xref: /freebsd/sys/kern/kern_exec.c (revision 8e648814b040d481f0429a2675e62af92b23bdd0)
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/capsicum.h>
37 #include <sys/systm.h>
38 #include <sys/eventhandler.h>
39 #include <sys/lock.h>
40 #include <sys/mutex.h>
41 #include <sys/sysproto.h>
42 #include <sys/signalvar.h>
43 #include <sys/kernel.h>
44 #include <sys/mount.h>
45 #include <sys/filedesc.h>
46 #include <sys/fcntl.h>
47 #include <sys/acct.h>
48 #include <sys/exec.h>
49 #include <sys/imgact.h>
50 #include <sys/imgact_elf.h>
51 #include <sys/wait.h>
52 #include <sys/malloc.h>
53 #include <sys/priv.h>
54 #include <sys/proc.h>
55 #include <sys/pioctl.h>
56 #include <sys/namei.h>
57 #include <sys/resourcevar.h>
58 #include <sys/rwlock.h>
59 #include <sys/sched.h>
60 #include <sys/sdt.h>
61 #include <sys/sf_buf.h>
62 #include <sys/syscallsubr.h>
63 #include <sys/sysent.h>
64 #include <sys/shm.h>
65 #include <sys/sysctl.h>
66 #include <sys/vnode.h>
67 #include <sys/stat.h>
68 #ifdef KTRACE
69 #include <sys/ktrace.h>
70 #endif
71 
72 #include <vm/vm.h>
73 #include <vm/vm_param.h>
74 #include <vm/pmap.h>
75 #include <vm/vm_page.h>
76 #include <vm/vm_map.h>
77 #include <vm/vm_kern.h>
78 #include <vm/vm_extern.h>
79 #include <vm/vm_object.h>
80 #include <vm/vm_pager.h>
81 
82 #ifdef	HWPMC_HOOKS
83 #include <sys/pmckern.h>
84 #endif
85 
86 #include <machine/reg.h>
87 
88 #include <security/audit/audit.h>
89 #include <security/mac/mac_framework.h>
90 
91 #ifdef KDTRACE_HOOKS
92 #include <sys/dtrace_bsd.h>
93 dtrace_execexit_func_t	dtrace_fasttrap_exec;
94 #endif
95 
96 SDT_PROVIDER_DECLARE(proc);
97 SDT_PROBE_DEFINE1(proc, kernel, , exec, "char *");
98 SDT_PROBE_DEFINE1(proc, kernel, , exec__failure, "int");
99 SDT_PROBE_DEFINE1(proc, kernel, , exec__success, "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     CTLFLAG_CAPRD, 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 disallow_high_osrel;
125 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW,
126     &disallow_high_osrel, 0,
127     "Disallow execution of binaries built for higher version of the world");
128 
129 static int map_at_zero = 0;
130 TUNABLE_INT("security.bsd.map_at_zero", &map_at_zero);
131 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RW, &map_at_zero, 0,
132     "Permit processes to map an object at virtual address 0.");
133 
134 static int
135 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
136 {
137 	struct proc *p;
138 	int error;
139 
140 	p = curproc;
141 #ifdef SCTL_MASK32
142 	if (req->flags & SCTL_MASK32) {
143 		unsigned int val;
144 		val = (unsigned int)p->p_sysent->sv_psstrings;
145 		error = SYSCTL_OUT(req, &val, sizeof(val));
146 	} else
147 #endif
148 		error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
149 		   sizeof(p->p_sysent->sv_psstrings));
150 	return error;
151 }
152 
153 static int
154 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
155 {
156 	struct proc *p;
157 	int error;
158 
159 	p = curproc;
160 #ifdef SCTL_MASK32
161 	if (req->flags & SCTL_MASK32) {
162 		unsigned int val;
163 		val = (unsigned int)p->p_sysent->sv_usrstack;
164 		error = SYSCTL_OUT(req, &val, sizeof(val));
165 	} else
166 #endif
167 		error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
168 		    sizeof(p->p_sysent->sv_usrstack));
169 	return error;
170 }
171 
172 static int
173 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
174 {
175 	struct proc *p;
176 
177 	p = curproc;
178 	return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
179 	    sizeof(p->p_sysent->sv_stackprot)));
180 }
181 
182 /*
183  * Each of the items is a pointer to a `const struct execsw', hence the
184  * double pointer here.
185  */
186 static const struct execsw **execsw;
187 
188 #ifndef _SYS_SYSPROTO_H_
189 struct execve_args {
190 	char    *fname;
191 	char    **argv;
192 	char    **envv;
193 };
194 #endif
195 
196 int
197 sys_execve(td, uap)
198 	struct thread *td;
199 	struct execve_args /* {
200 		char *fname;
201 		char **argv;
202 		char **envv;
203 	} */ *uap;
204 {
205 	int error;
206 	struct image_args args;
207 
208 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
209 	    uap->argv, uap->envv);
210 	if (error == 0)
211 		error = kern_execve(td, &args, NULL);
212 	return (error);
213 }
214 
215 #ifndef _SYS_SYSPROTO_H_
216 struct fexecve_args {
217 	int	fd;
218 	char	**argv;
219 	char	**envv;
220 }
221 #endif
222 int
223 sys_fexecve(struct thread *td, struct fexecve_args *uap)
224 {
225 	int error;
226 	struct image_args args;
227 
228 	error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
229 	    uap->argv, uap->envv);
230 	if (error == 0) {
231 		args.fd = uap->fd;
232 		error = kern_execve(td, &args, NULL);
233 	}
234 	return (error);
235 }
236 
237 #ifndef _SYS_SYSPROTO_H_
238 struct __mac_execve_args {
239 	char	*fname;
240 	char	**argv;
241 	char	**envv;
242 	struct mac	*mac_p;
243 };
244 #endif
245 
246 int
247 sys___mac_execve(td, uap)
248 	struct thread *td;
249 	struct __mac_execve_args /* {
250 		char *fname;
251 		char **argv;
252 		char **envv;
253 		struct mac *mac_p;
254 	} */ *uap;
255 {
256 #ifdef MAC
257 	int error;
258 	struct image_args args;
259 
260 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
261 	    uap->argv, uap->envv);
262 	if (error == 0)
263 		error = kern_execve(td, &args, uap->mac_p);
264 	return (error);
265 #else
266 	return (ENOSYS);
267 #endif
268 }
269 
270 /*
271  * XXX: kern_execve has the astonishing property of not always returning to
272  * the caller.  If sufficiently bad things happen during the call to
273  * do_execve(), it can end up calling exit1(); as a result, callers must
274  * avoid doing anything which they might need to undo (e.g., allocating
275  * memory).
