xref: /freebsd/sys/kern/kern_exec.c (revision 9fd69f37d28cfd7438cac3eeb45fe9dd46b4d7dd)
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 map_at_zero = 0;
126 TUNABLE_INT("security.bsd.map_at_zero", &map_at_zero);
127 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RW, &map_at_zero, 0,
128     "Permit processes to map an object at virtual address 0.");
129 
130 static int
131 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
132 {
133 	struct proc *p;
134 	int error;
135 
136 	p = curproc;
137 #ifdef SCTL_MASK32
138 	if (req->flags & SCTL_MASK32) {
139 		unsigned int val;
140 		val = (unsigned int)p->p_sysent->sv_psstrings;
141 		error = SYSCTL_OUT(req, &val, sizeof(val));
142 	} else
143 #endif
144 		error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
145 		   sizeof(p->p_sysent->sv_psstrings));
146 	return error;
147 }
148 
149 static int
150 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
151 {
152 	struct proc *p;
153 	int error;
154 
155 	p = curproc;
156 #ifdef SCTL_MASK32
157 	if (req->flags & SCTL_MASK32) {
158 		unsigned int val;
159 		val = (unsigned int)p->p_sysent->sv_usrstack;
160 		error = SYSCTL_OUT(req, &val, sizeof(val));
161 	} else
162 #endif
163 		error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
164 		    sizeof(p->p_sysent->sv_usrstack));
165 	return error;
166 }
167 
168 static int
169 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
170 {
171 	struct proc *p;
172 
173 	p = curproc;
174 	return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
175 	    sizeof(p->p_sysent->sv_stackprot)));
176 }
177 
178 /*
179  * Each of the items is a pointer to a `const struct execsw', hence the
180  * double pointer here.
181  */
182 static const struct execsw **execsw;
183 
184 #ifndef _SYS_SYSPROTO_H_
185 struct execve_args {
186 	char    *fname;
187 	char    **argv;
188 	char    **envv;
189 };
190 #endif
191 
192 int
193 execve(td, uap)
194 	struct thread *td;
195 	struct execve_args /* {
196 		char *fname;
197 		char **argv;
198 		char **envv;
199 	} */ *uap;
200 {
201 	int error;
202 	struct image_args args;
203 
204 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
205 	    uap->argv, uap->envv);
206 	if (error == 0)
207 		error = kern_execve(td, &args, NULL);
208 	return (error);
209 }
210 
211 #ifndef _SYS_SYSPROTO_H_
212 struct fexecve_args {
213 	int	fd;
214 	char	**argv;
215 	char	**envv;
216 }
217 #endif
218 int
219 fexecve(struct thread *td, struct fexecve_args *uap)
220 {
221 	int error;
222 	struct image_args args;
223 
224 	error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
225 	    uap->argv, uap->envv);
226 	if (error == 0) {
227 		args.fd = uap->fd;
228 		error = kern_execve(td, &args, NULL);
229 	}
230 	return (error);
231 }
232 
233 #ifndef _SYS_SYSPROTO_H_
234 struct __mac_execve_args {
235 	char	*fname;
236 	char	**argv;
237 	char	**envv;
238 	struct mac	*mac_p;
239 };
240 #endif
241 
242 int
243 __mac_execve(td, uap)
244 	struct thread *td;
245 	struct __mac_execve_args /* {
246 		char *fname;
247 		char **argv;
248 		char **envv;
249 		struct mac *mac_p;
250 	} */ *uap;
251 {
252 #ifdef MAC
253 	int error;
254 	struct image_args args;
255 
256 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
257 	    uap->argv, uap->envv);
258 	if (error == 0)
259 		error = kern_execve(td, &args, uap->mac_p);
260 	return (error);
261 #else
262 	return (ENOSYS);
263 #endif
264 }
265 
266 /*
267  * XXX: kern_execve has the astonishing property of not always returning to
268  * the caller.  If sufficiently bad things happen during the call to
269  * do_execve(), it can end up calling exit1(); as a result, callers must
270  * avoid doing anything which they might need to undo (e.g., allocating
271  * memory).
