xref: /freebsd/sys/kern/kern_exec.c (revision bc6940585f5f394c8c3fcff58c764a40f7294d4b)
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 
109 /* XXX This should be vm_size_t. */
110 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD,
111     NULL, 0, sysctl_kern_ps_strings, "LU", "");
112 
113 /* XXX This should be vm_size_t. */
114 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD,
115     NULL, 0, sysctl_kern_usrstack, "LU", "");
116 
117 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD,
118     NULL, 0, sysctl_kern_stackprot, "I", "");
119 
120 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
121 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
122     &ps_arg_cache_limit, 0, "");
123 
124 static int map_at_zero = 0;
125 TUNABLE_INT("security.bsd.map_at_zero", &map_at_zero);
126 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RW, &map_at_zero, 0,
127     "Permit processes to map an object at virtual address 0.");
128 
129 static int
130 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
131 {
132 	struct proc *p;
133 	int error;
134 
135 	p = curproc;
136 #ifdef SCTL_MASK32
137 	if (req->flags & SCTL_MASK32) {
138 		unsigned int val;
139 		val = (unsigned int)p->p_sysent->sv_psstrings;
140 		error = SYSCTL_OUT(req, &val, sizeof(val));
141 	} else
142 #endif
143 		error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
144 		   sizeof(p->p_sysent->sv_psstrings));
145 	return error;
146 }
147 
148 static int
149 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
150 {
151 	struct proc *p;
152 	int error;
153 
154 	p = curproc;
155 #ifdef SCTL_MASK32
156 	if (req->flags & SCTL_MASK32) {
157 		unsigned int val;
158 		val = (unsigned int)p->p_sysent->sv_usrstack;
159 		error = SYSCTL_OUT(req, &val, sizeof(val));
160 	} else
161 #endif
162 		error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
163 		    sizeof(p->p_sysent->sv_usrstack));
164 	return error;
165 }
166 
167 static int
168 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
169 {
170 	struct proc *p;
171 
172 	p = curproc;
173 	return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
174 	    sizeof(p->p_sysent->sv_stackprot)));
175 }
176 
177 /*
178  * Each of the items is a pointer to a `const struct execsw', hence the
179  * double pointer here.
180  */
181 static const struct execsw **execsw;
182 
183 #ifndef _SYS_SYSPROTO_H_
184 struct execve_args {
185 	char    *fname;
186 	char    **argv;
187 	char    **envv;
188 };
189 #endif
190 
191 int
192 execve(td, uap)
193 	struct thread *td;
194 	struct execve_args /* {
195 		char *fname;
196 		char **argv;
197 		char **envv;
198 	} */ *uap;
199 {
200 	int error;
201 	struct image_args args;
202 
203 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
204 	    uap->argv, uap->envv);
205 	if (error == 0)
206 		error = kern_execve(td, &args, NULL);
207 	return (error);
208 }
209 
210 #ifndef _SYS_SYSPROTO_H_
211 struct fexecve_args {
212 	int	fd;
213 	char	**argv;
214 	char	**envv;
215 }
216 #endif
217 int
218 fexecve(struct thread *td, struct fexecve_args *uap)
219 {
220 	int error;
221 	struct image_args args;
222 
223 	error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
224 	    uap->argv, uap->envv);
225 	if (error == 0) {
226 		args.fd = uap->fd;
227 		error = kern_execve(td, &args, NULL);
228 	}
229 	return (error);
230 }
231 
232 #ifndef _SYS_SYSPROTO_H_
233 struct __mac_execve_args {
234 	char	*fname;
235 	char	**argv;
236 	char	**envv;
237 	struct mac	*mac_p;
238 };
239 #endif
240 
241 int
242 __mac_execve(td, uap)
243 	struct thread *td;
244 	struct __mac_execve_args /* {
245 		char *fname;
246 		char **argv;
247 		char **envv;
248 		struct mac *mac_p;
249 	} */ *uap;
250 {
251 #ifdef MAC
252 	int error;
253 	struct image_args args;
254 
255 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
256 	    uap->argv, uap->envv);
257 	if (error == 0)
258 		error = kern_execve(td, &args, uap->mac_p);
259 	return (error);
260 #else
261 	return (ENOSYS);
262 #endif
263 }
264 
265 /*
266  * XXX: kern_execve has the astonishing property of not always returning to
267  * the caller.  If sufficiently bad things happen during the call to
268  * do_execve(), it can end up calling exit1(); as a result, callers must
269  * avoid doing anything which they might need to undo (e.g., allocating
270  * memory).
