xref: /freebsd/sys/kern/kern_exec.c (revision f5f47d5068fb97df18eb114a66ae8ef51a0b3c8c)
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_kdtrace.h"
33 #include "opt_ktrace.h"
34 #include "opt_vm.h"
35 
36 #include <sys/param.h>
37 #include <sys/capability.h>
38 #include <sys/systm.h>
39 #include <sys/capability.h>
40 #include <sys/eventhandler.h>
41 #include <sys/lock.h>
42 #include <sys/mutex.h>
43 #include <sys/sysproto.h>
44 #include <sys/signalvar.h>
45 #include <sys/kernel.h>
46 #include <sys/mount.h>
47 #include <sys/filedesc.h>
48 #include <sys/fcntl.h>
49 #include <sys/acct.h>
50 #include <sys/exec.h>
51 #include <sys/imgact.h>
52 #include <sys/imgact_elf.h>
53 #include <sys/wait.h>
54 #include <sys/malloc.h>
55 #include <sys/priv.h>
56 #include <sys/proc.h>
57 #include <sys/pioctl.h>
58 #include <sys/namei.h>
59 #include <sys/resourcevar.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_DEFINE(proc, kernel, , exec, exec);
99 SDT_PROBE_ARGTYPE(proc, kernel, , exec, 0, "char *");
100 SDT_PROBE_DEFINE(proc, kernel, , exec_failure, exec-failure);
101 SDT_PROBE_ARGTYPE(proc, kernel, , exec_failure, 0, "int");
102 SDT_PROBE_DEFINE(proc, kernel, , exec_success, exec-success);
103 SDT_PROBE_ARGTYPE(proc, kernel, , exec_success, 0, "char *");
104 
105 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
106 
107 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
108 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
109 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
110 static int do_execve(struct thread *td, struct image_args *args,
111     struct mac *mac_p);
112 
113 /* XXX This should be vm_size_t. */
114 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD,
115     NULL, 0, sysctl_kern_ps_strings, "LU", "");
116 
117 /* XXX This should be vm_size_t. */
118 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD|
119     CTLFLAG_CAPRD, NULL, 0, sysctl_kern_usrstack, "LU", "");
120 
121 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD,
122     NULL, 0, sysctl_kern_stackprot, "I", "");
123 
124 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
125 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
126     &ps_arg_cache_limit, 0, "");
127 
128 static int map_at_zero = 0;
129 TUNABLE_INT("security.bsd.map_at_zero", &map_at_zero);
130 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RW, &map_at_zero, 0,
131     "Permit processes to map an object at virtual address 0.");
132 
133 static int
134 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
135 {
136 	struct proc *p;
137 	int error;
138 
139 	p = curproc;
140 #ifdef SCTL_MASK32
141 	if (req->flags & SCTL_MASK32) {
142 		unsigned int val;
143 		val = (unsigned int)p->p_sysent->sv_psstrings;
144 		error = SYSCTL_OUT(req, &val, sizeof(val));
145 	} else
146 #endif
147 		error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
148 		   sizeof(p->p_sysent->sv_psstrings));
149 	return error;
150 }
151 
152 static int
153 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
154 {
155 	struct proc *p;
156 	int error;
157 
158 	p = curproc;
159 #ifdef SCTL_MASK32
160 	if (req->flags & SCTL_MASK32) {
161 		unsigned int val;
162 		val = (unsigned int)p->p_sysent->sv_usrstack;
163 		error = SYSCTL_OUT(req, &val, sizeof(val));
164 	} else
165 #endif
166 		error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
167 		    sizeof(p->p_sysent->sv_usrstack));
168 	return error;
169 }
170 
171 static int
172 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
173 {
174 	struct proc *p;
175 
176 	p = curproc;
177 	return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
178 	    sizeof(p->p_sysent->sv_stackprot)));
179 }
180 
181 /*
182  * Each of the items is a pointer to a `const struct execsw', hence the
183  * double pointer here.
184  */
185 static const struct execsw **execsw;
186 
187 #ifndef _SYS_SYSPROTO_H_
188 struct execve_args {
189 	char    *fname;
190 	char    **argv;
191 	char    **envv;
192 };
193 #endif
194 
195 int
196 sys_execve(td, uap)
197 	struct thread *td;
198 	struct execve_args /* {
199 		char *fname;
200 		char **argv;
201 		char **envv;
202 	} */ *uap;
203 {
204 	int error;
205 	struct image_args args;
206 
207 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
208 	    uap->argv, uap->envv);
209 	if (error == 0)
210 		error = kern_execve(td, &args, NULL);
211 	return (error);
212 }
213 
214 #ifndef _SYS_SYSPROTO_H_
215 struct fexecve_args {
216 	int	fd;
217 	char	**argv;
218 	char	**envv;
219 }
220 #endif
221 int
222 sys_fexecve(struct thread *td, struct fexecve_args *uap)
223 {
224 	int error;
225 	struct image_args args;
226 
227 	error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
228 	    uap->argv, uap->envv);
229 	if (error == 0) {
230 		args.fd = uap->fd;
231 		error = kern_execve(td, &args, NULL);
232 	}
233 	return (error);
234 }
235 
236 #ifndef _SYS_SYSPROTO_H_
237 struct __mac_execve_args {
238 	char	*fname;
239 	char	**argv;
240 	char	**envv;
241 	struct mac	*mac_p;
242 };
243 #endif
244 
245 int
246 sys___mac_execve(td, uap)
247 	struct thread *td;
248 	struct __mac_execve_args /* {
249 		char *fname;
250 		char **argv;
251 		char **envv;
252 		struct mac *mac_p;
253 	} */ *uap;
254 {
255 #ifdef MAC
256 	int error;
257 	struct image_args args;
258 
259 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
260 	    uap->argv, uap->envv);
261 	if (error == 0)
262 		error = kern_execve(td, &args, uap->mac_p);
263 	return (error);
264 #else
265 	return (ENOSYS);
266 #endif
267 }
268 
269 /*
270  * XXX: kern_execve has the astonishing property of not always returning to
271  * the caller.  If sufficiently bad things happen during the call to
272  * do_execve(), it can end up calling exit1(); as a result, callers must
273  * avoid doing anything which they might need to undo (e.g., allocating
274  * memory).
