xref: /freebsd/sys/kern/kern_exec.c (revision d5fc25e5d6c52b306312784663ccad85923a9c76)
1 /*-
2  * Copyright (c) 1993, David Greenman
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
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.
13  *
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
24  * SUCH DAMAGE.
25  */
26 
27 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD$");
29 
30 #include "opt_hwpmc_hooks.h"
31 #include "opt_kdtrace.h"
32 #include "opt_ktrace.h"
33 #include "opt_vm.h"
34 
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/eventhandler.h>
38 #include <sys/lock.h>
39 #include <sys/mutex.h>
40 #include <sys/sysproto.h>
41 #include <sys/signalvar.h>
42 #include <sys/kernel.h>
43 #include <sys/mount.h>
44 #include <sys/filedesc.h>
45 #include <sys/fcntl.h>
46 #include <sys/acct.h>
47 #include <sys/exec.h>
48 #include <sys/imgact.h>
49 #include <sys/imgact_elf.h>
50 #include <sys/wait.h>
51 #include <sys/malloc.h>
52 #include <sys/priv.h>
53 #include <sys/proc.h>
54 #include <sys/pioctl.h>
55 #include <sys/namei.h>
56 #include <sys/resourcevar.h>
57 #include <sys/sdt.h>
58 #include <sys/sf_buf.h>
59 #include <sys/syscallsubr.h>
60 #include <sys/sysent.h>
61 #include <sys/shm.h>
62 #include <sys/sysctl.h>
63 #include <sys/vnode.h>
64 #include <sys/stat.h>
65 #ifdef KTRACE
66 #include <sys/ktrace.h>
67 #endif
68 
69 #include <vm/vm.h>
70 #include <vm/vm_param.h>
71 #include <vm/pmap.h>
72 #include <vm/vm_page.h>
73 #include <vm/vm_map.h>
74 #include <vm/vm_kern.h>
75 #include <vm/vm_extern.h>
76 #include <vm/vm_object.h>
77 #include <vm/vm_pager.h>
78 
79 #ifdef	HWPMC_HOOKS
80 #include <sys/pmckern.h>
81 #endif
82 
83 #include <machine/reg.h>
84 
85 #include <security/audit/audit.h>
86 #include <security/mac/mac_framework.h>
87 
88 #ifdef KDTRACE_HOOKS
89 #include <sys/dtrace_bsd.h>
90 dtrace_execexit_func_t	dtrace_fasttrap_exec;
91 #endif
92 
93 SDT_PROVIDER_DECLARE(proc);
94 SDT_PROBE_DEFINE(proc, kernel, , exec);
95 SDT_PROBE_ARGTYPE(proc, kernel, , exec, 0, "char *");
96 SDT_PROBE_DEFINE(proc, kernel, , exec_failure);
97 SDT_PROBE_ARGTYPE(proc, kernel, , exec_failure, 0, "int");
98 SDT_PROBE_DEFINE(proc, kernel, , exec_success);
99 SDT_PROBE_ARGTYPE(proc, kernel, , exec_success, 0, "char *");
100 
101 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
102 
103 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
104 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
105 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
106 static int do_execve(struct thread *td, struct image_args *args,
107     struct mac *mac_p);
108 static void exec_free_args(struct image_args *);
109 
110 /* XXX This should be vm_size_t. */
111 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD,
112     NULL, 0, sysctl_kern_ps_strings, "LU", "");
113 
114 /* XXX This should be vm_size_t. */
115 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD,
116     NULL, 0, sysctl_kern_usrstack, "LU", "");
117 
118 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD,
119     NULL, 0, sysctl_kern_stackprot, "I", "");
120 
121 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
122 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
123     &ps_arg_cache_limit, 0, "");
124 
125 static int
126 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
127 {
128 	struct proc *p;
129 	int error;
130 
131 	p = curproc;
132 #ifdef SCTL_MASK32
133 	if (req->flags & SCTL_MASK32) {
134 		unsigned int val;
135 		val = (unsigned int)p->p_sysent->sv_psstrings;
136 		error = SYSCTL_OUT(req, &val, sizeof(val));
137 	} else
138 #endif
139 		error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
140 		   sizeof(p->p_sysent->sv_psstrings));
141 	return error;
142 }
143 
144 static int
145 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
146 {
147 	struct proc *p;
148 	int error;
149 
150 	p = curproc;
151 #ifdef SCTL_MASK32
152 	if (req->flags & SCTL_MASK32) {
153 		unsigned int val;
154 		val = (unsigned int)p->p_sysent->sv_usrstack;
155 		error = SYSCTL_OUT(req, &val, sizeof(val));
156 	} else
157 #endif
158 		error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
159 		    sizeof(p->p_sysent->sv_usrstack));
160 	return error;
161 }
162 
163 static int
164 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
165 {
166 	struct proc *p;
167 
168 	p = curproc;
169 	return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
170 	    sizeof(p->p_sysent->sv_stackprot)));
171 }
172 
173 /*
174  * Each of the items is a pointer to a `const struct execsw', hence the
175  * double pointer here.