271  */
272 int
273 kern_execve(td, args, mac_p)
274 	struct thread *td;
275 	struct image_args *args;
276 	struct mac *mac_p;
277 {
278 	struct proc *p = td->td_proc;
279 	int error;
280 
281 	AUDIT_ARG_ARGV(args->begin_argv, args->argc,
282 	    args->begin_envv - args->begin_argv);
283 	AUDIT_ARG_ENVV(args->begin_envv, args->envc,
284 	    args->endp - args->begin_envv);
285 	if (p->p_flag & P_HADTHREADS) {
286 		PROC_LOCK(p);
287 		if (thread_single(SINGLE_BOUNDARY)) {
288 			PROC_UNLOCK(p);
289 	       		exec_free_args(args);
290 			return (ERESTART);	/* Try again later. */
291 		}
292 		PROC_UNLOCK(p);
293 	}
294 
295 	error = do_execve(td, args, mac_p);
296 
297 	if (p->p_flag & P_HADTHREADS) {
298 		PROC_LOCK(p);
299 		/*
300 		 * If success, we upgrade to SINGLE_EXIT state to
301 		 * force other threads to suicide.
302 		 */
303 		if (error == 0)
304 			thread_single(SINGLE_EXIT);
305 		else
306 			thread_single_end();
307 		PROC_UNLOCK(p);
308 	}
309 
310 	return (error);
311 }
312 
313 /*
314  * In-kernel implementation of execve().  All arguments are assumed to be
315  * userspace pointers from the passed thread.
316  */
317 static int
318 do_execve(td, args, mac_p)
319 	struct thread *td;
320 	struct image_args *args;
321 	struct mac *mac_p;
322 {
323 	struct proc *p = td->td_proc;
324 	struct nameidata nd;
325 	struct ucred *newcred = NULL, *oldcred;
326 	struct uidinfo *euip;
327 	register_t *stack_base;
328 	int error, i;
329 	struct image_params image_params, *imgp;
330 	struct vattr attr;
331 	int (*img_first)(struct image_params *);
332 	struct pargs *oldargs = NULL, *newargs = NULL;
333 	struct sigacts *oldsigacts, *newsigacts;
334 #ifdef KTRACE
335 	struct vnode *tracevp = NULL;
336 	struct ucred *tracecred = NULL;
337 #endif
338 	struct vnode *textvp = NULL, *binvp = NULL;
339 	int credential_changing;
340 	int vfslocked;
341 	int textset;
342 #ifdef MAC
343 	struct label *interpvplabel = NULL;
344 	int will_transition;
345 #endif
346 #ifdef HWPMC_HOOKS
347 	struct pmckern_procexec pe;
348 #endif
349 	static const char fexecv_proc_title[] = "(fexecv)";
350 
351 	vfslocked = 0;
352 	imgp = &image_params;
353 
354 	/*
355 	 * Lock the process and set the P_INEXEC flag to indicate that
356 	 * it should be left alone until we're done here.  This is
357 	 * necessary to avoid race conditions - e.g. in ptrace() -
358 	 * that might allow a local user to illicitly obtain elevated
359 	 * privileges.
360 	 */
361 	PROC_LOCK(p);
362 	KASSERT((p->p_flag & P_INEXEC) == 0,
363 	    ("%s(): process already has P_INEXEC flag", __func__));
364 	p->p_flag |= P_INEXEC;
365 	PROC_UNLOCK(p);
366 
367 	/*
368 	 * Initialize part of the common data
369 	 */
370 	imgp->proc = p;
371 	imgp->execlabel = NULL;
372 	imgp->attr = &attr;
373 	imgp->entry_addr = 0;
374 	imgp->reloc_base = 0;
375 	imgp->vmspace_destroyed = 0;
376 	imgp->interpreted = 0;
377 	imgp->opened = 0;
378 	imgp->interpreter_name = args->buf;
379 	imgp->auxargs = NULL;
380 	imgp->vp = NULL;
381 	imgp->object = NULL;
382 	imgp->firstpage = NULL;
383 	imgp->ps_strings = 0;
384 	imgp->auxarg_size = 0;
385 	imgp->args = args;
386 	imgp->execpath = imgp->freepath = NULL;
387 	imgp->execpathp = 0;
388 
389 #ifdef MAC
390 	error = mac_execve_enter(imgp, mac_p);
391 	if (error)
392 		goto exec_fail;
393 #endif
394 
395 	imgp->image_header = NULL;
396 
397 	/*
398 	 * Translate the file name. namei() returns a vnode pointer
399 	 *	in ni_vp amoung other things.
400 	 *
401 	 * XXXAUDIT: It would be desirable to also audit the name of the
402 	 * interpreter if this is an interpreted binary.
403 	 */
404 	if (args->fname != NULL) {
405 		NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME
406 		    | MPSAFE | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
407 	}
408 
409 	SDT_PROBE(proc, kernel, , exec, args->fname, 0, 0, 0, 0 );
410 
411 interpret:
412 	if (args->fname != NULL) {
413 		error = namei(&nd);
414 		if (error)
415 			goto exec_fail;
416 
417 		vfslocked = NDHASGIANT(&nd);
418 		binvp  = nd.ni_vp;
419 		imgp->vp = binvp;
420 	} else {
421 		AUDIT_ARG_FD(args->fd);
422 		error = fgetvp(td, args->fd, &binvp);
423 		if (error)
424 			goto exec_fail;
425 		vfslocked = VFS_LOCK_GIANT(binvp->v_mount);
426 		vn_lock(binvp, LK_EXCLUSIVE | LK_RETRY);
427 		AUDIT_ARG_VNODE1(binvp);
428 		imgp->vp = binvp;
429 	}
430 
431 	/*
432 	 * Check file permissions (also 'opens' file)
433 	 */
434 	error = exec_check_permissions(imgp);
435 	if (error)
436 		goto exec_fail_dealloc;
437 
438 	imgp->object = imgp->vp->v_object;
439 	if (imgp->object != NULL)
440 		vm_object_reference(imgp->object);
441 
442 	/*
443 	 * Set VV_TEXT now so no one can write to the executable while we're
444 	 * activating it.
