xref: /freebsd/sys/kern/kern_exec.c (revision c0020399a650364d0134f79f3fa319f84064372d)
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_mac.h"
34 #include "opt_vm.h"
35 
36 #include <sys/param.h>
37 #include <sys/systm.h>
38 #include <sys/eventhandler.h>
39 #include <sys/lock.h>
40 #include <sys/mutex.h>
41 #include <sys/sysproto.h>
42 #include <sys/signalvar.h>
43 #include <sys/kernel.h>
44 #include <sys/mount.h>
45 #include <sys/filedesc.h>
46 #include <sys/fcntl.h>
47 #include <sys/acct.h>
48 #include <sys/exec.h>
49 #include <sys/imgact.h>
50 #include <sys/imgact_elf.h>
51 #include <sys/wait.h>
52 #include <sys/malloc.h>
53 #include <sys/priv.h>
54 #include <sys/proc.h>
55 #include <sys/pioctl.h>
56 #include <sys/namei.h>
57 #include <sys/resourcevar.h>
58 #include <sys/sdt.h>
59 #include <sys/sf_buf.h>
60 #include <sys/syscallsubr.h>
61 #include <sys/sysent.h>
62 #include <sys/shm.h>
63 #include <sys/sysctl.h>
64 #include <sys/vnode.h>
65 #include <sys/stat.h>
66 #ifdef KTRACE
67 #include <sys/ktrace.h>
68 #endif
69 
70 #include <vm/vm.h>
71 #include <vm/vm_param.h>
72 #include <vm/pmap.h>
73 #include <vm/vm_page.h>
74 #include <vm/vm_map.h>
75 #include <vm/vm_kern.h>
76 #include <vm/vm_extern.h>
77 #include <vm/vm_object.h>
78 #include <vm/vm_pager.h>
79 
80 #ifdef	HWPMC_HOOKS
81 #include <sys/pmckern.h>
82 #endif
83 
84 #include <machine/reg.h>
85 
86 #include <security/audit/audit.h>
87 #include <security/mac/mac_framework.h>
88 
89 #ifdef KDTRACE_HOOKS
90 #include <sys/dtrace_bsd.h>
91 dtrace_execexit_func_t	dtrace_fasttrap_exec;
92 #endif
93 
94 SDT_PROVIDER_DECLARE(proc);
95 SDT_PROBE_DEFINE(proc, kernel, , exec);
96 SDT_PROBE_ARGTYPE(proc, kernel, , exec, 0, "char *");
97 SDT_PROBE_DEFINE(proc, kernel, , exec_failure);
98 SDT_PROBE_ARGTYPE(proc, kernel, , exec_failure, 0, "int");
99 SDT_PROBE_DEFINE(proc, kernel, , exec_success);
100 SDT_PROBE_ARGTYPE(proc, kernel, , exec_success, 0, "char *");
101 
102 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
103 
104 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
105 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
106 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
107 static int do_execve(struct thread *td, struct image_args *args,
108     struct mac *mac_p);
109 static void exec_free_args(struct image_args *);
110 
111 /* XXX This should be vm_size_t. */
112 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD,
113     NULL, 0, sysctl_kern_ps_strings, "LU", "");
114 
115 /* XXX This should be vm_size_t. */
116 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD,
117     NULL, 0, sysctl_kern_usrstack, "LU", "");
118 
119 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD,
120     NULL, 0, sysctl_kern_stackprot, "I", "");
121 
122 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
123 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
124     &ps_arg_cache_limit, 0, "");
125 
126 static int
127 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
128 {
129 	struct proc *p;
130 	int error;
131 
132 	p = curproc;
133 #ifdef SCTL_MASK32
134 	if (req->flags & SCTL_MASK32) {
135 		unsigned int val;
136 		val = (unsigned int)p->p_sysent->sv_psstrings;
137 		error = SYSCTL_OUT(req, &val, sizeof(val));
138 	} else
139 #endif
140 		error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
141 		   sizeof(p->p_sysent->sv_psstrings));
142 	return error;
143 }
144 
145 static int
146 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
147 {
148 	struct proc *p;
149 	int error;
150 
151 	p = curproc;
152 #ifdef SCTL_MASK32
153 	if (req->flags & SCTL_MASK32) {
154 		unsigned int val;
155 		val = (unsigned int)p->p_sysent->sv_usrstack;
156 		error = SYSCTL_OUT(req, &val, sizeof(val));
157 	} else
158 #endif
159 		error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
160 		    sizeof(p->p_sysent->sv_usrstack));
161 	return error;
162 }
163 
164 static int
165 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
166 {
167 	struct proc *p;
168 
169 	p = curproc;
170 	return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
171 	    sizeof(p->p_sysent->sv_stackprot)));
172 }
173 
174 /*
175  * Each of the items is a pointer to a `const struct execsw', hence the
176  * double pointer here.
177  */
178 static const struct execsw **execsw;
179 
180 #ifndef _SYS_SYSPROTO_H_
181 struct execve_args {
182 	char    *fname;
183 	char    **argv;
184 	char    **envv;
185 };
186 #endif
187 
188 int
189 execve(td, uap)
190 	struct thread *td;
191 	struct execve_args /* {
192 		char *fname;
193 		char **argv;
194 		char **envv;
195 	} */ *uap;
196 {
197 	int error;
198 	struct image_args args;
199 
200 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
201 	    uap->argv, uap->envv);
202 	if (error == 0)
203 		error = kern_execve(td, &args, NULL);
204 	return (error);
205 }
206 
207 #ifndef _SYS_SYSPROTO_H_
208 struct fexecve_args {
209 	int	fd;
210 	char	**argv;
211 	char	**envv;
212 }
213 #endif
214 int
215 fexecve(struct thread *td, struct fexecve_args *uap)
216 {
217 	int error;
218 	struct image_args args;
219 
220 	error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
221 	    uap->argv, uap->envv);
222 	if (error == 0) {
223 		args.fd = uap->fd;
224 		error = kern_execve(td, &args, NULL);
225 	}
226 	return (error);
227 }
228 
229 #ifndef _SYS_SYSPROTO_H_
230 struct __mac_execve_args {
231 	char	*fname;
232 	char	**argv;
233 	char	**envv;
234 	struct mac	*mac_p;
235 };
236 #endif
237 
238 int
239 __mac_execve(td, uap)
240 	struct thread *td;
241 	struct __mac_execve_args /* {
242 		char *fname;
243 		char **argv;
244 		char **envv;
245 		struct mac *mac_p;
246 	} */ *uap;
247 {
248 #ifdef MAC
249 	int error;
250 	struct image_args args;
251 
252 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
253 	    uap->argv, uap->envv);
254 	if (error == 0)
255 		error = kern_execve(td, &args, uap->mac_p);
256 	return (error);
257 #else
258 	return (ENOSYS);
259 #endif
260 }
261 
262 /*
263  * XXX: kern_execve has the astonishing property of not always returning to
264  * the caller.  If sufficiently bad things happen during the call to
265  * do_execve(), it can end up calling exit1(); as a result, callers must
266  * avoid doing anything which they might need to undo (e.g., allocating
267  * memory).