276  */
277 int
278 kern_execve(td, args, mac_p)
279 	struct thread *td;
280 	struct image_args *args;
281 	struct mac *mac_p;
282 {
283 	struct proc *p = td->td_proc;
284 	struct vmspace *oldvmspace;
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 	KASSERT((td->td_pflags & TDP_EXECVMSPC) == 0, ("nested execve"));
302 	oldvmspace = td->td_proc->p_vmspace;
303 	error = do_execve(td, args, mac_p);
304 
305 	if (p->p_flag & P_HADTHREADS) {
306 		PROC_LOCK(p);
307 		/*
308 		 * If success, we upgrade to SINGLE_EXIT state to
309 		 * force other threads to suicide.
310 		 */
311 		if (error == 0)
312 			thread_single(SINGLE_EXIT);
313 		else
314 			thread_single_end();
315 		PROC_UNLOCK(p);
316 	}
317 	if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
318 		KASSERT(td->td_proc->p_vmspace != oldvmspace,
319 		    ("oldvmspace still used"));
320 		vmspace_free(oldvmspace);
321 		td->td_pflags &= ~TDP_EXECVMSPC;
322 	}
323 
324 	return (error);
325 }
326 
327 /*
328  * In-kernel implementation of execve().  All arguments are assumed to be
329  * userspace pointers from the passed thread.
330  */
331 static int
332 do_execve(td, args, mac_p)
333 	struct thread *td;
334 	struct image_args *args;
335 	struct mac *mac_p;
336 {
337 	struct proc *p = td->td_proc;
338 	struct nameidata nd;
339 	struct ucred *newcred = NULL, *oldcred;
340 	struct uidinfo *euip;
341 	register_t *stack_base;
342 	int error, i;
343 	struct image_params image_params, *imgp;
344 	struct vattr attr;
345 	int (*img_first)(struct image_params *);
346 	struct pargs *oldargs = NULL, *newargs = NULL;
347 	struct sigacts *oldsigacts, *newsigacts;
348 #ifdef KTRACE
349 	struct vnode *tracevp = NULL;
350 	struct ucred *tracecred = NULL;
351 #endif
352 	struct vnode *textvp = NULL, *binvp = NULL;
353 	cap_rights_t rights;
354 	int credential_changing;
355 	int textset;
356 #ifdef MAC
357 	struct label *interpvplabel = NULL;
358 	int will_transition;
359 #endif
360 #ifdef HWPMC_HOOKS
361 	struct pmckern_procexec pe;
362 #endif
363 	static const char fexecv_proc_title[] = "(fexecv)";
364 
365 	imgp = &image_params;
366 
367 	/*
368 	 * Lock the process and set the P_INEXEC flag to indicate that
369 	 * it should be left alone until we're done here.  This is
370 	 * necessary to avoid race conditions - e.g. in ptrace() -
371 	 * that might allow a local user to illicitly obtain elevated
372 	 * privileges.
373 	 */
374 	PROC_LOCK(p);
375 	KASSERT((p->p_flag & P_INEXEC) == 0,
376 	    ("%s(): process already has P_INEXEC flag", __func__));
377 	p->p_flag |= P_INEXEC;
378 	PROC_UNLOCK(p);
379 
380 	/*
381 	 * Initialize part of the common data
382 	 */
383 	imgp->proc = p;
384 	imgp->execlabel = NULL;
385 	imgp->attr = &attr;
386 	imgp->entry_addr = 0;
387 	imgp->reloc_base = 0;
388 	imgp->vmspace_destroyed = 0;
389 	imgp->interpreted = 0;
390 	imgp->opened = 0;
391 	imgp->interpreter_name = NULL;
392 	imgp->auxargs = NULL;
393 	imgp->vp = NULL;
394 	imgp->object = NULL;
395 	imgp->firstpage = NULL;
396 	imgp->ps_strings = 0;
397 	imgp->auxarg_size = 0;
398 	imgp->args = args;
399 	imgp->execpath = imgp->freepath = NULL;
400 	imgp->execpathp = 0;
401 	imgp->canary = 0;
402 	imgp->canarylen = 0;
403 	imgp->pagesizes = 0;
404 	imgp->pagesizeslen = 0;
405 	imgp->stack_prot = 0;
406 
407 #ifdef MAC
408 	error = mac_execve_enter(imgp, mac_p);
409 	if (error)
410 		goto exec_fail;
411 #endif
412 
413 	imgp->image_header = NULL;
414 
415 	/*
416 	 * Translate the file name. namei() returns a vnode pointer
417 	 *	in ni_vp amoung other things.
418 	 *
419 	 * XXXAUDIT: It would be desirable to also audit the name of the
420 	 * interpreter if this is an interpreted binary.
421 	 */
422 	if (args->fname != NULL) {
423 		NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME
424 		    | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
425 	}
426 
427 	SDT_PROBE(proc, kernel, , exec, args->fname, 0, 0, 0, 0 );
428 
429 interpret:
430 	if (args->fname != NULL) {
431 #ifdef CAPABILITY_MODE
432 		/*
433 		 * While capability mode can't reach this point via direct
434 		 * path arguments to execve(), we also don't allow
435 		 * interpreters to be used in capability mode (for now).
436 		 * Catch indirect lookups and return a permissions error.
