xref: /freebsd/sys/kern/kern_exec.c (revision ab0b9f6b3073e6c4d1dfbf07444d7db67a189a96)
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_capsicum.h"
31 #include "opt_hwpmc_hooks.h"
32 #include "opt_ktrace.h"
33 #include "opt_vm.h"
34 
35 #include <sys/param.h>
36 #include <sys/capability.h>
37 #include <sys/systm.h>
38 #include <sys/capability.h>
39 #include <sys/eventhandler.h>
40 #include <sys/lock.h>
41 #include <sys/mutex.h>
42 #include <sys/sysproto.h>
43 #include <sys/signalvar.h>
44 #include <sys/kernel.h>
45 #include <sys/mount.h>
46 #include <sys/filedesc.h>
47 #include <sys/fcntl.h>
48 #include <sys/acct.h>
49 #include <sys/exec.h>
50 #include <sys/imgact.h>
51 #include <sys/imgact_elf.h>
52 #include <sys/wait.h>
53 #include <sys/malloc.h>
54 #include <sys/priv.h>
55 #include <sys/proc.h>
56 #include <sys/pioctl.h>
57 #include <sys/namei.h>
58 #include <sys/resourcevar.h>
59 #include <sys/rwlock.h>
60 #include <sys/sched.h>
61 #include <sys/sdt.h>
62 #include <sys/sf_buf.h>
63 #include <sys/syscallsubr.h>
64 #include <sys/sysent.h>
65 #include <sys/shm.h>
66 #include <sys/sysctl.h>
67 #include <sys/vnode.h>
68 #include <sys/stat.h>
69 #ifdef KTRACE
70 #include <sys/ktrace.h>
71 #endif
72 
73 #include <vm/vm.h>
74 #include <vm/vm_param.h>
75 #include <vm/pmap.h>
76 #include <vm/vm_page.h>
77 #include <vm/vm_map.h>
78 #include <vm/vm_kern.h>
79 #include <vm/vm_extern.h>
80 #include <vm/vm_object.h>
81 #include <vm/vm_pager.h>
82 
83 #ifdef	HWPMC_HOOKS
84 #include <sys/pmckern.h>
85 #endif
86 
87 #include <machine/reg.h>
88 
89 #include <security/audit/audit.h>
90 #include <security/mac/mac_framework.h>
91 
92 #ifdef KDTRACE_HOOKS
93 #include <sys/dtrace_bsd.h>
94 dtrace_execexit_func_t	dtrace_fasttrap_exec;
95 #endif
96 
97 SDT_PROVIDER_DECLARE(proc);
98 SDT_PROBE_DEFINE1(proc, kernel, , exec, "char *");
99 SDT_PROBE_DEFINE1(proc, kernel, , exec__failure, "int");
100 SDT_PROBE_DEFINE1(proc, kernel, , exec__success, "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 
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     CTLFLAG_CAPRD, 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 disallow_high_osrel;
126 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW,
127     &disallow_high_osrel, 0,
128     "Disallow execution of binaries built for higher version of the world");
129 
130 static int map_at_zero = 0;
131 TUNABLE_INT("security.bsd.map_at_zero", &map_at_zero);
132 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RW, &map_at_zero, 0,
133     "Permit processes to map an object at virtual address 0.");
134 
135 static int
136 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
137 {
138 	struct proc *p;
139 	int error;
140 
141 	p = curproc;
142 #ifdef SCTL_MASK32
143 	if (req->flags & SCTL_MASK32) {
144 		unsigned int val;
145 		val = (unsigned int)p->p_sysent->sv_psstrings;
146 		error = SYSCTL_OUT(req, &val, sizeof(val));
147 	} else
148 #endif
149 		error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
150 		   sizeof(p->p_sysent->sv_psstrings));
151 	return error;
152 }
153 
154 static int
155 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
156 {
157 	struct proc *p;
158 	int error;
159 
160 	p = curproc;
161 #ifdef SCTL_MASK32
162 	if (req->flags & SCTL_MASK32) {
163 		unsigned int val;
164 		val = (unsigned int)p->p_sysent->sv_usrstack;
165 		error = SYSCTL_OUT(req, &val, sizeof(val));
166 	} else
167 #endif
168 		error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
169 		    sizeof(p->p_sysent->sv_usrstack));
170 	return error;
171 }
172 
173 static int
174 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
175 {
176 	struct proc *p;
177 
178 	p = curproc;
179 	return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
180 	    sizeof(p->p_sysent->sv_stackprot)));
181 }
182 
183 /*
184  * Each of the items is a pointer to a `const struct execsw', hence the
185  * double pointer here.
186  */
187 static const struct execsw **execsw;
188 
189 #ifndef _SYS_SYSPROTO_H_
190 struct execve_args {
191 	char    *fname;
192 	char    **argv;
193 	char    **envv;
194 };
195 #endif
196 
197 int
198 sys_execve(td, uap)
199 	struct thread *td;
200 	struct execve_args /* {
201 		char *fname;
202 		char **argv;
203 		char **envv;
204 	} */ *uap;
205 {
206 	int error;
207 	struct image_args args;
208 
209 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
210 	    uap->argv, uap->envv);
211 	if (error == 0)
212 		error = kern_execve(td, &args, NULL);
213 	return (error);
214 }
215 
216 #ifndef _SYS_SYSPROTO_H_
217 struct fexecve_args {
218 	int	fd;
219 	char	**argv;
220 	char	**envv;
221 }
222 #endif
223 int
224 sys_fexecve(struct thread *td, struct fexecve_args *uap)
225 {
226 	int error;
227 	struct image_args args;
228 
229 	error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
230 	    uap->argv, uap->envv);
231 	if (error == 0) {
232 		args.fd = uap->fd;
233 		error = kern_execve(td, &args, NULL);
234 	}
235 	return (error);
236 }
237 
238 #ifndef _SYS_SYSPROTO_H_
239 struct __mac_execve_args {
240 	char	*fname;
241 	char	**argv;
242 	char	**envv;
243 	struct mac	*mac_p;
244 };
245 #endif
246 
247 int
248 sys___mac_execve(td, uap)
249 	struct thread *td;
250 	struct __mac_execve_args /* {
251 		char *fname;
252 		char **argv;
253 		char **envv;
254 		struct mac *mac_p;
255 	} */ *uap;
256 {
257 #ifdef MAC
258 	int error;
259 	struct image_args args;
260 
261 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
262 	    uap->argv, uap->envv);
263 	if (error == 0)
264 		error = kern_execve(td, &args, uap->mac_p);
265 	return (error);
266 #else
267 	return (ENOSYS);
268 #endif
269 }
270 
271 /*
272  * XXX: kern_execve has the astonishing property of not always returning to
273  * the caller.  If sufficiently bad things happen during the call to
274  * do_execve(), it can end up calling exit1(); as a result, callers must
275  * avoid doing anything which they might need to undo (e.g., allocating
276  * memory).