176  */
177 static const struct execsw **execsw;
178 
179 #ifndef _SYS_SYSPROTO_H_
180 struct execve_args {
181 	char    *fname;
182 	char    **argv;
183 	char    **envv;
184 };
185 #endif
186 
187 int
188 execve(td, uap)
189 	struct thread *td;
190 	struct execve_args /* {
191 		char *fname;
192 		char **argv;
193 		char **envv;
194 	} */ *uap;
195 {
196 	int error;
197 	struct image_args args;
198 
199 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
200 	    uap->argv, uap->envv);
201 	if (error == 0)
202 		error = kern_execve(td, &args, NULL);
203 	return (error);
204 }
205 
206 #ifndef _SYS_SYSPROTO_H_
207 struct fexecve_args {
208 	int	fd;
209 	char	**argv;
210 	char	**envv;
211 }
212 #endif
213 int
214 fexecve(struct thread *td, struct fexecve_args *uap)
215 {
216 	int error;
217 	struct image_args args;
218 
219 	error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
220 	    uap->argv, uap->envv);
221 	if (error == 0) {
222 		args.fd = uap->fd;
223 		error = kern_execve(td, &args, NULL);
224 	}
225 	return (error);
226 }
227 
228 #ifndef _SYS_SYSPROTO_H_
229 struct __mac_execve_args {
230 	char	*fname;
231 	char	**argv;
232 	char	**envv;
233 	struct mac	*mac_p;
234 };
235 #endif
236 
237 int
238 __mac_execve(td, uap)
239 	struct thread *td;
240 	struct __mac_execve_args /* {
241 		char *fname;
242 		char **argv;
243 		char **envv;
244 		struct mac *mac_p;
245 	} */ *uap;
246 {
247 #ifdef MAC
248 	int error;
249 	struct image_args args;
250 
251 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
252 	    uap->argv, uap->envv);
253 	if (error == 0)
254 		error = kern_execve(td, &args, uap->mac_p);
255 	return (error);
256 #else
257 	return (ENOSYS);
258 #endif
259 }
260 
261 /*
262  * XXX: kern_execve has the astonishing property of not always returning to
263  * the caller.  If sufficiently bad things happen during the call to
264  * do_execve(), it can end up calling exit1(); as a result, callers must
265  * avoid doing anything which they might need to undo (e.g., allocating
266  * memory).
267  */
268 int
269 kern_execve(td, args, mac_p)
270 	struct thread *td;
271 	struct image_args *args;
272 	struct mac *mac_p;
273 {
274 	struct proc *p = td->td_proc;
275 	int error;
276 
277 	AUDIT_ARG(argv, args->begin_argv, args->argc,
278 	    args->begin_envv - args->begin_argv);
279 	AUDIT_ARG(envv, args->begin_envv, args->envc,
280 	    args->endp - args->begin_envv);
281 	if (p->p_flag & P_HADTHREADS) {
282 		PROC_LOCK(p);
283 		if (thread_single(SINGLE_BOUNDARY)) {
284 			PROC_UNLOCK(p);
285 	       		exec_free_args(args);
286 			return (ERESTART);	/* Try again later. */
287 		}
288 		PROC_UNLOCK(p);
289 	}
290 
291 	error = do_execve(td, args, mac_p);
292 
293 	if (p->p_flag & P_HADTHREADS) {
294 		PROC_LOCK(p);
295 		/*
296 		 * If success, we upgrade to SINGLE_EXIT state to
297 		 * force other threads to suicide.
298 		 */
299 		if (error == 0)
300 			thread_single(SINGLE_EXIT);
301 		else
302 			thread_single_end();
303 		PROC_UNLOCK(p);
304 	}
305 
306 	return (error);
307 }
308 
309 /*
310  * In-kernel implementation of execve().  All arguments are assumed to be
311  * userspace pointers from the passed thread.
312  */
313 static int
314 do_execve(td, args, mac_p)
315 	struct thread *td;
316 	struct image_args *args;
317 	struct mac *mac_p;
318 {
319 	struct proc *p = td->td_proc;
320 	struct nameidata nd;
321 	struct ucred *newcred = NULL, *oldcred;
322 	struct uidinfo *euip;
323 	register_t *stack_base;
324 	int error, len = 0, i;
325 	struct image_params image_params, *imgp;
326 	struct vattr attr;
327 	int (*img_first)(struct image_params *);
328 	struct pargs *oldargs = NULL, *newargs = NULL;
329 	struct sigacts *oldsigacts, *newsigacts;
330 #ifdef KTRACE
331 	struct vnode *tracevp = NULL;
332 	struct ucred *tracecred = NULL;
333 #endif
334 	struct vnode *textvp = NULL, *binvp = NULL;
335 	int credential_changing;
336 	int vfslocked;
337 	int textset;
338 #ifdef MAC
339 	struct label *interpvplabel = NULL;
340 	int will_transition;
341 #endif
342 #ifdef HWPMC_HOOKS
343 	struct pmckern_procexec pe;
344 #endif
345 	static const char fexecv_proc_title[] = "(fexecv)";
346 
347 	vfslocked = 0;
348 	imgp = &image_params;
349 
350 	/*
351 	 * Lock the process and set the P_INEXEC flag to indicate that
352 	 * it should be left alone until we're done here.  This is
353 	 * necessary to avoid race conditions - e.g. in ptrace() -
354 	 * that might allow a local user to illicitly obtain elevated
355 	 * privileges.
356 	 */
357 	PROC_LOCK(p);
358 	KASSERT((p->p_flag & P_INEXEC) == 0,
359 	    ("%s(): process already has P_INEXEC flag", __func__));
360 	p->p_flag |= P_INEXEC;
361 	PROC_UNLOCK(p);
362 
363 	/*
364 	 * Initialize part of the common data
365 	 */
366 	imgp->proc = p;
367 	imgp->execlabel = NULL;
368 	imgp->attr = &attr;
369 	imgp->entry_addr = 0;
370 	imgp->vmspace_destroyed = 0;
371 	imgp->interpreted = 0;
372 	imgp->opened = 0;
373 	imgp->interpreter_name = args->buf + PATH_MAX + ARG_MAX;
374 	imgp->auxargs = NULL;
375 	imgp->vp = NULL;
376 	imgp->object = NULL;
377 	imgp->firstpage = NULL;
378 	imgp->ps_strings = 0;
379 	imgp->auxarg_size = 0;
380 	imgp->args = args;
381 	imgp->execpath = imgp->freepath = NULL;
382 	imgp->execpathp = 0;
383 
384 #ifdef MAC
385 	error = mac_execve_enter(imgp, mac_p);
386 	if (error)
387 		goto exec_fail;
388 #endif
389 
390 	imgp->image_header = NULL;
391 
392 	/*
393 	 * Translate the file name. namei() returns a vnode pointer
394 	 *	in ni_vp amoung other things.
395 	 *
396 	 * XXXAUDIT: It would be desirable to also audit the name of the
397 	 * interpreter if this is an interpreted binary.