437 		 */
438 		if (IN_CAPABILITY_MODE(td)) {
439 			error = ECAPMODE;
440 			goto exec_fail;
441 		}
442 #endif
443 		error = namei(&nd);
444 		if (error)
445 			goto exec_fail;
446 
447 		binvp  = nd.ni_vp;
448 		imgp->vp = binvp;
449 	} else {
450 		AUDIT_ARG_FD(args->fd);
451 		/*
452 		 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
453 		 */
454 		error = fgetvp_exec(td, args->fd,
455 		    cap_rights_init(&rights, CAP_FEXECVE), &binvp);
456 		if (error)
457 			goto exec_fail;
458 		vn_lock(binvp, LK_EXCLUSIVE | LK_RETRY);
459 		AUDIT_ARG_VNODE1(binvp);
460 		imgp->vp = binvp;
461 	}
462 
463 	/*
464 	 * Check file permissions (also 'opens' file)
465 	 */
466 	error = exec_check_permissions(imgp);
467 	if (error)
468 		goto exec_fail_dealloc;
469 
470 	imgp->object = imgp->vp->v_object;
471 	if (imgp->object != NULL)
472 		vm_object_reference(imgp->object);
473 
474 	/*
475 	 * Set VV_TEXT now so no one can write to the executable while we're
476 	 * activating it.
477 	 *
478 	 * Remember if this was set before and unset it in case this is not
479 	 * actually an executable image.
480 	 */
481 	textset = VOP_IS_TEXT(imgp->vp);
482 	VOP_SET_TEXT(imgp->vp);
483 
484 	error = exec_map_first_page(imgp);
485 	if (error)
486 		goto exec_fail_dealloc;
487 
488 	imgp->proc->p_osrel = 0;
489 	/*
490 	 *	If the current process has a special image activator it
491 	 *	wants to try first, call it.   For example, emulating shell
492 	 *	scripts differently.
493 	 */
494 	error = -1;
495 	if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
496 		error = img_first(imgp);
497 
498 	/*
499 	 *	Loop through the list of image activators, calling each one.
500 	 *	An activator returns -1 if there is no match, 0 on success,
501 	 *	and an error otherwise.
502 	 */
503 	for (i = 0; error == -1 && execsw[i]; ++i) {
504 		if (execsw[i]->ex_imgact == NULL ||
505 		    execsw[i]->ex_imgact == img_first) {
506 			continue;
507 		}
508 		error = (*execsw[i]->ex_imgact)(imgp);
509 	}
510 
511 	if (error) {
512 		if (error == -1) {
513 			if (textset == 0)
514 				VOP_UNSET_TEXT(imgp->vp);
515 			error = ENOEXEC;
516 		}
517 		goto exec_fail_dealloc;
518 	}
519 
520 	/*
521 	 * Special interpreter operation, cleanup and loop up to try to
522 	 * activate the interpreter.
523 	 */
524 	if (imgp->interpreted) {
525 		exec_unmap_first_page(imgp);
526 		/*
527 		 * VV_TEXT needs to be unset for scripts.  There is a short
528 		 * period before we determine that something is a script where
529 		 * VV_TEXT will be set. The vnode lock is held over this
530 		 * entire period so nothing should illegitimately be blocked.
531 		 */
532 		VOP_UNSET_TEXT(imgp->vp);
533 		/* free name buffer and old vnode */
534 		if (args->fname != NULL)
535 			NDFREE(&nd, NDF_ONLY_PNBUF);
536 #ifdef MAC
537 		mac_execve_interpreter_enter(binvp, &interpvplabel);
538 #endif
539 		if (imgp->opened) {
540 			VOP_CLOSE(binvp, FREAD, td->td_ucred, td);
541 			imgp->opened = 0;
542 		}
543 		vput(binvp);
544 		vm_object_deallocate(imgp->object);
545 		imgp->object = NULL;
546 		/* set new name to that of the interpreter */
547 		NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
548 		    UIO_SYSSPACE, imgp->interpreter_name, td);
549 		args->fname = imgp->interpreter_name;
550 		goto interpret;
551 	}
552 
553 	/*
554 	 * NB: We unlock the vnode here because it is believed that none
555 	 * of the sv_copyout_strings/sv_fixup operations require the vnode.
556 	 */
557 	VOP_UNLOCK(imgp->vp, 0);
558 
559 	/*
560 	 * Do the best to calculate the full path to the image file.
561 	 */
562 	if (imgp->auxargs != NULL &&
563 	    ((args->fname != NULL && args->fname[0] == '/') ||
564 	     vn_fullpath(td, imgp->vp, &imgp->execpath, &imgp->freepath) != 0))
565 		imgp->execpath = args->fname;
566 
567 	if (disallow_high_osrel &&
568 	    P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
569 		error = ENOEXEC;
570 		uprintf("Osrel %d for image %s too high\n", p->p_osrel,
571 		    imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
572 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
573 		goto exec_fail_dealloc;
574 	}
575 
576 	/*
577 	 * Copy out strings (args and env) and initialize stack base
578 	 */
579 	if (p->p_sysent->sv_copyout_strings)
580 		stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
581 	else
582 		stack_base = exec_copyout_strings(imgp);
583 
584 	/*
585 	 * If custom stack fixup routine present for this process
586 	 * let it do the stack setup.
587 	 * Else stuff argument count as first item on stack
588 	 */
589 	if (p->p_sysent->sv_fixup != NULL)
590 		(*p->p_sysent->sv_fixup)(&stack_base, imgp);
591 	else
592 		suword(--stack_base, imgp->args->argc);
593 
594 	/*
595 	 * For security and other reasons, the file descriptor table cannot
596 	 * be shared after an exec.
597 	 */
598 	fdunshare(p, td);
599 
600 	/*
601 	 * Malloc things before we need locks.
602 	 */
603 	newcred = crget();
604 	euip = uifind(attr.va_uid);
605 	i = imgp->args->begin_envv - imgp->args->begin_argv;
606 	/* Cache arguments if they fit inside our allowance */
607 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
608 		newargs = pargs_alloc(i);
609 		bcopy(imgp->args->begin_argv, newargs->ar_args, i);
610 	}
611 
612 	/* close files on exec */
613 	fdcloseexec(td);
614 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
615 
616 	/* Get a reference to the vnode prior to locking the proc */
617 	VREF(binvp);
618 
619 	/*
620 	 * For security and other reasons, signal handlers cannot
621 	 * be shared after an exec. The new process gets a copy of the old
622 	 * handlers. In execsigs(), the new process will have its signals
623 	 * reset.