277  */
278 int
279 kern_execve(td, args, mac_p)
280 	struct thread *td;
281 	struct image_args *args;
282 	struct mac *mac_p;
283 {
284 	struct proc *p = td->td_proc;
285 	int error;
286 
287 	AUDIT_ARG_ARGV(args->begin_argv, args->argc,
288 	    args->begin_envv - args->begin_argv);
289 	AUDIT_ARG_ENVV(args->begin_envv, args->envc,
290 	    args->endp - args->begin_envv);
291 	if (p->p_flag & P_HADTHREADS) {
292 		PROC_LOCK(p);
293 		if (thread_single(SINGLE_BOUNDARY)) {
294 			PROC_UNLOCK(p);
295 	       		exec_free_args(args);
296 			return (ERESTART);	/* Try again later. */
297 		}
298 		PROC_UNLOCK(p);
299 	}
300 
301 	error = do_execve(td, args, mac_p);
302 
303 	if (p->p_flag & P_HADTHREADS) {
304 		PROC_LOCK(p);
305 		/*
306 		 * If success, we upgrade to SINGLE_EXIT state to
307 		 * force other threads to suicide.
308 		 */
309 		if (error == 0)
310 			thread_single(SINGLE_EXIT);
311 		else
312 			thread_single_end();
313 		PROC_UNLOCK(p);
314 	}
315 
316 	return (error);
317 }
318 
319 /*
320  * In-kernel implementation of execve().  All arguments are assumed to be
321  * userspace pointers from the passed thread.
322  */
323 static int
324 do_execve(td, args, mac_p)
325 	struct thread *td;
326 	struct image_args *args;
327 	struct mac *mac_p;
328 {
329 	struct proc *p = td->td_proc;
330 	struct nameidata nd;
331 	struct ucred *newcred = NULL, *oldcred;
332 	struct uidinfo *euip;
333 	register_t *stack_base;
334 	int error, i;
335 	struct image_params image_params, *imgp;
336 	struct vattr attr;
337 	int (*img_first)(struct image_params *);
338 	struct pargs *oldargs = NULL, *newargs = NULL;
339 	struct sigacts *oldsigacts, *newsigacts;
340 #ifdef KTRACE
341 	struct vnode *tracevp = NULL;
342 	struct ucred *tracecred = NULL;
343 #endif
344 	struct vnode *textvp = NULL, *binvp = NULL;
345 	cap_rights_t rights;
346 	int credential_changing;
347 	int textset;
348 #ifdef MAC
349 	struct label *interpvplabel = NULL;
350 	int will_transition;
351 #endif
352 #ifdef HWPMC_HOOKS
353 	struct pmckern_procexec pe;
354 #endif
355 	static const char fexecv_proc_title[] = "(fexecv)";
356 
357 	imgp = &image_params;
358 
359 	/*
360 	 * Lock the process and set the P_INEXEC flag to indicate that
361 	 * it should be left alone until we're done here.  This is
362 	 * necessary to avoid race conditions - e.g. in ptrace() -
363 	 * that might allow a local user to illicitly obtain elevated
364 	 * privileges.
365 	 */
366 	PROC_LOCK(p);
367 	KASSERT((p->p_flag & P_INEXEC) == 0,
368 	    ("%s(): process already has P_INEXEC flag", __func__));
369 	p->p_flag |= P_INEXEC;
370 	PROC_UNLOCK(p);
371 
372 	/*
373 	 * Initialize part of the common data
374 	 */
375 	imgp->proc = p;
376 	imgp->execlabel = NULL;
377 	imgp->attr = &attr;
378 	imgp->entry_addr = 0;
379 	imgp->reloc_base = 0;
380 	imgp->vmspace_destroyed = 0;
381 	imgp->interpreted = 0;
382 	imgp->opened = 0;
383 	imgp->interpreter_name = NULL;
384 	imgp->auxargs = NULL;
385 	imgp->vp = NULL;
386 	imgp->object = NULL;
387 	imgp->firstpage = NULL;
388 	imgp->ps_strings = 0;
389 	imgp->auxarg_size = 0;
390 	imgp->args = args;
391 	imgp->execpath = imgp->freepath = NULL;
392 	imgp->execpathp = 0;
393 	imgp->canary = 0;
394 	imgp->canarylen = 0;
395 	imgp->pagesizes = 0;
396 	imgp->pagesizeslen = 0;
397 	imgp->stack_prot = 0;
398 
399 #ifdef MAC
400 	error = mac_execve_enter(imgp, mac_p);
401 	if (error)
402 		goto exec_fail;
403 #endif
404 
405 	imgp->image_header = NULL;
406 
407 	/*
408 	 * Translate the file name. namei() returns a vnode pointer
409 	 *	in ni_vp amoung other things.
410 	 *
411 	 * XXXAUDIT: It would be desirable to also audit the name of the
412 	 * interpreter if this is an interpreted binary.
413 	 */
414 	if (args->fname != NULL) {
415 		NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME
416 		    | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
417 	}
418 
419 	SDT_PROBE(proc, kernel, , exec, args->fname, 0, 0, 0, 0 );
420 
421 interpret:
422 	if (args->fname != NULL) {
423 #ifdef CAPABILITY_MODE
424 		/*
425 		 * While capability mode can't reach this point via direct
426 		 * path arguments to execve(), we also don't allow
427 		 * interpreters to be used in capability mode (for now).
428 		 * Catch indirect lookups and return a permissions error.