398 	 */
399 	if (args->fname != NULL) {
400 		NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME
401 		    | MPSAFE | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
402 	}
403 
404 	SDT_PROBE(proc, kernel, , exec, args->fname, 0, 0, 0, 0 );
405 
406 interpret:
407 	if (args->fname != NULL) {
408 		error = namei(&nd);
409 		if (error)
410 			goto exec_fail;
411 
412 		vfslocked = NDHASGIANT(&nd);
413 		binvp  = nd.ni_vp;
414 		imgp->vp = binvp;
415 	} else {
416 		AUDIT_ARG(fd, args->fd);
417 		error = fgetvp(td, args->fd, &binvp);
418 		if (error)
419 			goto exec_fail;
420 		vfslocked = VFS_LOCK_GIANT(binvp->v_mount);
421 		vn_lock(binvp, LK_EXCLUSIVE | LK_RETRY);
422 		AUDIT_ARG(vnode, binvp, ARG_VNODE1);
423 		imgp->vp = binvp;
424 	}
425 
426 	/*
427 	 * Check file permissions (also 'opens' file)
428 	 */
429 	error = exec_check_permissions(imgp);
430 	if (error)
431 		goto exec_fail_dealloc;
432 
433 	imgp->object = imgp->vp->v_object;
434 	if (imgp->object != NULL)
435 		vm_object_reference(imgp->object);
436 
437 	/*
438 	 * Set VV_TEXT now so no one can write to the executable while we're
439 	 * activating it.
440 	 *
441 	 * Remember if this was set before and unset it in case this is not
442 	 * actually an executable image.
443 	 */
444 	textset = imgp->vp->v_vflag & VV_TEXT;
445 	imgp->vp->v_vflag |= VV_TEXT;
446 
447 	error = exec_map_first_page(imgp);
448 	if (error)
449 		goto exec_fail_dealloc;
450 
451 	imgp->proc->p_osrel = 0;
452 	/*
453 	 *	If the current process has a special image activator it
454 	 *	wants to try first, call it.   For example, emulating shell
455 	 *	scripts differently.
456 	 */
457 	error = -1;
458 	if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
459 		error = img_first(imgp);
460 
461 	/*
462 	 *	Loop through the list of image activators, calling each one.
463 	 *	An activator returns -1 if there is no match, 0 on success,
464 	 *	and an error otherwise.
465 	 */
466 	for (i = 0; error == -1 && execsw[i]; ++i) {
467 		if (execsw[i]->ex_imgact == NULL ||
468 		    execsw[i]->ex_imgact == img_first) {
469 			continue;
470 		}
471 		error = (*execsw[i]->ex_imgact)(imgp);
472 	}
473 
474 	if (error) {
475 		if (error == -1) {
476 			if (textset == 0)
477 				imgp->vp->v_vflag &= ~VV_TEXT;
478 			error = ENOEXEC;
479 		}
480 		goto exec_fail_dealloc;
481 	}
482 
483 	/*
484 	 * Special interpreter operation, cleanup and loop up to try to
485 	 * activate the interpreter.
486 	 */
487 	if (imgp->interpreted) {
488 		exec_unmap_first_page(imgp);
489 		/*
490 		 * VV_TEXT needs to be unset for scripts.  There is a short
491 		 * period before we determine that something is a script where
492 		 * VV_TEXT will be set. The vnode lock is held over this
493 		 * entire period so nothing should illegitimately be blocked.
494 		 */
495 		imgp->vp->v_vflag &= ~VV_TEXT;
496 		/* free name buffer and old vnode */
497 		if (args->fname != NULL)
498 			NDFREE(&nd, NDF_ONLY_PNBUF);
499 #ifdef MAC
500 		mac_execve_interpreter_enter(binvp, &interpvplabel);
501 #endif
502 		if (imgp->opened) {
503 			VOP_CLOSE(binvp, FREAD, td->td_ucred, td);
504 			imgp->opened = 0;
505 		}
506 		vput(binvp);
507 		vm_object_deallocate(imgp->object);
508 		imgp->object = NULL;
509 		VFS_UNLOCK_GIANT(vfslocked);
510 		vfslocked = 0;
511 		/* set new name to that of the interpreter */
512 		NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME | MPSAFE,
513 		    UIO_SYSSPACE, imgp->interpreter_name, td);
514 		args->fname = imgp->interpreter_name;
515 		goto interpret;
516 	}
517 
518 	/*
519 	 * NB: We unlock the vnode here because it is believed that none
520 	 * of the sv_copyout_strings/sv_fixup operations require the vnode.
521 	 */
522 	VOP_UNLOCK(imgp->vp, 0);
523 
524 	/*
525 	 * Do the best to calculate the full path to the image file.
526 	 */
527 	if (imgp->auxargs != NULL &&
528 	    ((args->fname != NULL && args->fname[0] == '/') ||
529 	     vn_fullpath(td, imgp->vp, &imgp->execpath, &imgp->freepath) != 0))
530 		imgp->execpath = args->fname;
531 
532 	/*
533 	 * Copy out strings (args and env) and initialize stack base
534 	 */
535 	if (p->p_sysent->sv_copyout_strings)
536 		stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
537 	else
538 		stack_base = exec_copyout_strings(imgp);
539 
540 	/*
541 	 * If custom stack fixup routine present for this process
542 	 * let it do the stack setup.
543 	 * Else stuff argument count as first item on stack
544 	 */
545 	if (p->p_sysent->sv_fixup != NULL)
546 		(*p->p_sysent->sv_fixup)(&stack_base, imgp);
547 	else
548 		suword(--stack_base, imgp->args->argc);
549 
550 	/*
551 	 * For security and other reasons, the file descriptor table cannot
552 	 * be shared after an exec.
553 	 */
554 	fdunshare(p, td);
555 
556 	/*
557 	 * Malloc things before we need locks.