445 	 *
446 	 * Remember if this was set before and unset it in case this is not
447 	 * actually an executable image.
448 	 */
449 	textset = imgp->vp->v_vflag & VV_TEXT;
450 	imgp->vp->v_vflag |= VV_TEXT;
451 
452 	error = exec_map_first_page(imgp);
453 	if (error)
454 		goto exec_fail_dealloc;
455 
456 	imgp->proc->p_osrel = 0;
457 	/*
458 	 *	If the current process has a special image activator it
459 	 *	wants to try first, call it.   For example, emulating shell
460 	 *	scripts differently.
461 	 */
462 	error = -1;
463 	if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
464 		error = img_first(imgp);
465 
466 	/*
467 	 *	Loop through the list of image activators, calling each one.
468 	 *	An activator returns -1 if there is no match, 0 on success,
469 	 *	and an error otherwise.
470 	 */
471 	for (i = 0; error == -1 && execsw[i]; ++i) {
472 		if (execsw[i]->ex_imgact == NULL ||
473 		    execsw[i]->ex_imgact == img_first) {
474 			continue;
475 		}
476 		error = (*execsw[i]->ex_imgact)(imgp);
477 	}
478 
479 	if (error) {
480 		if (error == -1) {
481 			if (textset == 0)
482 				imgp->vp->v_vflag &= ~VV_TEXT;
483 			error = ENOEXEC;
484 		}
485 		goto exec_fail_dealloc;
486 	}
487 
488 	/*
489 	 * Special interpreter operation, cleanup and loop up to try to
490 	 * activate the interpreter.
491 	 */
492 	if (imgp->interpreted) {
493 		exec_unmap_first_page(imgp);
494 		/*
495 		 * VV_TEXT needs to be unset for scripts.  There is a short
496 		 * period before we determine that something is a script where
497 		 * VV_TEXT will be set. The vnode lock is held over this
498 		 * entire period so nothing should illegitimately be blocked.
499 		 */
500 		imgp->vp->v_vflag &= ~VV_TEXT;
501 		/* free name buffer and old vnode */
502 		if (args->fname != NULL)
503 			NDFREE(&nd, NDF_ONLY_PNBUF);
504 #ifdef MAC
505 		mac_execve_interpreter_enter(binvp, &interpvplabel);
506 #endif
507 		if (imgp->opened) {
508 			VOP_CLOSE(binvp, FREAD, td->td_ucred, td);
509 			imgp->opened = 0;
510 		}
511 		vput(binvp);
512 		vm_object_deallocate(imgp->object);
513 		imgp->object = NULL;
514 		VFS_UNLOCK_GIANT(vfslocked);
515 		vfslocked = 0;
516 		/* set new name to that of the interpreter */
517 		NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME | MPSAFE,
518 		    UIO_SYSSPACE, imgp->interpreter_name, td);
519 		args->fname = imgp->interpreter_name;
520 		goto interpret;
521 	}
522 
523 	/*
524 	 * NB: We unlock the vnode here because it is believed that none
525 	 * of the sv_copyout_strings/sv_fixup operations require the vnode.
526 	 */
527 	VOP_UNLOCK(imgp->vp, 0);
528 
529 	/*
530 	 * Do the best to calculate the full path to the image file.
531 	 */
532 	if (imgp->auxargs != NULL &&
533 	    ((args->fname != NULL && args->fname[0] == '/') ||
534 	     vn_fullpath(td, imgp->vp, &imgp->execpath, &imgp->freepath) != 0))
535 		imgp->execpath = args->fname;
536 
537 	/*
538 	 * Copy out strings (args and env) and initialize stack base
539 	 */
540 	if (p->p_sysent->sv_copyout_strings)
541 		stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
542 	else
543 		stack_base = exec_copyout_strings(imgp);
544 
545 	/*
546 	 * If custom stack fixup routine present for this process
547 	 * let it do the stack setup.
548 	 * Else stuff argument count as first item on stack
549 	 */
550 	if (p->p_sysent->sv_fixup != NULL)
551 		(*p->p_sysent->sv_fixup)(&stack_base, imgp);
552 	else
553 		suword(--stack_base, imgp->args->argc);
554 
555 	/*
556 	 * For security and other reasons, the file descriptor table cannot
557 	 * be shared after an exec.
558 	 */
559 	fdunshare(p, td);
560 
561 	/*
562 	 * Malloc things before we need locks.