429 		 */
430 		if (IN_CAPABILITY_MODE(td)) {
431 			error = ECAPMODE;
432 			goto exec_fail;
433 		}
434 #endif
435 		error = namei(&nd);
436 		if (error)
437 			goto exec_fail;
438 
439 		binvp  = nd.ni_vp;
440 		imgp->vp = binvp;
441 	} else {
442 		AUDIT_ARG_FD(args->fd);
443 		/*
444 		 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
445 		 */
446 		error = fgetvp_exec(td, args->fd,
447 		    cap_rights_init(&rights, CAP_FEXECVE), &binvp);
448 		if (error)
449 			goto exec_fail;
450 		vn_lock(binvp, LK_EXCLUSIVE | LK_RETRY);
451 		AUDIT_ARG_VNODE1(binvp);
452 		imgp->vp = binvp;
453 	}
454 
455 	/*
456 	 * Check file permissions (also 'opens' file)
457 	 */
458 	error = exec_check_permissions(imgp);
459 	if (error)
460 		goto exec_fail_dealloc;
461 
462 	imgp->object = imgp->vp->v_object;
463 	if (imgp->object != NULL)
464 		vm_object_reference(imgp->object);
465 
466 	/*
467 	 * Set VV_TEXT now so no one can write to the executable while we're
468 	 * activating it.
469 	 *
470 	 * Remember if this was set before and unset it in case this is not
471 	 * actually an executable image.
472 	 */
473 	textset = VOP_IS_TEXT(imgp->vp);
474 	VOP_SET_TEXT(imgp->vp);
475 
476 	error = exec_map_first_page(imgp);
477 	if (error)
478 		goto exec_fail_dealloc;
479 
480 	imgp->proc->p_osrel = 0;
481 	/*
482 	 *	If the current process has a special image activator it
483 	 *	wants to try first, call it.   For example, emulating shell
484 	 *	scripts differently.
485 	 */
486 	error = -1;
487 	if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
488 		error = img_first(imgp);
489 
490 	/*
491 	 *	Loop through the list of image activators, calling each one.
492 	 *	An activator returns -1 if there is no match, 0 on success,
493 	 *	and an error otherwise.
494 	 */
495 	for (i = 0; error == -1 && execsw[i]; ++i) {
496 		if (execsw[i]->ex_imgact == NULL ||
497 		    execsw[i]->ex_imgact == img_first) {
498 			continue;
499 		}
500 		error = (*execsw[i]->ex_imgact)(imgp);
501 	}
502 
503 	if (error) {
504 		if (error == -1) {
505 			if (textset == 0)
506 				VOP_UNSET_TEXT(imgp->vp);
507 			error = ENOEXEC;
508 		}
509 		goto exec_fail_dealloc;
510 	}
511 
512 	/*
513 	 * Special interpreter operation, cleanup and loop up to try to
514 	 * activate the interpreter.
515 	 */
516 	if (imgp->interpreted) {
517 		exec_unmap_first_page(imgp);
518 		/*
519 		 * VV_TEXT needs to be unset for scripts.  There is a short
520 		 * period before we determine that something is a script where
521 		 * VV_TEXT will be set. The vnode lock is held over this
522 		 * entire period so nothing should illegitimately be blocked.
523 		 */
524 		VOP_UNSET_TEXT(imgp->vp);
525 		/* free name buffer and old vnode */
526 		if (args->fname != NULL)
527 			NDFREE(&nd, NDF_ONLY_PNBUF);
528 #ifdef MAC
529 		mac_execve_interpreter_enter(binvp, &interpvplabel);
530 #endif
531 		if (imgp->opened) {
532 			VOP_CLOSE(binvp, FREAD, td->td_ucred, td);
533 			imgp->opened = 0;
534 		}
535 		vput(binvp);
536 		vm_object_deallocate(imgp->object);
537 		imgp->object = NULL;
538 		/* set new name to that of the interpreter */
539 		NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
540 		    UIO_SYSSPACE, imgp->interpreter_name, td);
541 		args->fname = imgp->interpreter_name;
542 		goto interpret;
543 	}
544 
545 	/*
546 	 * NB: We unlock the vnode here because it is believed that none
547 	 * of the sv_copyout_strings/sv_fixup operations require the vnode.
548 	 */
549 	VOP_UNLOCK(imgp->vp, 0);
550 
551 	/*
552 	 * Do the best to calculate the full path to the image file.
553 	 */
554 	if (imgp->auxargs != NULL &&
555 	    ((args->fname != NULL && args->fname[0] == '/') ||
556 	     vn_fullpath(td, imgp->vp, &imgp->execpath, &imgp->freepath) != 0))
557 		imgp->execpath = args->fname;
558 
559 	if (disallow_high_osrel &&
560 	    P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
561 		error = ENOEXEC;
562 		uprintf("Osrel %d for image %s too high\n", p->p_osrel,
563 		    imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
564 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
565 		goto exec_fail_dealloc;
566 	}
567 
568 	/*
569 	 * Copy out strings (args and env) and initialize stack base
570 	 */
571 	if (p->p_sysent->sv_copyout_strings)
572 		stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
573 	else
574 		stack_base = exec_copyout_strings(imgp);
575 
576 	/*
577 	 * If custom stack fixup routine present for this process
578 	 * let it do the stack setup.
579 	 * Else stuff argument count as first item on stack
580 	 */
581 	if (p->p_sysent->sv_fixup != NULL)
582 		(*p->p_sysent->sv_fixup)(&stack_base, imgp);
583 	else
584 		suword(--stack_base, imgp->args->argc);
585 
586 	/*
587 	 * For security and other reasons, the file descriptor table cannot
588 	 * be shared after an exec.
589 	 */
590 	fdunshare(p, td);
591 
592 	/*
593 	 * Malloc things before we need locks.
594 	 */
595 	newcred = crget();
596 	euip = uifind(attr.va_uid);
597 	i = imgp->args->begin_envv - imgp->args->begin_argv;
598 	/* Cache arguments if they fit inside our allowance */
599 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
600 		newargs = pargs_alloc(i);
601 		bcopy(imgp->args->begin_argv, newargs->ar_args, i);
602 	}
603 
604 	/* close files on exec */
605 	fdcloseexec(td);
606 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
607 
608 	/* Get a reference to the vnode prior to locking the proc */
609 	VREF(binvp);
610 
611 	/*
612 	 * For security and other reasons, signal handlers cannot
613 	 * be shared after an exec. The new process gets a copy of the old
614 	 * handlers. In execsigs(), the new process will have its signals
615 	 * reset.