558 	 */
559 	newcred = crget();
560 	euip = uifind(attr.va_uid);
561 	i = imgp->args->begin_envv - imgp->args->begin_argv;
562 	/* Cache arguments if they fit inside our allowance */
563 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
564 		newargs = pargs_alloc(i);
565 		bcopy(imgp->args->begin_argv, newargs->ar_args, i);
566 	}
567 
568 	/* close files on exec */
569 	fdcloseexec(td);
570 	vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
571 
572 	/* Get a reference to the vnode prior to locking the proc */
573 	VREF(binvp);
574 
575 	/*
576 	 * For security and other reasons, signal handlers cannot
577 	 * be shared after an exec. The new process gets a copy of the old
578 	 * handlers. In execsigs(), the new process will have its signals
579 	 * reset.
580 	 */
581 	PROC_LOCK(p);
582 	if (sigacts_shared(p->p_sigacts)) {
583 		oldsigacts = p->p_sigacts;
584 		PROC_UNLOCK(p);
585 		newsigacts = sigacts_alloc();
586 		sigacts_copy(newsigacts, oldsigacts);
587 		PROC_LOCK(p);
588 		p->p_sigacts = newsigacts;
589 	} else
590 		oldsigacts = NULL;
591 
592 	/* Stop profiling */
593 	stopprofclock(p);
594 
595 	/* reset caught signals */
596 	execsigs(p);
597 
598 	/* name this process - nameiexec(p, ndp) */
599 	if (args->fname) {
600 		len = min(nd.ni_cnd.cn_namelen,MAXCOMLEN);
601 		bcopy(nd.ni_cnd.cn_nameptr, p->p_comm, len);
602 	} else {
603 		if (vn_commname(binvp, p->p_comm, MAXCOMLEN + 1) == 0)
604 			len = MAXCOMLEN;
605 		else {
606 			len = sizeof(fexecv_proc_title);
607 			bcopy(fexecv_proc_title, p->p_comm, len);
608 		}
609 	}
610 	p->p_comm[len] = 0;
611 	bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
612 
613 	/*
614 	 * mark as execed, wakeup the process that vforked (if any) and tell
615 	 * it that it now has its own resources back
616 	 */
617 	p->p_flag |= P_EXEC;
618 	if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
619 		p->p_flag &= ~P_PPWAIT;
620 		cv_broadcast(&p->p_pwait);
621 	}
622 
623 	/*
624 	 * Implement image setuid/setgid.
625 	 *
626 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
627 	 * the process is being traced.
628 	 *
629 	 * XXXMAC: For the time being, use NOSUID to also prohibit
630 	 * transitions on the file system.
631 	 */
632 	oldcred = p->p_ucred;
633 	credential_changing = 0;
634 	credential_changing |= (attr.va_mode & S_ISUID) && oldcred->cr_uid !=
635 	    attr.va_uid;
636 	credential_changing |= (attr.va_mode & S_ISGID) && oldcred->cr_gid !=
637 	    attr.va_gid;
638 #ifdef MAC
639 	will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
640 	    interpvplabel, imgp);
641 	credential_changing |= will_transition;
642 #endif
643 
644 	if (credential_changing &&
645 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
646 	    (p->p_flag & P_TRACED) == 0) {
647 		/*
648 		 * Turn off syscall tracing for set-id programs, except for
649 		 * root.  Record any set-id flags first to make sure that
650 		 * we do not regain any tracing during a possible block.
651 		 */
652 		setsugid(p);
653 
654 #ifdef KTRACE
655 		if (p->p_tracevp != NULL &&
656 		    priv_check_cred(oldcred, PRIV_DEBUG_DIFFCRED, 0)) {
657 			mtx_lock(&ktrace_mtx);
658 			p->p_traceflag = 0;
659 			tracevp = p->p_tracevp;
660 			p->p_tracevp = NULL;
661 			tracecred = p->p_tracecred;
662 			p->p_tracecred = NULL;
663 			mtx_unlock(&ktrace_mtx);
664 		}
665 #endif
666 		/*
667 		 * Close any file descriptors 0..2 that reference procfs,
668 		 * then make sure file descriptors 0..2 are in use.
669 		 *
670 		 * setugidsafety() may call closef() and then pfind()
671 		 * which may grab the process lock.
672 		 * fdcheckstd() may call falloc() which may block to
673 		 * allocate memory, so temporarily drop the process lock.
674 		 */
675 		PROC_UNLOCK(p);
676 		setugidsafety(td);
677 		VOP_UNLOCK(imgp->vp, 0);
678 		error = fdcheckstd(td);
679 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
680 		if (error != 0)
681 			goto done1;
682 		PROC_LOCK(p);
683 		/*
684 		 * Set the new credentials.
685 		 */
686 		crcopy(newcred, oldcred);
687 		if (attr.va_mode & S_ISUID)
688 			change_euid(newcred, euip);
689 		if (attr.va_mode & S_ISGID)
690 			change_egid(newcred, attr.va_gid);
691 #ifdef MAC
692 		if (will_transition) {
693 			mac_vnode_execve_transition(oldcred, newcred, imgp->vp,
694 			    interpvplabel, imgp);
695 		}
696 #endif
697 		/*
698 		 * Implement correct POSIX saved-id behavior.
699 		 *
700 		 * XXXMAC: Note that the current logic will save the
701 		 * uid and gid if a MAC domain transition occurs, even
702 		 * though maybe it shouldn't.
703 		 */
704 		change_svuid(newcred, newcred->cr_uid);
705 		change_svgid(newcred, newcred->cr_gid);
706 		p->p_ucred = newcred;
707 		newcred = NULL;
708 	} else {
709 		if (oldcred->cr_uid == oldcred->cr_ruid &&
710 		    oldcred->cr_gid == oldcred->cr_rgid)
711 			p->p_flag &= ~P_SUGID;
712 		/*
713 		 * Implement correct POSIX saved-id behavior.
714 		 *
715 		 * XXX: It's not clear that the existing behavior is
716 		 * POSIX-compliant.  A number of sources indicate that the
717 		 * saved uid/gid should only be updated if the new ruid is
718 		 * not equal to the old ruid, or the new euid is not equal
719 		 * to the old euid and the new euid is not equal to the old
720 		 * ruid.  The FreeBSD code always updates the saved uid/gid.