268  */
269 int
270 kern_execve(td, args, mac_p)
271 	struct thread *td;
272 	struct image_args *args;
273 	struct mac *mac_p;
274 {
275 	struct proc *p = td->td_proc;
276 	int error;
277 
278 	AUDIT_ARG(argv, args->begin_argv, args->argc,
279 	    args->begin_envv - args->begin_argv);
280 	AUDIT_ARG(envv, args->begin_envv, args->envc,
281 	    args->endp - args->begin_envv);
282 	if (p->p_flag & P_HADTHREADS) {
283 		PROC_LOCK(p);
284 		if (thread_single(SINGLE_BOUNDARY)) {
285 			PROC_UNLOCK(p);
286 	       		exec_free_args(args);
287 			return (ERESTART);	/* Try again later. */
288 		}
289 		PROC_UNLOCK(p);
290 	}
291 
292 	error = do_execve(td, args, mac_p);
293 
294 	if (p->p_flag & P_HADTHREADS) {
295 		PROC_LOCK(p);
296 		/*
297 		 * If success, we upgrade to SINGLE_EXIT state to
298 		 * force other threads to suicide.
299 		 */
300 		if (error == 0)
301 			thread_single(SINGLE_EXIT);
302 		else
303 			thread_single_end();
304 		PROC_UNLOCK(p);
305 	}
306 
307 	return (error);
308 }
309 
310 /*
311  * In-kernel implementation of execve().  All arguments are assumed to be
312  * userspace pointers from the passed thread.
313  */
314 static int
315 do_execve(td, args, mac_p)
316 	struct thread *td;
317 	struct image_args *args;
318 	struct mac *mac_p;
319 {
320 	struct proc *p = td->td_proc;
321 	struct nameidata nd;
322 	struct ucred *newcred = NULL, *oldcred;
323 	struct uidinfo *euip;
324 	register_t *stack_base;
325 	int error, len = 0, i;
326 	struct image_params image_params, *imgp;
327 	struct vattr attr;
328 	int (*img_first)(struct image_params *);
329 	struct pargs *oldargs = NULL, *newargs = NULL;
330 	struct sigacts *oldsigacts, *newsigacts;
331 #ifdef KTRACE
332 	struct vnode *tracevp = NULL;
333 	struct ucred *tracecred = NULL;
334 #endif
335 	struct vnode *textvp = NULL, *binvp = NULL;
336 	int credential_changing;
337 	int vfslocked;
338 	int textset;
339 #ifdef MAC
340 	struct label *interpvplabel = NULL;
341 	int will_transition;
342 #endif
343 #ifdef HWPMC_HOOKS
344 	struct pmckern_procexec pe;
345 #endif
346 	static const char fexecv_proc_title[] = "(fexecv)";
347 
348 	vfslocked = 0;
349 	imgp = &image_params;
350 
351 	/*
352 	 * Lock the process and set the P_INEXEC flag to indicate that
353 	 * it should be left alone until we're done here.  This is
354 	 * necessary to avoid race conditions - e.g. in ptrace() -
355 	 * that might allow a local user to illicitly obtain elevated
356 	 * privileges.
357 	 */
358 	PROC_LOCK(p);
359 	KASSERT((p->p_flag & P_INEXEC) == 0,
360 	    ("%s(): process already has P_INEXEC flag", __func__));
361 	p->p_flag |= P_INEXEC;
362 	PROC_UNLOCK(p);
363 
364 	/*
365 	 * Initialize part of the common data
366 	 */
367 	imgp->proc = p;
368 	imgp->execlabel = NULL;
369 	imgp->attr = &attr;
370 	imgp->entry_addr = 0;
371 	imgp->vmspace_destroyed = 0;
372 	imgp->interpreted = 0;
373 	imgp->opened = 0;
374 	imgp->interpreter_name = args->buf + PATH_MAX + ARG_MAX;
375 	imgp->auxargs = NULL;
376 	imgp->vp = NULL;
377 	imgp->object = NULL;
378 	imgp->firstpage = NULL;
379 	imgp->ps_strings = 0;
380 	imgp->auxarg_size = 0;
381 	imgp->args = args;
382 	imgp->execpath = imgp->freepath = NULL;
383 	imgp->execpathp = 0;
384 
385 #ifdef MAC
386 	error = mac_execve_enter(imgp, mac_p);
387 	if (error)
388 		goto exec_fail;
389 #endif
390 
391 	imgp->image_header = NULL;
392 
393 	/*
394 	 * Translate the file name. namei() returns a vnode pointer
395 	 *	in ni_vp amoung other things.
396 	 *
397 	 * XXXAUDIT: It would be desirable to also audit the name of the
398 	 * interpreter if this is an interpreted binary.
