #include <vm/vm_extern.h>
79 #include <vm/vm_object.h>
80 #include <vm/vm_pager.h>
83 #include <sys/pmckern.h>
86 #include <machine/reg.h>
88 #include <security/audit/audit.h>
89 #include <security/mac/mac_framework.h>
92 #include <sys/dtrace_bsd.h>
93 dtrace_execexit_func_t dtrace_fasttrap_exec;
96 SDT_PROVIDER_DECLARE(proc);
97 SDT_PROBE_DEFINE(proc, kernel, , exec, exec);
98 SDT_PROBE_ARGTYPE(proc, kernel, , exec, 0, "char *");
99 SDT_PROBE_DEFINE(proc, kernel, , exec_failure, exec-failure);
100 SDT_PROBE_ARGTYPE(proc, kernel, , exec_failure, 0, "int");
101 SDT_PROBE_DEFINE(proc, kernel, , exec_success, exec-success);
102 SDT_PROBE_ARGTYPE(proc, kernel, , exec_success, 0, "char *");
104 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
106 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
107 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
108 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
109 static int do_execve(struct thread *td, struct image_args *args,
112 /* XXX This should be vm_size_t. */
113 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD,
114 NULL, 0, sysctl_kern_ps_strings, "LU", "");
116 /* XXX This should be vm_size_t. */
117 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD|
118 CTLFLAG_CAPRD, NULL, 0, sysctl_kern_usrstack, "LU", "");
120 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD,
121 NULL, 0, sysctl_kern_stackprot, "I", "");
123 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
124 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
125 &ps_arg_cache_limit, 0, "");
127 static int map_at_zero = 0;
128 TUNABLE_INT("security.bsd.map_at_zero", &map_at_zero);
129 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RW, &map_at_zero, 0,
130 "Permit processes to map an object at virtual address 0.");
133 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
140 if (req->flags & SCTL_MASK32) {
142 val = (unsigned int)p->p_sysent->sv_psstrings;
143 error = SYSCTL_OUT(req, &val, sizeof(val));
146 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
147 sizeof(p->p_sysent->sv_psstrings));
152 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
159 if (req->flags & SCTL_MASK32) {
161 val = (unsigned int)p->p_sysent->sv_usrstack;
162 error = SYSCTL_OUT(req, &val, sizeof(val));
165 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
166 sizeof(p->p_sysent->sv_usrstack));
171 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
176 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
177 sizeof(p->p_sysent->sv_stackprot)));
181 * Each of the items is a pointer to a `const struct execsw', hence the
182 * double pointer here.
184 static const struct execsw **execsw;
186 #ifndef _SYS_SYSPROTO_H_
197 struct execve_args /* {
204 struct image_args args;
206 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
207 uap->argv, uap->envv);
209 error = kern_execve(td, &args, NULL);
213 #ifndef _SYS_SYSPROTO_H_
214 struct fexecve_args {
221 sys_fexecve(struct thread *td, struct fexecve_args *uap)
224 struct image_args args;
226 error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
227 uap->argv, uap->envv);
230 error = kern_execve(td, &args, NULL);
235 #ifndef _SYS_SYSPROTO_H_
236 struct __mac_execve_args {
245 sys___mac_execve(td, uap)
247 struct __mac_execve_args /* {
256 struct image_args args;
258 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
259 uap->argv, uap->envv);
261 error = kern_execve(td, &args, uap->mac_p);
269 * XXX: kern_execve has the astonishing property of not always returning to
270 * the caller. If sufficiently bad things happen during the call to
271 * do_execve(), it can end up calling exit1(); as a result, callers must
272 * avoid doing anything which they might need to undo (e.g., allocating
276 kern_execve(td, args, mac_p)
278 struct image_args *args;
281 struct proc *p = td->td_proc;
284 AUDIT_ARG_ARGV(args->begin_argv, args->argc,
285 args->begin_envv - args->begin_argv);
286 AUDIT_ARG_ENVV(args->begin_envv, args->envc,
287 args->endp - args->begin_envv);
288 if (p->p_flag & P_HADTHREADS) {
290 if (thread_single(SINGLE_BOUNDARY)) {
292 exec_free_args(args);
293 return (ERESTART); /* Try again later. */
298 error = do_execve(td, args, mac_p);
300 if (p->p_flag & P_HADTHREADS) {
303 * If success, we upgrade to SINGLE_EXIT state to
304 * force other threads to suicide.