721 		 * Also, this code uses the new (replaced) euid and egid as
722 		 * the source, which may or may not be the right ones to use.
723 		 */
724 		if (oldcred->cr_svuid != oldcred->cr_uid ||
725 		    oldcred->cr_svgid != oldcred->cr_gid) {
726 			crcopy(newcred, oldcred);
727 			change_svuid(newcred, newcred->cr_uid);
728 			change_svgid(newcred, newcred->cr_gid);
729 			p->p_ucred = newcred;
730 			newcred = NULL;
731 		}
732 	}
733 
734 	/*
735 	 * Store the vp for use in procfs.  This vnode was referenced prior
736 	 * to locking the proc lock.
737 	 */
738 	textvp = p->p_textvp;
739 	p->p_textvp = binvp;
740 
741 #ifdef KDTRACE_HOOKS
742 	/*
743 	 * Tell the DTrace fasttrap provider about the exec if it
744 	 * has declared an interest.
745 	 */
746 	if (dtrace_fasttrap_exec)
747 		dtrace_fasttrap_exec(p);
748 #endif
749 
750 	/*
751 	 * Notify others that we exec'd, and clear the P_INEXEC flag
752 	 * as we're now a bona fide freshly-execed process.
753 	 */
754 	KNOTE_LOCKED(&p->p_klist, NOTE_EXEC);
755 	p->p_flag &= ~P_INEXEC;
756 
757 	/*
758 	 * If tracing the process, trap to debugger so breakpoints
759 	 * can be set before the program executes.
760 	 * Use tdsignal to deliver signal to current thread, use
761 	 * psignal may cause the signal to be delivered to wrong thread
762 	 * because that thread will exit, remember we are going to enter
763 	 * single thread mode.
764 	 */
765 	if (p->p_flag & P_TRACED)
766 		tdsignal(p, td, SIGTRAP, NULL);
767 
768 	/* clear "fork but no exec" flag, as we _are_ execing */
769 	p->p_acflag &= ~AFORK;
770 
771 	/*
772 	 * Free any previous argument cache and replace it with
773 	 * the new argument cache, if any.
774 	 */
775 	oldargs = p->p_args;
776 	p->p_args = newargs;
777 	newargs = NULL;
778 
779 #ifdef	HWPMC_HOOKS
780 	/*
781 	 * Check if system-wide sampling is in effect or if the
782 	 * current process is using PMCs.  If so, do exec() time
783 	 * processing.  This processing needs to happen AFTER the
784 	 * P_INEXEC flag is cleared.
785 	 *
786 	 * The proc lock needs to be released before taking the PMC
787 	 * SX.
788 	 */
789 	if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
790 		PROC_UNLOCK(p);
791 		pe.pm_credentialschanged = credential_changing;
792 		pe.pm_entryaddr = imgp->entry_addr;
793 
794 		PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
795 	} else
796 		PROC_UNLOCK(p);
797 #else  /* !HWPMC_HOOKS */
798 	PROC_UNLOCK(p);
799 #endif
800 
801 	/* Set values passed into the program in registers. */
802 	if (p->p_sysent->sv_setregs)
803 		(*p->p_sysent->sv_setregs)(td, imgp->entry_addr,
804 		    (u_long)(uintptr_t)stack_base, imgp->ps_strings);
805 	else
806 		exec_setregs(td, imgp->entry_addr,
807 		    (u_long)(uintptr_t)stack_base, imgp->ps_strings);
808 
809 	vfs_mark_atime(imgp->vp, td->td_ucred);
810 
811 	SDT_PROBE(proc, kernel, , exec_success, args->fname, 0, 0, 0, 0);
812 
813 done1:
814 	/*
815 	 * Free any resources malloc'd earlier that we didn't use.
816 	 */
817 	uifree(euip);
818 	if (newcred == NULL)
819 		crfree(oldcred);
820 	else
821 		crfree(newcred);
822 	VOP_UNLOCK(imgp->vp, 0);
823 
824 	/*
825 	 * Handle deferred decrement of ref counts.
826 	 */
827 	if (textvp != NULL) {
828 		int tvfslocked;
829 
830 		tvfslocked = VFS_LOCK_GIANT(textvp->v_mount);
831 		vrele(textvp);
832 		VFS_UNLOCK_GIANT(tvfslocked);
833 	}
834 	if (binvp && error != 0)
835 		vrele(binvp);
836 #ifdef KTRACE
837 	if (tracevp != NULL) {
838 		int tvfslocked;
839 
840 		tvfslocked = VFS_LOCK_GIANT(tracevp->v_mount);
841 		vrele(tracevp);
842 		VFS_UNLOCK_GIANT(tvfslocked);
843 	}
844 	if (tracecred != NULL)
845 		crfree(tracecred);
846 #endif
847 	vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
848 	pargs_drop(oldargs);
849 	pargs_drop(newargs);
850 	if (oldsigacts != NULL)
851 		sigacts_free(oldsigacts);
852 
853 exec_fail_dealloc:
854 
855 	/*
856 	 * free various allocated resources
857 	 */
858 	if (imgp->firstpage != NULL)
859 		exec_unmap_first_page(imgp);
860 
861 	if (imgp->vp != NULL) {
862 		if (args->fname)
863 			NDFREE(&nd, NDF_ONLY_PNBUF);
864 		if (imgp->opened)
865 			VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
866 		vput(imgp->vp);
867 	}
868 
869 	if (imgp->object != NULL)
870 		vm_object_deallocate(imgp->object);
871 
872 	free(imgp->freepath, M_TEMP);
873 
874 	if (error == 0) {
875 		/*
876 		 * Stop the process here if its stop event mask has
877 		 * the S_EXEC bit set.