275  */
276 int
277 kern_execve(td, args, mac_p)
278 	struct thread *td;
279 	struct image_args *args;
280 	struct mac *mac_p;
281 {
282 	struct proc *p = td->td_proc;
283 	int error;
284 
285 	AUDIT_ARG_ARGV(args->begin_argv, args->argc,
286 	    args->begin_envv - args->begin_argv);
287 	AUDIT_ARG_ENVV(args->begin_envv, args->envc,
288 	    args->endp - args->begin_envv);
289 	if (p->p_flag & P_HADTHREADS) {
290 		PROC_LOCK(p);
291 		if (thread_single(SINGLE_BOUNDARY)) {
292 			PROC_UNLOCK(p);
293 	       		exec_free_args(args);
294 			return (ERESTART);	/* Try again later. */
295 		}
296 		PROC_UNLOCK(p);
297 	}
298 
299 	error = do_execve(td, args, mac_p);
300 
301 	if (p->p_flag & P_HADTHREADS) {
302 		PROC_LOCK(p);
303 		/*
304 		 * If success, we upgrade to SINGLE_EXIT state to
305 		 * force other threads to suicide.
306 		 */
307 		if (error == 0)
308 			thread_single(SINGLE_EXIT);
309 		else
310 			thread_single_end();
311 		PROC_UNLOCK(p);
312 	}
313 
314 	return (error);
315 }
316 
317 /*
318  * In-kernel implementation of execve().  All arguments are assumed to be
319  * userspace pointers from the passed thread.
320  */
321 static int
322 do_execve(td, args, mac_p)
323 	struct thread *td;
324 	struct image_args *args;
325 	struct mac *mac_p;
326 {
327 	struct proc *p = td->td_proc;
328 	struct nameidata nd;
329 	struct ucred *newcred = NULL, *oldcred;
330 	struct uidinfo *euip;
331 	register_t *stack_base;
332 	int error, i;
333 	struct image_params image_params, *imgp;
334 	struct vattr attr;
335 	int (*img_first)(struct image_params *);
336 	struct pargs *oldargs = NULL, *newargs = NULL;
337 	struct sigacts *oldsigacts, *newsigacts;
338 #ifdef KTRACE
339 	struct vnode *tracevp = NULL;
340 	struct ucred *tracecred = NULL;
341 #endif
342 	struct vnode *textvp = NULL, *binvp = NULL;
343 	int credential_changing;
344 	int vfslocked;
345 	int textset;
346 #ifdef MAC
347 	struct label *interpvplabel = NULL;
348 	int will_transition;
349 #endif
350 #ifdef HWPMC_HOOKS
351 	struct pmckern_procexec pe;
352 #endif
353 	static const char fexecv_proc_title[] = "(fexecv)";
354 
355 	vfslocked = 0;
356 	imgp = &image_params;
357 
358 	/*
359 	 * Lock the process and set the P_INEXEC flag to indicate that
360 	 * it should be left alone until we're done here.  This is
361 	 * necessary to avoid race conditions - e.g. in ptrace() -
362 	 * that might allow a local user to illicitly obtain elevated
363 	 * privileges.
364 	 */
365 	PROC_LOCK(p);
366 	KASSERT((p->p_flag & P_INEXEC) == 0,
367 	    ("%s(): process already has P_INEXEC flag", __func__));
368 	p->p_flag |= P_INEXEC;
369 	PROC_UNLOCK(p);
370 
371 	/*
372 	 * Initialize part of the common data
373 	 */
374 	imgp->proc = p;
375 	imgp->execlabel = NULL;
376 	imgp->attr = &attr;
377 	imgp->entry_addr = 0;
378 	imgp->reloc_base = 0;
379 	imgp->vmspace_destroyed = 0;
380 	imgp->interpreted = 0;
381 	imgp->opened = 0;
382 	imgp->interpreter_name = NULL;
383 	imgp->auxargs = NULL;
384 	imgp->vp = NULL;
385 	imgp->object = NULL;
386 	imgp->firstpage = NULL;
387 	imgp->ps_strings = 0;
388 	imgp->auxarg_size = 0;
389 	imgp->args = args;
390 	imgp->execpath = imgp->freepath = NULL;
391 	imgp->execpathp = 0;
392 	imgp->canary = 0;
393 	imgp->canarylen = 0;
394 	imgp->pagesizes = 0;
395 	imgp->pagesizeslen = 0;
396 	imgp->stack_prot = 0;
397 
398 #ifdef MAC
399 	error = mac_execve_enter(imgp, mac_p);
400 	if (error)
401 		goto exec_fail;
402 #endif
403 
404 	imgp->image_header = NULL;
405 
406 	/*
407 	 * Translate the file name. namei() returns a vnode pointer
408 	 *	in ni_vp amoung other things.
409 	 *
410 	 * XXXAUDIT: It would be desirable to also audit the name of the
411 	 * interpreter if this is an interpreted binary.
412 	 */
413 	if (args->fname != NULL) {
414 		NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME
415 		    | MPSAFE | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
416 	}
417 
418 	SDT_PROBE(proc, kernel, , exec, args->fname, 0, 0, 0, 0 );
419 
420 interpret:
421 	if (args->fname != NULL) {
422 #ifdef CAPABILITY_MODE
423 		/*
424 		 * While capability mode can't reach this point via direct
425 		 * path arguments to execve(), we also don't allow
426 		 * interpreters to be used in capability mode (for now).