624 	 */
625 	PROC_LOCK(p);
626 	oldcred = crcopysafe(p, newcred);
627 	if (sigacts_shared(p->p_sigacts)) {
628 		oldsigacts = p->p_sigacts;
629 		PROC_UNLOCK(p);
630 		newsigacts = sigacts_alloc();
631 		sigacts_copy(newsigacts, oldsigacts);
632 		PROC_LOCK(p);
633 		p->p_sigacts = newsigacts;
634 	} else
635 		oldsigacts = NULL;
636 
637 	/* Stop profiling */
638 	stopprofclock(p);
639 
640 	/* reset caught signals */
641 	execsigs(p);
642 
643 	/* name this process - nameiexec(p, ndp) */
644 	bzero(p->p_comm, sizeof(p->p_comm));
645 	if (args->fname)
646 		bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
647 		    min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
648 	else if (vn_commname(binvp, p->p_comm, sizeof(p->p_comm)) != 0)
649 		bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
650 	bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
651 #ifdef KTR
652 	sched_clear_tdname(td);
653 #endif
654 
655 	/*
656 	 * mark as execed, wakeup the process that vforked (if any) and tell
657 	 * it that it now has its own resources back
658 	 */
659 	p->p_flag |= P_EXEC;
660 	if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
661 		p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
662 		cv_broadcast(&p->p_pwait);
663 	}
664 
665 	/*
666 	 * Implement image setuid/setgid.
667 	 *
668 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
669 	 * the process is being traced.
670 	 *
671 	 * We disable setuid/setgid/etc in compatibility mode on the basis
672 	 * that most setugid applications are not written with that
673 	 * environment in mind, and will therefore almost certainly operate
674 	 * incorrectly. In principle there's no reason that setugid
675 	 * applications might not be useful in capability mode, so we may want
676 	 * to reconsider this conservative design choice in the future.
677 	 *
678 	 * XXXMAC: For the time being, use NOSUID to also prohibit
679 	 * transitions on the file system.
680 	 */
681 	credential_changing = 0;
682 	credential_changing |= (attr.va_mode & S_ISUID) && oldcred->cr_uid !=
683 	    attr.va_uid;
684 	credential_changing |= (attr.va_mode & S_ISGID) && oldcred->cr_gid !=
685 	    attr.va_gid;
686 #ifdef MAC
687 	will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
688 	    interpvplabel, imgp);
689 	credential_changing |= will_transition;
690 #endif
691 
692 	if (credential_changing &&
693 #ifdef CAPABILITY_MODE
694 	    ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
695 #endif
696 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
697 	    (p->p_flag & P_TRACED) == 0) {
698 		/*
699 		 * Turn off syscall tracing for set-id programs, except for
700 		 * root.  Record any set-id flags first to make sure that
701 		 * we do not regain any tracing during a possible block.
702 		 */
703 		setsugid(p);
704 
705 #ifdef KTRACE
706 		if (p->p_tracecred != NULL &&
707 		    priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED, 0))
708 			ktrprocexec(p, &tracecred, &tracevp);
709 #endif
710 		/*
711 		 * Close any file descriptors 0..2 that reference procfs,
712 		 * then make sure file descriptors 0..2 are in use.
713 		 *
714 		 * setugidsafety() may call closef() and then pfind()
715 		 * which may grab the process lock.
716 		 * fdcheckstd() may call falloc() which may block to
717 		 * allocate memory, so temporarily drop the process lock.
718 		 */
719 		PROC_UNLOCK(p);
720 		VOP_UNLOCK(imgp->vp, 0);
721 		setugidsafety(td);
722 		error = fdcheckstd(td);
723 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
724 		if (error != 0)
725 			goto done1;
726 		PROC_LOCK(p);
727 		/*
728 		 * Set the new credentials.
729 		 */
730 		if (attr.va_mode & S_ISUID)
731 			change_euid(newcred, euip);
732 		if (attr.va_mode & S_ISGID)
733 			change_egid(newcred, attr.va_gid);
734 #ifdef MAC
735 		if (will_transition) {
736 			mac_vnode_execve_transition(oldcred, newcred, imgp->vp,
737 			    interpvplabel, imgp);
738 		}
739 #endif
740 		/*
741 		 * Implement correct POSIX saved-id behavior.
742 		 *
743 		 * XXXMAC: Note that the current logic will save the
744 		 * uid and gid if a MAC domain transition occurs, even
745 		 * though maybe it shouldn't.
746 		 */
747 		change_svuid(newcred, newcred->cr_uid);
748 		change_svgid(newcred, newcred->cr_gid);
749 		p->p_ucred = newcred;
750 		newcred = NULL;
751 	} else {
752 		if (oldcred->cr_uid == oldcred->cr_ruid &&
753 		    oldcred->cr_gid == oldcred->cr_rgid)
754 			p->p_flag &= ~P_SUGID;
755 		/*
756 		 * Implement correct POSIX saved-id behavior.
757 		 *
758 		 * XXX: It's not clear that the existing behavior is
759 		 * POSIX-compliant.  A number of sources indicate that the
760 		 * saved uid/gid should only be updated if the new ruid is
761 		 * not equal to the old ruid, or the new euid is not equal
762 		 * to the old euid and the new euid is not equal to the old
763 		 * ruid.  The FreeBSD code always updates the saved uid/gid.
764 		 * Also, this code uses the new (replaced) euid and egid as
765 		 * the source, which may or may not be the right ones to use.
766 		 */
767 		if (oldcred->cr_svuid != oldcred->cr_uid ||
768 		    oldcred->cr_svgid != oldcred->cr_gid) {
769 			change_svuid(newcred, newcred->cr_uid);
770 			change_svgid(newcred, newcred->cr_gid);
771 			p->p_ucred = newcred;
772 			newcred = NULL;
773 		}
774 	}
775 
776 	/*
777 	 * Store the vp for use in procfs.  This vnode was referenced prior
778 	 * to locking the proc lock.