563 	 */
564 	newcred = crget();
565 	euip = uifind(attr.va_uid);
566 	i = imgp->args->begin_envv - imgp->args->begin_argv;
567 	/* Cache arguments if they fit inside our allowance */
568 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
569 		newargs = pargs_alloc(i);
570 		bcopy(imgp->args->begin_argv, newargs->ar_args, i);
571 	}
572 
573 	/* close files on exec */
574 	fdcloseexec(td);
575 	vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
576 
577 	/* Get a reference to the vnode prior to locking the proc */
578 	VREF(binvp);
579 
580 	/*
581 	 * For security and other reasons, signal handlers cannot
582 	 * be shared after an exec. The new process gets a copy of the old
583 	 * handlers. In execsigs(), the new process will have its signals
584 	 * reset.
585 	 */
586 	PROC_LOCK(p);
587 	oldcred = crcopysafe(p, newcred);
588 	if (sigacts_shared(p->p_sigacts)) {
589 		oldsigacts = p->p_sigacts;
590 		PROC_UNLOCK(p);
591 		newsigacts = sigacts_alloc();
592 		sigacts_copy(newsigacts, oldsigacts);
593 		PROC_LOCK(p);
594 		p->p_sigacts = newsigacts;
595 	} else
596 		oldsigacts = NULL;
597 
598 	/* Stop profiling */
599 	stopprofclock(p);
600 
601 	/* reset caught signals */
602 	execsigs(p);
603 
604 	/* name this process - nameiexec(p, ndp) */
605 	bzero(p->p_comm, sizeof(p->p_comm));
606 	if (args->fname)
607 		bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
608 		    min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
609 	else if (vn_commname(binvp, p->p_comm, sizeof(p->p_comm)) != 0)
610 		bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
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 	credential_changing = 0;
633 	credential_changing |= (attr.va_mode & S_ISUID) && oldcred->cr_uid !=
634 	    attr.va_uid;
635 	credential_changing |= (attr.va_mode & S_ISGID) && oldcred->cr_gid !=
636 	    attr.va_gid;
637 #ifdef MAC
638 	will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
639 	    interpvplabel, imgp);
640 	credential_changing |= will_transition;
641 #endif
642 
643 	if (credential_changing &&
644 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
645 	    (p->p_flag & P_TRACED) == 0) {
646 		/*
647 		 * Turn off syscall tracing for set-id programs, except for
648 		 * root.  Record any set-id flags first to make sure that
649 		 * we do not regain any tracing during a possible block.
650 		 */
651 		setsugid(p);
652 
653 #ifdef KTRACE
654 		if (p->p_tracevp != NULL &&
655 		    priv_check_cred(oldcred, PRIV_DEBUG_DIFFCRED, 0)) {
656 			mtx_lock(&ktrace_mtx);
657 			p->p_traceflag = 0;
658 			tracevp = p->p_tracevp;
659 			p->p_tracevp = NULL;
660 			tracecred = p->p_tracecred;
661 			p->p_tracecred = NULL;
662 			mtx_unlock(&ktrace_mtx);
663 		}
664 #endif
665 		/*
666 		 * Close any file descriptors 0..2 that reference procfs,
667 		 * then make sure file descriptors 0..2 are in use.
668 		 *
669 		 * setugidsafety() may call closef() and then pfind()
670 		 * which may grab the process lock.
671 		 * fdcheckstd() may call falloc() which may block to
672 		 * allocate memory, so temporarily drop the process lock.
673 		 */
674 		PROC_UNLOCK(p);
675 		VOP_UNLOCK(imgp->vp, 0);
676 		setugidsafety(td);
677 		error = fdcheckstd(td);
678 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
679 		if (error != 0)
680 			goto done1;
681 		PROC_LOCK(p);
682 		/*
683 		 * Set the new credentials.
684 		 */
685 		if (attr.va_mode & S_ISUID)
686 			change_euid(newcred, euip);
687 		if (attr.va_mode & S_ISGID)
688 			change_egid(newcred, attr.va_gid);
689 #ifdef MAC
690 		if (will_transition) {
691 			mac_vnode_execve_transition(oldcred, newcred, imgp->vp,
692 			    interpvplabel, imgp);
693 		}
694 #endif
695 		/*
696 		 * Implement correct POSIX saved-id behavior.
697 		 *
698 		 * XXXMAC: Note that the current logic will save the
699 		 * uid and gid if a MAC domain transition occurs, even
700 		 * though maybe it shouldn't.
701 		 */
702 		change_svuid(newcred, newcred->cr_uid);
703 		change_svgid(newcred, newcred->cr_gid);
704 		p->p_ucred = newcred;
705 		newcred = NULL;
706 	} else {
707 		if (oldcred->cr_uid == oldcred->cr_ruid &&
708 		    oldcred->cr_gid == oldcred->cr_rgid)
709 			p->p_flag &= ~P_SUGID;
710 		/*
711 		 * Implement correct POSIX saved-id behavior.