399 	 */
400 	if (args->fname != NULL) {
401 		NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME
402 		    | MPSAFE | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
403 	}
404 
405 	SDT_PROBE(proc, kernel, , exec, args->fname, 0, 0, 0, 0 );
406 
407 interpret:
408 	if (args->fname != NULL) {
409 		error = namei(&nd);
410 		if (error)
411 			goto exec_fail;
412 
413 		vfslocked = NDHASGIANT(&nd);
414 		binvp  = nd.ni_vp;
415 		imgp->vp = binvp;
416 	} else {
417 		AUDIT_ARG(fd, args->fd);
418 		error = fgetvp(td, args->fd, &binvp);
419 		if (error)
420 			goto exec_fail;
421 		vfslocked = VFS_LOCK_GIANT(binvp->v_mount);
422 		vn_lock(binvp, LK_EXCLUSIVE | LK_RETRY);
423 		AUDIT_ARG(vnode, binvp, ARG_VNODE1);
424 		imgp->vp = binvp;
425 	}
426 
427 	/*
428 	 * Check file permissions (also 'opens' file)
429 	 */
430 	error = exec_check_permissions(imgp);
431 	if (error)
432 		goto exec_fail_dealloc;
433 
434 	imgp->object = imgp->vp->v_object;
435 	if (imgp->object != NULL)
436 		vm_object_reference(imgp->object);
437 
438 	/*
439 	 * Set VV_TEXT now so no one can write to the executable while we're
440 	 * activating it.
441 	 *
442 	 * Remember if this was set before and unset it in case this is not
443 	 * actually an executable image.
444 	 */
445 	textset = imgp->vp->v_vflag & VV_TEXT;
446 	imgp->vp->v_vflag |= VV_TEXT;
447 
448 	error = exec_map_first_page(imgp);
449 	if (error)
450 		goto exec_fail_dealloc;
451 
452 	imgp->proc->p_osrel = 0;
453 	/*
454 	 *	If the current process has a special image activator it
455 	 *	wants to try first, call it.   For example, emulating shell
456 	 *	scripts differently.
457 	 */
458 	error = -1;
459 	if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
460 		error = img_first(imgp);
461 
462 	/*
463 	 *	Loop through the list of image activators, calling each one.
464 	 *	An activator returns -1 if there is no match, 0 on success,
465 	 *	and an error otherwise.
466 	 */
467 	for (i = 0; error == -1 && execsw[i]; ++i) {
468 		if (execsw[i]->ex_imgact == NULL ||
469 		    execsw[i]->ex_imgact == img_first) {
470 			continue;
471 		}
472 		error = (*execsw[i]->ex_imgact)(imgp);
473 	}
474 
475 	if (error) {
476 		if (error == -1) {
477 			if (textset == 0)
478 				imgp->vp->v_vflag &= ~VV_TEXT;
479 			error = ENOEXEC;
480 		}
481 		goto exec_fail_dealloc;
482 	}
483 
484 	/*
485 	 * Special interpreter operation, cleanup and loop up to try to
486 	 * activate the interpreter.
487 	 */
488 	if (imgp->interpreted) {
489 		exec_unmap_first_page(imgp);
490 		/*
491 		 * VV_TEXT needs to be unset for scripts.  There is a short
492 		 * period before we determine that something is a script where
493 		 * VV_TEXT will be set. The vnode lock is held over this
494 		 * entire period so nothing should illegitimately be blocked.
495 		 */
496 		imgp->vp->v_vflag &= ~VV_TEXT;
497 		/* free name buffer and old vnode */
498 		if (args->fname != NULL)
499 			NDFREE(&nd, NDF_ONLY_PNBUF);
500 #ifdef MAC
501 		mac_execve_interpreter_enter(binvp, &interpvplabel);
502 #endif
503 		if (imgp->opened) {
504 			VOP_CLOSE(binvp, FREAD, td->td_ucred, td);
505 			imgp->opened = 0;
506 		}
507 		vput(binvp);
508 		vm_object_deallocate(imgp->object);
509 		imgp->object = NULL;
510 		VFS_UNLOCK_GIANT(vfslocked);
511 		vfslocked = 0;
512 		/* set new name to that of the interpreter */
513 		NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME | MPSAFE,
514 		    UIO_SYSSPACE, imgp->interpreter_name, td);
515 		args->fname = imgp->interpreter_name;
516 		goto interpret;
517 	}
518 
519 	/*
520 	 * NB: We unlock the vnode here because it is believed that none
521 	 * of the sv_copyout_strings/sv_fixup operations require the vnode.
522 	 */
523 	VOP_UNLOCK(imgp->vp, 0);
524 
525 	/*
526 	 * Do the best to calculate the full path to the image file.
527 	 */
528 	if (imgp->auxargs != NULL &&
529 	    ((args->fname != NULL && args->fname[0] == '/') ||
530 	     vn_fullpath(td, imgp->vp, &imgp->execpath, &imgp->freepath) != 0))
531 		imgp->execpath = args->fname;
532 
533 	/*
534 	 * Copy out strings (args and env) and initialize stack base
535 	 */
536 	if (p->p_sysent->sv_copyout_strings)
537 		stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
538 	else
539 		stack_base = exec_copyout_strings(imgp);
540 
541 	/*
542 	 * If custom stack fixup routine present for this process
543 	 * let it do the stack setup.
544 	 * Else stuff argument count as first item on stack
545 	 */
546 	if (p->p_sysent->sv_fixup != NULL)
547 		(*p->p_sysent->sv_fixup)(&stack_base, imgp);
548 	else
549 		suword(--stack_base, imgp->args->argc);
550 
551 	/*
552 	 * For security and other reasons, the file descriptor table cannot
553 	 * be shared after an exec.
554 	 */
555 	fdunshare(p, td);
556 
557 	/*
558 	 * Malloc things before we need locks.