272  */
273 int
274 kern_execve(td, args, mac_p)
275 	struct thread *td;
276 	struct image_args *args;
277 	struct mac *mac_p;
278 {
279 	struct proc *p = td->td_proc;
280 	int error;
281 
282 	AUDIT_ARG_ARGV(args->begin_argv, args->argc,
283 	    args->begin_envv - args->begin_argv);
284 	AUDIT_ARG_ENVV(args->begin_envv, args->envc,
285 	    args->endp - args->begin_envv);
286 	if (p->p_flag & P_HADTHREADS) {
287 		PROC_LOCK(p);
288 		if (thread_single(SINGLE_BOUNDARY)) {
289 			PROC_UNLOCK(p);
290 	       		exec_free_args(args);
291 			return (ERESTART);	/* Try again later. */
292 		}
293 		PROC_UNLOCK(p);
294 	}
295 
296 	error = do_execve(td, args, mac_p);
297 
298 	if (p->p_flag & P_HADTHREADS) {
299 		PROC_LOCK(p);
300 		/*
301 		 * If success, we upgrade to SINGLE_EXIT state to
302 		 * force other threads to suicide.
303 		 */
304 		if (error == 0)
305 			thread_single(SINGLE_EXIT);
306 		else
307 			thread_single_end();
308 		PROC_UNLOCK(p);
309 	}
310 
311 	return (error);
312 }
313 
314 /*
315  * In-kernel implementation of execve().  All arguments are assumed to be
316  * userspace pointers from the passed thread.
317  */
318 static int
319 do_execve(td, args, mac_p)
320 	struct thread *td;
321 	struct image_args *args;
322 	struct mac *mac_p;
323 {
324 	struct proc *p = td->td_proc;
325 	struct nameidata nd;
326 	struct ucred *newcred = NULL, *oldcred;
327 	struct uidinfo *euip;
328 	register_t *stack_base;
329 	int error, i;
330 	struct image_params image_params, *imgp;
331 	struct vattr attr;
332 	int (*img_first)(struct image_params *);
333 	struct pargs *oldargs = NULL, *newargs = NULL;
334 	struct sigacts *oldsigacts, *newsigacts;
335 #ifdef KTRACE
336 	struct vnode *tracevp = NULL;
337 	struct ucred *tracecred = NULL;
338 #endif
339 	struct vnode *textvp = NULL, *binvp = NULL;
340 	int credential_changing;
341 	int vfslocked;
342 	int textset;
343 #ifdef MAC
344 	struct label *interpvplabel = NULL;
345 	int will_transition;
346 #endif
347 #ifdef HWPMC_HOOKS
348 	struct pmckern_procexec pe;
349 #endif
350 	static const char fexecv_proc_title[] = "(fexecv)";
351 
352 	vfslocked = 0;
353 	imgp = &image_params;
354 
355 	/*
356 	 * Lock the process and set the P_INEXEC flag to indicate that
357 	 * it should be left alone until we're done here.  This is
358 	 * necessary to avoid race conditions - e.g. in ptrace() -
359 	 * that might allow a local user to illicitly obtain elevated
360 	 * privileges.
361 	 */
362 	PROC_LOCK(p);
363 	KASSERT((p->p_flag & P_INEXEC) == 0,
364 	    ("%s(): process already has P_INEXEC flag", __func__));
365 	p->p_flag |= P_INEXEC;
366 	PROC_UNLOCK(p);
367 
368 	/*
369 	 * Initialize part of the common data
370 	 */
371 	imgp->proc = p;
372 	imgp->execlabel = NULL;
373 	imgp->attr = &attr;
374 	imgp->entry_addr = 0;
375 	imgp->vmspace_destroyed = 0;
376 	imgp->interpreted = 0;
377 	imgp->opened = 0;
378 	imgp->interpreter_name = args->buf + PATH_MAX + ARG_MAX;
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 debugger so breakpoints
756 	 * can be set before the program executes.
757 	 * Use tdsignal to deliver signal to current thread, use
758 	 * psignal may cause the signal to be delivered to wrong thread
759 	 * because that thread will exit, remember we are going to enter
760 	 * single thread mode.
761 	 */
762 	if (p->p_flag & P_TRACED)
763 		tdsignal(p, td, SIGTRAP, NULL);
764 
765 	/* clear "fork but no exec" flag, as we _are_ execing */
766 	p->p_acflag &= ~AFORK;
767 
768 	/*
769 	 * Free any previous argument cache and replace it with
770 	 * the new argument cache, if any.