712 		 *
713 		 * XXX: It's not clear that the existing behavior is
714 		 * POSIX-compliant.  A number of sources indicate that the
715 		 * saved uid/gid should only be updated if the new ruid is
716 		 * not equal to the old ruid, or the new euid is not equal
717 		 * to the old euid and the new euid is not equal to the old
718 		 * ruid.  The FreeBSD code always updates the saved uid/gid.
719 		 * Also, this code uses the new (replaced) euid and egid as
720 		 * the source, which may or may not be the right ones to use.
721 		 */
722 		if (oldcred->cr_svuid != oldcred->cr_uid ||
723 		    oldcred->cr_svgid != oldcred->cr_gid) {
724 			change_svuid(newcred, newcred->cr_uid);
725 			change_svgid(newcred, newcred->cr_gid);
726 			p->p_ucred = newcred;
727 			newcred = NULL;
728 		}
729 	}
730 
731 	/*
732 	 * Store the vp for use in procfs.  This vnode was referenced prior
733 	 * to locking the proc lock.
734 	 */
735 	textvp = p->p_textvp;
736 	p->p_textvp = binvp;
737 
738 #ifdef KDTRACE_HOOKS
739 	/*
740 	 * Tell the DTrace fasttrap provider about the exec if it
741 	 * has declared an interest.
742 	 */
743 	if (dtrace_fasttrap_exec)
744 		dtrace_fasttrap_exec(p);
745 #endif
746 
747 	/*
748 	 * Notify others that we exec'd, and clear the P_INEXEC flag
749 	 * as we're now a bona fide freshly-execed process.
750 	 */
751 	KNOTE_LOCKED(&p->p_klist, NOTE_EXEC);
752 	p->p_flag &= ~P_INEXEC;
753 
754 	/*
755 	 * If tracing the process, trap to the debugger so that
756 	 * breakpoints can be set before the program executes.  We
757 	 * have to use tdsignal() to deliver the signal to the current
758 	 * thread since any other threads in this process will exit if
759 	 * execve() succeeds.
760 	 */
761 	if (p->p_flag & P_TRACED)
762 		tdsignal(td, SIGTRAP);
763 
764 	/* clear "fork but no exec" flag, as we _are_ execing */
765 	p->p_acflag &= ~AFORK;
766 
767 	/*
768 	 * Free any previous argument cache and replace it with
769 	 * the new argument cache, if any.
770 	 */
771 	oldargs = p->p_args;
772 	p->p_args = newargs;
773 	newargs = NULL;
774 
775 #ifdef	HWPMC_HOOKS
776 	/*
777 	 * Check if system-wide sampling is in effect or if the
778 	 * current process is using PMCs.  If so, do exec() time
779 	 * processing.  This processing needs to happen AFTER the
780 	 * P_INEXEC flag is cleared.
781 	 *
782 	 * The proc lock needs to be released before taking the PMC
783 	 * SX.
784 	 */
785 	if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
786 		PROC_UNLOCK(p);
787 		VOP_UNLOCK(imgp->vp, 0);
788 		pe.pm_credentialschanged = credential_changing;
789 		pe.pm_entryaddr = imgp->entry_addr;
790 
791 		PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
792 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
793 	} else
794 		PROC_UNLOCK(p);
795 #else  /* !HWPMC_HOOKS */
796 	PROC_UNLOCK(p);
797 #endif
798 
799 	/* Set values passed into the program in registers. */
800 	if (p->p_sysent->sv_setregs)
801 		(*p->p_sysent->sv_setregs)(td, imgp,
802 		    (u_long)(uintptr_t)stack_base);
803 	else
804 		exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base);
805 
806 	vfs_mark_atime(imgp->vp, td->td_ucred);
807 
808 	SDT_PROBE(proc, kernel, , exec_success, args->fname, 0, 0, 0, 0);
809 
810 done1:
811 	/*
812 	 * Free any resources malloc'd earlier that we didn't use.
813 	 */
814 	uifree(euip);
815 	if (newcred == NULL)
816 		crfree(oldcred);
817 	else
818 		crfree(newcred);
819 	VOP_UNLOCK(imgp->vp, 0);
820 
821 	/*
822 	 * Handle deferred decrement of ref counts.