427 		 * Catch indirect lookups and return a permissions error.
428 		 */
429 		if (IN_CAPABILITY_MODE(td)) {
430 			error = ECAPMODE;
431 			goto exec_fail;
432 		}
433 #endif
434 		error = namei(&nd);
435 		if (error)
436 			goto exec_fail;
437 
438 		vfslocked = NDHASGIANT(&nd);
439 		binvp  = nd.ni_vp;
440 		imgp->vp = binvp;
441 	} else {
442 		AUDIT_ARG_FD(args->fd);
443 		/*
444 		 * Some might argue that CAP_READ and/or CAP_MMAP should also
445 		 * be required here; such arguments will be entertained.
446 		 *
447 		 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
448 		 */
449 		error = fgetvp_exec(td, args->fd, CAP_FEXECVE, &binvp);
450 		if (error)
451 			goto exec_fail;
452 		vfslocked = VFS_LOCK_GIANT(binvp->v_mount);
453 		vn_lock(binvp, LK_EXCLUSIVE | LK_RETRY);
454 		AUDIT_ARG_VNODE1(binvp);
455 		imgp->vp = binvp;
456 	}
457 
458 	/*
459 	 * Check file permissions (also 'opens' file)
460 	 */
461 	error = exec_check_permissions(imgp);
462 	if (error)
463 		goto exec_fail_dealloc;
464 
465 	imgp->object = imgp->vp->v_object;
466 	if (imgp->object != NULL)
467 		vm_object_reference(imgp->object);
468 
469 	/*
470 	 * Set VV_TEXT now so no one can write to the executable while we're
471 	 * activating it.
472 	 *
473 	 * Remember if this was set before and unset it in case this is not
474 	 * actually an executable image.
475 	 */
476 	textset = imgp->vp->v_vflag & VV_TEXT;
477 	ASSERT_VOP_ELOCKED(imgp->vp, "vv_text");
478 	imgp->vp->v_vflag |= VV_TEXT;
479 
480 	error = exec_map_first_page(imgp);
481 	if (error)
482 		goto exec_fail_dealloc;
483 
484 	imgp->proc->p_osrel = 0;
485 	/*
486 	 *	If the current process has a special image activator it
487 	 *	wants to try first, call it.   For example, emulating shell
488 	 *	scripts differently.
489 	 */
490 	error = -1;
491 	if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
492 		error = img_first(imgp);
493 
494 	/*
495 	 *	Loop through the list of image activators, calling each one.
496 	 *	An activator returns -1 if there is no match, 0 on success,
497 	 *	and an error otherwise.
498 	 */
499 	for (i = 0; error == -1 && execsw[i]; ++i) {
500 		if (execsw[i]->ex_imgact == NULL ||
501 		    execsw[i]->ex_imgact == img_first) {
502 			continue;
503 		}
504 		error = (*execsw[i]->ex_imgact)(imgp);
505 	}
506 
507 	if (error) {
508 		if (error == -1) {
509 			if (textset == 0) {
510 				ASSERT_VOP_ELOCKED(imgp->vp, "vv_text");
511 				imgp->vp->v_vflag &= ~VV_TEXT;
512 			}
513 			error = ENOEXEC;
514 		}
515 		goto exec_fail_dealloc;
516 	}
517 
518 	/*
519 	 * Special interpreter operation, cleanup and loop up to try to
520 	 * activate the interpreter.
521 	 */
522 	if (imgp->interpreted) {
523 		exec_unmap_first_page(imgp);
524 		/*
525 		 * VV_TEXT needs to be unset for scripts.  There is a short
526 		 * period before we determine that something is a script where
527 		 * VV_TEXT will be set. The vnode lock is held over this
528 		 * entire period so nothing should illegitimately be blocked.
529 		 */
530 		imgp->vp->v_vflag &= ~VV_TEXT;
531 		/* free name buffer and old vnode */
532 		if (args->fname != NULL)
533 			NDFREE(&nd, NDF_ONLY_PNBUF);
534 #ifdef MAC
535 		mac_execve_interpreter_enter(binvp, &interpvplabel);
536 #endif
537 		if (imgp->opened) {
538 			VOP_CLOSE(binvp, FREAD, td->td_ucred, td);
539 			imgp->opened = 0;
540 		}
541 		vput(binvp);
542 		vm_object_deallocate(imgp->object);
543 		imgp->object = NULL;
544 		VFS_UNLOCK_GIANT(vfslocked);
545 		vfslocked = 0;
546 		/* set new name to that of the interpreter */
547 		NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME | MPSAFE,
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 	/*
568 	 * Copy out strings (args and env) and initialize stack base
569 	 */
570 	if (p->p_sysent->sv_copyout_strings)
571 		stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
572 	else
573 		stack_base = exec_copyout_strings(imgp);
574 
575 	/*
576 	 * If custom stack fixup routine present for this process
577 	 * let it do the stack setup.
578 	 * Else stuff argument count as first item on stack
579 	 */
580 	if (p->p_sysent->sv_fixup != NULL)
581 		(*p->p_sysent->sv_fixup)(&stack_base, imgp);
582 	else
583 		suword(--stack_base, imgp->args->argc);
584 
585 	/*
586 	 * For security and other reasons, the file descriptor table cannot
587 	 * be shared after an exec.
588 	 */
589 	fdunshare(p, td);
590 
591 	/*
592 	 * Malloc things before we need locks.
593 	 */
594 	newcred = crget();
595 	euip = uifind(attr.va_uid);
596 	i = imgp->args->begin_envv - imgp->args->begin_argv;
597 	/* Cache arguments if they fit inside our allowance */
598 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
599 		newargs = pargs_alloc(i);
600 		bcopy(imgp->args->begin_argv, newargs->ar_args, i);
601 	}
602 
603 	/* close files on exec */
604 	fdcloseexec(td);
605 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
606 
607 	/* Get a reference to the vnode prior to locking the proc */
608 	VREF(binvp);
609 
610 	/*
611 	 * For security and other reasons, signal handlers cannot
612 	 * be shared after an exec. The new process gets a copy of the old
613 	 * handlers. In execsigs(), the new process will have its signals
614 	 * reset.