779 	 */
780 	textvp = p->p_textvp;
781 	p->p_textvp = binvp;
782 
783 #ifdef KDTRACE_HOOKS
784 	/*
785 	 * Tell the DTrace fasttrap provider about the exec if it
786 	 * has declared an interest.
787 	 */
788 	if (dtrace_fasttrap_exec)
789 		dtrace_fasttrap_exec(p);
790 #endif
791 
792 	/*
793 	 * Notify others that we exec'd, and clear the P_INEXEC flag
794 	 * as we're now a bona fide freshly-execed process.
795 	 */
796 	KNOTE_LOCKED(&p->p_klist, NOTE_EXEC);
797 	p->p_flag &= ~P_INEXEC;
798 
799 	/* clear "fork but no exec" flag, as we _are_ execing */
800 	p->p_acflag &= ~AFORK;
801 
802 	/*
803 	 * Free any previous argument cache and replace it with
804 	 * the new argument cache, if any.
805 	 */
806 	oldargs = p->p_args;
807 	p->p_args = newargs;
808 	newargs = NULL;
809 
810 #ifdef	HWPMC_HOOKS
811 	/*
812 	 * Check if system-wide sampling is in effect or if the
813 	 * current process is using PMCs.  If so, do exec() time
814 	 * processing.  This processing needs to happen AFTER the
815 	 * P_INEXEC flag is cleared.
816 	 *
817 	 * The proc lock needs to be released before taking the PMC
818 	 * SX.
819 	 */
820 	if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
821 		PROC_UNLOCK(p);
822 		VOP_UNLOCK(imgp->vp, 0);
823 		pe.pm_credentialschanged = credential_changing;
824 		pe.pm_entryaddr = imgp->entry_addr;
825 
826 		PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
827 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
828 	} else
829 		PROC_UNLOCK(p);
830 #else  /* !HWPMC_HOOKS */
831 	PROC_UNLOCK(p);
832 #endif
833 
834 	/* Set values passed into the program in registers. */
835 	if (p->p_sysent->sv_setregs)
836 		(*p->p_sysent->sv_setregs)(td, imgp,
837 		    (u_long)(uintptr_t)stack_base);
838 	else
839 		exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base);
840 
841 	vfs_mark_atime(imgp->vp, td->td_ucred);
842 
843 	SDT_PROBE(proc, kernel, , exec__success, args->fname, 0, 0, 0, 0);
844 
845 done1:
846 	/*
847 	 * Free any resources malloc'd earlier that we didn't use.
848 	 */
849 	uifree(euip);
850 	if (newcred == NULL)
851 		crfree(oldcred);
852 	else
853 		crfree(newcred);
854 	VOP_UNLOCK(imgp->vp, 0);
855 
856 	/*
857 	 * Handle deferred decrement of ref counts.
858 	 */
859 	if (textvp != NULL)
860 		vrele(textvp);
861 	if (binvp && error != 0)
862 		vrele(binvp);
863 #ifdef KTRACE
864 	if (tracevp != NULL)
865 		vrele(tracevp);
866 	if (tracecred != NULL)
867 		crfree(tracecred);
868 #endif
869 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
870 	pargs_drop(oldargs);
871 	pargs_drop(newargs);
872 	if (oldsigacts != NULL)
873 		sigacts_free(oldsigacts);
874 
875 exec_fail_dealloc:
876 
877 	/*
878 	 * free various allocated resources
879 	 */
880 	if (imgp->firstpage != NULL)
881 		exec_unmap_first_page(imgp);
882 
883 	if (imgp->vp != NULL) {
884 		if (args->fname)
885 			NDFREE(&nd, NDF_ONLY_PNBUF);
886 		if (imgp->opened)
887 			VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
888 		vput(imgp->vp);
889 	}
890 
891 	if (imgp->object != NULL)
892 		vm_object_deallocate(imgp->object);
893 
894 	free(imgp->freepath, M_TEMP);
895 
896 	if (error == 0) {
897 		PROC_LOCK(p);
898 		td->td_dbgflags |= TDB_EXEC;
899 		PROC_UNLOCK(p);
900 
901 		/*
902 		 * Stop the process here if its stop event mask has
903 		 * the S_EXEC bit set.
904 		 */
905 		STOPEVENT(p, S_EXEC, 0);
906 		goto done2;
907 	}
908 
909 exec_fail:
910 	/* we're done here, clear P_INEXEC */
911 	PROC_LOCK(p);
912 	p->p_flag &= ~P_INEXEC;
913 	PROC_UNLOCK(p);
914 
915 	SDT_PROBE(proc, kernel, , exec__failure, error, 0, 0, 0, 0);
916 
917 done2:
918 #ifdef MAC
919 	mac_execve_exit(imgp);
920 	mac_execve_interpreter_exit(interpvplabel);
921 #endif
922 	exec_free_args(args);
923 
924 	if (error && imgp->vmspace_destroyed) {
925 		/* sorry, no more process anymore. exit gracefully */
926 		exit1(td, W_EXITCODE(0, SIGABRT));
927 		/* NOT REACHED */
928 	}
929 
930 #ifdef KTRACE
931 	if (error == 0)
932 		ktrprocctor(p);
933 #endif
934 
935 	return (error);
936 }
937 
938 int
939 exec_map_first_page(imgp)
940 	struct image_params *imgp;
941 {
942 	int rv, i;
943 	int initial_pagein;
944 	vm_page_t ma[VM_INITIAL_PAGEIN];
945 	vm_object_t object;
946 
947 	if (imgp->firstpage != NULL)
948 		exec_unmap_first_page(imgp);
949 
950 	object = imgp->vp->v_object;
951 	if (object == NULL)
952 		return (EACCES);
953 	VM_OBJECT_WLOCK(object);
954 #if VM_NRESERVLEVEL > 0
955 	if ((object->flags & OBJ_COLORED) == 0) {
956 		object->flags |= OBJ_COLORED;
957 		object->pg_color = 0;
958 	}
959 #endif
960 	ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL);
961 	if (ma[0]->valid != VM_PAGE_BITS_ALL) {
962 		initial_pagein = VM_INITIAL_PAGEIN;
963 		if (initial_pagein > object->size)
964 			initial_pagein = object->size;
965 		for (i = 1; i < initial_pagein; i++) {
966 			if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) {
967 				if (ma[i]->valid)
968 					break;
969 				if (vm_page_tryxbusy(ma[i]))