559 	 */
560 	newcred = crget();
561 	euip = uifind(attr.va_uid);
562 	i = imgp->args->begin_envv - imgp->args->begin_argv;
563 	/* Cache arguments if they fit inside our allowance */
564 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
565 		newargs = pargs_alloc(i);
566 		bcopy(imgp->args->begin_argv, newargs->ar_args, i);
567 	}
568 
569 	/* close files on exec */
570 	fdcloseexec(td);
571 	vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
572 
573 	/* Get a reference to the vnode prior to locking the proc */
574 	VREF(binvp);
575 
576 	/*
577 	 * For security and other reasons, signal handlers cannot
578 	 * be shared after an exec. The new process gets a copy of the old
579 	 * handlers. In execsigs(), the new process will have its signals
580 	 * reset.
581 	 */
582 	PROC_LOCK(p);
583 	if (sigacts_shared(p->p_sigacts)) {
584 		oldsigacts = p->p_sigacts;
585 		PROC_UNLOCK(p);
586 		newsigacts = sigacts_alloc();
587 		sigacts_copy(newsigacts, oldsigacts);
588 		PROC_LOCK(p);
589 		p->p_sigacts = newsigacts;
590 	} else
591 		oldsigacts = NULL;
592 
593 	/* Stop profiling */
594 	stopprofclock(p);
595 
596 	/* reset caught signals */
597 	execsigs(p);
598 
599 	/* name this process - nameiexec(p, ndp) */
600 	if (args->fname) {
601 		len = min(nd.ni_cnd.cn_namelen,MAXCOMLEN);
602 		bcopy(nd.ni_cnd.cn_nameptr, p->p_comm, len);
603 	} else {
604 		if (vn_commname(binvp, p->p_comm, MAXCOMLEN + 1) == 0)
605 			len = MAXCOMLEN;
606 		else {
607 			len = sizeof(fexecv_proc_title);
608 			bcopy(fexecv_proc_title, p->p_comm, len);
609 		}
610 	}
611 	p->p_comm[len] = 0;
612 	bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
613 
614 	/*
615 	 * mark as execed, wakeup the process that vforked (if any) and tell
616 	 * it that it now has its own resources back
617 	 */
618 	p->p_flag |= P_EXEC;
619 	if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
620 		p->p_flag &= ~P_PPWAIT;
621 		cv_broadcast(&p->p_pwait);
622 	}
623 
624 	/*
625 	 * Implement image setuid/setgid.
626 	 *
627 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
628 	 * the process is being traced.
629 	 *
630 	 * XXXMAC: For the time being, use NOSUID to also prohibit
631 	 * transitions on the file system.
632 	 */
633 	oldcred = p->p_ucred;
634 	credential_changing = 0;
635 	credential_changing |= (attr.va_mode & S_ISUID) && oldcred->cr_uid !=
636 	    attr.va_uid;
637 	credential_changing |= (attr.va_mode & S_ISGID) && oldcred->cr_gid !=
638 	    attr.va_gid;
639 #ifdef MAC
640 	will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
641 	    interpvplabel, imgp);
642 	credential_changing |= will_transition;
643 #endif
644 
645 	if (credential_changing &&
646 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
647 	    (p->p_flag & P_TRACED) == 0) {
648 		/*
649 		 * Turn off syscall tracing for set-id programs, except for
650 		 * root.  Record any set-id flags first to make sure that
651 		 * we do not regain any tracing during a possible block.
652 		 */
653 		setsugid(p);
654 
655 #ifdef KTRACE
656 		if (p->p_tracevp != NULL &&
657 		    priv_check_cred(oldcred, PRIV_DEBUG_DIFFCRED, 0)) {
658 			mtx_lock(&ktrace_mtx);
659 			p->p_traceflag = 0;
660 			tracevp = p->p_tracevp;
661 			p->p_tracevp = NULL;
662 			tracecred = p->p_tracecred;
663 			p->p_tracecred = NULL;
664 			mtx_unlock(&ktrace_mtx);
665 		}
666 #endif
667 		/*
668 		 * Close any file descriptors 0..2 that reference procfs,
669 		 * then make sure file descriptors 0..2 are in use.
670 		 *
671 		 * setugidsafety() may call closef() and then pfind()
672 		 * which may grab the process lock.
673 		 * fdcheckstd() may call falloc() which may block to
674 		 * allocate memory, so temporarily drop the process lock.
675 		 */
676 		PROC_UNLOCK(p);
677 		setugidsafety(td);
678 		VOP_UNLOCK(imgp->vp, 0);
679 		error = fdcheckstd(td);
680 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
681 		if (error != 0)
682 			goto done1;
683 		PROC_LOCK(p);
684 		/*
685 		 * Set the new credentials.
686 		 */
687 		crcopy(newcred, oldcred);
688 		if (attr.va_mode & S_ISUID)
689 			change_euid(newcred, euip);
690 		if (attr.va_mode & S_ISGID)
691 			change_egid(newcred, attr.va_gid);
692 #ifdef MAC
693 		if (will_transition) {
694 			mac_vnode_execve_transition(oldcred, newcred, imgp->vp,
695 			    interpvplabel, imgp);
696 		}
697 #endif
698 		/*
699 		 * Implement correct POSIX saved-id behavior.
700 		 *
701 		 * XXXMAC: Note that the current logic will save the
702 		 * uid and gid if a MAC domain transition occurs, even
703 		 * though maybe it shouldn't.
704 		 */
705 		change_svuid(newcred, newcred->cr_uid);
706 		change_svgid(newcred, newcred->cr_gid);
707 		p->p_ucred = newcred;
708 		newcred = NULL;
709 	} else {
710 		if (oldcred->cr_uid == oldcred->cr_ruid &&
711 		    oldcred->cr_gid == oldcred->cr_rgid)
712 			p->p_flag &= ~P_SUGID;
713 		/*
714 		 * Implement correct POSIX saved-id behavior.
715 		 *
716 		 * XXX: It's not clear that the existing behavior is
717 		 * POSIX-compliant.  A number of sources indicate that the
718 		 * saved uid/gid should only be updated if the new ruid is
719 		 * not equal to the old ruid, or the new euid is not equal
720 		 * to the old euid and the new euid is not equal to the old
721 		 * ruid.  The FreeBSD code always updates the saved uid/gid.