615 	 */
616 	PROC_LOCK(p);
617 	oldcred = crcopysafe(p, newcred);
618 	if (sigacts_shared(p->p_sigacts)) {
619 		oldsigacts = p->p_sigacts;
620 		PROC_UNLOCK(p);
621 		newsigacts = sigacts_alloc();
622 		sigacts_copy(newsigacts, oldsigacts);
623 		PROC_LOCK(p);
624 		p->p_sigacts = newsigacts;
625 	} else
626 		oldsigacts = NULL;
627 
628 	/* Stop profiling */
629 	stopprofclock(p);
630 
631 	/* reset caught signals */
632 	execsigs(p);
633 
634 	/* name this process - nameiexec(p, ndp) */
635 	bzero(p->p_comm, sizeof(p->p_comm));
636 	if (args->fname)
637 		bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
638 		    min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
639 	else if (vn_commname(binvp, p->p_comm, sizeof(p->p_comm)) != 0)
640 		bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
641 	bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
642 #ifdef KTR
643 	sched_clear_tdname(td);
644 #endif
645 
646 	/*
647 	 * mark as execed, wakeup the process that vforked (if any) and tell
648 	 * it that it now has its own resources back
649 	 */
650 	p->p_flag |= P_EXEC;
651 	if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
652 		p->p_flag &= ~P_PPWAIT;
653 		cv_broadcast(&p->p_pwait);
654 	}
655 
656 	/*
657 	 * Implement image setuid/setgid.
658 	 *
659 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
660 	 * the process is being traced.
661 	 *
662 	 * We disable setuid/setgid/etc in compatibility mode on the basis
663 	 * that most setugid applications are not written with that
664 	 * environment in mind, and will therefore almost certainly operate
665 	 * incorrectly. In principle there's no reason that setugid
666 	 * applications might not be useful in capability mode, so we may want
667 	 * to reconsider this conservative design choice in the future.
668 	 *
669 	 * XXXMAC: For the time being, use NOSUID to also prohibit
670 	 * transitions on the file system.
671 	 */
672 	credential_changing = 0;
673 	credential_changing |= (attr.va_mode & S_ISUID) && oldcred->cr_uid !=
674 	    attr.va_uid;
675 	credential_changing |= (attr.va_mode & S_ISGID) && oldcred->cr_gid !=
676 	    attr.va_gid;
677 #ifdef MAC
678 	will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
679 	    interpvplabel, imgp);
680 	credential_changing |= will_transition;
681 #endif
682 
683 	if (credential_changing &&
684 #ifdef CAPABILITY_MODE
685 	    ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
686 #endif
687 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
688 	    (p->p_flag & P_TRACED) == 0) {
689 		/*
690 		 * Turn off syscall tracing for set-id programs, except for
691 		 * root.  Record any set-id flags first to make sure that
692 		 * we do not regain any tracing during a possible block.
693 		 */
694 		setsugid(p);
695 
696 #ifdef KTRACE
697 		if (priv_check_cred(oldcred, PRIV_DEBUG_DIFFCRED, 0))
698 			ktrprocexec(p, &tracecred, &tracevp);
699 #endif
700 		/*
701 		 * Close any file descriptors 0..2 that reference procfs,
702 		 * then make sure file descriptors 0..2 are in use.
703 		 *
704 		 * setugidsafety() may call closef() and then pfind()
705 		 * which may grab the process lock.
706 		 * fdcheckstd() may call falloc() which may block to
707 		 * allocate memory, so temporarily drop the process lock.
708 		 */
709 		PROC_UNLOCK(p);
710 		VOP_UNLOCK(imgp->vp, 0);
711 		setugidsafety(td);
712 		error = fdcheckstd(td);
713 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
714 		if (error != 0)
715 			goto done1;
716 		PROC_LOCK(p);
717 		/*
718 		 * Set the new credentials.
719 		 */
720 		if (attr.va_mode & S_ISUID)
721 			change_euid(newcred, euip);
722 		if (attr.va_mode & S_ISGID)
723 			change_egid(newcred, attr.va_gid);
724 #ifdef MAC
725 		if (will_transition) {
726 			mac_vnode_execve_transition(oldcred, newcred, imgp->vp,
727 			    interpvplabel, imgp);
728 		}
729 #endif
730 		/*
731 		 * Implement correct POSIX saved-id behavior.
732 		 *
733 		 * XXXMAC: Note that the current logic will save the
734 		 * uid and gid if a MAC domain transition occurs, even
735 		 * though maybe it shouldn't.
736 		 */
737 		change_svuid(newcred, newcred->cr_uid);
738 		change_svgid(newcred, newcred->cr_gid);
739 		p->p_ucred = newcred;
740 		newcred = NULL;
741 	} else {
742 		if (oldcred->cr_uid == oldcred->cr_ruid &&
743 		    oldcred->cr_gid == oldcred->cr_rgid)
744 			p->p_flag &= ~P_SUGID;
745 		/*
746 		 * Implement correct POSIX saved-id behavior.
747 		 *
748 		 * XXX: It's not clear that the existing behavior is
749 		 * POSIX-compliant.  A number of sources indicate that the
750 		 * saved uid/gid should only be updated if the new ruid is
751 		 * not equal to the old ruid, or the new euid is not equal
752 		 * to the old euid and the new euid is not equal to the old
753 		 * ruid.  The FreeBSD code always updates the saved uid/gid.
754 		 * Also, this code uses the new (replaced) euid and egid as
755 		 * the source, which may or may not be the right ones to use.