307 thread_single(SINGLE_EXIT);
317 * In-kernel implementation of execve(). All arguments are assumed to be
318 * userspace pointers from the passed thread.
321 do_execve(td, args, mac_p)
323 struct image_args *args;
326 struct proc *p = td->td_proc;
328 struct ucred *newcred = NULL, *oldcred;
329 struct uidinfo *euip;
330 register_t *stack_base;
332 struct image_params image_params, *imgp;
334 int (*img_first)(struct image_params *);
335 struct pargs *oldargs = NULL, *newargs = NULL;
336 struct sigacts *oldsigacts, *newsigacts;
338 struct vnode *tracevp = NULL;
339 struct ucred *tracecred = NULL;
341 struct vnode *textvp = NULL, *binvp = NULL;
342 int credential_changing;
346 struct label *interpvplabel = NULL;
350 struct pmckern_procexec pe;
352 static const char fexecv_proc_title[] = "(fexecv)";
355 imgp = &image_params;
358 * Lock the process and set the P_INEXEC flag to indicate that
359 * it should be left alone until we're done here. This is
360 * necessary to avoid race conditions - e.g. in ptrace() -
361 * that might allow a local user to illicitly obtain elevated
365 KASSERT((p->p_flag & P_INEXEC) == 0,
366 ("%s(): process already has P_INEXEC flag", __func__));
367 p->p_flag |= P_INEXEC;
371 * Initialize part of the common data
374 imgp->execlabel = NULL;
376 imgp->entry_addr = 0;
377 imgp->reloc_base = 0;
378 imgp->vmspace_destroyed = 0;
379 imgp->interpreted = 0;
381 imgp->interpreter_name = NULL;
382 imgp->auxargs = NULL;
385 imgp->firstpage = NULL;
386 imgp->ps_strings = 0;
387 imgp->auxarg_size = 0;
389 imgp->execpath = imgp->freepath = NULL;
394 imgp->pagesizeslen = 0;
395 imgp->stack_prot = 0;
398 error = mac_execve_enter(imgp, mac_p);
403 imgp->image_header = NULL;
406 * Translate the file name. namei() returns a vnode pointer
407 * in ni_vp amoung other things.
409 * XXXAUDIT: It would be desirable to also audit the name of the
410 * interpreter if this is an interpreted binary.
412 if (args->fname != NULL) {
413 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME
414 | MPSAFE | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
417 SDT_PROBE(proc, kernel, , exec, args->fname, 0, 0, 0, 0 );
420 if (args->fname != NULL) {
421 #ifdef CAPABILITY_MODE
423 * While capability mode can't reach this point via direct
424 * path arguments to execve(), we also don't allow
425 * interpreters to be used in capability mode (for now).
426 * Catch indirect lookups and return a permissions error.
428 if (IN_CAPABILITY_MODE(td)) {
437 vfslocked = NDHASGIANT(&nd);
441 AUDIT_ARG_FD(args->fd);
443 * Some might argue that CAP_READ and/or CAP_MMAP should also
444 * be required here; such arguments will be entertained.
446 error = fgetvp_read(td, args->fd, CAP_FEXECVE, &binvp);
449 vfslocked = VFS_LOCK_GIANT(binvp->v_mount);
450 vn_lock(binvp, LK_EXCLUSIVE | LK_RETRY);
451 AUDIT_ARG_VNODE1(binvp);
456 * Check file permissions (also 'opens' file)
458 error = exec_check_permissions(imgp);
460 goto exec_fail_dealloc;
462 imgp->object = imgp->vp->v_object;
463 if (imgp->object != NULL)
464 vm_object_reference(imgp->object);
467 * Set VV_TEXT now so no one can write to the executable while we're
470 * Remember if this was set before and unset it in case this is not
471 * actually an executable image.
473 textset = imgp->vp->v_vflag & VV_TEXT;
474 imgp->vp->v_vflag |= VV_TEXT;
476 error = exec_map_first_page(imgp);
478 goto exec_fail_dealloc;
480 imgp->proc->p_osrel = 0;
482 * If the current process has a special image activator it
483 * wants to try first, call it. For example, emulating shell
484 * scripts differently.
487 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
488 error = img_first(imgp);
491 * Loop through the list of image activators, calling each one.