616 	 */
617 	PROC_LOCK(p);
618 	oldcred = crcopysafe(p, newcred);
619 	if (sigacts_shared(p->p_sigacts)) {
620 		oldsigacts = p->p_sigacts;
621 		PROC_UNLOCK(p);
622 		newsigacts = sigacts_alloc();
623 		sigacts_copy(newsigacts, oldsigacts);
624 		PROC_LOCK(p);
625 		p->p_sigacts = newsigacts;
626 	} else
627 		oldsigacts = NULL;
628 
629 	/* Stop profiling */
630 	stopprofclock(p);
631 
632 	/* reset caught signals */
633 	execsigs(p);
634 
635 	/* name this process - nameiexec(p, ndp) */
636 	bzero(p->p_comm, sizeof(p->p_comm));
637 	if (args->fname)
638 		bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
639 		    min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
640 	else if (vn_commname(binvp, p->p_comm, sizeof(p->p_comm)) != 0)
641 		bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
642 	bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
643 #ifdef KTR
644 	sched_clear_tdname(td);
645 #endif
646 
647 	/*
648 	 * mark as execed, wakeup the process that vforked (if any) and tell
649 	 * it that it now has its own resources back
650 	 */
651 	p->p_flag |= P_EXEC;
652 	if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
653 		p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
654 		cv_broadcast(&p->p_pwait);
655 	}
656 
657 	/*
658 	 * Implement image setuid/setgid.
659 	 *
660 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
661 	 * the process is being traced.
662 	 *
663 	 * We disable setuid/setgid/etc in compatibility mode on the basis
664 	 * that most setugid applications are not written with that
665 	 * environment in mind, and will therefore almost certainly operate
666 	 * incorrectly. In principle there's no reason that setugid
667 	 * applications might not be useful in capability mode, so we may want
668 	 * to reconsider this conservative design choice in the future.
669 	 *
670 	 * XXXMAC: For the time being, use NOSUID to also prohibit
671 	 * transitions on the file system.
672 	 */
673 	credential_changing = 0;
674 	credential_changing |= (attr.va_mode & S_ISUID) && oldcred->cr_uid !=
675 	    attr.va_uid;
676 	credential_changing |= (attr.va_mode & S_ISGID) && oldcred->cr_gid !=
677 	    attr.va_gid;
678 #ifdef MAC
679 	will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
680 	    interpvplabel, imgp);
681 	credential_changing |= will_transition;
682 #endif
683 
684 	if (credential_changing &&
685 #ifdef CAPABILITY_MODE
686 	    ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
687 #endif
688 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
689 	    (p->p_flag & P_TRACED) == 0) {
690 		/*
691 		 * Turn off syscall tracing for set-id programs, except for
692 		 * root.  Record any set-id flags first to make sure that
693 		 * we do not regain any tracing during a possible block.
694 		 */
695 		setsugid(p);
696 
697 #ifdef KTRACE
698 		if (p->p_tracecred != NULL &&
699 		    priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED, 0))
700 			ktrprocexec(p, &tracecred, &tracevp);
701 #endif
702 		/*
703 		 * Close any file descriptors 0..2 that reference procfs,
704 		 * then make sure file descriptors 0..2 are in use.
705 		 *
706 		 * setugidsafety() may call closef() and then pfind()
707 		 * which may grab the process lock.
708 		 * fdcheckstd() may call falloc() which may block to
709 		 * allocate memory, so temporarily drop the process lock.
710 		 */
711 		PROC_UNLOCK(p);
712 		VOP_UNLOCK(imgp->vp, 0);
713 		setugidsafety(td);
714 		error = fdcheckstd(td);
715 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
716 		if (error != 0)
717 			goto done1;
718 		PROC_LOCK(p);
719 		/*
720 		 * Set the new credentials.
721 		 */
722 		if (attr.va_mode & S_ISUID)
723 			change_euid(newcred, euip);
724 		if (attr.va_mode & S_ISGID)
725 			change_egid(newcred, attr.va_gid);
726 #ifdef MAC
727 		if (will_transition) {
728 			mac_vnode_execve_transition(oldcred, newcred, imgp->vp,
729 			    interpvplabel, imgp);
730 		}
731 #endif
732 		/*
733 		 * Implement correct POSIX saved-id behavior.
734 		 *
735 		 * XXXMAC: Note that the current logic will save the
736 		 * uid and gid if a MAC domain transition occurs, even
737 		 * though maybe it shouldn't.
738 		 */
739 		change_svuid(newcred, newcred->cr_uid);
740 		change_svgid(newcred, newcred->cr_gid);
741 		p->p_ucred = newcred;
742 		newcred = NULL;
743 	} else {
744 		if (oldcred->cr_uid == oldcred->cr_ruid &&
745 		    oldcred->cr_gid == oldcred->cr_rgid)
746 			p->p_flag &= ~P_SUGID;
747 		/*
748 		 * Implement correct POSIX saved-id behavior.
749 		 *
750 		 * XXX: It's not clear that the existing behavior is
751 		 * POSIX-compliant.  A number of sources indicate that the
752 		 * saved uid/gid should only be updated if the new ruid is
753 		 * not equal to the old ruid, or the new euid is not equal
754 		 * to the old euid and the new euid is not equal to the old
755 		 * ruid.  The FreeBSD code always updates the saved uid/gid.
756 		 * Also, this code uses the new (replaced) euid and egid as
757 		 * the source, which may or may not be the right ones to use.
758 		 */
759 		if (oldcred->cr_svuid != oldcred->cr_uid ||
760 		    oldcred->cr_svgid != oldcred->cr_gid) {
761 			change_svuid(newcred, newcred->cr_uid);
762 			change_svgid(newcred, newcred->cr_gid);
763 			p->p_ucred = newcred;
764 			newcred = NULL;
765 		}
766 	}
767 
768 	/*
769 	 * Store the vp for use in procfs.  This vnode was referenced prior
770 	 * to locking the proc lock.