771 	 */
772 	oldargs = p->p_args;
773 	p->p_args = newargs;
774 	newargs = NULL;
775 
776 #ifdef	HWPMC_HOOKS
777 	/*
778 	 * Check if system-wide sampling is in effect or if the
779 	 * current process is using PMCs.  If so, do exec() time
780 	 * processing.  This processing needs to happen AFTER the
781 	 * P_INEXEC flag is cleared.
782 	 *
783 	 * The proc lock needs to be released before taking the PMC
784 	 * SX.
785 	 */
786 	if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
787 		PROC_UNLOCK(p);
788 		VOP_UNLOCK(imgp->vp, 0);
789 		pe.pm_credentialschanged = credential_changing;
790 		pe.pm_entryaddr = imgp->entry_addr;
791 
792 		PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
793 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
794 	} else
795 		PROC_UNLOCK(p);
796 #else  /* !HWPMC_HOOKS */
797 	PROC_UNLOCK(p);
798 #endif
799 
800 	/* Set values passed into the program in registers. */
801 	if (p->p_sysent->sv_setregs)
802 		(*p->p_sysent->sv_setregs)(td, imgp->entry_addr,
803 		    (u_long)(uintptr_t)stack_base, imgp->ps_strings);
804 	else
805 		exec_setregs(td, imgp->entry_addr,
806 		    (u_long)(uintptr_t)stack_base, imgp->ps_strings);
807 
808 	vfs_mark_atime(imgp->vp, td->td_ucred);
809 
810 	SDT_PROBE(proc, kernel, , exec_success, args->fname, 0, 0, 0, 0);
811 
812 done1:
813 	/*
814 	 * Free any resources malloc'd earlier that we didn't use.
815 	 */
816 	uifree(euip);
817 	if (newcred == NULL)
818 		crfree(oldcred);
819 	else
820 		crfree(newcred);
821 	VOP_UNLOCK(imgp->vp, 0);
822 
823 	/*
824 	 * Handle deferred decrement of ref counts.
825 	 */
826 	if (textvp != NULL) {
827 		int tvfslocked;
828 
829 		tvfslocked = VFS_LOCK_GIANT(textvp->v_mount);
830 		vrele(textvp);
831 		VFS_UNLOCK_GIANT(tvfslocked);
832 	}
833 	if (binvp && error != 0)
834 		vrele(binvp);
835 #ifdef KTRACE
836 	if (tracevp != NULL) {
837 		int tvfslocked;
838 
839 		tvfslocked = VFS_LOCK_GIANT(tracevp->v_mount);
840 		vrele(tracevp);
841 		VFS_UNLOCK_GIANT(tvfslocked);
842 	}
843 	if (tracecred != NULL)
844 		crfree(tracecred);
845 #endif
846 	vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
847 	pargs_drop(oldargs);
848 	pargs_drop(newargs);
849 	if (oldsigacts != NULL)
850 		sigacts_free(oldsigacts);
851 
852 exec_fail_dealloc:
853 
854 	/*
855 	 * free various allocated resources
856 	 */
857 	if (imgp->firstpage != NULL)
858 		exec_unmap_first_page(imgp);
859 
860 	if (imgp->vp != NULL) {
861 		if (args->fname)
862 			NDFREE(&nd, NDF_ONLY_PNBUF);
863 		if (imgp->opened)
864 			VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
865 		vput(imgp->vp);
866 	}
867 
868 	if (imgp->object != NULL)
869 		vm_object_deallocate(imgp->object);
870 
871 	free(imgp->freepath, M_TEMP);
872 
873 	if (error == 0) {
874 		/*
875 		 * Stop the process here if its stop event mask has
876 		 * the S_EXEC bit set.