756 		 */
757 		if (oldcred->cr_svuid != oldcred->cr_uid ||
758 		    oldcred->cr_svgid != oldcred->cr_gid) {
759 			change_svuid(newcred, newcred->cr_uid);
760 			change_svgid(newcred, newcred->cr_gid);
761 			p->p_ucred = newcred;
762 			newcred = NULL;
763 		}
764 	}
765 
766 	/*
767 	 * Store the vp for use in procfs.  This vnode was referenced prior
768 	 * to locking the proc lock.
769 	 */
770 	textvp = p->p_textvp;
771 	p->p_textvp = binvp;
772 
773 #ifdef KDTRACE_HOOKS
774 	/*
775 	 * Tell the DTrace fasttrap provider about the exec if it
776 	 * has declared an interest.
777 	 */
778 	if (dtrace_fasttrap_exec)
779 		dtrace_fasttrap_exec(p);
780 #endif
781 
782 	/*
783 	 * Notify others that we exec'd, and clear the P_INEXEC flag
784 	 * as we're now a bona fide freshly-execed process.
785 	 */
786 	KNOTE_LOCKED(&p->p_klist, NOTE_EXEC);
787 	p->p_flag &= ~P_INEXEC;
788 
789 	/* clear "fork but no exec" flag, as we _are_ execing */
790 	p->p_acflag &= ~AFORK;
791 
792 	/*
793 	 * Free any previous argument cache and replace it with
794 	 * the new argument cache, if any.
795 	 */
796 	oldargs = p->p_args;
797 	p->p_args = newargs;
798 	newargs = NULL;
799 
800 #ifdef	HWPMC_HOOKS
801 	/*
802 	 * Check if system-wide sampling is in effect or if the
803 	 * current process is using PMCs.  If so, do exec() time
804 	 * processing.  This processing needs to happen AFTER the
805 	 * P_INEXEC flag is cleared.
806 	 *
807 	 * The proc lock needs to be released before taking the PMC
808 	 * SX.
809 	 */
810 	if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
811 		PROC_UNLOCK(p);
812 		VOP_UNLOCK(imgp->vp, 0);
813 		pe.pm_credentialschanged = credential_changing;
814 		pe.pm_entryaddr = imgp->entry_addr;
815 
816 		PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
817 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
818 	} else
819 		PROC_UNLOCK(p);
820 #else  /* !HWPMC_HOOKS */
821 	PROC_UNLOCK(p);
822 #endif
823 
824 	/* Set values passed into the program in registers. */
825 	if (p->p_sysent->sv_setregs)
826 		(*p->p_sysent->sv_setregs)(td, imgp,
827 		    (u_long)(uintptr_t)stack_base);
828 	else
829 		exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base);
830 
831 	vfs_mark_atime(imgp->vp, td->td_ucred);
832 
833 	SDT_PROBE(proc, kernel, , exec_success, args->fname, 0, 0, 0, 0);
834 
835 done1:
836 	/*
837 	 * Free any resources malloc'd earlier that we didn't use.
838 	 */
839 	uifree(euip);
840 	if (newcred == NULL)
841 		crfree(oldcred);
842 	else
843 		crfree(newcred);
844 	VOP_UNLOCK(imgp->vp, 0);
845 
846 	/*
847 	 * Handle deferred decrement of ref counts.
848 	 */
849 	if (textvp != NULL) {
850 		int tvfslocked;
851 
852 		tvfslocked = VFS_LOCK_GIANT(textvp->v_mount);
853 		vrele(textvp);
854 		VFS_UNLOCK_GIANT(tvfslocked);
855 	}
856 	if (binvp && error != 0)
857 		vrele(binvp);
858 #ifdef KTRACE
859 	if (tracevp != NULL) {
860 		int tvfslocked;
861 
862 		tvfslocked = VFS_LOCK_GIANT(tracevp->v_mount);
863 		vrele(tracevp);
864 		VFS_UNLOCK_GIANT(tvfslocked);
865 	}
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 	VFS_UNLOCK_GIANT(vfslocked);
923 	exec_free_args(args);
924 
925 	if (error && imgp->vmspace_destroyed) {
926 		/* sorry, no more process anymore. exit gracefully */
927 		exit1(td, W_EXITCODE(0, SIGABRT));
928 		/* NOT REACHED */
929 	}
930 
931 #ifdef KTRACE
932 	if (error == 0)
933 		ktrprocctor(p);
934 #endif
935 
936 	return (error);
937 }
938 
939 int
940 exec_map_first_page(imgp)
941 	struct image_params *imgp;
942 {
943 	int rv, i;
944 	int initial_pagein;
945 	vm_page_t ma[VM_INITIAL_PAGEIN];
946 	vm_object_t object;
947 
948 	if (imgp->firstpage != NULL)
949 		exec_unmap_first_page(imgp);
950 
951 	object = imgp->vp->v_object;
952 	if (object == NULL)
953 		return (EACCES);
954 	VM_OBJECT_LOCK(object);
955 #if VM_NRESERVLEVEL > 0
956 	if ((object->flags & OBJ_COLORED) == 0) {
957 		object->flags |= OBJ_COLORED;
958 		object->pg_color = 0;
959 	}
960 #endif
961 	ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
962 	if (ma[0]->valid != VM_PAGE_BITS_ALL) {
963 		initial_pagein = VM_INITIAL_PAGEIN;
964 		if (initial_pagein > object->size)
965 			initial_pagein = object->size;
966 		for (i = 1; i < initial_pagein; i++) {
967 			if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) {
968 				if (ma[i]->valid)
969 					break;
970 				if ((ma[i]->oflags & VPO_BUSY) || ma[i]->busy)
971 					break;
972 				vm_page_busy(ma[i]);
973 			} else {
974 				ma[i] = vm_page_alloc(object, i,
975 				    VM_ALLOC_NORMAL | VM_ALLOC_IFNOTCACHED);
976 				if (ma[i] == NULL)
977 					break;
978 			}
979 		}
980 		initial_pagein = i;
981 		rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