722 		 * Also, this code uses the new (replaced) euid and egid as
723 		 * the source, which may or may not be the right ones to use.
724 		 */
725 		if (oldcred->cr_svuid != oldcred->cr_uid ||
726 		    oldcred->cr_svgid != oldcred->cr_gid) {
727 			crcopy(newcred, oldcred);
728 			change_svuid(newcred, newcred->cr_uid);
729 			change_svgid(newcred, newcred->cr_gid);
730 			p->p_ucred = newcred;
731 			newcred = NULL;
732 		}
733 	}
734 
735 	/*
736 	 * Store the vp for use in procfs.  This vnode was referenced prior
737 	 * to locking the proc lock.
738 	 */
739 	textvp = p->p_textvp;
740 	p->p_textvp = binvp;
741 
742 #ifdef KDTRACE_HOOKS
743 	/*
744 	 * Tell the DTrace fasttrap provider about the exec if it
745 	 * has declared an interest.
746 	 */
747 	if (dtrace_fasttrap_exec)
748 		dtrace_fasttrap_exec(p);
749 #endif
750 
751 	/*
752 	 * Notify others that we exec'd, and clear the P_INEXEC flag
753 	 * as we're now a bona fide freshly-execed process.
754 	 */
755 	KNOTE_LOCKED(&p->p_klist, NOTE_EXEC);
756 	p->p_flag &= ~P_INEXEC;
757 
758 	/*
759 	 * If tracing the process, trap to debugger so breakpoints
760 	 * can be set before the program executes.
761 	 * Use tdsignal to deliver signal to current thread, use
762 	 * psignal may cause the signal to be delivered to wrong thread
763 	 * because that thread will exit, remember we are going to enter
764 	 * single thread mode.
765 	 */
766 	if (p->p_flag & P_TRACED)
767 		tdsignal(p, td, SIGTRAP, NULL);
768 
769 	/* clear "fork but no exec" flag, as we _are_ execing */
770 	p->p_acflag &= ~AFORK;
771 
772 	/*
773 	 * Free any previous argument cache and replace it with
774 	 * the new argument cache, if any.
775 	 */
776 	oldargs = p->p_args;
777 	p->p_args = newargs;
778 	newargs = NULL;
779 
780 #ifdef	HWPMC_HOOKS
781 	/*
782 	 * Check if system-wide sampling is in effect or if the
783 	 * current process is using PMCs.  If so, do exec() time
784 	 * processing.  This processing needs to happen AFTER the
785 	 * P_INEXEC flag is cleared.
786 	 *
787 	 * The proc lock needs to be released before taking the PMC
788 	 * SX.
789 	 */
790 	if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
791 		PROC_UNLOCK(p);
792 		pe.pm_credentialschanged = credential_changing;
793 		pe.pm_entryaddr = imgp->entry_addr;
794 
795 		PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
796 	} else
797 		PROC_UNLOCK(p);
798 #else  /* !HWPMC_HOOKS */
799 	PROC_UNLOCK(p);
800 #endif
801 
802 	/* Set values passed into the program in registers. */
803 	if (p->p_sysent->sv_setregs)
804 		(*p->p_sysent->sv_setregs)(td, imgp->entry_addr,
805 		    (u_long)(uintptr_t)stack_base, imgp->ps_strings);
806 	else
807 		exec_setregs(td, imgp->entry_addr,
808 		    (u_long)(uintptr_t)stack_base, imgp->ps_strings);
809 
810 	vfs_mark_atime(imgp->vp, td->td_ucred);
811 
812 	SDT_PROBE(proc, kernel, , exec_success, args->fname, 0, 0, 0, 0);
813 
814 done1:
815 	/*
816 	 * Free any resources malloc'd earlier that we didn't use.
817 	 */
818 	uifree(euip);
819 	if (newcred == NULL)
820 		crfree(oldcred);
821 	else
822 		crfree(newcred);
823 	VOP_UNLOCK(imgp->vp, 0);
824 
825 	/*
826 	 * Handle deferred decrement of ref counts.
827 	 */
828 	if (textvp != NULL) {
829 		int tvfslocked;
830 
831 		tvfslocked = VFS_LOCK_GIANT(textvp->v_mount);
832 		vrele(textvp);
833 		VFS_UNLOCK_GIANT(tvfslocked);
834 	}
835 	if (binvp && error != 0)
836 		vrele(binvp);
837 #ifdef KTRACE
838 	if (tracevp != NULL) {
839 		int tvfslocked;
840 
841 		tvfslocked = VFS_LOCK_GIANT(tracevp->v_mount);
842 		vrele(tracevp);
843 		VFS_UNLOCK_GIANT(tvfslocked);
844 	}
845 	if (tracecred != NULL)
846 		crfree(tracecred);
847 #endif
848 	vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
849 	pargs_drop(oldargs);
850 	pargs_drop(newargs);
851 	if (oldsigacts != NULL)
852 		sigacts_free(oldsigacts);
853 
854 exec_fail_dealloc:
855 
856 	/*
857 	 * free various allocated resources
858 	 */
859 	if (imgp->firstpage != NULL)
860 		exec_unmap_first_page(imgp);
861 
862 	if (imgp->vp != NULL) {
863 		if (args->fname)
864 			NDFREE(&nd, NDF_ONLY_PNBUF);
865 		if (imgp->opened)
866 			VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
867 		vput(imgp->vp);
868 	}
869 
870 	if (imgp->object != NULL)
871 		vm_object_deallocate(imgp->object);
872 
873 	free(imgp->freepath, M_TEMP);
874 
875 	if (error == 0) {
876 		/*
877 		 * Stop the process here if its stop event mask has
878 		 * the S_EXEC bit set.