492 * An activator returns -1 if there is no match, 0 on success,
493 * and an error otherwise.
495 for (i = 0; error == -1 && execsw[i]; ++i) {
496 if (execsw[i]->ex_imgact == NULL ||
497 execsw[i]->ex_imgact == img_first) {
500 error = (*execsw[i]->ex_imgact)(imgp);
506 imgp->vp->v_vflag &= ~VV_TEXT;
509 goto exec_fail_dealloc;
513 * Special interpreter operation, cleanup and loop up to try to
514 * activate the interpreter.
516 if (imgp->interpreted) {
517 exec_unmap_first_page(imgp);
519 * VV_TEXT needs to be unset for scripts. There is a short
520 * period before we determine that something is a script where
521 * VV_TEXT will be set. The vnode lock is held over this
522 * entire period so nothing should illegitimately be blocked.
524 imgp->vp->v_vflag &= ~VV_TEXT;
525 /* free name buffer and old vnode */
526 if (args->fname != NULL)
527 NDFREE(&nd, NDF_ONLY_PNBUF);
529 mac_execve_interpreter_enter(binvp, &interpvplabel);
532 VOP_CLOSE(binvp, FREAD, td->td_ucred, td);
536 vm_object_deallocate(imgp->object);
538 VFS_UNLOCK_GIANT(vfslocked);
540 /* set new name to that of the interpreter */
541 NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME | MPSAFE,
542 UIO_SYSSPACE, imgp->interpreter_name, td);
543 args->fname = imgp->interpreter_name;
548 * NB: We unlock the vnode here because it is believed that none
549 * of the sv_copyout_strings/sv_fixup operations require the vnode.
551 VOP_UNLOCK(imgp->vp, 0);
554 * Do the best to calculate the full path to the image file.
556 if (imgp->auxargs != NULL &&
557 ((args->fname != NULL && args->fname[0] == '/') ||
558 vn_fullpath(td, imgp->vp, &imgp->execpath, &imgp->freepath) != 0))
559 imgp->execpath = args->fname;
562 * Copy out strings (args and env) and initialize stack base
564 if (p->p_sysent->sv_copyout_strings)
565 stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
567 stack_base = exec_copyout_strings(imgp);
570 * If custom stack fixup routine present for this process
571 * let it do the stack setup.
572 * Else stuff argument count as first item on stack
574 if (p->p_sysent->sv_fixup != NULL)
575 (*p->p_sysent->sv_fixup)(&stack_base, imgp);
577 suword(--stack_base, imgp->args->argc);
580 * For security and other reasons, the file descriptor table cannot
581 * be shared after an exec.
586 * Malloc things before we need locks.
589 euip = uifind(attr.va_uid);
590 i = imgp->args->begin_envv - imgp->args->begin_argv;
591 /* Cache arguments if they fit inside our allowance */
592 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
593 newargs = pargs_alloc(i);
594 bcopy(imgp->args->begin_argv, newargs->ar_args, i);
597 /* close files on exec */
599 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
601 /* Get a reference to the vnode prior to locking the proc */
605 * For security and other reasons, signal handlers cannot
606 * be shared after an exec. The new process gets a copy of the old
607 * handlers. In execsigs(), the new process will have its signals
611 oldcred = crcopysafe(p, newcred);
612 if (sigacts_shared(p->p_sigacts)) {
613 oldsigacts = p->p_sigacts;
615 newsigacts = sigacts_alloc();
616 sigacts_copy(newsigacts, oldsigacts);
618 p->p_sigacts = newsigacts;
625 /* reset caught signals */
628 /* name this process - nameiexec(p, ndp) */
629 bzero(p->p_comm, sizeof(p->p_comm));
631 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
632 min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
633 else if (vn_commname(binvp, p->p_comm, sizeof(p->p_comm)) != 0)
634 bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
635 bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
638 * mark as execed, wakeup the process that vforked (if any) and tell
639 * it that it now has its own resources back
642 if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
643 p->p_flag &= ~P_PPWAIT;
644 cv_broadcast(&p->p_pwait);
648 * Implement image setuid/setgid.
650 * Don't honor setuid/setgid if the filesystem prohibits it or if
651 * the process is being traced.