771 	 */
772 	textvp = p->p_textvp;
773 	p->p_textvp = binvp;
774 
775 #ifdef KDTRACE_HOOKS
776 	/*
777 	 * Tell the DTrace fasttrap provider about the exec if it
778 	 * has declared an interest.
779 	 */
780 	if (dtrace_fasttrap_exec)
781 		dtrace_fasttrap_exec(p);
782 #endif
783 
784 	/*
785 	 * Notify others that we exec'd, and clear the P_INEXEC flag
786 	 * as we're now a bona fide freshly-execed process.
787 	 */
788 	KNOTE_LOCKED(&p->p_klist, NOTE_EXEC);
789 	p->p_flag &= ~P_INEXEC;
790 
791 	/* clear "fork but no exec" flag, as we _are_ execing */
792 	p->p_acflag &= ~AFORK;
793 
794 	/*
795 	 * Free any previous argument cache and replace it with
796 	 * the new argument cache, if any.
797 	 */
798 	oldargs = p->p_args;
799 	p->p_args = newargs;
800 	newargs = NULL;
801 
802 #ifdef	HWPMC_HOOKS
803 	/*
804 	 * Check if system-wide sampling is in effect or if the
805 	 * current process is using PMCs.  If so, do exec() time
806 	 * processing.  This processing needs to happen AFTER the
807 	 * P_INEXEC flag is cleared.
808 	 *
809 	 * The proc lock needs to be released before taking the PMC
810 	 * SX.
811 	 */
812 	if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
813 		PROC_UNLOCK(p);
814 		VOP_UNLOCK(imgp->vp, 0);
815 		pe.pm_credentialschanged = credential_changing;
816 		pe.pm_entryaddr = imgp->entry_addr;
817 
818 		PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
819 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
820 	} else
821 		PROC_UNLOCK(p);
822 #else  /* !HWPMC_HOOKS */
823 	PROC_UNLOCK(p);
824 #endif
825 
826 	/* Set values passed into the program in registers. */
827 	if (p->p_sysent->sv_setregs)
828 		(*p->p_sysent->sv_setregs)(td, imgp,
829 		    (u_long)(uintptr_t)stack_base);
830 	else
831 		exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base);
832 
833 	vfs_mark_atime(imgp->vp, td->td_ucred);
834 
835 	SDT_PROBE(proc, kernel, , exec__success, args->fname, 0, 0, 0, 0);
836 
837 done1:
838 	/*
839 	 * Free any resources malloc'd earlier that we didn't use.
840 	 */
841 	uifree(euip);
842 	if (newcred == NULL)
843 		crfree(oldcred);
844 	else
845 		crfree(newcred);
846 	VOP_UNLOCK(imgp->vp, 0);
847 
848 	/*
849 	 * Handle deferred decrement of ref counts.
850 	 */
851 	if (textvp != NULL)
852 		vrele(textvp);
853 	if (binvp && error != 0)
854 		vrele(binvp);
855 #ifdef KTRACE
856 	if (tracevp != NULL)
857 		vrele(tracevp);
858 	if (tracecred != NULL)
859 		crfree(tracecred);
860 #endif
861 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
862 	pargs_drop(oldargs);
863 	pargs_drop(newargs);
864 	if (oldsigacts != NULL)
865 		sigacts_free(oldsigacts);
866 
867 exec_fail_dealloc:
868 
869 	/*
870 	 * free various allocated resources
871 	 */
872 	if (imgp->firstpage != NULL)
873 		exec_unmap_first_page(imgp);
874 
875 	if (imgp->vp != NULL) {
876 		if (args->fname)
877 			NDFREE(&nd, NDF_ONLY_PNBUF);
878 		if (imgp->opened)
879 			VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
880 		vput(imgp->vp);
881 	}
882 
883 	if (imgp->object != NULL)
884 		vm_object_deallocate(imgp->object);
885 
886 	free(imgp->freepath, M_TEMP);
887 
888 	if (error == 0) {
889 		PROC_LOCK(p);
890 		td->td_dbgflags |= TDB_EXEC;
891 		PROC_UNLOCK(p);
892 
893 		/*
894 		 * Stop the process here if its stop event mask has
895 		 * the S_EXEC bit set.
896 		 */
897 		STOPEVENT(p, S_EXEC, 0);
898 		goto done2;
899 	}
900 
901 exec_fail:
902 	/* we're done here, clear P_INEXEC */
903 	PROC_LOCK(p);
904 	p->p_flag &= ~P_INEXEC;
905 	PROC_UNLOCK(p);
906 
907 	SDT_PROBE(proc, kernel, , exec__failure, error, 0, 0, 0, 0);
908 
909 done2:
910 #ifdef MAC
911 	mac_execve_exit(imgp);
912 	mac_execve_interpreter_exit(interpvplabel);
913 #endif
914 	exec_free_args(args);
915 
916 	if (error && imgp->vmspace_destroyed) {
917 		/* sorry, no more process anymore. exit gracefully */
918 		exit1(td, W_EXITCODE(0, SIGABRT));
919 		/* NOT REACHED */
920 	}
921 
922 #ifdef KTRACE
923 	if (error == 0)
924 		ktrprocctor(p);
925 #endif
926 
927 	return (error);
928 }
929 
930 int
931 exec_map_first_page(imgp)
932 	struct image_params *imgp;
933 {
934 	int rv, i;
935 	int initial_pagein;
936 	vm_page_t ma[VM_INITIAL_PAGEIN];
937 	vm_object_t object;
938 
939 	if (imgp->firstpage != NULL)
940 		exec_unmap_first_page(imgp);
941 
942 	object = imgp->vp->v_object;
943 	if (object == NULL)
944 		return (EACCES);
945 	VM_OBJECT_WLOCK(object);
946 #if VM_NRESERVLEVEL > 0
947 	if ((object->flags & OBJ_COLORED) == 0) {
948 		object->flags |= OBJ_COLORED;
949 		object->pg_color = 0;
950 	}
951 #endif
952 	ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL);
953 	if (ma[0]->valid != VM_PAGE_BITS_ALL) {
954 		initial_pagein = VM_INITIAL_PAGEIN;
955 		if (initial_pagein > object->size)
956 			initial_pagein = object->size;
957 		for (i = 1; i < initial_pagein; i++) {
958 			if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) {
959 				if (ma[i]->valid)
960 					break;
961 				if (vm_page_tryxbusy(ma[i]))