877 		 */
878 		STOPEVENT(p, S_EXEC, 0);
879 		goto done2;
880 	}
881 
882 exec_fail:
883 	/* we're done here, clear P_INEXEC */
884 	PROC_LOCK(p);
885 	p->p_flag &= ~P_INEXEC;
886 	PROC_UNLOCK(p);
887 
888 	SDT_PROBE(proc, kernel, , exec_failure, error, 0, 0, 0, 0);
889 
890 done2:
891 #ifdef MAC
892 	mac_execve_exit(imgp);
893 	mac_execve_interpreter_exit(interpvplabel);
894 #endif
895 	VFS_UNLOCK_GIANT(vfslocked);
896 	exec_free_args(args);
897 
898 	if (error && imgp->vmspace_destroyed) {
899 		/* sorry, no more process anymore. exit gracefully */
900 		exit1(td, W_EXITCODE(0, SIGABRT));
901 		/* NOT REACHED */
902 	}
903 	return (error);
904 }
905 
906 int
907 exec_map_first_page(imgp)
908 	struct image_params *imgp;
909 {
910 	int rv, i;
911 	int initial_pagein;
912 	vm_page_t ma[VM_INITIAL_PAGEIN];
913 	vm_object_t object;
914 
915 	if (imgp->firstpage != NULL)
916 		exec_unmap_first_page(imgp);
917 
918 	object = imgp->vp->v_object;
919 	if (object == NULL)
920 		return (EACCES);
921 	VM_OBJECT_LOCK(object);
922 #if VM_NRESERVLEVEL > 0
923 	if ((object->flags & OBJ_COLORED) == 0) {
924 		object->flags |= OBJ_COLORED;
925 		object->pg_color = 0;
926 	}
927 #endif
928 	ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
929 	if (ma[0]->valid != VM_PAGE_BITS_ALL) {
930 		initial_pagein = VM_INITIAL_PAGEIN;
931 		if (initial_pagein > object->size)
932 			initial_pagein = object->size;
933 		for (i = 1; i < initial_pagein; i++) {
934 			if ((ma[i] = vm_page_lookup(object, i)) != NULL) {
935 				if (ma[i]->valid)
936 					break;
937 				if ((ma[i]->oflags & VPO_BUSY) || ma[i]->busy)
938 					break;
939 				vm_page_busy(ma[i]);
940 			} else {
941 				ma[i] = vm_page_alloc(object, i,
942 				    VM_ALLOC_NORMAL | VM_ALLOC_IFNOTCACHED);
943 				if (ma[i] == NULL)
944 					break;
945 			}
946 		}
947 		initial_pagein = i;
948 		rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
949 		ma[0] = vm_page_lookup(object, 0);
950 		if ((rv != VM_PAGER_OK) || (ma[0] == NULL)) {
951 			if (ma[0]) {
952 				vm_page_lock_queues();
953 				vm_page_free(ma[0]);
954 				vm_page_unlock_queues();
955 			}
956 			VM_OBJECT_UNLOCK(object);
957 			return (EIO);
958 		}
959 	}
960 	vm_page_lock_queues();
961 	vm_page_hold(ma[0]);
962 	vm_page_unlock_queues();
963 	vm_page_wakeup(ma[0]);
964 	VM_OBJECT_UNLOCK(object);
965 
966 	imgp->firstpage = sf_buf_alloc(ma[0], 0);
967 	imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
968 
969 	return (0);
970 }
971 
972 void
973 exec_unmap_first_page(imgp)
974 	struct image_params *imgp;
975 {
976 	vm_page_t m;
977 
978 	if (imgp->firstpage != NULL) {
979 		m = sf_buf_page(imgp->firstpage);
980 		sf_buf_free(imgp->firstpage);
981 		imgp->firstpage = NULL;
982 		vm_page_lock_queues();
983 		vm_page_unhold(m);
984 		vm_page_unlock_queues();
985 	}
986 }
987 
988 /*
989  * Destroy old address space, and allocate a new stack
990  *	The new stack is only SGROWSIZ large because it is grown
991  *	automatically in trap.c.