653 * We disable setuid/setgid/etc in compatibility mode on the basis
654 * that most setugid applications are not written with that
655 * environment in mind, and will therefore almost certainly operate
656 * incorrectly. In principle there's no reason that setugid
657 * applications might not be useful in capability mode, so we may want
658 * to reconsider this conservative design choice in the future.
660 * XXXMAC: For the time being, use NOSUID to also prohibit
661 * transitions on the file system.
663 credential_changing = 0;
664 credential_changing |= (attr.va_mode & S_ISUID) && oldcred->cr_uid !=
666 credential_changing |= (attr.va_mode & S_ISGID) && oldcred->cr_gid !=
669 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
670 interpvplabel, imgp);
671 credential_changing |= will_transition;
674 if (credential_changing &&
675 #ifdef CAPABILITY_MODE
676 ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
678 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
679 (p->p_flag & P_TRACED) == 0) {
681 * Turn off syscall tracing for set-id programs, except for
682 * root. Record any set-id flags first to make sure that
683 * we do not regain any tracing during a possible block.
688 if (priv_check_cred(oldcred, PRIV_DEBUG_DIFFCRED, 0))
689 ktrprocexec(p, &tracecred, &tracevp);
692 * Close any file descriptors 0..2 that reference procfs,
693 * then make sure file descriptors 0..2 are in use.
695 * setugidsafety() may call closef() and then pfind()
696 * which may grab the process lock.
697 * fdcheckstd() may call falloc() which may block to
698 * allocate memory, so temporarily drop the process lock.
701 VOP_UNLOCK(imgp->vp, 0);
703 error = fdcheckstd(td);
704 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
709 * Set the new credentials.
711 if (attr.va_mode & S_ISUID)
712 change_euid(newcred, euip);
713 if (attr.va_mode & S_ISGID)
714 change_egid(newcred, attr.va_gid);
716 if (will_transition) {
717 mac_vnode_execve_transition(oldcred, newcred, imgp->vp,
718 interpvplabel, imgp);
722 * Implement correct POSIX saved-id behavior.
724 * XXXMAC: Note that the current logic will save the
725 * uid and gid if a MAC domain transition occurs, even
726 * though maybe it shouldn't.
728 change_svuid(newcred, newcred->cr_uid);
729 change_svgid(newcred, newcred->cr_gid);
730 p->p_ucred = newcred;
733 if (oldcred->cr_uid == oldcred->cr_ruid &&
734 oldcred->cr_gid == oldcred->cr_rgid)
735 p->p_flag &= ~P_SUGID;
737 * Implement correct POSIX saved-id behavior.
739 * XXX: It's not clear that the existing behavior is
740 * POSIX-compliant. A number of sources indicate that the
741 * saved uid/gid should only be updated if the new ruid is
742 * not equal to the old ruid, or the new euid is not equal
743 * to the old euid and the new euid is not equal to the old
744 * ruid. The FreeBSD code always updates the saved uid/gid.
745 * Also, this code uses the new (replaced) euid and egid as
746 * the source, which may or may not be the right ones to use.
748 if (oldcred->cr_svuid != oldcred->cr_uid ||
749 oldcred->cr_svgid != oldcred->cr_gid) {
750 change_svuid(newcred, newcred->cr_uid);
751 change_svgid(newcred, newcred->cr_gid);
752 p->p_ucred = newcred;
758 * Store the vp for use in procfs. This vnode was referenced prior
759 * to locking the proc lock.
761 textvp = p->p_textvp;
766 * Tell the DTrace fasttrap provider about the exec if it
767 * has declared an interest.
769 if (dtrace_fasttrap_exec)
770 dtrace_fasttrap_exec(p);
774 * Notify others that we exec'd, and clear the P_INEXEC flag
775 * as we're now a bona fide freshly-execed process.
777 KNOTE_LOCKED(&p->p_klist, NOTE_EXEC);
778 p->p_flag &= ~P_INEXEC;
781 * If tracing the process, trap to the debugger so that
782 * breakpoints can be set before the program executes. We
783 * have to use tdsignal() to deliver the signal to the current
784 * thread since any other threads in this process will exit if
787 if (p->p_flag & P_TRACED)
788 tdsignal(td, SIGTRAP);
790 /* clear "fork but no exec" flag, as we _are_ execing */
791 p->p_acflag &= ~AFORK;
794 * Free any previous argument cache and replace it with
795 * the new argument cache, if any.