962 					break;
963 			} else {
964 				ma[i] = vm_page_alloc(object, i,
965 				    VM_ALLOC_NORMAL | VM_ALLOC_IFNOTCACHED);
966 				if (ma[i] == NULL)
967 					break;
968 			}
969 		}
970 		initial_pagein = i;
971 		rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
972 		ma[0] = vm_page_lookup(object, 0);
973 		if ((rv != VM_PAGER_OK) || (ma[0] == NULL)) {
974 			if (ma[0] != NULL) {
975 				vm_page_lock(ma[0]);
976 				vm_page_free(ma[0]);
977 				vm_page_unlock(ma[0]);
978 			}
979 			VM_OBJECT_WUNLOCK(object);
980 			return (EIO);
981 		}
982 	}
983 	vm_page_xunbusy(ma[0]);
984 	vm_page_lock(ma[0]);
985 	vm_page_hold(ma[0]);
986 	vm_page_unlock(ma[0]);
987 	VM_OBJECT_WUNLOCK(object);
988 
989 	imgp->firstpage = sf_buf_alloc(ma[0], 0);
990 	imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
991 
992 	return (0);
993 }
994 
995 void
996 exec_unmap_first_page(imgp)
997 	struct image_params *imgp;
998 {
999 	vm_page_t m;
1000 
1001 	if (imgp->firstpage != NULL) {
1002 		m = sf_buf_page(imgp->firstpage);
1003 		sf_buf_free(imgp->firstpage);
1004 		imgp->firstpage = NULL;
1005 		vm_page_lock(m);
1006 		vm_page_unhold(m);
1007 		vm_page_unlock(m);
1008 	}
1009 }
1010 
1011 /*
1012  * Destroy old address space, and allocate a new stack
1013  *	The new stack is only SGROWSIZ large because it is grown
1014  *	automatically in trap.c.
1015  */
1016 int
1017 exec_new_vmspace(imgp, sv)
1018 	struct image_params *imgp;
1019 	struct sysentvec *sv;
1020 {
1021 	int error;
1022 	struct proc *p = imgp->proc;
1023 	struct vmspace *vmspace = p->p_vmspace;
1024 	vm_object_t obj;
1025 	vm_offset_t sv_minuser, stack_addr;
1026 	vm_map_t map;
1027 	u_long ssiz;
1028 
1029 	imgp->vmspace_destroyed = 1;
1030 	imgp->sysent = sv;
1031 
1032 	/* May be called with Giant held */
1033 	EVENTHANDLER_INVOKE(process_exec, p, imgp);
1034 
1035 	/*
1036 	 * Blow away entire process VM, if address space not shared,
1037 	 * otherwise, create a new VM space so that other threads are
1038 	 * not disrupted
1039 	 */
1040 	map = &vmspace->vm_map;
1041 	if (map_at_zero)
1042 		sv_minuser = sv->sv_minuser;
1043 	else
1044 		sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1045 	if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser &&
1046 	    vm_map_max(map) == sv->sv_maxuser) {
1047 		shmexit(vmspace);
1048 		pmap_remove_pages(vmspace_pmap(vmspace));
1049 		vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1050 	} else {
1051 		error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1052 		if (error)
1053 			return (error);
1054 		vmspace = p->p_vmspace;
1055 		map = &vmspace->vm_map;
1056 	}
1057 
1058 	/* Map a shared page */
1059 	obj = sv->sv_shared_page_obj;
1060 	if (obj != NULL) {
1061 		vm_object_reference(obj);
1062 		error = vm_map_fixed(map, obj, 0,
1063 		    sv->sv_shared_page_base, sv->sv_shared_page_len,
1064 		    VM_PROT_READ | VM_PROT_EXECUTE,
1065 		    VM_PROT_READ | VM_PROT_EXECUTE,
1066 		    MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1067 		if (error) {
1068 			vm_object_deallocate(obj);
1069 			return (error);
1070 		}
1071 	}
1072 
1073 	/* Allocate a new stack */
1074 	if (sv->sv_maxssiz != NULL)
1075 		ssiz = *sv->sv_maxssiz;
1076 	else
1077 		ssiz = maxssiz;
1078 	stack_addr = sv->sv_usrstack - ssiz;
1079 	error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1080 	    obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1081 		sv->sv_stackprot,
1082 	    VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1083 	if (error)
1084 		return (error);
1085 
1086 #ifdef __ia64__
1087 	/* Allocate a new register stack */
1088 	stack_addr = IA64_BACKINGSTORE;
1089 	error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1090 	    sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_UP);
1091 	if (error)
1092 		return (error);
1093 #endif
1094 
1095 	/* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
1096 	 * VM_STACK case, but they are still used to monitor the size of the
1097 	 * process stack so we can check the stack rlimit.
1098 	 */
1099 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1100 	vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - ssiz;
1101 
1102 	return (0);
1103 }
1104 
1105 /*
1106  * Copy out argument and environment strings from the old process address
1107  * space into the temporary string buffer.
1108  */
1109 int
1110 exec_copyin_args(struct image_args *args, char *fname,
1111     enum uio_seg segflg, char **argv, char **envv)
1112 {
1113 	char *argp, *envp;
1114 	int error;
1115 	size_t length;
1116 
1117 	bzero(args, sizeof(*args));
1118 	if (argv == NULL)
1119 		return (EFAULT);
1120 
1121 	/*
1122 	 * Allocate demand-paged memory for the file name, argument, and
1123 	 * environment strings.
1124 	 */
1125 	error = exec_alloc_args(args);
1126 	if (error != 0)
1127 		return (error);
1128 
1129 	/*
1130 	 * Copy the file name.
1131 	 */
1132 	if (fname != NULL) {
1133 		args->fname = args->buf;
1134 		error = (segflg == UIO_SYSSPACE) ?