970 					break;
971 			} else {
972 				ma[i] = vm_page_alloc(object, i,
973 				    VM_ALLOC_NORMAL | VM_ALLOC_IFNOTCACHED);
974 				if (ma[i] == NULL)
975 					break;
976 			}
977 		}
978 		initial_pagein = i;
979 		rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
980 		ma[0] = vm_page_lookup(object, 0);
981 		if ((rv != VM_PAGER_OK) || (ma[0] == NULL)) {
982 			if (ma[0] != NULL) {
983 				vm_page_lock(ma[0]);
984 				vm_page_free(ma[0]);
985 				vm_page_unlock(ma[0]);
986 			}
987 			VM_OBJECT_WUNLOCK(object);
988 			return (EIO);
989 		}
990 	}
991 	vm_page_xunbusy(ma[0]);
992 	vm_page_lock(ma[0]);
993 	vm_page_hold(ma[0]);
994 	vm_page_unlock(ma[0]);
995 	VM_OBJECT_WUNLOCK(object);
996 
997 	imgp->firstpage = sf_buf_alloc(ma[0], 0);
998 	imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
999 
1000 	return (0);
1001 }
1002 
1003 void
1004 exec_unmap_first_page(imgp)
1005 	struct image_params *imgp;
1006 {
1007 	vm_page_t m;
1008 
1009 	if (imgp->firstpage != NULL) {
1010 		m = sf_buf_page(imgp->firstpage);
1011 		sf_buf_free(imgp->firstpage);
1012 		imgp->firstpage = NULL;
1013 		vm_page_lock(m);
1014 		vm_page_unhold(m);
1015 		vm_page_unlock(m);
1016 	}
1017 }
1018 
1019 /*
1020  * Destroy old address space, and allocate a new stack
1021  *	The new stack is only SGROWSIZ large because it is grown
1022  *	automatically in trap.c.
1023  */
1024 int
1025 exec_new_vmspace(imgp, sv)
1026 	struct image_params *imgp;
1027 	struct sysentvec *sv;
1028 {
1029 	int error;
1030 	struct proc *p = imgp->proc;
1031 	struct vmspace *vmspace = p->p_vmspace;
1032 	vm_object_t obj;
1033 	vm_offset_t sv_minuser, stack_addr;
1034 	vm_map_t map;
1035 	u_long ssiz;
1036 
1037 	imgp->vmspace_destroyed = 1;
1038 	imgp->sysent = sv;
1039 
1040 	/* May be called with Giant held */
1041 	EVENTHANDLER_INVOKE(process_exec, p, imgp);
1042 
1043 	/*
1044 	 * Blow away entire process VM, if address space not shared,
1045 	 * otherwise, create a new VM space so that other threads are
1046 	 * not disrupted
1047 	 */
1048 	map = &vmspace->vm_map;
1049 	if (map_at_zero)
1050 		sv_minuser = sv->sv_minuser;
1051 	else
1052 		sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1053 	if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser &&
1054 	    vm_map_max(map) == sv->sv_maxuser) {
1055 		shmexit(vmspace);
1056 		pmap_remove_pages(vmspace_pmap(vmspace));
1057 		vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1058 	} else {
1059 		error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1060 		if (error)
1061 			return (error);
1062 		vmspace = p->p_vmspace;
1063 		map = &vmspace->vm_map;
1064 	}
1065 
1066 	/* Map a shared page */
1067 	obj = sv->sv_shared_page_obj;
1068 	if (obj != NULL) {
1069 		vm_object_reference(obj);
1070 		error = vm_map_fixed(map, obj, 0,
1071 		    sv->sv_shared_page_base, sv->sv_shared_page_len,
1072 		    VM_PROT_READ | VM_PROT_EXECUTE,
1073 		    VM_PROT_READ | VM_PROT_EXECUTE,
1074 		    MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1075 		if (error) {
1076 			vm_object_deallocate(obj);
1077 			return (error);
1078 		}
1079 	}
1080 
1081 	/* Allocate a new stack */
1082 	if (sv->sv_maxssiz != NULL)
1083 		ssiz = *sv->sv_maxssiz;
1084 	else
1085 		ssiz = maxssiz;
1086 	stack_addr = sv->sv_usrstack - ssiz;
1087 	error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1088 	    obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1089 		sv->sv_stackprot,
1090 	    VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1091 	if (error)
1092 		return (error);
1093 
1094 #ifdef __ia64__
1095 	/* Allocate a new register stack */
1096 	stack_addr = IA64_BACKINGSTORE;
1097 	error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1098 	    sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_UP);
1099 	if (error)
1100 		return (error);
1101 #endif
1102 
1103 	/*
1104 	 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they
1105 	 * are still used to enforce the stack rlimit on the process stack.
1106 	 */
1107 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1108 	vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - ssiz;
1109 
1110 	return (0);
1111 }
1112 
1113 /*
1114  * Copy out argument and environment strings from the old process address
1115  * space into the temporary string buffer.
1116  */
1117 int
1118 exec_copyin_args(struct image_args *args, char *fname,
1119     enum uio_seg segflg, char **argv, char **envv)
1120 {
1121 	char *argp, *envp;
1122 	int error;
1123 	size_t length;
1124 
1125 	bzero(args, sizeof(*args));
1126 	if (argv == NULL)
1127 		return (EFAULT);
1128 
1129 	/*
1130 	 * Allocate demand-paged memory for the file name, argument, and
1131 	 * environment strings.
1132 	 */
1133 	error = exec_alloc_args(args);
1134 	if (error != 0)
1135 		return (error);
1136 
1137 	/*
1138 	 * Copy the file name.
1139 	 */
1140 	if (fname != NULL) {
1141 		args->fname = args->buf;
1142 		error = (segflg == UIO_SYSSPACE) ?