879 		 */
880 		STOPEVENT(p, S_EXEC, 0);
881 		goto done2;
882 	}
883 
884 exec_fail:
885 	/* we're done here, clear P_INEXEC */
886 	PROC_LOCK(p);
887 	p->p_flag &= ~P_INEXEC;
888 	PROC_UNLOCK(p);
889 
890 	SDT_PROBE(proc, kernel, , exec_failure, error, 0, 0, 0, 0);
891 
892 done2:
893 #ifdef MAC
894 	mac_execve_exit(imgp);
895 	mac_execve_interpreter_exit(interpvplabel);
896 #endif
897 	VFS_UNLOCK_GIANT(vfslocked);
898 	exec_free_args(args);
899 
900 	if (error && imgp->vmspace_destroyed) {
901 		/* sorry, no more process anymore. exit gracefully */
902 		exit1(td, W_EXITCODE(0, SIGABRT));
903 		/* NOT REACHED */
904 	}
905 	return (error);
906 }
907 
908 int
909 exec_map_first_page(imgp)
910 	struct image_params *imgp;
911 {
912 	int rv, i;
913 	int initial_pagein;
914 	vm_page_t ma[VM_INITIAL_PAGEIN];
915 	vm_object_t object;
916 
917 	if (imgp->firstpage != NULL)
918 		exec_unmap_first_page(imgp);
919 
920 	object = imgp->vp->v_object;
921 	if (object == NULL)
922 		return (EACCES);
923 	VM_OBJECT_LOCK(object);
924 #if VM_NRESERVLEVEL > 0
925 	if ((object->flags & OBJ_COLORED) == 0) {
926 		object->flags |= OBJ_COLORED;
927 		object->pg_color = 0;
928 	}
929 #endif
930 	ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
931 	if ((ma[0]->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL) {
932 		initial_pagein = VM_INITIAL_PAGEIN;
933 		if (initial_pagein > object->size)
934 			initial_pagein = object->size;
935 		for (i = 1; i < initial_pagein; i++) {
936 			if ((ma[i] = vm_page_lookup(object, i)) != NULL) {
937 				if (ma[i]->valid)
938 					break;
939 				if ((ma[i]->oflags & VPO_BUSY) || ma[i]->busy)
940 					break;
941 				vm_page_busy(ma[i]);
942 			} else {
943 				ma[i] = vm_page_alloc(object, i,
944 				    VM_ALLOC_NORMAL | VM_ALLOC_IFNOTCACHED);
945 				if (ma[i] == NULL)
946 					break;
947 			}
948 		}
949 		initial_pagein = i;
950 		rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
951 		ma[0] = vm_page_lookup(object, 0);
952 		if ((rv != VM_PAGER_OK) || (ma[0] == NULL) ||
953 		    (ma[0]->valid == 0)) {
954 			if (ma[0]) {
955 				vm_page_lock_queues();
956 				vm_page_free(ma[0]);
957 				vm_page_unlock_queues();
958 			}
959 			VM_OBJECT_UNLOCK(object);
960 			return (EIO);
961 		}
962 	}
963 	vm_page_lock_queues();
964 	vm_page_hold(ma[0]);
965 	vm_page_unlock_queues();
966 	vm_page_wakeup(ma[0]);
967 	VM_OBJECT_UNLOCK(object);
968 
969 	imgp->firstpage = sf_buf_alloc(ma[0], 0);
970 	imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
971 
972 	return (0);
973 }
974 
975 void
976 exec_unmap_first_page(imgp)
977 	struct image_params *imgp;
978 {
979 	vm_page_t m;
980 
981 	if (imgp->firstpage != NULL) {
982 		m = sf_buf_page(imgp->firstpage);
983 		sf_buf_free(imgp->firstpage);
984 		imgp->firstpage = NULL;
985 		vm_page_lock_queues();
986 		vm_page_unhold(m);
987 		vm_page_unlock_queues();
988 	}
989 }
990 
991 /*
992  * Destroy old address space, and allocate a new stack
993  *	The new stack is only SGROWSIZ large because it is grown
994  *	automatically in trap.c.
995  */
996 int
997 exec_new_vmspace(imgp, sv)
998 	struct image_params *imgp;
999 	struct sysentvec *sv;
1000 {
1001 	int error;
1002 	struct proc *p = imgp->proc;
1003 	struct vmspace *vmspace = p->p_vmspace;
1004 	vm_offset_t stack_addr;
1005 	vm_map_t map;
1006 	u_long ssiz;
1007 
1008 	imgp->vmspace_destroyed = 1;
1009 	imgp->sysent = sv;
1010 
1011 	/* May be called with Giant held */
1012 	EVENTHANDLER_INVOKE(process_exec, p, imgp);
1013 
1014 	/*
1015 	 * Blow away entire process VM, if address space not shared,
1016 	 * otherwise, create a new VM space so that other threads are
1017 	 * not disrupted
1018 	 */
1019 	map = &vmspace->vm_map;
1020 	if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv->sv_minuser &&
1021 	    vm_map_max(map) == sv->sv_maxuser) {
1022 		shmexit(vmspace);
1023 		pmap_remove_pages(vmspace_pmap(vmspace));
1024 		vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1025 	} else {
1026 		error = vmspace_exec(p, sv->sv_minuser, sv->sv_maxuser);
1027 		if (error)
1028 			return (error);
1029 		vmspace = p->p_vmspace;
1030 		map = &vmspace->vm_map;
1031 	}
1032 
1033 	/* Allocate a new stack */
1034 	if (sv->sv_maxssiz != NULL)
1035 		ssiz = *sv->sv_maxssiz;
1036 	else
1037 		ssiz = maxssiz;
1038 	stack_addr = sv->sv_usrstack - ssiz;
1039 	error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1040 	    sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1041 	if (error)
1042 		return (error);
1043 
1044 #ifdef __ia64__
1045 	/* Allocate a new register stack */
1046 	stack_addr = IA64_BACKINGSTORE;
1047 	error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1048 	    sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_UP);
1049 	if (error)
1050 		return (error);
1051 #endif
1052 
1053 	/* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
1054 	 * VM_STACK case, but they are still used to monitor the size of the
1055 	 * process stack so we can check the stack rlimit.
1056 	 */
1057 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1058 	vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - ssiz;
1059 
1060 	return (0);
1061 }
1062 
1063 /*
1064  * Copy out argument and environment strings from the old process address
1065  * space into the temporary string buffer.