878 		 */
879 		STOPEVENT(p, S_EXEC, 0);
880 		goto done2;
881 	}
882 
883 exec_fail:
884 	/* we're done here, clear P_INEXEC */
885 	PROC_LOCK(p);
886 	p->p_flag &= ~P_INEXEC;
887 	PROC_UNLOCK(p);
888 
889 	SDT_PROBE(proc, kernel, , exec_failure, error, 0, 0, 0, 0);
890 
891 done2:
892 #ifdef MAC
893 	mac_execve_exit(imgp);
894 	mac_execve_interpreter_exit(interpvplabel);
895 #endif
896 	VFS_UNLOCK_GIANT(vfslocked);
897 	exec_free_args(args);
898 
899 	if (error && imgp->vmspace_destroyed) {
900 		/* sorry, no more process anymore. exit gracefully */
901 		exit1(td, W_EXITCODE(0, SIGABRT));
902 		/* NOT REACHED */
903 	}
904 	return (error);
905 }
906 
907 int
908 exec_map_first_page(imgp)
909 	struct image_params *imgp;
910 {
911 	int rv, i;
912 	int initial_pagein;
913 	vm_page_t ma[VM_INITIAL_PAGEIN];
914 	vm_object_t object;
915 
916 	if (imgp->firstpage != NULL)
917 		exec_unmap_first_page(imgp);
918 
919 	object = imgp->vp->v_object;
920 	if (object == NULL)
921 		return (EACCES);
922 	VM_OBJECT_LOCK(object);
923 #if VM_NRESERVLEVEL > 0
924 	if ((object->flags & OBJ_COLORED) == 0) {
925 		object->flags |= OBJ_COLORED;
926 		object->pg_color = 0;
927 	}
928 #endif
929 	ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
930 	if ((ma[0]->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) {
931 		initial_pagein = VM_INITIAL_PAGEIN;
932 		if (initial_pagein > object->size)
933 			initial_pagein = object->size;
934 		for (i = 1; i < initial_pagein; i++) {
935 			if ((ma[i] = vm_page_lookup(object, i)) != NULL) {
936 				if (ma[i]->valid)
937 					break;
938 				if ((ma[i]->oflags & VPO_BUSY) || ma[i]->busy)
939 					break;
940 				vm_page_busy(ma[i]);
941 			} else {
942 				ma[i] = vm_page_alloc(object, i,
943 				    VM_ALLOC_NORMAL | VM_ALLOC_IFNOTCACHED);
944 				if (ma[i] == NULL)
945 					break;
946 			}
947 		}
948 		initial_pagein = i;
949 		rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
950 		ma[0] = vm_page_lookup(object, 0);
951 		if ((rv != VM_PAGER_OK) || (ma[0] == NULL) ||
952 		    (ma[0]->valid == 0)) {
953 			if (ma[0]) {
954 				vm_page_lock_queues();
955 				vm_page_free(ma[0]);
956 				vm_page_unlock_queues();
957 			}
958 			VM_OBJECT_UNLOCK(object);
959 			return (EIO);
960 		}
961 	}
962 	vm_page_lock_queues();
963 	vm_page_hold(ma[0]);
964 	vm_page_unlock_queues();
965 	vm_page_wakeup(ma[0]);
966 	VM_OBJECT_UNLOCK(object);
967 
968 	imgp->firstpage = sf_buf_alloc(ma[0], 0);
969 	imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
970 
971 	return (0);
972 }
973 
974 void
975 exec_unmap_first_page(imgp)
976 	struct image_params *imgp;
977 {
978 	vm_page_t m;
979 
980 	if (imgp->firstpage != NULL) {
981 		m = sf_buf_page(imgp->firstpage);
982 		sf_buf_free(imgp->firstpage);
983 		imgp->firstpage = NULL;
984 		vm_page_lock_queues();
985 		vm_page_unhold(m);
986 		vm_page_unlock_queues();
987 	}
988 }
989 
990 /*
991  * Destroy old address space, and allocate a new stack
992  *	The new stack is only SGROWSIZ large because it is grown
993  *	automatically in trap.c.
994  */
995 int
996 exec_new_vmspace(imgp, sv)
997 	struct image_params *imgp;
998 	struct sysentvec *sv;
999 {
1000 	int error;
1001 	struct proc *p = imgp->proc;
1002 	struct vmspace *vmspace = p->p_vmspace;
1003 	vm_offset_t stack_addr;
1004 	vm_map_t map;
1005 	u_long ssiz;
1006 
1007 	imgp->vmspace_destroyed = 1;
1008 	imgp->sysent = sv;
1009 
1010 	/* May be called with Giant held */
1011 	EVENTHANDLER_INVOKE(process_exec, p, imgp);
1012 
1013 	/*
1014 	 * Blow away entire process VM, if address space not shared,
1015 	 * otherwise, create a new VM space so that other threads are
1016 	 * not disrupted
1017 	 */
1018 	map = &vmspace->vm_map;
1019 	if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv->sv_minuser &&
1020 	    vm_map_max(map) == sv->sv_maxuser) {
1021 		shmexit(vmspace);
1022 		pmap_remove_pages(vmspace_pmap(vmspace));
1023 		vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1024 	} else {
1025 		error = vmspace_exec(p, sv->sv_minuser, sv->sv_maxuser);
1026 		if (error)
1027 			return (error);
1028 		vmspace = p->p_vmspace;
1029 		map = &vmspace->vm_map;
1030 	}
1031 
1032 	/* Allocate a new stack */
1033 	if (sv->sv_maxssiz != NULL)
1034 		ssiz = *sv->sv_maxssiz;
1035 	else
1036 		ssiz = maxssiz;
1037 	stack_addr = sv->sv_usrstack - ssiz;
1038 	error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1039 	    sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1040 	if (error)
1041 		return (error);
1042 
1043 #ifdef __ia64__
1044 	/* Allocate a new register stack */
1045 	stack_addr = IA64_BACKINGSTORE;
1046 	error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1047 	    sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_UP);
1048 	if (error)
1049 		return (error);
1050 #endif
1051 
1052 	/* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
1053 	 * VM_STACK case, but they are still used to monitor the size of the
1054 	 * process stack so we can check the stack rlimit.