992  */
993 int
994 exec_new_vmspace(imgp, sv)
995 	struct image_params *imgp;
996 	struct sysentvec *sv;
997 {
998 	int error;
999 	struct proc *p = imgp->proc;
1000 	struct vmspace *vmspace = p->p_vmspace;
1001 	vm_offset_t sv_minuser, stack_addr;
1002 	vm_map_t map;
1003 	u_long ssiz;
1004 
1005 	imgp->vmspace_destroyed = 1;
1006 	imgp->sysent = sv;
1007 
1008 	/* May be called with Giant held */
1009 	EVENTHANDLER_INVOKE(process_exec, p, imgp);
1010 
1011 	/*
1012 	 * Blow away entire process VM, if address space not shared,
1013 	 * otherwise, create a new VM space so that other threads are
1014 	 * not disrupted
1015 	 */
1016 	map = &vmspace->vm_map;
1017 	if (map_at_zero)
1018 		sv_minuser = sv->sv_minuser;
1019 	else
1020 		sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1021 	if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser &&
1022 	    vm_map_max(map) == sv->sv_maxuser) {
1023 		shmexit(vmspace);
1024 		pmap_remove_pages(vmspace_pmap(vmspace));
1025 		vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1026 	} else {
1027 		error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1028 		if (error)
1029 			return (error);
1030 		vmspace = p->p_vmspace;
1031 		map = &vmspace->vm_map;
1032 	}
1033 
1034 	/* Allocate a new stack */
1035 	if (sv->sv_maxssiz != NULL)
1036 		ssiz = *sv->sv_maxssiz;
1037 	else
1038 		ssiz = maxssiz;
1039 	stack_addr = sv->sv_usrstack - ssiz;
1040 	error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1041 	    sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1042 	if (error)
1043 		return (error);
1044 
1045 #ifdef __ia64__
1046 	/* Allocate a new register stack */
1047 	stack_addr = IA64_BACKINGSTORE;
1048 	error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1049 	    sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_UP);
1050 	if (error)
1051 		return (error);
1052 #endif
1053 
1054 	/* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
1055 	 * VM_STACK case, but they are still used to monitor the size of the
1056 	 * process stack so we can check the stack rlimit.
1057 	 */
1058 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1059 	vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - ssiz;
1060 
1061 	return (0);
1062 }
1063 
1064 /*
1065  * Copy out argument and environment strings from the old process address
1066  * space into the temporary string buffer.
1067  */
1068 int
1069 exec_copyin_args(struct image_args *args, char *fname,
1070     enum uio_seg segflg, char **argv, char **envv)
1071 {
1072 	char *argp, *envp;
1073 	int error;
1074 	size_t length;
1075 
1076 	bzero(args, sizeof(*args));
1077 	if (argv == NULL)
1078 		return (EFAULT);
1079 	/*
1080 	 * Allocate temporary demand zeroed space for argument and
1081 	 *	environment strings:
1082 	 *
1083 	 * o ARG_MAX for argument and environment;
1084 	 * o MAXSHELLCMDLEN for the name of interpreters.
1085 	 */
1086 	args->buf = (char *) kmem_alloc_wait(exec_map,
1087 	    PATH_MAX + ARG_MAX + MAXSHELLCMDLEN);
1088 	if (args->buf == NULL)
1089 		return (ENOMEM);
1090 	args->begin_argv = args->buf;
1091 	args->endp = args->begin_argv;
1092 	args->stringspace = ARG_MAX;
1093 	/*
1094 	 * Copy the file name.
1095 	 */
1096 	if (fname != NULL) {
1097 		args->fname = args->buf + ARG_MAX;
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 		args->fname = NULL;
1105 
1106 	/*
1107 	 * extract arguments first
1108 	 */
1109 	while ((argp = (caddr_t) (intptr_t) fuword(argv++))) {
1110 		if (argp == (caddr_t) -1) {
1111 			error = EFAULT;
1112 			goto err_exit;
1113 		}
1114 		if ((error = copyinstr(argp, args->endp,
1115 		    args->stringspace, &length))) {
1116 			if (error == ENAMETOOLONG)
1117 				error = E2BIG;
1118 			goto err_exit;
1119 		}
1120 		args->stringspace -= length;
1121 		args->endp += length;
1122 		args->argc++;
1123 	}
1124 
1125 	args->begin_envv = args->endp;
1126 
1127 	/*
1128 	 * extract environment strings
1129 	 */
1130 	if (envv) {
1131 		while ((envp = (caddr_t)(intptr_t)fuword(envv++))) {
1132 			if (envp == (caddr_t)-1) {
1133 				error = EFAULT;
1134 				goto err_exit;
1135 			}
1136 			if ((error = copyinstr(envp, args->endp,
1137 			    args->stringspace, &length))) {
1138 				if (error == ENAMETOOLONG)
1139 					error = E2BIG;
1140 				goto err_exit;
1141 			}
1142 			args->stringspace -= length;
1143 			args->endp += length;
1144 			args->envc++;
1145 		}
1146 	}
1147 
1148 	return (0);
1149 
1150 err_exit:
1151 	exec_free_args(args);
1152 	return (error);
1153 }
1154 
1155 static void
1156 exec_free_args(struct image_args *args)
1157 {
1158 
1159 	if (args->buf) {
1160 		kmem_free_wakeup(exec_map, (vm_offset_t)args->buf,
1161 		    PATH_MAX + ARG_MAX + MAXSHELLCMDLEN);
1162 		args->buf = NULL;
1163 	}
1164 }
1165 
1166 /*
1167  * Copy strings out to the new process address space, constructing new arg
1168  * and env vector tables. Return a pointer to the base so that it can be used
1169  * as the initial stack pointer.