982 		ma[0] = vm_page_lookup(object, 0);
983 		if ((rv != VM_PAGER_OK) || (ma[0] == NULL)) {
984 			if (ma[0] != NULL) {
985 				vm_page_lock(ma[0]);
986 				vm_page_free(ma[0]);
987 				vm_page_unlock(ma[0]);
988 			}
989 			VM_OBJECT_UNLOCK(object);
990 			return (EIO);
991 		}
992 	}
993 	vm_page_lock(ma[0]);
994 	vm_page_hold(ma[0]);
995 	vm_page_unlock(ma[0]);
996 	vm_page_wakeup(ma[0]);
997 	VM_OBJECT_UNLOCK(object);
998 
999 	imgp->firstpage = sf_buf_alloc(ma[0], 0);
1000 	imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
1001 
1002 	return (0);
1003 }
1004 
1005 void
1006 exec_unmap_first_page(imgp)
1007 	struct image_params *imgp;
1008 {
1009 	vm_page_t m;
1010 
1011 	if (imgp->firstpage != NULL) {
1012 		m = sf_buf_page(imgp->firstpage);
1013 		sf_buf_free(imgp->firstpage);
1014 		imgp->firstpage = NULL;
1015 		vm_page_lock(m);
1016 		vm_page_unhold(m);
1017 		vm_page_unlock(m);
1018 	}
1019 }
1020 
1021 /*
1022  * Destroy old address space, and allocate a new stack
1023  *	The new stack is only SGROWSIZ large because it is grown
1024  *	automatically in trap.c.
1025  */
1026 int
1027 exec_new_vmspace(imgp, sv)
1028 	struct image_params *imgp;
1029 	struct sysentvec *sv;
1030 {
1031 	int error;
1032 	struct proc *p = imgp->proc;
1033 	struct vmspace *vmspace = p->p_vmspace;
1034 	vm_object_t obj;
1035 	vm_offset_t sv_minuser, stack_addr;
1036 	vm_map_t map;
1037 	u_long ssiz;
1038 
1039 	imgp->vmspace_destroyed = 1;
1040 	imgp->sysent = sv;
1041 
1042 	/* May be called with Giant held */
1043 	EVENTHANDLER_INVOKE(process_exec, p, imgp);
1044 
1045 	/*
1046 	 * Blow away entire process VM, if address space not shared,
1047 	 * otherwise, create a new VM space so that other threads are
1048 	 * not disrupted
1049 	 */
1050 	map = &vmspace->vm_map;
1051 	if (map_at_zero)
1052 		sv_minuser = sv->sv_minuser;
1053 	else
1054 		sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1055 	if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser &&
1056 	    vm_map_max(map) == sv->sv_maxuser) {
1057 		shmexit(vmspace);
1058 		pmap_remove_pages(vmspace_pmap(vmspace));
1059 		vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1060 	} else {
1061 		error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1062 		if (error)
1063 			return (error);
1064 		vmspace = p->p_vmspace;
1065 		map = &vmspace->vm_map;
1066 	}
1067 
1068 	/* Map a shared page */
1069 	obj = sv->sv_shared_page_obj;
1070 	if (obj != NULL) {
1071 		vm_object_reference(obj);
1072 		error = vm_map_fixed(map, obj, 0,
1073 		    sv->sv_shared_page_base, sv->sv_shared_page_len,
1074 		    VM_PROT_READ | VM_PROT_EXECUTE, VM_PROT_ALL,
1075 		    MAP_COPY_ON_WRITE | MAP_ACC_NO_CHARGE);
1076 		if (error) {
1077 			vm_object_deallocate(obj);
1078 			return (error);
1079 		}
1080 	}
1081 
1082 	/* Allocate a new stack */
1083 	if (sv->sv_maxssiz != NULL)
1084 		ssiz = *sv->sv_maxssiz;
1085 	else
1086 		ssiz = maxssiz;
1087 	stack_addr = sv->sv_usrstack - ssiz;
1088 	error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1089 	    obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1090 		sv->sv_stackprot,
1091 	    VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1092 	if (error)
1093 		return (error);
1094 
1095 #ifdef __ia64__
1096 	/* Allocate a new register stack */
1097 	stack_addr = IA64_BACKINGSTORE;
1098 	error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1099 	    sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_UP);
1100 	if (error)
1101 		return (error);
1102 #endif
1103 
1104 	/* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
1105 	 * VM_STACK case, but they are still used to monitor the size of the
1106 	 * process stack so we can check the stack rlimit.
1107 	 */
1108 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1109 	vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - ssiz;
1110 
1111 	return (0);
1112 }
1113 
1114 /*
1115  * Copy out argument and environment strings from the old process address
1116  * space into the temporary string buffer.
1117  */
1118 int
1119 exec_copyin_args(struct image_args *args, char *fname,
1120     enum uio_seg segflg, char **argv, char **envv)
1121 {
1122 	char *argp, *envp;
1123 	int error;
1124 	size_t length;
1125 
1126 	bzero(args, sizeof(*args));
1127 	if (argv == NULL)
1128 		return (EFAULT);
1129 
1130 	/*
1131 	 * Allocate demand-paged memory for the file name, argument, and
1132 	 * environment strings.
1133 	 */
1134 	error = exec_alloc_args(args);
1135 	if (error != 0)
1136 		return (error);
1137 
1138 	/*
1139 	 * Copy the file name.
1140 	 */
1141 	if (fname != NULL) {
1142 		args->fname = args->buf;
1143 		error = (segflg == UIO_SYSSPACE) ?