803 * Check if system-wide sampling is in effect or if the
804 * current process is using PMCs. If so, do exec() time
805 * processing. This processing needs to happen AFTER the
806 * P_INEXEC flag is cleared.
808 * The proc lock needs to be released before taking the PMC
811 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
813 VOP_UNLOCK(imgp->vp, 0);
814 pe.pm_credentialschanged = credential_changing;
815 pe.pm_entryaddr = imgp->entry_addr;
817 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
818 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
821 #else /* !HWPMC_HOOKS */
825 /* Set values passed into the program in registers. */
826 if (p->p_sysent->sv_setregs)
827 (*p->p_sysent->sv_setregs)(td, imgp,
828 (u_long)(uintptr_t)stack_base);
830 exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base);
832 vfs_mark_atime(imgp->vp, td->td_ucred);
834 SDT_PROBE(proc, kernel, , exec_success, args->fname, 0, 0, 0, 0);
838 * Free any resources malloc'd earlier that we didn't use.
845 VOP_UNLOCK(imgp->vp, 0);
848 * Handle deferred decrement of ref counts.
850 if (textvp != NULL) {
853 tvfslocked = VFS_LOCK_GIANT(textvp->v_mount);
855 VFS_UNLOCK_GIANT(tvfslocked);
857 if (binvp && error != 0)
860 if (tracevp != NULL) {
863 tvfslocked = VFS_LOCK_GIANT(tracevp->v_mount);
865 VFS_UNLOCK_GIANT(tvfslocked);
867 if (tracecred != NULL)
870 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
873 if (oldsigacts != NULL)
874 sigacts_free(oldsigacts);
879 * free various allocated resources
881 if (imgp->firstpage != NULL)
882 exec_unmap_first_page(imgp);
884 if (imgp->vp != NULL) {
886 NDFREE(&nd, NDF_ONLY_PNBUF);
888 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
892 if (imgp->object != NULL)
893 vm_object_deallocate(imgp->object);
895 free(imgp->freepath, M_TEMP);
899 td->td_dbgflags |= TDB_EXEC;
903 * Stop the process here if its stop event mask has
904 * the S_EXEC bit set.
906 STOPEVENT(p, S_EXEC, 0);
911 /* we're done here, clear P_INEXEC */
913 p->p_flag &= ~P_INEXEC;
916 SDT_PROBE(proc, kernel, , exec_failure, error, 0, 0, 0, 0);
920 mac_execve_exit(imgp);
921 mac_execve_interpreter_exit(interpvplabel);
923 VFS_UNLOCK_GIANT(vfslocked);
924 exec_free_args(args);
926 if (error && imgp->vmspace_destroyed) {
927 /* sorry, no more process anymore. exit gracefully */
928 exit1(td, W_EXITCODE(0, SIGABRT));
941 exec_map_first_page(imgp)
942 struct image_params *imgp;
946 vm_page_t ma[VM_INITIAL_PAGEIN];
949 if (imgp->firstpage != NULL)
950 exec_unmap_first_page(imgp);
952 object = imgp->vp->v_object;
955 VM_OBJECT_LOCK(object);
956 #if VM_NRESERVLEVEL > 0
957 if ((object->flags & OBJ_COLORED) == 0) {
958 object->flags |= OBJ_COLORED;
959 object->pg_color = 0;
962 ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
963 if (ma[0]->valid != VM_PAGE_BITS_ALL) {
964 initial_pagein = VM_INITIAL_PAGEIN;
965 if (initial_pagein > object->size)
966 initial_pagein = object->size;
967 for (i = 1; i < initial_pagein; i++) {
968 if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) {
971 if ((ma[i]->oflags & VPO_BUSY) || ma[i]->busy)
975 ma[i] = vm_page_alloc(object, i,
976 VM_ALLOC_NORMAL | VM_ALLOC_IFNOTCACHED);
982 rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
983 ma[0] = vm_page_lookup(object, 0);
984 if ((rv != VM_PAGER_OK) || (ma[0] == NULL)) {
988 vm_page_unlock(ma[0]);
990 VM_OBJECT_UNLOCK(object);
996 vm_page_unlock(ma[0]);
997 vm_page_wakeup(ma[0]);
998 VM_OBJECT_UNLOCK(object);
1000 imgp->firstpage = sf_buf_alloc(ma[0], 0);
1001 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
1007 exec_unmap_first_page(imgp)
1008 struct image_params *imgp;
1012 if (imgp->firstpage != NULL) {
1013 m = sf_buf_page(imgp->firstpage);
1014 sf_buf_free(imgp->firstpage);
1015 imgp->firstpage = NULL;
1023 * Destroy old address space, and allocate a new stack
1024 * The new stack is only SGROWSIZ large because it is grown
1025 * automatically in trap.c.