823 	 */
824 	if (textvp != NULL) {
825 		int tvfslocked;
826 
827 		tvfslocked = VFS_LOCK_GIANT(textvp->v_mount);
828 		vrele(textvp);
829 		VFS_UNLOCK_GIANT(tvfslocked);
830 	}
831 	if (binvp && error != 0)
832 		vrele(binvp);
833 #ifdef KTRACE
834 	if (tracevp != NULL) {
835 		int tvfslocked;
836 
837 		tvfslocked = VFS_LOCK_GIANT(tracevp->v_mount);
838 		vrele(tracevp);
839 		VFS_UNLOCK_GIANT(tvfslocked);
840 	}
841 	if (tracecred != NULL)
842 		crfree(tracecred);
843 #endif
844 	vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
845 	pargs_drop(oldargs);
846 	pargs_drop(newargs);
847 	if (oldsigacts != NULL)
848 		sigacts_free(oldsigacts);
849 
850 exec_fail_dealloc:
851 
852 	/*
853 	 * free various allocated resources
854 	 */
855 	if (imgp->firstpage != NULL)
856 		exec_unmap_first_page(imgp);
857 
858 	if (imgp->vp != NULL) {
859 		if (args->fname)
860 			NDFREE(&nd, NDF_ONLY_PNBUF);
861 		if (imgp->opened)
862 			VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
863 		vput(imgp->vp);
864 	}
865 
866 	if (imgp->object != NULL)
867 		vm_object_deallocate(imgp->object);
868 
869 	free(imgp->freepath, M_TEMP);
870 
871 	if (error == 0) {
872 		PROC_LOCK(p);
873 		td->td_dbgflags |= TDB_EXEC;
874 		PROC_UNLOCK(p);
875 
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) {
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_next(ma[i - 1])) != 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 			if (ma[0] != NULL) {
954 				vm_page_lock(ma[0]);
955 				vm_page_free(ma[0]);
956 				vm_page_unlock(ma[0]);
957 			}
958 			VM_OBJECT_UNLOCK(object);
959 			return (EIO);
960 		}
961 	}
962 	vm_page_lock(ma[0]);
963 	vm_page_hold(ma[0]);
964 	vm_page_unlock(ma[0]);
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(m);
985 		vm_page_unhold(m);
986 		vm_page_unlock(m);
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 sv_minuser, 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 (map_at_zero)
1020 		sv_minuser = sv->sv_minuser;
1021 	else
1022 		sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1023 	if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser &&
1024 	    vm_map_max(map) == sv->sv_maxuser) {
1025 		shmexit(vmspace);
1026 		pmap_remove_pages(vmspace_pmap(vmspace));
1027 		vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1028 	} else {
1029 		error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1030 		if (error)
1031 			return (error);
1032 		vmspace = p->p_vmspace;
1033 		map = &vmspace->vm_map;
1034 	}
1035 
1036 	/* Allocate a new stack */
1037 	if (sv->sv_maxssiz != NULL)
1038 		ssiz = *sv->sv_maxssiz;
1039 	else
1040 		ssiz = maxssiz;
1041 	stack_addr = sv->sv_usrstack - ssiz;
1042 	error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1043 	    sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1044 	if (error)
1045 		return (error);
1046 
1047 #ifdef __ia64__
1048 	/* Allocate a new register stack */
1049 	stack_addr = IA64_BACKINGSTORE;
1050 	error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1051 	    sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_UP);
1052 	if (error)
1053 		return (error);
1054 #endif
1055 
1056 	/* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
1057 	 * VM_STACK case, but they are still used to monitor the size of the
1058 	 * process stack so we can check the stack rlimit.
1059 	 */
1060 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1061 	vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - ssiz;
1062 
1063 	return (0);
1064 }
1065 
1066 /*
1067  * Copy out argument and environment strings from the old process address
1068  * space into the temporary string buffer.
1069  */
1070 int
1071 exec_copyin_args(struct image_args *args, char *fname,
1072     enum uio_seg segflg, char **argv, char **envv)
1073 {
1074 	char *argp, *envp;
1075 	int error;
1076 	size_t length;
1077 
1078 	bzero(args, sizeof(*args));
1079 	if (argv == NULL)
1080 		return (EFAULT);
1081 	/*
1082 	 * Allocate temporary demand zeroed space for argument and
1083 	 *	environment strings:
1084 	 *
1085 	 * o ARG_MAX for argument and environment;
1086 	 * o MAXSHELLCMDLEN for the name of interpreters.
1087 	 */
1088 	args->buf = (char *) kmem_alloc_wait(exec_map,
1089 	    PATH_MAX + ARG_MAX + MAXSHELLCMDLEN);
1090 	if (args->buf == NULL)
1091 		return (ENOMEM);
1092 
1093 	/*
1094 	 * Copy the file name.
1095 	 */
1096 	if (fname != NULL) {
1097 		args->fname = args->buf + MAXSHELLCMDLEN;
1098 		error = (segflg == UIO_SYSSPACE) ?