1066  */
1067 int
1068 exec_copyin_args(struct image_args *args, char *fname,
1069     enum uio_seg segflg, char **argv, char **envv)
1070 {
1071 	char *argp, *envp;
1072 	int error;
1073 	size_t length;
1074 
1075 	bzero(args, sizeof(*args));
1076 	if (argv == NULL)
1077 		return (EFAULT);
1078 	/*
1079 	 * Allocate temporary demand zeroed space for argument and
1080 	 *	environment strings:
1081 	 *
1082 	 * o ARG_MAX for argument and environment;
1083 	 * o MAXSHELLCMDLEN for the name of interpreters.
1084 	 */
1085 	args->buf = (char *) kmem_alloc_wait(exec_map,
1086 	    PATH_MAX + ARG_MAX + MAXSHELLCMDLEN);
1087 	if (args->buf == NULL)
1088 		return (ENOMEM);
1089 	args->begin_argv = args->buf;
1090 	args->endp = args->begin_argv;
1091 	args->stringspace = ARG_MAX;
1092 	/*
1093 	 * Copy the file name.
1094 	 */
1095 	if (fname != NULL) {
1096 		args->fname = args->buf + ARG_MAX;
1097 		error = (segflg == UIO_SYSSPACE) ?
1098 		    copystr(fname, args->fname, PATH_MAX, &length) :
1099 		    copyinstr(fname, args->fname, PATH_MAX, &length);
1100 		if (error != 0)
1101 			goto err_exit;
1102 	} else
1103 		args->fname = NULL;
1104 
1105 	/*
1106 	 * extract arguments first
1107 	 */
1108 	while ((argp = (caddr_t) (intptr_t) fuword(argv++))) {
1109 		if (argp == (caddr_t) -1) {
1110 			error = EFAULT;
1111 			goto err_exit;
1112 		}
1113 		if ((error = copyinstr(argp, args->endp,
1114 		    args->stringspace, &length))) {
1115 			if (error == ENAMETOOLONG)
1116 				error = E2BIG;
1117 			goto err_exit;
1118 		}
1119 		args->stringspace -= length;
1120 		args->endp += length;
1121 		args->argc++;
1122 	}
1123 
1124 	args->begin_envv = args->endp;
1125 
1126 	/*
1127 	 * extract environment strings
1128 	 */
1129 	if (envv) {
1130 		while ((envp = (caddr_t)(intptr_t)fuword(envv++))) {
1131 			if (envp == (caddr_t)-1) {
1132 				error = EFAULT;
1133 				goto err_exit;
1134 			}
1135 			if ((error = copyinstr(envp, args->endp,
1136 			    args->stringspace, &length))) {
1137 				if (error == ENAMETOOLONG)
1138 					error = E2BIG;
1139 				goto err_exit;
1140 			}
1141 			args->stringspace -= length;
1142 			args->endp += length;
1143 			args->envc++;
1144 		}
1145 	}
1146 
1147 	return (0);
1148 
1149 err_exit:
1150 	exec_free_args(args);
1151 	return (error);
1152 }
1153 
1154 static void
1155 exec_free_args(struct image_args *args)
1156 {
1157 
1158 	if (args->buf) {
1159 		kmem_free_wakeup(exec_map, (vm_offset_t)args->buf,
1160 		    PATH_MAX + ARG_MAX + MAXSHELLCMDLEN);
1161 		args->buf = NULL;
1162 	}
1163 }
1164 
1165 /*
1166  * Copy strings out to the new process address space, constructing new arg
1167  * and env vector tables. Return a pointer to the base so that it can be used
1168  * as the initial stack pointer.
1169  */
1170 register_t *
1171 exec_copyout_strings(imgp)
1172 	struct image_params *imgp;
1173 {
1174 	int argc, envc;
1175 	char **vectp;
1176 	char *stringp, *destp;
1177 	register_t *stack_base;
1178 	struct ps_strings *arginfo;
1179 	struct proc *p;
1180 	size_t execpath_len;
1181 	int szsigcode;
1182 
1183 	/*
1184 	 * Calculate string base and vector table pointers.
1185 	 * Also deal with signal trampoline code for this exec type.
1186 	 */
1187 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
1188 		execpath_len = strlen(imgp->execpath) + 1;
1189 	else
1190 		execpath_len = 0;
1191 	p = imgp->proc;
1192 	szsigcode = 0;
1193 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1194 	if (p->p_sysent->sv_szsigcode != NULL)
1195 		szsigcode = *(p->p_sysent->sv_szsigcode);
1196 	destp =	(caddr_t)arginfo - szsigcode - SPARE_USRSPACE -
1197 	    roundup(execpath_len, sizeof(char *)) -
1198 	    roundup((ARG_MAX - imgp->args->stringspace), sizeof(char *));
1199 
1200 	/*
1201 	 * install sigcode
1202 	 */
1203 	if (szsigcode)
1204 		copyout(p->p_sysent->sv_sigcode, ((caddr_t)arginfo -
1205 		    szsigcode), szsigcode);
1206 
1207 	/*
1208 	 * Copy the image path for the rtld.
1209 	 */
1210 	if (execpath_len != 0) {
1211 		imgp->execpathp = (uintptr_t)arginfo - szsigcode - execpath_len;
1212 		copyout(imgp->execpath, (void *)imgp->execpathp,
1213 		    execpath_len);
1214 	}
1215 
1216 	/*
1217 	 * If we have a valid auxargs ptr, prepare some room
1218 	 * on the stack.
1219 	 */
1220 	if (imgp->auxargs) {
1221 		/*
1222 		 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1223 		 * lower compatibility.
1224 		 */
1225 		imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1226 		    (AT_COUNT * 2);
1227 		/*
1228 		 * The '+ 2' is for the null pointers at the end of each of
1229 		 * the arg and env vector sets,and imgp->auxarg_size is room
1230 		 * for argument of Runtime loader.