1028 exec_new_vmspace(imgp, sv)
1029 struct image_params *imgp;
1030 struct sysentvec *sv;
1033 struct proc *p = imgp->proc;
1034 struct vmspace *vmspace = p->p_vmspace;
1036 vm_offset_t sv_minuser, stack_addr;
1040 imgp->vmspace_destroyed = 1;
1043 /* May be called with Giant held */
1044 EVENTHANDLER_INVOKE(process_exec, p, imgp);
1047 * Blow away entire process VM, if address space not shared,
1048 * otherwise, create a new VM space so that other threads are
1051 map = &vmspace->vm_map;
1053 sv_minuser = sv->sv_minuser;
1055 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1056 if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser &&
1057 vm_map_max(map) == sv->sv_maxuser) {
1059 pmap_remove_pages(vmspace_pmap(vmspace));
1060 vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1062 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1065 vmspace = p->p_vmspace;
1066 map = &vmspace->vm_map;
1069 /* Map a shared page */
1070 obj = sv->sv_shared_page_obj;
1072 vm_object_reference(obj);
1073 error = vm_map_fixed(map, obj, 0,
1074 sv->sv_shared_page_base, sv->sv_shared_page_len,
1075 VM_PROT_READ | VM_PROT_EXECUTE, VM_PROT_ALL,
1076 MAP_COPY_ON_WRITE | MAP_ACC_NO_CHARGE);
1078 vm_object_deallocate(obj);
1083 /* Allocate a new stack */
1084 if (sv->sv_maxssiz != NULL)
1085 ssiz = *sv->sv_maxssiz;
1088 stack_addr = sv->sv_usrstack - ssiz;
1089 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1090 obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1092 VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1097 /* Allocate a new register stack */
1098 stack_addr = IA64_BACKINGSTORE;
1099 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1100 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_UP);
1105 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
1106 * VM_STACK case, but they are still used to monitor the size of the
1107 * process stack so we can check the stack rlimit.
1109 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1110 vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - ssiz;
1116 * Copy out argument and environment strings from the old process address
1117 * space into the temporary string buffer.
1120 exec_copyin_args(struct image_args *args, char *fname,
1121 enum uio_seg segflg, char **argv, char **envv)
1127 bzero(args, sizeof(*args));
1132 * Allocate demand-paged memory for the file name, argument, and
1133 * environment strings.
1135 error = exec_alloc_args(args);
1140 * Copy the file name.
1142 if (fname != NULL) {
1143 args->fname = args->buf;
1144 error = (segflg == UIO_SYSSPACE) ?
1145 copystr(fname, args->fname, PATH_MAX, &length) :
1146 copyinstr(fname, args->fname, PATH_MAX, &length);
1152 args->begin_argv = args->buf + length;
1153 args->endp = args->begin_argv;
1154 args->stringspace = ARG_MAX;
1157 * extract arguments first
1159 while ((argp = (caddr_t) (intptr_t) fuword(argv++))) {
1160 if (argp == (caddr_t) -1) {
1164 if ((error = copyinstr(argp, args->endp,
1165 args->stringspace, &length))) {
1166 if (error == ENAMETOOLONG)
1170 args->stringspace -= length;
1171 args->endp += length;
1175 args->begin_envv = args->endp;
1178 * extract environment strings
1181 while ((envp = (caddr_t)(intptr_t)fuword(envv++))) {
1182 if (envp == (caddr_t)-1) {
1186 if ((error = copyinstr(envp, args->endp,
1187 args->stringspace, &length))) {
1188 if (error == ENAMETOOLONG)
1192 args->stringspace -= length;
1193 args->endp += length;
1201 exec_free_args(args);
1206 * Allocate temporary demand-paged, zero-filled memory for the file name,
1207 * argument, and environment strings. Returns zero if the allocation succeeds
1208 * and ENOMEM otherwise.