1170  */
1171 register_t *
1172 exec_copyout_strings(imgp)
1173 	struct image_params *imgp;
1174 {
1175 	int argc, envc;
1176 	char **vectp;
1177 	char *stringp, *destp;
1178 	register_t *stack_base;
1179 	struct ps_strings *arginfo;
1180 	struct proc *p;
1181 	size_t execpath_len;
1182 	int szsigcode;
1183 
1184 	/*
1185 	 * Calculate string base and vector table pointers.
1186 	 * Also deal with signal trampoline code for this exec type.
1187 	 */
1188 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
1189 		execpath_len = strlen(imgp->execpath) + 1;
1190 	else
1191 		execpath_len = 0;
1192 	p = imgp->proc;
1193 	szsigcode = 0;
1194 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1195 	if (p->p_sysent->sv_szsigcode != NULL)
1196 		szsigcode = *(p->p_sysent->sv_szsigcode);
1197 	destp =	(caddr_t)arginfo - szsigcode - SPARE_USRSPACE -
1198 	    roundup(execpath_len, sizeof(char *)) -
1199 	    roundup((ARG_MAX - imgp->args->stringspace), sizeof(char *));
1200 
1201 	/*
1202 	 * install sigcode
1203 	 */
1204 	if (szsigcode)
1205 		copyout(p->p_sysent->sv_sigcode, ((caddr_t)arginfo -
1206 		    szsigcode), szsigcode);
1207 
1208 	/*
1209 	 * Copy the image path for the rtld.
1210 	 */
1211 	if (execpath_len != 0) {
1212 		imgp->execpathp = (uintptr_t)arginfo - szsigcode - execpath_len;
1213 		copyout(imgp->execpath, (void *)imgp->execpathp,
1214 		    execpath_len);
1215 	}
1216 
1217 	/*
1218 	 * If we have a valid auxargs ptr, prepare some room
1219 	 * on the stack.
1220 	 */
1221 	if (imgp->auxargs) {
1222 		/*
1223 		 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1224 		 * lower compatibility.
1225 		 */
1226 		imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1227 		    (AT_COUNT * 2);
1228 		/*
1229 		 * The '+ 2' is for the null pointers at the end of each of
1230 		 * the arg and env vector sets,and imgp->auxarg_size is room
1231 		 * for argument of Runtime loader.
1232 		 */
1233 		vectp = (char **)(destp - (imgp->args->argc +
1234 		    imgp->args->envc + 2 + imgp->auxarg_size + execpath_len) *
1235 		    sizeof(char *));
1236 	} else {
1237 		/*
1238 		 * The '+ 2' is for the null pointers at the end of each of
1239 		 * the arg and env vector sets
1240 		 */
1241 		vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc + 2) *
1242 		    sizeof(char *));
1243 	}
1244 
1245 	/*
1246 	 * vectp also becomes our initial stack base
1247 	 */
1248 	stack_base = (register_t *)vectp;
1249 
1250 	stringp = imgp->args->begin_argv;
1251 	argc = imgp->args->argc;
1252 	envc = imgp->args->envc;
1253 
1254 	/*
1255 	 * Copy out strings - arguments and environment.
1256 	 */
1257 	copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
1258 
1259 	/*
1260 	 * Fill in "ps_strings" struct for ps, w, etc.
1261 	 */
1262 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1263 	suword(&arginfo->ps_nargvstr, argc);
1264 
1265 	/*
1266 	 * Fill in argument portion of vector table.
1267 	 */
1268 	for (; argc > 0; --argc) {
1269 		suword(vectp++, (long)(intptr_t)destp);
1270 		while (*stringp++ != 0)
1271 			destp++;
1272 		destp++;
1273 	}
1274 
1275 	/* a null vector table pointer separates the argp's from the envp's */
1276 	suword(vectp++, 0);
1277 
1278 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1279 	suword(&arginfo->ps_nenvstr, envc);
1280 
1281 	/*
1282 	 * Fill in environment portion of vector table.