1143 		    copystr(fname, args->fname, PATH_MAX, &length) :
1144 		    copyinstr(fname, args->fname, PATH_MAX, &length);
1145 		if (error != 0)
1146 			goto err_exit;
1147 	} else
1148 		length = 0;
1149 
1150 	args->begin_argv = args->buf + length;
1151 	args->endp = args->begin_argv;
1152 	args->stringspace = ARG_MAX;
1153 
1154 	/*
1155 	 * extract arguments first
1156 	 */
1157 	while ((argp = (caddr_t) (intptr_t) fuword(argv++))) {
1158 		if (argp == (caddr_t) -1) {
1159 			error = EFAULT;
1160 			goto err_exit;
1161 		}
1162 		if ((error = copyinstr(argp, args->endp,
1163 		    args->stringspace, &length))) {
1164 			if (error == ENAMETOOLONG)
1165 				error = E2BIG;
1166 			goto err_exit;
1167 		}
1168 		args->stringspace -= length;
1169 		args->endp += length;
1170 		args->argc++;
1171 	}
1172 
1173 	args->begin_envv = args->endp;
1174 
1175 	/*
1176 	 * extract environment strings
1177 	 */
1178 	if (envv) {
1179 		while ((envp = (caddr_t)(intptr_t)fuword(envv++))) {
1180 			if (envp == (caddr_t)-1) {
1181 				error = EFAULT;
1182 				goto err_exit;
1183 			}
1184 			if ((error = copyinstr(envp, args->endp,
1185 			    args->stringspace, &length))) {
1186 				if (error == ENAMETOOLONG)
1187 					error = E2BIG;
1188 				goto err_exit;
1189 			}
1190 			args->stringspace -= length;
1191 			args->endp += length;
1192 			args->envc++;
1193 		}
1194 	}
1195 
1196 	return (0);
1197 
1198 err_exit:
1199 	exec_free_args(args);
1200 	return (error);
1201 }
1202 
1203 /*
1204  * Allocate temporary demand-paged, zero-filled memory for the file name,
1205  * argument, and environment strings.  Returns zero if the allocation succeeds
1206  * and ENOMEM otherwise.
1207  */
1208 int
1209 exec_alloc_args(struct image_args *args)
1210 {
1211 
1212 	args->buf = (char *)kmap_alloc_wait(exec_map, PATH_MAX + ARG_MAX);
1213 	return (args->buf != NULL ? 0 : ENOMEM);
1214 }
1215 
1216 void
1217 exec_free_args(struct image_args *args)
1218 {
1219 
1220 	if (args->buf != NULL) {
1221 		kmap_free_wakeup(exec_map, (vm_offset_t)args->buf,
1222 		    PATH_MAX + ARG_MAX);
1223 		args->buf = NULL;
1224 	}
1225 	if (args->fname_buf != NULL) {
1226 		free(args->fname_buf, M_TEMP);
1227 		args->fname_buf = NULL;
1228 	}
1229 }
1230 
1231 /*
1232  * Copy strings out to the new process address space, constructing new arg
1233  * and env vector tables. Return a pointer to the base so that it can be used
1234  * as the initial stack pointer.
1235  */
1236 register_t *
1237 exec_copyout_strings(imgp)
1238 	struct image_params *imgp;
1239 {
1240 	int argc, envc;
1241 	char **vectp;
1242 	char *stringp;
1243 	uintptr_t destp;
1244 	register_t *stack_base;
1245 	struct ps_strings *arginfo;
1246 	struct proc *p;
1247 	size_t execpath_len;
1248 	int szsigcode, szps;
1249 	char canary[sizeof(long) * 8];
1250 
1251 	szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1252 	/*
1253 	 * Calculate string base and vector table pointers.
1254 	 * Also deal with signal trampoline code for this exec type.
1255 	 */
1256 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
1257 		execpath_len = strlen(imgp->execpath) + 1;
1258 	else
1259 		execpath_len = 0;
1260 	p = imgp->proc;
1261 	szsigcode = 0;
1262 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1263 	if (p->p_sysent->sv_sigcode_base == 0) {
1264 		if (p->p_sysent->sv_szsigcode != NULL)
1265 			szsigcode = *(p->p_sysent->sv_szsigcode);
1266 	}
1267 	destp =	(uintptr_t)arginfo;
1268 
1269 	/*
1270 	 * install sigcode
1271 	 */
1272 	if (szsigcode != 0) {
1273 		destp -= szsigcode;
1274 		destp = rounddown2(destp, sizeof(void *));
1275 		copyout(p->p_sysent->sv_sigcode, (void *)destp, szsigcode);
1276 	}
1277 
1278 	/*
1279 	 * Copy the image path for the rtld.
1280 	 */
1281 	if (execpath_len != 0) {
1282 		destp -= execpath_len;
1283 		imgp->execpathp = destp;
1284 		copyout(imgp->execpath, (void *)destp, execpath_len);
1285 	}
1286 
1287 	/*
1288 	 * Prepare the canary for SSP.
1289 	 */
1290 	arc4rand(canary, sizeof(canary), 0);
1291 	destp -= sizeof(canary);
1292 	imgp->canary = destp;
1293 	copyout(canary, (void *)destp, sizeof(canary));
1294 	imgp->canarylen = sizeof(canary);
1295 
1296 	/*
1297 	 * Prepare the pagesizes array.
1298 	 */
1299 	destp -= szps;
1300 	destp = rounddown2(destp, sizeof(void *));
1301 	imgp->pagesizes = destp;
1302 	copyout(pagesizes, (void *)destp, szps);
1303 	imgp->pagesizeslen = szps;
1304 
1305 	destp -= ARG_MAX - imgp->args->stringspace;
1306 	destp = rounddown2(destp, sizeof(void *));
1307 
1308 	/*
1309 	 * If we have a valid auxargs ptr, prepare some room
1310 	 * on the stack.
1311 	 */
1312 	if (imgp->auxargs) {
1313 		/*
1314 		 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1315 		 * lower compatibility.
1316 		 */
1317 		imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1318 		    (AT_COUNT * 2);
1319 		/*
1320 		 * The '+ 2' is for the null pointers at the end of each of
1321 		 * the arg and env vector sets,and imgp->auxarg_size is room
1322 		 * for argument of Runtime loader.