1135 		    copystr(fname, args->fname, PATH_MAX, &length) :
1136 		    copyinstr(fname, args->fname, PATH_MAX, &length);
1137 		if (error != 0)
1138 			goto err_exit;
1139 	} else
1140 		length = 0;
1141 
1142 	args->begin_argv = args->buf + length;
1143 	args->endp = args->begin_argv;
1144 	args->stringspace = ARG_MAX;
1145 
1146 	/*
1147 	 * extract arguments first
1148 	 */
1149 	while ((argp = (caddr_t) (intptr_t) fuword(argv++))) {
1150 		if (argp == (caddr_t) -1) {
1151 			error = EFAULT;
1152 			goto err_exit;
1153 		}
1154 		if ((error = copyinstr(argp, args->endp,
1155 		    args->stringspace, &length))) {
1156 			if (error == ENAMETOOLONG)
1157 				error = E2BIG;
1158 			goto err_exit;
1159 		}
1160 		args->stringspace -= length;
1161 		args->endp += length;
1162 		args->argc++;
1163 	}
1164 
1165 	args->begin_envv = args->endp;
1166 
1167 	/*
1168 	 * extract environment strings
1169 	 */
1170 	if (envv) {
1171 		while ((envp = (caddr_t)(intptr_t)fuword(envv++))) {
1172 			if (envp == (caddr_t)-1) {
1173 				error = EFAULT;
1174 				goto err_exit;
1175 			}
1176 			if ((error = copyinstr(envp, args->endp,
1177 			    args->stringspace, &length))) {
1178 				if (error == ENAMETOOLONG)
1179 					error = E2BIG;
1180 				goto err_exit;
1181 			}
1182 			args->stringspace -= length;
1183 			args->endp += length;
1184 			args->envc++;
1185 		}
1186 	}
1187 
1188 	return (0);
1189 
1190 err_exit:
1191 	exec_free_args(args);
1192 	return (error);
1193 }
1194 
1195 /*
1196  * Allocate temporary demand-paged, zero-filled memory for the file name,
1197  * argument, and environment strings.  Returns zero if the allocation succeeds
1198  * and ENOMEM otherwise.
1199  */
1200 int
1201 exec_alloc_args(struct image_args *args)
1202 {
1203 
1204 	args->buf = (char *)kmap_alloc_wait(exec_map, PATH_MAX + ARG_MAX);
1205 	return (args->buf != NULL ? 0 : ENOMEM);
1206 }
1207 
1208 void
1209 exec_free_args(struct image_args *args)
1210 {
1211 
1212 	if (args->buf != NULL) {
1213 		kmap_free_wakeup(exec_map, (vm_offset_t)args->buf,
1214 		    PATH_MAX + ARG_MAX);
1215 		args->buf = NULL;
1216 	}
1217 	if (args->fname_buf != NULL) {
1218 		free(args->fname_buf, M_TEMP);
1219 		args->fname_buf = NULL;
1220 	}
1221 }
1222 
1223 /*
1224  * Copy strings out to the new process address space, constructing new arg
1225  * and env vector tables. Return a pointer to the base so that it can be used
1226  * as the initial stack pointer.
1227  */
1228 register_t *
1229 exec_copyout_strings(imgp)
1230 	struct image_params *imgp;
1231 {
1232 	int argc, envc;
1233 	char **vectp;
1234 	char *stringp, *destp;
1235 	register_t *stack_base;
1236 	struct ps_strings *arginfo;
1237 	struct proc *p;
1238 	size_t execpath_len;
1239 	int szsigcode, szps;
1240 	char canary[sizeof(long) * 8];
1241 
1242 	szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1243 	/*
1244 	 * Calculate string base and vector table pointers.
1245 	 * Also deal with signal trampoline code for this exec type.
1246 	 */
1247 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
1248 		execpath_len = strlen(imgp->execpath) + 1;
1249 	else
1250 		execpath_len = 0;
1251 	p = imgp->proc;
1252 	szsigcode = 0;
1253 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1254 	if (p->p_sysent->sv_sigcode_base == 0) {
1255 		if (p->p_sysent->sv_szsigcode != NULL)
1256 			szsigcode = *(p->p_sysent->sv_szsigcode);
1257 	}
1258 	destp =	(caddr_t)arginfo - szsigcode - SPARE_USRSPACE -
1259 	    roundup(execpath_len, sizeof(char *)) -
1260 	    roundup(sizeof(canary), sizeof(char *)) -
1261 	    roundup(szps, sizeof(char *)) -
1262 	    roundup((ARG_MAX - imgp->args->stringspace), sizeof(char *));
1263 
1264 	/*
1265 	 * install sigcode
1266 	 */
1267 	if (szsigcode != 0)
1268 		copyout(p->p_sysent->sv_sigcode, ((caddr_t)arginfo -
1269 		    szsigcode), szsigcode);
1270 
1271 	/*
1272 	 * Copy the image path for the rtld.
1273 	 */
1274 	if (execpath_len != 0) {
1275 		imgp->execpathp = (uintptr_t)arginfo - szsigcode - execpath_len;
1276 		copyout(imgp->execpath, (void *)imgp->execpathp,
1277 		    execpath_len);
1278 	}
1279 
1280 	/*
1281 	 * Prepare the canary for SSP.
1282 	 */
1283 	arc4rand(canary, sizeof(canary), 0);
1284 	imgp->canary = (uintptr_t)arginfo - szsigcode - execpath_len -
1285 	    sizeof(canary);
1286 	copyout(canary, (void *)imgp->canary, sizeof(canary));
1287 	imgp->canarylen = sizeof(canary);
1288 
1289 	/*
1290 	 * Prepare the pagesizes array.
1291 	 */
1292 	imgp->pagesizes = (uintptr_t)arginfo - szsigcode - execpath_len -
1293 	    roundup(sizeof(canary), sizeof(char *)) - szps;
1294 	copyout(pagesizes, (void *)imgp->pagesizes, szps);
1295 	imgp->pagesizeslen = szps;
1296 
1297 	/*
1298 	 * If we have a valid auxargs ptr, prepare some room
1299 	 * on the stack.
1300 	 */
1301 	if (imgp->auxargs) {
1302 		/*
1303 		 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1304 		 * lower compatibility.
1305 		 */
1306 		imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1307 		    (AT_COUNT * 2);
1308 		/*
1309 		 * The '+ 2' is for the null pointers at the end of each of
1310 		 * the arg and env vector sets,and imgp->auxarg_size is room
1311 		 * for argument of Runtime loader.