1211 exec_alloc_args(struct image_args *args)
1214 args->buf = (char *)kmem_alloc_wait(exec_map, PATH_MAX + ARG_MAX);
1215 return (args->buf != NULL ? 0 : ENOMEM);
1219 exec_free_args(struct image_args *args)
1222 if (args->buf != NULL) {
1223 kmem_free_wakeup(exec_map, (vm_offset_t)args->buf,
1224 PATH_MAX + ARG_MAX);
1227 if (args->fname_buf != NULL) {
1228 free(args->fname_buf, M_TEMP);
1229 args->fname_buf = NULL;
1234 * Copy strings out to the new process address space, constructing new arg
1235 * and env vector tables. Return a pointer to the base so that it can be used
1236 * as the initial stack pointer.
1239 exec_copyout_strings(imgp)
1240 struct image_params *imgp;
1244 char *stringp, *destp;
1245 register_t *stack_base;
1246 struct ps_strings *arginfo;
1248 size_t execpath_len;
1249 int szsigcode, szps;
1250 char canary[sizeof(long) * 8];
1252 szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1254 * Calculate string base and vector table pointers.
1255 * Also deal with signal trampoline code for this exec type.
1257 if (imgp->execpath != NULL && imgp->auxargs != NULL)
1258 execpath_len = strlen(imgp->execpath) + 1;
1263 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1264 if (p->p_sysent->sv_sigcode_base == 0) {
1265 if (p->p_sysent->sv_szsigcode != NULL)
1266 szsigcode = *(p->p_sysent->sv_szsigcode);
1268 destp = (caddr_t)arginfo - szsigcode - SPARE_USRSPACE -
1269 roundup(execpath_len, sizeof(char *)) -
1270 roundup(sizeof(canary), sizeof(char *)) -
1271 roundup(szps, sizeof(char *)) -
1272 roundup((ARG_MAX - imgp->args->stringspace), sizeof(char *));
1278 copyout(p->p_sysent->sv_sigcode, ((caddr_t)arginfo -
1279 szsigcode), szsigcode);
1282 * Copy the image path for the rtld.
1284 if (execpath_len != 0) {
1285 imgp->execpathp = (uintptr_t)arginfo - szsigcode - execpath_len;
1286 copyout(imgp->execpath, (void *)imgp->execpathp,
1291 * Prepare the canary for SSP.
1293 arc4rand(canary, sizeof(canary), 0);
1294 imgp->canary = (uintptr_t)arginfo - szsigcode - execpath_len -
1296 copyout(canary, (void *)imgp->canary, sizeof(canary));
1297 imgp->canarylen = sizeof(canary);
1300 * Prepare the pagesizes array.
1302 imgp->pagesizes = (uintptr_t)arginfo - szsigcode - execpath_len -
1303 roundup(sizeof(canary), sizeof(char *)) - szps;
1304 copyout(pagesizes, (void *)imgp->pagesizes, szps);
1305 imgp->pagesizeslen = szps;
1308 * If we have a valid auxargs ptr, prepare some room
1311 if (imgp->auxargs) {
1313 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1314 * lower compatibility.
1316 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1319 * The '+ 2' is for the null pointers at the end of each of
1320 * the arg and env vector sets,and imgp->auxarg_size is room
1321 * for argument of Runtime loader.
1323 vectp = (char **)(destp - (imgp->args->argc +
1324 imgp->args->envc + 2 + imgp->auxarg_size)
1328 * The '+ 2' is for the null pointers at the end of each of
1329 * the arg and env vector sets
1331 vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc + 2) *
1336 * vectp also becomes our initial stack base
1338 stack_base = (register_t *)vectp;
1340 stringp = imgp->args->begin_argv;
1341 argc = imgp->args->argc;
1342 envc = imgp->args->envc;
1345 * Copy out strings - arguments and environment.
1347 copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
1350 * Fill in "ps_strings" struct for ps, w, etc.
1352 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1353 suword32(&arginfo->ps_nargvstr, argc);
1356 * Fill in argument portion of vector table.
1358 for (; argc > 0; --argc) {
1359 suword(vectp++, (long)(intptr_t)destp);
1360 while (*stringp++ != 0)
1365 /* a null vector table pointer separates the argp's from the envp's */
1368 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1369 suword32(&arginfo->ps_nenvstr, envc);
1372 * Fill in environment portion of vector table.