1099 		    copystr(fname, args->fname, PATH_MAX, &length) :
1100 		    copyinstr(fname, args->fname, PATH_MAX, &length);
1101 		if (error != 0)
1102 			goto err_exit;
1103 	} else
1104 		length = 0;
1105 
1106 	args->begin_argv = args->buf + MAXSHELLCMDLEN + length;
1107 	args->endp = args->begin_argv;
1108 	args->stringspace = ARG_MAX;
1109 
1110 	/*
1111 	 * extract arguments first
1112 	 */
1113 	while ((argp = (caddr_t) (intptr_t) fuword(argv++))) {
1114 		if (argp == (caddr_t) -1) {
1115 			error = EFAULT;
1116 			goto err_exit;
1117 		}
1118 		if ((error = copyinstr(argp, args->endp,
1119 		    args->stringspace, &length))) {
1120 			if (error == ENAMETOOLONG)
1121 				error = E2BIG;
1122 			goto err_exit;
1123 		}
1124 		args->stringspace -= length;
1125 		args->endp += length;
1126 		args->argc++;
1127 	}
1128 
1129 	args->begin_envv = args->endp;
1130 
1131 	/*
1132 	 * extract environment strings
1133 	 */
1134 	if (envv) {
1135 		while ((envp = (caddr_t)(intptr_t)fuword(envv++))) {
1136 			if (envp == (caddr_t)-1) {
1137 				error = EFAULT;
1138 				goto err_exit;
1139 			}
1140 			if ((error = copyinstr(envp, args->endp,
1141 			    args->stringspace, &length))) {
1142 				if (error == ENAMETOOLONG)
1143 					error = E2BIG;
1144 				goto err_exit;
1145 			}
1146 			args->stringspace -= length;
1147 			args->endp += length;
1148 			args->envc++;
1149 		}
1150 	}
1151 
1152 	return (0);
1153 
1154 err_exit:
1155 	exec_free_args(args);
1156 	return (error);
1157 }
1158 
1159 void
1160 exec_free_args(struct image_args *args)
1161 {
1162 
1163 	if (args->buf != NULL) {
1164 		kmem_free_wakeup(exec_map, (vm_offset_t)args->buf,
1165 		    PATH_MAX + ARG_MAX + MAXSHELLCMDLEN);
1166 		args->buf = NULL;
1167 	}
1168 }
1169 
1170 /*
1171  * Copy strings out to the new process address space, constructing new arg
1172  * and env vector tables. Return a pointer to the base so that it can be used
1173  * as the initial stack pointer.
1174  */
1175 register_t *
1176 exec_copyout_strings(imgp)
1177 	struct image_params *imgp;
1178 {
1179 	int argc, envc;
1180 	char **vectp;
1181 	char *stringp, *destp;
1182 	register_t *stack_base;
1183 	struct ps_strings *arginfo;
1184 	struct proc *p;
1185 	size_t execpath_len;
1186 	int szsigcode;
1187 
1188 	/*
1189 	 * Calculate string base and vector table pointers.
1190 	 * Also deal with signal trampoline code for this exec type.
1191 	 */
1192 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
1193 		execpath_len = strlen(imgp->execpath) + 1;
1194 	else
1195 		execpath_len = 0;
1196 	p = imgp->proc;
1197 	szsigcode = 0;
1198 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1199 	if (p->p_sysent->sv_szsigcode != NULL)
1200 		szsigcode = *(p->p_sysent->sv_szsigcode);
1201 	destp =	(caddr_t)arginfo - szsigcode - SPARE_USRSPACE -
1202 	    roundup(execpath_len, sizeof(char *)) -
1203 	    roundup((ARG_MAX - imgp->args->stringspace), sizeof(char *));
1204 
1205 	/*
1206 	 * install sigcode
1207 	 */
1208 	if (szsigcode)
1209 		copyout(p->p_sysent->sv_sigcode, ((caddr_t)arginfo -
1210 		    szsigcode), szsigcode);
1211 
1212 	/*
1213 	 * Copy the image path for the rtld.
1214 	 */
1215 	if (execpath_len != 0) {
1216 		imgp->execpathp = (uintptr_t)arginfo - szsigcode - execpath_len;
1217 		copyout(imgp->execpath, (void *)imgp->execpathp,
1218 		    execpath_len);
1219 	}
1220 
1221 	/*
1222 	 * If we have a valid auxargs ptr, prepare some room
1223 	 * on the stack.
1224 	 */
1225 	if (imgp->auxargs) {
1226 		/*
1227 		 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1228 		 * lower compatibility.
1229 		 */
1230 		imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1231 		    (AT_COUNT * 2);
1232 		/*
1233 		 * The '+ 2' is for the null pointers at the end of each of
1234 		 * the arg and env vector sets,and imgp->auxarg_size is room
1235 		 * for argument of Runtime loader.
1236 		 */
1237 		vectp = (char **)(destp - (imgp->args->argc +
1238 		    imgp->args->envc + 2 + imgp->auxarg_size + execpath_len) *
1239 		    sizeof(char *));
1240 	} else {
1241 		/*
1242 		 * The '+ 2' is for the null pointers at the end of each of
1243 		 * the arg and env vector sets
1244 		 */
1245 		vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc + 2) *
1246 		    sizeof(char *));
1247 	}
1248 
1249 	/*
1250 	 * vectp also becomes our initial stack base
1251 	 */
1252 	stack_base = (register_t *)vectp;
1253 
1254 	stringp = imgp->args->begin_argv;
1255 	argc = imgp->args->argc;
1256 	envc = imgp->args->envc;
1257 
1258 	/*
1259 	 * Copy out strings - arguments and environment.
1260 	 */
1261 	copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
1262 
1263 	/*
1264 	 * Fill in "ps_strings" struct for ps, w, etc.