1144 		    copystr(fname, args->fname, PATH_MAX, &length) :
1145 		    copyinstr(fname, args->fname, PATH_MAX, &length);
1146 		if (error != 0)
1147 			goto err_exit;
1148 	} else
1149 		length = 0;
1150 
1151 	args->begin_argv = args->buf + length;
1152 	args->endp = args->begin_argv;
1153 	args->stringspace = ARG_MAX;
1154 
1155 	/*
1156 	 * extract arguments first
1157 	 */
1158 	while ((argp = (caddr_t) (intptr_t) fuword(argv++))) {
1159 		if (argp == (caddr_t) -1) {
1160 			error = EFAULT;
1161 			goto err_exit;
1162 		}
1163 		if ((error = copyinstr(argp, args->endp,
1164 		    args->stringspace, &length))) {
1165 			if (error == ENAMETOOLONG)
1166 				error = E2BIG;
1167 			goto err_exit;
1168 		}
1169 		args->stringspace -= length;
1170 		args->endp += length;
1171 		args->argc++;
1172 	}
1173 
1174 	args->begin_envv = args->endp;
1175 
1176 	/*
1177 	 * extract environment strings
1178 	 */
1179 	if (envv) {
1180 		while ((envp = (caddr_t)(intptr_t)fuword(envv++))) {
1181 			if (envp == (caddr_t)-1) {
1182 				error = EFAULT;
1183 				goto err_exit;
1184 			}
1185 			if ((error = copyinstr(envp, args->endp,
1186 			    args->stringspace, &length))) {
1187 				if (error == ENAMETOOLONG)
1188 					error = E2BIG;
1189 				goto err_exit;
1190 			}
1191 			args->stringspace -= length;
1192 			args->endp += length;
1193 			args->envc++;
1194 		}
1195 	}
1196 
1197 	return (0);
1198 
1199 err_exit:
1200 	exec_free_args(args);
1201 	return (error);
1202 }
1203 
1204 /*
1205  * Allocate temporary demand-paged, zero-filled memory for the file name,
1206  * argument, and environment strings.  Returns zero if the allocation succeeds
1207  * and ENOMEM otherwise.
1208  */
1209 int
1210 exec_alloc_args(struct image_args *args)
1211 {
1212 
1213 	args->buf = (char *)kmem_alloc_wait(exec_map, PATH_MAX + ARG_MAX);
1214 	return (args->buf != NULL ? 0 : ENOMEM);
1215 }
1216 
1217 void
1218 exec_free_args(struct image_args *args)
1219 {
1220 
1221 	if (args->buf != NULL) {
1222 		kmem_free_wakeup(exec_map, (vm_offset_t)args->buf,
1223 		    PATH_MAX + ARG_MAX);
1224 		args->buf = NULL;
1225 	}
1226 	if (args->fname_buf != NULL) {
1227 		free(args->fname_buf, M_TEMP);
1228 		args->fname_buf = NULL;
1229 	}
1230 }
1231 
1232 /*
1233  * Copy strings out to the new process address space, constructing new arg
1234  * and env vector tables. Return a pointer to the base so that it can be used
1235  * as the initial stack pointer.
1236  */
1237 register_t *
1238 exec_copyout_strings(imgp)
1239 	struct image_params *imgp;
1240 {
1241 	int argc, envc;
1242 	char **vectp;
1243 	char *stringp, *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 =	(caddr_t)arginfo - szsigcode - SPARE_USRSPACE -
1268 	    roundup(execpath_len, sizeof(char *)) -
1269 	    roundup(sizeof(canary), sizeof(char *)) -
1270 	    roundup(szps, sizeof(char *)) -
1271 	    roundup((ARG_MAX - imgp->args->stringspace), sizeof(char *));
1272 
1273 	/*
1274 	 * install sigcode
1275 	 */
1276 	if (szsigcode != 0)
1277 		copyout(p->p_sysent->sv_sigcode, ((caddr_t)arginfo -
1278 		    szsigcode), szsigcode);
1279 
1280 	/*
1281 	 * Copy the image path for the rtld.
1282 	 */
1283 	if (execpath_len != 0) {
1284 		imgp->execpathp = (uintptr_t)arginfo - szsigcode - execpath_len;
1285 		copyout(imgp->execpath, (void *)imgp->execpathp,
1286 		    execpath_len);
1287 	}
1288 
1289 	/*
1290 	 * Prepare the canary for SSP.
1291 	 */
1292 	arc4rand(canary, sizeof(canary), 0);
1293 	imgp->canary = (uintptr_t)arginfo - szsigcode - execpath_len -
1294 	    sizeof(canary);
1295 	copyout(canary, (void *)imgp->canary, sizeof(canary));
1296 	imgp->canarylen = sizeof(canary);
1297 
1298 	/*
1299 	 * Prepare the pagesizes array.
1300 	 */
1301 	imgp->pagesizes = (uintptr_t)arginfo - szsigcode - execpath_len -
1302 	    roundup(sizeof(canary), sizeof(char *)) - szps;
1303 	copyout(pagesizes, (void *)imgp->pagesizes, szps);
1304 	imgp->pagesizeslen = szps;
1305 
1306 	/*
1307 	 * If we have a valid auxargs ptr, prepare some room
1308 	 * on the stack.
1309 	 */
1310 	if (imgp->auxargs) {
1311 		/*
1312 		 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1313 		 * lower compatibility.
1314 		 */
1315 		imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1316 		    (AT_COUNT * 2);
1317 		/*
1318 		 * The '+ 2' is for the null pointers at the end of each of
1319 		 * the arg and env vector sets,and imgp->auxarg_size is room
1320 		 * for argument of Runtime loader.