1312 		 */
1313 		vectp = (char **)(destp - (imgp->args->argc +
1314 		    imgp->args->envc + 2 + imgp->auxarg_size)
1315 		    * sizeof(char *));
1316 	} else {
1317 		/*
1318 		 * The '+ 2' is for the null pointers at the end of each of
1319 		 * the arg and env vector sets
1320 		 */
1321 		vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc + 2) *
1322 		    sizeof(char *));
1323 	}
1324 
1325 	/*
1326 	 * vectp also becomes our initial stack base
1327 	 */
1328 	stack_base = (register_t *)vectp;
1329 
1330 	stringp = imgp->args->begin_argv;
1331 	argc = imgp->args->argc;
1332 	envc = imgp->args->envc;
1333 
1334 	/*
1335 	 * Copy out strings - arguments and environment.
1336 	 */
1337 	copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
1338 
1339 	/*
1340 	 * Fill in "ps_strings" struct for ps, w, etc.
1341 	 */
1342 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1343 	suword32(&arginfo->ps_nargvstr, argc);
1344 
1345 	/*
1346 	 * Fill in argument portion of vector table.
1347 	 */
1348 	for (; argc > 0; --argc) {
1349 		suword(vectp++, (long)(intptr_t)destp);
1350 		while (*stringp++ != 0)
1351 			destp++;
1352 		destp++;
1353 	}
1354 
1355 	/* a null vector table pointer separates the argp's from the envp's */
1356 	suword(vectp++, 0);
1357 
1358 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1359 	suword32(&arginfo->ps_nenvstr, envc);
1360 
1361 	/*
1362 	 * Fill in environment portion of vector table.
1363 	 */
1364 	for (; envc > 0; --envc) {
1365 		suword(vectp++, (long)(intptr_t)destp);
1366 		while (*stringp++ != 0)
1367 			destp++;
1368 		destp++;
1369 	}
1370 
1371 	/* end of vector table is a null pointer */
1372 	suword(vectp, 0);
1373 
1374 	return (stack_base);
1375 }
1376 
1377 /*
1378  * Check permissions of file to execute.
1379  *	Called with imgp->vp locked.
1380  *	Return 0 for success or error code on failure.
1381  */
1382 int
1383 exec_check_permissions(imgp)
1384 	struct image_params *imgp;
1385 {
1386 	struct vnode *vp = imgp->vp;
1387 	struct vattr *attr = imgp->attr;
1388 	struct thread *td;
1389 	int error, writecount;
1390 
1391 	td = curthread;
1392 
1393 	/* Get file attributes */
1394 	error = VOP_GETATTR(vp, attr, td->td_ucred);
1395 	if (error)
1396 		return (error);
1397 
1398 #ifdef MAC
1399 	error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1400 	if (error)
1401 		return (error);
1402 #endif
1403 
1404 	/*
1405 	 * 1) Check if file execution is disabled for the filesystem that
1406 	 *    this file resides on.
1407 	 * 2) Ensure that at least one execute bit is on. Otherwise, a
1408 	 *    privileged user will always succeed, and we don't want this
1409 	 *    to happen unless the file really is executable.
1410 	 * 3) Ensure that the file is a regular file.
1411 	 */
1412 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1413 	    (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1414 	    (attr->va_type != VREG))
1415 		return (EACCES);
1416 
1417 	/*
1418 	 * Zero length files can't be exec'd
1419 	 */
1420 	if (attr->va_size == 0)
1421 		return (ENOEXEC);
1422 
1423 	/*
1424 	 *  Check for execute permission to file based on current credentials.
1425 	 */
1426 	error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1427 	if (error)
1428 		return (error);
1429 
1430 	/*
1431 	 * Check number of open-for-writes on the file and deny execution
1432 	 * if there are any.
1433 	 */
1434 	error = VOP_GET_WRITECOUNT(vp, &writecount);
1435 	if (error != 0)
1436 		return (error);
1437 	if (writecount != 0)
1438 		return (ETXTBSY);
1439 
1440 	/*
1441 	 * Call filesystem specific open routine (which does nothing in the
1442 	 * general case).
1443 	 */
1444 	error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1445 	if (error == 0)
1446 		imgp->opened = 1;
1447 	return (error);
1448 }
1449 
1450 /*
1451  * Exec handler registration
1452  */
1453 int
1454 exec_register(execsw_arg)
1455 	const struct execsw *execsw_arg;
1456 {
1457 	const struct execsw **es, **xs, **newexecsw;
1458 	int count = 2;	/* New slot and trailing NULL */
1459 
1460 	if (execsw)
1461 		for (es = execsw; *es; es++)
1462 			count++;
1463 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1464 	if (newexecsw == NULL)
1465 		return (ENOMEM);
1466 	xs = newexecsw;
1467 	if (execsw)
1468 		for (es = execsw; *es; es++)
1469 			*xs++ = *es;
1470 	*xs++ = execsw_arg;
1471 	*xs = NULL;
1472 	if (execsw)
1473 		free(execsw, M_TEMP);
1474 	execsw = newexecsw;
1475 	return (0);
1476 }
1477 
1478 int
1479 exec_unregister(execsw_arg)
1480 	const struct execsw *execsw_arg;
1481 {
1482 	const struct execsw **es, **xs, **newexecsw;
1483 	int count = 1;
1484 
1485 	if (execsw == NULL)
1486 		panic("unregister with no handlers left?\n");
1487 
1488 	for (es = execsw; *es; es++) {
1489 		if (*es == execsw_arg)
1490 			break;
1491 	}
1492 	if (*es == NULL)
1493 		return (ENOENT);
1494 	for (es = execsw; *es; es++)
1495 		if (*es != execsw_arg)
1496 			count++;
1497 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1498 	if (newexecsw == NULL)
1499 		return (ENOMEM);
1500 	xs = newexecsw;
1501 	for (es = execsw; *es; es++)
1502 		if (*es != execsw_arg)
1503 			*xs++ = *es;
1504 	*xs = NULL;
1505 	if (execsw)
1506 		free(execsw, M_TEMP);
1507 	execsw = newexecsw;
1508 	return (0);
1509 }
1510