1231 		 */
1232 		vectp = (char **)(destp - (imgp->args->argc +
1233 		    imgp->args->envc + 2 + imgp->auxarg_size + execpath_len) *
1234 		    sizeof(char *));
1235 	} else {
1236 		/*
1237 		 * The '+ 2' is for the null pointers at the end of each of
1238 		 * the arg and env vector sets
1239 		 */
1240 		vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc + 2) *
1241 		    sizeof(char *));
1242 	}
1243 
1244 	/*
1245 	 * vectp also becomes our initial stack base
1246 	 */
1247 	stack_base = (register_t *)vectp;
1248 
1249 	stringp = imgp->args->begin_argv;
1250 	argc = imgp->args->argc;
1251 	envc = imgp->args->envc;
1252 
1253 	/*
1254 	 * Copy out strings - arguments and environment.
1255 	 */
1256 	copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
1257 
1258 	/*
1259 	 * Fill in "ps_strings" struct for ps, w, etc.
1260 	 */
1261 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1262 	suword(&arginfo->ps_nargvstr, argc);
1263 
1264 	/*
1265 	 * Fill in argument portion of vector table.
1266 	 */
1267 	for (; argc > 0; --argc) {
1268 		suword(vectp++, (long)(intptr_t)destp);
1269 		while (*stringp++ != 0)
1270 			destp++;
1271 		destp++;
1272 	}
1273 
1274 	/* a null vector table pointer separates the argp's from the envp's */
1275 	suword(vectp++, 0);
1276 
1277 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1278 	suword(&arginfo->ps_nenvstr, envc);
1279 
1280 	/*
1281 	 * Fill in environment portion of vector table.
1282 	 */
1283 	for (; envc > 0; --envc) {
1284 		suword(vectp++, (long)(intptr_t)destp);
1285 		while (*stringp++ != 0)
1286 			destp++;
1287 		destp++;
1288 	}
1289 
1290 	/* end of vector table is a null pointer */
1291 	suword(vectp, 0);
1292 
1293 	return (stack_base);
1294 }
1295 
1296 /*
1297  * Check permissions of file to execute.
1298  *	Called with imgp->vp locked.
1299  *	Return 0 for success or error code on failure.
1300  */
1301 int
1302 exec_check_permissions(imgp)
1303 	struct image_params *imgp;
1304 {
1305 	struct vnode *vp = imgp->vp;
1306 	struct vattr *attr = imgp->attr;
1307 	struct thread *td;
1308 	int error;
1309 
1310 	td = curthread;
1311 
1312 	/* Get file attributes */
1313 	error = VOP_GETATTR(vp, attr, td->td_ucred);
1314 	if (error)
1315 		return (error);
1316 
1317 #ifdef MAC
1318 	error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1319 	if (error)
1320 		return (error);
1321 #endif
1322 
1323 	/*
1324 	 * 1) Check if file execution is disabled for the filesystem that this
1325 	 *	file resides on.
1326 	 * 2) Insure that at least one execute bit is on - otherwise root
1327 	 *	will always succeed, and we don't want to happen unless the
1328 	 *	file really is executable.
1329 	 * 3) Insure that the file is a regular file.
1330 	 */
1331 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1332 	    ((attr->va_mode & 0111) == 0) ||
1333 	    (attr->va_type != VREG))
1334 		return (EACCES);
1335 
1336 	/*
1337 	 * Zero length files can't be exec'd
1338 	 */
1339 	if (attr->va_size == 0)
1340 		return (ENOEXEC);
1341 
1342 	/*
1343 	 *  Check for execute permission to file based on current credentials.
1344 	 */
1345 	error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1346 	if (error)
1347 		return (error);
1348 
1349 	/*
1350 	 * Check number of open-for-writes on the file and deny execution
1351 	 * if there are any.
1352 	 */
1353 	if (vp->v_writecount)
1354 		return (ETXTBSY);
1355 
1356 	/*
1357 	 * Call filesystem specific open routine (which does nothing in the
1358 	 * general case).
1359 	 */
1360 	error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1361 	if (error == 0)
1362 		imgp->opened = 1;
1363 	return (error);
1364 }
1365 
1366 /*
1367  * Exec handler registration
1368  */
1369 int
1370 exec_register(execsw_arg)
1371 	const struct execsw *execsw_arg;
1372 {
1373 	const struct execsw **es, **xs, **newexecsw;
1374 	int count = 2;	/* New slot and trailing NULL */
1375 
1376 	if (execsw)
1377 		for (es = execsw; *es; es++)
1378 			count++;
1379 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1380 	if (newexecsw == NULL)
1381 		return (ENOMEM);
1382 	xs = newexecsw;
1383 	if (execsw)
1384 		for (es = execsw; *es; es++)
1385 			*xs++ = *es;
1386 	*xs++ = execsw_arg;
1387 	*xs = NULL;
1388 	if (execsw)
1389 		free(execsw, M_TEMP);
1390 	execsw = newexecsw;
1391 	return (0);
1392 }
1393 
1394 int
1395 exec_unregister(execsw_arg)
1396 	const struct execsw *execsw_arg;
1397 {
1398 	const struct execsw **es, **xs, **newexecsw;
1399 	int count = 1;
1400 
1401 	if (execsw == NULL)
1402 		panic("unregister with no handlers left?\n");
1403 
1404 	for (es = execsw; *es; es++) {
1405 		if (*es == execsw_arg)
1406 			break;
1407 	}
1408 	if (*es == NULL)
1409 		return (ENOENT);
1410 	for (es = execsw; *es; es++)
1411 		if (*es != execsw_arg)
1412 			count++;
1413 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1414 	if (newexecsw == NULL)
1415 		return (ENOMEM);
1416 	xs = newexecsw;
1417 	for (es = execsw; *es; es++)
1418 		if (*es != execsw_arg)
1419 			*xs++ = *es;
1420 	*xs = NULL;
1421 	if (execsw)
1422 		free(execsw, M_TEMP);
1423 	execsw = newexecsw;
1424 	return (0);
1425 }
1426