1321 		 */
1322 		vectp = (char **)(destp - (imgp->args->argc +
1323 		    imgp->args->envc + 2 + imgp->auxarg_size)
1324 		    * sizeof(char *));
1325 	} else {
1326 		/*
1327 		 * The '+ 2' is for the null pointers at the end of each of
1328 		 * the arg and env vector sets
1329 		 */
1330 		vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc + 2) *
1331 		    sizeof(char *));
1332 	}
1333 
1334 	/*
1335 	 * vectp also becomes our initial stack base
1336 	 */
1337 	stack_base = (register_t *)vectp;
1338 
1339 	stringp = imgp->args->begin_argv;
1340 	argc = imgp->args->argc;
1341 	envc = imgp->args->envc;
1342 
1343 	/*
1344 	 * Copy out strings - arguments and environment.
1345 	 */
1346 	copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
1347 
1348 	/*
1349 	 * Fill in "ps_strings" struct for ps, w, etc.
1350 	 */
1351 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1352 	suword32(&arginfo->ps_nargvstr, argc);
1353 
1354 	/*
1355 	 * Fill in argument portion of vector table.
1356 	 */
1357 	for (; argc > 0; --argc) {
1358 		suword(vectp++, (long)(intptr_t)destp);
1359 		while (*stringp++ != 0)
1360 			destp++;
1361 		destp++;
1362 	}
1363 
1364 	/* a null vector table pointer separates the argp's from the envp's */
1365 	suword(vectp++, 0);
1366 
1367 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1368 	suword32(&arginfo->ps_nenvstr, envc);
1369 
1370 	/*
1371 	 * Fill in environment portion of vector table.
1372 	 */
1373 	for (; envc > 0; --envc) {
1374 		suword(vectp++, (long)(intptr_t)destp);
1375 		while (*stringp++ != 0)
1376 			destp++;
1377 		destp++;
1378 	}
1379 
1380 	/* end of vector table is a null pointer */
1381 	suword(vectp, 0);
1382 
1383 	return (stack_base);
1384 }
1385 
1386 /*
1387  * Check permissions of file to execute.
1388  *	Called with imgp->vp locked.
1389  *	Return 0 for success or error code on failure.
1390  */
1391 int
1392 exec_check_permissions(imgp)
1393 	struct image_params *imgp;
1394 {
1395 	struct vnode *vp = imgp->vp;
1396 	struct vattr *attr = imgp->attr;
1397 	struct thread *td;
1398 	int error;
1399 
1400 	td = curthread;
1401 
1402 	/* Get file attributes */
1403 	error = VOP_GETATTR(vp, attr, td->td_ucred);
1404 	if (error)
1405 		return (error);
1406 
1407 #ifdef MAC
1408 	error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1409 	if (error)
1410 		return (error);
1411 #endif
1412 
1413 	/*
1414 	 * 1) Check if file execution is disabled for the filesystem that
1415 	 *    this file resides on.
1416 	 * 2) Ensure that at least one execute bit is on. Otherwise, a
1417 	 *    privileged user will always succeed, and we don't want this
1418 	 *    to happen unless the file really is executable.
1419 	 * 3) Ensure that the file is a regular file.
1420 	 */
1421 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1422 	    (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1423 	    (attr->va_type != VREG))
1424 		return (EACCES);
1425 
1426 	/*
1427 	 * Zero length files can't be exec'd
1428 	 */
1429 	if (attr->va_size == 0)
1430 		return (ENOEXEC);
1431 
1432 	/*
1433 	 *  Check for execute permission to file based on current credentials.
1434 	 */
1435 	error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1436 	if (error)
1437 		return (error);
1438 
1439 	/*
1440 	 * Check number of open-for-writes on the file and deny execution
1441 	 * if there are any.
1442 	 */
1443 	if (vp->v_writecount)
1444 		return (ETXTBSY);
1445 
1446 	/*
1447 	 * Call filesystem specific open routine (which does nothing in the
1448 	 * general case).
1449 	 */
1450 	error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1451 	if (error == 0)
1452 		imgp->opened = 1;
1453 	return (error);
1454 }
1455 
1456 /*
1457  * Exec handler registration
1458  */
1459 int
1460 exec_register(execsw_arg)
1461 	const struct execsw *execsw_arg;
1462 {
1463 	const struct execsw **es, **xs, **newexecsw;
1464 	int count = 2;	/* New slot and trailing NULL */
1465 
1466 	if (execsw)
1467 		for (es = execsw; *es; es++)
1468 			count++;
1469 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1470 	if (newexecsw == NULL)
1471 		return (ENOMEM);
1472 	xs = newexecsw;
1473 	if (execsw)
1474 		for (es = execsw; *es; es++)
1475 			*xs++ = *es;
1476 	*xs++ = execsw_arg;
1477 	*xs = NULL;
1478 	if (execsw)
1479 		free(execsw, M_TEMP);
1480 	execsw = newexecsw;
1481 	return (0);
1482 }
1483 
1484 int
1485 exec_unregister(execsw_arg)
1486 	const struct execsw *execsw_arg;
1487 {
1488 	const struct execsw **es, **xs, **newexecsw;
1489 	int count = 1;
1490 
1491 	if (execsw == NULL)
1492 		panic("unregister with no handlers left?\n");
1493 
1494 	for (es = execsw; *es; es++) {
1495 		if (*es == execsw_arg)
1496 			break;
1497 	}
1498 	if (*es == NULL)
1499 		return (ENOENT);
1500 	for (es = execsw; *es; es++)
1501 		if (*es != execsw_arg)
1502 			count++;
1503 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1504 	if (newexecsw == NULL)
1505 		return (ENOMEM);
1506 	xs = newexecsw;
1507 	for (es = execsw; *es; es++)
1508 		if (*es != execsw_arg)
1509 			*xs++ = *es;
1510 	*xs = NULL;
1511 	if (execsw)
1512 		free(execsw, M_TEMP);
1513 	execsw = newexecsw;
1514 	return (0);
1515 }
1516