1055 	 */
1056 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1057 	vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - ssiz;
1058 
1059 	return (0);
1060 }
1061 
1062 /*
1063  * Copy out argument and environment strings from the old process address
1064  * space into the temporary string buffer.
1065  */
1066 int
1067 exec_copyin_args(struct image_args *args, char *fname,
1068     enum uio_seg segflg, char **argv, char **envv)
1069 {
1070 	char *argp, *envp;
1071 	int error;
1072 	size_t length;
1073 
1074 	bzero(args, sizeof(*args));
1075 	if (argv == NULL)
1076 		return (EFAULT);
1077 	/*
1078 	 * Allocate temporary demand zeroed space for argument and
1079 	 *	environment strings:
1080 	 *
1081 	 * o ARG_MAX for argument and environment;
1082 	 * o MAXSHELLCMDLEN for the name of interpreters.
1083 	 */
1084 	args->buf = (char *) kmem_alloc_wait(exec_map,
1085 	    PATH_MAX + ARG_MAX + MAXSHELLCMDLEN);
1086 	if (args->buf == NULL)
1087 		return (ENOMEM);
1088 	args->begin_argv = args->buf;
1089 	args->endp = args->begin_argv;
1090 	args->stringspace = ARG_MAX;
1091 	/*
1092 	 * Copy the file name.
1093 	 */
1094 	if (fname != NULL) {
1095 		args->fname = args->buf + ARG_MAX;
1096 		error = (segflg == UIO_SYSSPACE) ?
1097 		    copystr(fname, args->fname, PATH_MAX, &length) :
1098 		    copyinstr(fname, args->fname, PATH_MAX, &length);
1099 		if (error != 0)
1100 			goto err_exit;
1101 	} else
1102 		args->fname = NULL;
1103 
1104 	/*
1105 	 * extract arguments first
1106 	 */
1107 	while ((argp = (caddr_t) (intptr_t) fuword(argv++))) {
1108 		if (argp == (caddr_t) -1) {
1109 			error = EFAULT;
1110 			goto err_exit;
1111 		}
1112 		if ((error = copyinstr(argp, args->endp,
1113 		    args->stringspace, &length))) {
1114 			if (error == ENAMETOOLONG)
1115 				error = E2BIG;
1116 			goto err_exit;
1117 		}
1118 		args->stringspace -= length;
1119 		args->endp += length;
1120 		args->argc++;
1121 	}
1122 
1123 	args->begin_envv = args->endp;
1124 
1125 	/*
1126 	 * extract environment strings
1127 	 */
1128 	if (envv) {
1129 		while ((envp = (caddr_t)(intptr_t)fuword(envv++))) {
1130 			if (envp == (caddr_t)-1) {
1131 				error = EFAULT;
1132 				goto err_exit;
1133 			}
1134 			if ((error = copyinstr(envp, args->endp,
1135 			    args->stringspace, &length))) {
1136 				if (error == ENAMETOOLONG)
1137 					error = E2BIG;
1138 				goto err_exit;
1139 			}
1140 			args->stringspace -= length;
1141 			args->endp += length;
1142 			args->envc++;
1143 		}
1144 	}
1145 
1146 	return (0);
1147 
1148 err_exit:
1149 	exec_free_args(args);
1150 	return (error);
1151 }
1152 
1153 static void
1154 exec_free_args(struct image_args *args)
1155 {
1156 
1157 	if (args->buf) {
1158 		kmem_free_wakeup(exec_map, (vm_offset_t)args->buf,
1159 		    PATH_MAX + ARG_MAX + MAXSHELLCMDLEN);
1160 		args->buf = NULL;
1161 	}
1162 }
1163 
1164 /*
1165  * Copy strings out to the new process address space, constructing new arg
1166  * and env vector tables. Return a pointer to the base so that it can be used
1167  * as the initial stack pointer.
1168  */
1169 register_t *
1170 exec_copyout_strings(imgp)
1171 	struct image_params *imgp;
1172 {
1173 	int argc, envc;
1174 	char **vectp;
1175 	char *stringp, *destp;
1176 	register_t *stack_base;
1177 	struct ps_strings *arginfo;
1178 	struct proc *p;
1179 	size_t execpath_len;
1180 	int szsigcode;
1181 
1182 	/*
1183 	 * Calculate string base and vector table pointers.
1184 	 * Also deal with signal trampoline code for this exec type.
1185 	 */
1186 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
1187 		execpath_len = strlen(imgp->execpath) + 1;
1188 	else
1189 		execpath_len = 0;
1190 	p = imgp->proc;
1191 	szsigcode = 0;
1192 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1193 	if (p->p_sysent->sv_szsigcode != NULL)
1194 		szsigcode = *(p->p_sysent->sv_szsigcode);
1195 	destp =	(caddr_t)arginfo - szsigcode - SPARE_USRSPACE -
1196 	    roundup(execpath_len, sizeof(char *)) -
1197 	    roundup((ARG_MAX - imgp->args->stringspace), sizeof(char *));
1198 
1199 	/*
1200 	 * install sigcode
1201 	 */
1202 	if (szsigcode)
1203 		copyout(p->p_sysent->sv_sigcode, ((caddr_t)arginfo -
1204 		    szsigcode), szsigcode);
1205 
1206 	/*
1207 	 * Copy the image path for the rtld.
1208 	 */
1209 	if (execpath_len != 0) {
1210 		imgp->execpathp = (uintptr_t)arginfo - szsigcode - execpath_len;
1211 		copyout(imgp->execpath, (void *)imgp->execpathp,
1212 		    execpath_len);
1213 	}
1214 
1215 	/*
1216 	 * If we have a valid auxargs ptr, prepare some room
1217 	 * on the stack.
1218 	 */
1219 	if (imgp->auxargs) {
1220 		/*
1221 		 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1222 		 * lower compatibility.
1223 		 */
1224 		imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1225 		    (AT_COUNT * 2);
1226 		/*
1227 		 * The '+ 2' is for the null pointers at the end of each of
1228 		 * the arg and env vector sets,and imgp->auxarg_size is room
1229 		 * for argument of Runtime loader.