1265 	 */
1266 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1267 	suword32(&arginfo->ps_nargvstr, argc);
1268 
1269 	/*
1270 	 * Fill in argument portion of vector table.
1271 	 */
1272 	for (; argc > 0; --argc) {
1273 		suword(vectp++, (long)(intptr_t)destp);
1274 		while (*stringp++ != 0)
1275 			destp++;
1276 		destp++;
1277 	}
1278 
1279 	/* a null vector table pointer separates the argp's from the envp's */
1280 	suword(vectp++, 0);
1281 
1282 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1283 	suword32(&arginfo->ps_nenvstr, envc);
1284 
1285 	/*
1286 	 * Fill in environment portion of vector table.
1287 	 */
1288 	for (; envc > 0; --envc) {
1289 		suword(vectp++, (long)(intptr_t)destp);
1290 		while (*stringp++ != 0)
1291 			destp++;
1292 		destp++;
1293 	}
1294 
1295 	/* end of vector table is a null pointer */
1296 	suword(vectp, 0);
1297 
1298 	return (stack_base);
1299 }
1300 
1301 /*
1302  * Check permissions of file to execute.
1303  *	Called with imgp->vp locked.
1304  *	Return 0 for success or error code on failure.
1305  */
1306 int
1307 exec_check_permissions(imgp)
1308 	struct image_params *imgp;
1309 {
1310 	struct vnode *vp = imgp->vp;
1311 	struct vattr *attr = imgp->attr;
1312 	struct thread *td;
1313 	int error;
1314 
1315 	td = curthread;
1316 
1317 	/* Get file attributes */
1318 	error = VOP_GETATTR(vp, attr, td->td_ucred);
1319 	if (error)
1320 		return (error);
1321 
1322 #ifdef MAC
1323 	error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1324 	if (error)
1325 		return (error);
1326 #endif
1327 
1328 	/*
1329 	 * 1) Check if file execution is disabled for the filesystem that this
1330 	 *	file resides on.
1331 	 * 2) Insure that at least one execute bit is on - otherwise root
1332 	 *	will always succeed, and we don't want to happen unless the
1333 	 *	file really is executable.
1334 	 * 3) Insure that the file is a regular file.
1335 	 */
1336 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1337 	    ((attr->va_mode & 0111) == 0) ||
1338 	    (attr->va_type != VREG))
1339 		return (EACCES);
1340 
1341 	/*
1342 	 * Zero length files can't be exec'd
1343 	 */
1344 	if (attr->va_size == 0)
1345 		return (ENOEXEC);
1346 
1347 	/*
1348 	 *  Check for execute permission to file based on current credentials.
1349 	 */
1350 	error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1351 	if (error)
1352 		return (error);
1353 
1354 	/*
1355 	 * Check number of open-for-writes on the file and deny execution
1356 	 * if there are any.
1357 	 */
1358 	if (vp->v_writecount)
1359 		return (ETXTBSY);
1360 
1361 	/*
1362 	 * Call filesystem specific open routine (which does nothing in the
1363 	 * general case).
1364 	 */
1365 	error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1366 	if (error == 0)
1367 		imgp->opened = 1;
1368 	return (error);
1369 }
1370 
1371 /*
1372  * Exec handler registration
1373  */
1374 int
1375 exec_register(execsw_arg)
1376 	const struct execsw *execsw_arg;
1377 {
1378 	const struct execsw **es, **xs, **newexecsw;
1379 	int count = 2;	/* New slot and trailing NULL */
1380 
1381 	if (execsw)
1382 		for (es = execsw; *es; es++)
1383 			count++;
1384 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1385 	if (newexecsw == NULL)
1386 		return (ENOMEM);
1387 	xs = newexecsw;
1388 	if (execsw)
1389 		for (es = execsw; *es; es++)
1390 			*xs++ = *es;
1391 	*xs++ = execsw_arg;
1392 	*xs = NULL;
1393 	if (execsw)
1394 		free(execsw, M_TEMP);
1395 	execsw = newexecsw;
1396 	return (0);
1397 }
1398 
1399 int
1400 exec_unregister(execsw_arg)
1401 	const struct execsw *execsw_arg;
1402 {
1403 	const struct execsw **es, **xs, **newexecsw;
1404 	int count = 1;
1405 
1406 	if (execsw == NULL)
1407 		panic("unregister with no handlers left?\n");
1408 
1409 	for (es = execsw; *es; es++) {
1410 		if (*es == execsw_arg)
1411 			break;
1412 	}
1413 	if (*es == NULL)
1414 		return (ENOENT);
1415 	for (es = execsw; *es; es++)
1416 		if (*es != execsw_arg)
1417 			count++;
1418 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1419 	if (newexecsw == NULL)
1420 		return (ENOMEM);
1421 	xs = newexecsw;
1422 	for (es = execsw; *es; es++)
1423 		if (*es != execsw_arg)
1424 			*xs++ = *es;
1425 	*xs = NULL;
1426 	if (execsw)
1427 		free(execsw, M_TEMP);
1428 	execsw = newexecsw;
1429 	return (0);
1430 }
1431