1374 for (; envc > 0; --envc) {
1375 suword(vectp++, (long)(intptr_t)destp);
1376 while (*stringp++ != 0)
1381 /* end of vector table is a null pointer */
1384 return (stack_base);
1388 * Check permissions of file to execute.
1389 * Called with imgp->vp locked.
1390 * Return 0 for success or error code on failure.
1393 exec_check_permissions(imgp)
1394 struct image_params *imgp;
1396 struct vnode *vp = imgp->vp;
1397 struct vattr *attr = imgp->attr;
1403 /* Get file attributes */
1404 error = VOP_GETATTR(vp, attr, td->td_ucred);
1409 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1415 * 1) Check if file execution is disabled for the filesystem that
1416 * this file resides on.
1417 * 2) Ensure that at least one execute bit is on. Otherwise, a
1418 * privileged user will always succeed, and we don't want this
1419 * to happen unless the file really is executable.
1420 * 3) Ensure that the file is a regular file.
1422 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1423 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1424 (attr->va_type != VREG))
1428 * Zero length files can't be exec'd
1430 if (attr->va_size == 0)
1434 * Check for execute permission to file based on current credentials.
1436 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1441 * Check number of open-for-writes on the file and deny execution
1444 if (vp->v_writecount)
1448 * Call filesystem specific open routine (which does nothing in the
1451 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1458 * Exec handler registration
1461 exec_register(execsw_arg)
1462 const struct execsw *execsw_arg;
1464 const struct execsw **es, **xs, **newexecsw;
1465 int count = 2; /* New slot and trailing NULL */
1468 for (es = execsw; *es; es++)
1470 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1471 if (newexecsw == NULL)
1475 for (es = execsw; *es; es++)
1480 free(execsw, M_TEMP);
1486 exec_unregister(execsw_arg)
1487 const struct execsw *execsw_arg;
1489 const struct execsw **es, **xs, **newexecsw;
1493 panic("unregister with no handlers left?\n");
1495 for (es = execsw; *es; es++) {
1496 if (*es == execsw_arg)
1501 for (es = execsw; *es; es++)
1502 if (*es != execsw_arg)
1504 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1505 if (newexecsw == NULL)
1508 for (es = execsw; *es; es++)
1509 if (*es != execsw_arg)
1513 free(execsw, M_TEMP);
1518 static vm_object_t shared_page_obj;
1519 static int shared_page_free;
1522 shared_page_fill(int size, int align, const char *data)
1529 VM_OBJECT_LOCK(shared_page_obj);
1530 m = vm_page_grab(shared_page_obj, 0, VM_ALLOC_RETRY);
1531 res = roundup(shared_page_free, align);
1532 if (res + size >= IDX_TO_OFF(shared_page_obj->size))
1535 VM_OBJECT_UNLOCK(shared_page_obj);
1536 s = sf_buf_alloc(m, SFB_DEFAULT);
1538 bcopy(data, (void *)(sk + res), size);
1539 shared_page_free = res + size;
1541 VM_OBJECT_LOCK(shared_page_obj);
1544 VM_OBJECT_UNLOCK(shared_page_obj);
1549 shared_page_init(void *dummy __unused)
1553 shared_page_obj = vm_pager_allocate(OBJT_PHYS, 0, PAGE_SIZE,
1554 VM_PROT_DEFAULT, 0, NULL);
1555 VM_OBJECT_LOCK(shared_page_obj);
1556 m = vm_page_grab(shared_page_obj, 0, VM_ALLOC_RETRY | VM_ALLOC_NOBUSY |
1558 m->valid = VM_PAGE_BITS_ALL;
1559 VM_OBJECT_UNLOCK(shared_page_obj);
1562 SYSINIT(shp, SI_SUB_EXEC, SI_ORDER_FIRST, (sysinit_cfunc_t)shared_page_init,
1566 exec_sysvec_init(void *param)
1568 struct sysentvec *sv;
1570 sv = (struct sysentvec *)param;
1572 if ((sv->sv_flags & SV_SHP) == 0)
1574 sv->sv_shared_page_obj = shared_page_obj;
1575 sv->sv_sigcode_base = sv->sv_shared_page_base +
1576 shared_page_fill(*(sv->sv_szsigcode), 16, sv->sv_sigcode);