1323 		 */
1324 		vectp = (char **)(destp - (imgp->args->argc +
1325 		    imgp->args->envc + 2 + imgp->auxarg_size)
1326 		    * sizeof(char *));
1327 	} else {
1328 		/*
1329 		 * The '+ 2' is for the null pointers at the end of each of
1330 		 * the arg and env vector sets
1331 		 */
1332 		vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc
1333 		    + 2) * sizeof(char *));
1334 	}
1335 
1336 	/*
1337 	 * vectp also becomes our initial stack base
1338 	 */
1339 	stack_base = (register_t *)vectp;
1340 
1341 	stringp = imgp->args->begin_argv;
1342 	argc = imgp->args->argc;
1343 	envc = imgp->args->envc;
1344 
1345 	/*
1346 	 * Copy out strings - arguments and environment.
1347 	 */
1348 	copyout(stringp, (void *)destp, ARG_MAX - imgp->args->stringspace);
1349 
1350 	/*
1351 	 * Fill in "ps_strings" struct for ps, w, etc.
1352 	 */
1353 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1354 	suword32(&arginfo->ps_nargvstr, argc);
1355 
1356 	/*
1357 	 * Fill in argument portion of vector table.
1358 	 */
1359 	for (; argc > 0; --argc) {
1360 		suword(vectp++, (long)(intptr_t)destp);
1361 		while (*stringp++ != 0)
1362 			destp++;
1363 		destp++;
1364 	}
1365 
1366 	/* a null vector table pointer separates the argp's from the envp's */
1367 	suword(vectp++, 0);
1368 
1369 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1370 	suword32(&arginfo->ps_nenvstr, envc);
1371 
1372 	/*
1373 	 * Fill in environment portion of vector table.
1374 	 */
1375 	for (; envc > 0; --envc) {
1376 		suword(vectp++, (long)(intptr_t)destp);
1377 		while (*stringp++ != 0)
1378 			destp++;
1379 		destp++;
1380 	}
1381 
1382 	/* end of vector table is a null pointer */
1383 	suword(vectp, 0);
1384 
1385 	return (stack_base);
1386 }
1387 
1388 /*
1389  * Check permissions of file to execute.
1390  *	Called with imgp->vp locked.
1391  *	Return 0 for success or error code on failure.
1392  */
1393 int
1394 exec_check_permissions(imgp)
1395 	struct image_params *imgp;
1396 {
1397 	struct vnode *vp = imgp->vp;
1398 	struct vattr *attr = imgp->attr;
1399 	struct thread *td;
1400 	int error, writecount;
1401 
1402 	td = curthread;
1403 
1404 	/* Get file attributes */
1405 	error = VOP_GETATTR(vp, attr, td->td_ucred);
1406 	if (error)
1407 		return (error);
1408 
1409 #ifdef MAC
1410 	error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1411 	if (error)
1412 		return (error);
1413 #endif
1414 
1415 	/*
1416 	 * 1) Check if file execution is disabled for the filesystem that
1417 	 *    this file resides on.
1418 	 * 2) Ensure that at least one execute bit is on. Otherwise, a
1419 	 *    privileged user will always succeed, and we don't want this
1420 	 *    to happen unless the file really is executable.
1421 	 * 3) Ensure that the file is a regular file.
1422 	 */
1423 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1424 	    (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1425 	    (attr->va_type != VREG))
1426 		return (EACCES);
1427 
1428 	/*
1429 	 * Zero length files can't be exec'd
1430 	 */
1431 	if (attr->va_size == 0)
1432 		return (ENOEXEC);
1433 
1434 	/*
1435 	 *  Check for execute permission to file based on current credentials.
1436 	 */
1437 	error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1438 	if (error)
1439 		return (error);
1440 
1441 	/*
1442 	 * Check number of open-for-writes on the file and deny execution
1443 	 * if there are any.
1444 	 */
1445 	error = VOP_GET_WRITECOUNT(vp, &writecount);
1446 	if (error != 0)
1447 		return (error);
1448 	if (writecount != 0)
1449 		return (ETXTBSY);
1450 
1451 	/*
1452 	 * Call filesystem specific open routine (which does nothing in the
1453 	 * general case).
1454 	 */
1455 	error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1456 	if (error == 0)
1457 		imgp->opened = 1;
1458 	return (error);
1459 }
1460 
1461 /*
1462  * Exec handler registration
1463  */
1464 int
1465 exec_register(execsw_arg)
1466 	const struct execsw *execsw_arg;
1467 {
1468 	const struct execsw **es, **xs, **newexecsw;
1469 	int count = 2;	/* New slot and trailing NULL */
1470 
1471 	if (execsw)
1472 		for (es = execsw; *es; es++)
1473 			count++;
1474 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1475 	if (newexecsw == NULL)
1476 		return (ENOMEM);
1477 	xs = newexecsw;
1478 	if (execsw)
1479 		for (es = execsw; *es; es++)
1480 			*xs++ = *es;
1481 	*xs++ = execsw_arg;
1482 	*xs = NULL;
1483 	if (execsw)
1484 		free(execsw, M_TEMP);
1485 	execsw = newexecsw;
1486 	return (0);
1487 }
1488 
1489 int
1490 exec_unregister(execsw_arg)
1491 	const struct execsw *execsw_arg;
1492 {
1493 	const struct execsw **es, **xs, **newexecsw;
1494 	int count = 1;
1495 
1496 	if (execsw == NULL)
1497 		panic("unregister with no handlers left?\n");
1498 
1499 	for (es = execsw; *es; es++) {
1500 		if (*es == execsw_arg)
1501 			break;
1502 	}
1503 	if (*es == NULL)
1504 		return (ENOENT);
1505 	for (es = execsw; *es; es++)
1506 		if (*es != execsw_arg)
1507 			count++;
1508 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1509 	if (newexecsw == NULL)
1510 		return (ENOMEM);
1511 	xs = newexecsw;
1512 	for (es = execsw; *es; es++)
1513 		if (*es != execsw_arg)
1514 			*xs++ = *es;
1515 	*xs = NULL;
1516 	if (execsw)
1517 		free(execsw, M_TEMP);
1518 	execsw = newexecsw;
1519 	return (0);
1520 }
1521