1230 		 */
1231 		vectp = (char **)(destp - (imgp->args->argc +
1232 		    imgp->args->envc + 2 + imgp->auxarg_size + execpath_len) *
1233 		    sizeof(char *));
1234 	} else {
1235 		/*
1236 		 * The '+ 2' is for the null pointers at the end of each of
1237 		 * the arg and env vector sets
1238 		 */
1239 		vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc + 2) *
1240 		    sizeof(char *));
1241 	}
1242 
1243 	/*
1244 	 * vectp also becomes our initial stack base
1245 	 */
1246 	stack_base = (register_t *)vectp;
1247 
1248 	stringp = imgp->args->begin_argv;
1249 	argc = imgp->args->argc;
1250 	envc = imgp->args->envc;
1251 
1252 	/*
1253 	 * Copy out strings - arguments and environment.
1254 	 */
1255 	copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
1256 
1257 	/*
1258 	 * Fill in "ps_strings" struct for ps, w, etc.
1259 	 */
1260 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1261 	suword(&arginfo->ps_nargvstr, argc);
1262 
1263 	/*
1264 	 * Fill in argument portion of vector table.
1265 	 */
1266 	for (; argc > 0; --argc) {
1267 		suword(vectp++, (long)(intptr_t)destp);
1268 		while (*stringp++ != 0)
1269 			destp++;
1270 		destp++;
1271 	}
1272 
1273 	/* a null vector table pointer separates the argp's from the envp's */
1274 	suword(vectp++, 0);
1275 
1276 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1277 	suword(&arginfo->ps_nenvstr, envc);
1278 
1279 	/*
1280 	 * Fill in environment portion of vector table.
1281 	 */
1282 	for (; envc > 0; --envc) {
1283 		suword(vectp++, (long)(intptr_t)destp);
1284 		while (*stringp++ != 0)
1285 			destp++;
1286 		destp++;
1287 	}
1288 
1289 	/* end of vector table is a null pointer */
1290 	suword(vectp, 0);
1291 
1292 	return (stack_base);
1293 }
1294 
1295 /*
1296  * Check permissions of file to execute.
1297  *	Called with imgp->vp locked.
1298  *	Return 0 for success or error code on failure.
1299  */
1300 int
1301 exec_check_permissions(imgp)
1302 	struct image_params *imgp;
1303 {
1304 	struct vnode *vp = imgp->vp;
1305 	struct vattr *attr = imgp->attr;
1306 	struct thread *td;
1307 	int error;
1308 
1309 	td = curthread;
1310 
1311 	/* Get file attributes */
1312 	error = VOP_GETATTR(vp, attr, td->td_ucred);
1313 	if (error)
1314 		return (error);
1315 
1316 #ifdef MAC
1317 	error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1318 	if (error)
1319 		return (error);
1320 #endif
1321 
1322 	/*
1323 	 * 1) Check if file execution is disabled for the filesystem that this
1324 	 *	file resides on.
1325 	 * 2) Insure that at least one execute bit is on - otherwise root
1326 	 *	will always succeed, and we don't want to happen unless the
1327 	 *	file really is executable.
1328 	 * 3) Insure that the file is a regular file.
1329 	 */
1330 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1331 	    ((attr->va_mode & 0111) == 0) ||
1332 	    (attr->va_type != VREG))
1333 		return (EACCES);
1334 
1335 	/*
1336 	 * Zero length files can't be exec'd
1337 	 */
1338 	if (attr->va_size == 0)
1339 		return (ENOEXEC);
1340 
1341 	/*
1342 	 *  Check for execute permission to file based on current credentials.
1343 	 */
1344 	error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1345 	if (error)
1346 		return (error);
1347 
1348 	/*
1349 	 * Check number of open-for-writes on the file and deny execution
1350 	 * if there are any.
1351 	 */
1352 	if (vp->v_writecount)
1353 		return (ETXTBSY);
1354 
1355 	/*
1356 	 * Call filesystem specific open routine (which does nothing in the
1357 	 * general case).
1358 	 */
1359 	error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1360 	if (error == 0)
1361 		imgp->opened = 1;
1362 	return (error);
1363 }
1364 
1365 /*
1366  * Exec handler registration
1367  */
1368 int
1369 exec_register(execsw_arg)
1370 	const struct execsw *execsw_arg;
1371 {
1372 	const struct execsw **es, **xs, **newexecsw;
1373 	int count = 2;	/* New slot and trailing NULL */
1374 
1375 	if (execsw)
1376 		for (es = execsw; *es; es++)
1377 			count++;
1378 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1379 	if (newexecsw == NULL)
1380 		return (ENOMEM);
1381 	xs = newexecsw;
1382 	if (execsw)
1383 		for (es = execsw; *es; es++)
1384 			*xs++ = *es;
1385 	*xs++ = execsw_arg;
1386 	*xs = NULL;
1387 	if (execsw)
1388 		free(execsw, M_TEMP);
1389 	execsw = newexecsw;
1390 	return (0);
1391 }
1392 
1393 int
1394 exec_unregister(execsw_arg)
1395 	const struct execsw *execsw_arg;
1396 {
1397 	const struct execsw **es, **xs, **newexecsw;
1398 	int count = 1;
1399 
1400 	if (execsw == NULL)
1401 		panic("unregister with no handlers left?\n");
1402 
1403 	for (es = execsw; *es; es++) {
1404 		if (*es == execsw_arg)
1405 			break;
1406 	}
1407 	if (*es == NULL)
1408 		return (ENOENT);
1409 	for (es = execsw; *es; es++)
1410 		if (*es != execsw_arg)
1411 			count++;
1412 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1413 	if (newexecsw == NULL)
1414 		return (ENOMEM);
1415 	xs = newexecsw;
1416 	for (es = execsw; *es; es++)
1417 		if (*es != execsw_arg)
1418 			*xs++ = *es;
1419 	*xs = NULL;
1420 	if (execsw)
1421 		free(execsw, M_TEMP);
1422 	execsw = newexecsw;
1423 	return (0);
1424 }
1425