1283 	 */
1284 	for (; envc > 0; --envc) {
1285 		suword(vectp++, (long)(intptr_t)destp);
1286 		while (*stringp++ != 0)
1287 			destp++;
1288 		destp++;
1289 	}
1290 
1291 	/* end of vector table is a null pointer */
1292 	suword(vectp, 0);
1293 
1294 	return (stack_base);
1295 }
1296 
1297 /*
1298  * Check permissions of file to execute.
1299  *	Called with imgp->vp locked.
1300  *	Return 0 for success or error code on failure.
1301  */
1302 int
1303 exec_check_permissions(imgp)
1304 	struct image_params *imgp;
1305 {
1306 	struct vnode *vp = imgp->vp;
1307 	struct vattr *attr = imgp->attr;
1308 	struct thread *td;
1309 	int error;
1310 
1311 	td = curthread;
1312 
1313 	/* Get file attributes */
1314 	error = VOP_GETATTR(vp, attr, td->td_ucred);
1315 	if (error)
1316 		return (error);
1317 
1318 #ifdef MAC
1319 	error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1320 	if (error)
1321 		return (error);
1322 #endif
1323 
1324 	/*
1325 	 * 1) Check if file execution is disabled for the filesystem that this
1326 	 *	file resides on.
1327 	 * 2) Insure that at least one execute bit is on - otherwise root
1328 	 *	will always succeed, and we don't want to happen unless the
1329 	 *	file really is executable.
1330 	 * 3) Insure that the file is a regular file.
1331 	 */
1332 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1333 	    ((attr->va_mode & 0111) == 0) ||
1334 	    (attr->va_type != VREG))
1335 		return (EACCES);
1336 
1337 	/*
1338 	 * Zero length files can't be exec'd
1339 	 */
1340 	if (attr->va_size == 0)
1341 		return (ENOEXEC);
1342 
1343 	/*
1344 	 *  Check for execute permission to file based on current credentials.
1345 	 */
1346 	error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1347 	if (error)
1348 		return (error);
1349 
1350 	/*
1351 	 * Check number of open-for-writes on the file and deny execution
1352 	 * if there are any.
1353 	 */
1354 	if (vp->v_writecount)
1355 		return (ETXTBSY);
1356 
1357 	/*
1358 	 * Call filesystem specific open routine (which does nothing in the
1359 	 * general case).
1360 	 */
1361 	error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1362 	if (error == 0)
1363 		imgp->opened = 1;
1364 	return (error);
1365 }
1366 
1367 /*
1368  * Exec handler registration
1369  */
1370 int
1371 exec_register(execsw_arg)
1372 	const struct execsw *execsw_arg;
1373 {
1374 	const struct execsw **es, **xs, **newexecsw;
1375 	int count = 2;	/* New slot and trailing NULL */
1376 
1377 	if (execsw)
1378 		for (es = execsw; *es; es++)
1379 			count++;
1380 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1381 	if (newexecsw == NULL)
1382 		return (ENOMEM);
1383 	xs = newexecsw;
1384 	if (execsw)
1385 		for (es = execsw; *es; es++)
1386 			*xs++ = *es;
1387 	*xs++ = execsw_arg;
1388 	*xs = NULL;
1389 	if (execsw)
1390 		free(execsw, M_TEMP);
1391 	execsw = newexecsw;
1392 	return (0);
1393 }
1394 
1395 int
1396 exec_unregister(execsw_arg)
1397 	const struct execsw *execsw_arg;
1398 {
1399 	const struct execsw **es, **xs, **newexecsw;
1400 	int count = 1;
1401 
1402 	if (execsw == NULL)
1403 		panic("unregister with no handlers left?\n");
1404 
1405 	for (es = execsw; *es; es++) {
1406 		if (*es == execsw_arg)
1407 			break;
1408 	}
1409 	if (*es == NULL)
1410 		return (ENOENT);
1411 	for (es = execsw; *es; es++)
1412 		if (*es != execsw_arg)
1413 			count++;
1414 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1415 	if (newexecsw == NULL)
1416 		return (ENOMEM);
1417 	xs = newexecsw;
1418 	for (es = execsw; *es; es++)
1419 		if (*es != execsw_arg)
1420 			*xs++ = *es;
1421 	*xs = NULL;
1422 	if (execsw)
1423 		free(execsw, M_TEMP);
1424 	execsw = newexecsw;
1425 	return (0);
1426 }
1427