xref: /freebsd/sys/kern/kern_exec.c (revision a0e793cbf1951d07fc47a0d9ea389d7dacba5213)
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 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RWTUN, &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 	struct vmspace *oldvmspace;
284 	int error;
285 
286 	AUDIT_ARG_ARGV(args->begin_argv, args->argc,
287 	    args->begin_envv - args->begin_argv);
288 	AUDIT_ARG_ENVV(args->begin_envv, args->envc,
289 	    args->endp - args->begin_envv);
290 	if (p->p_flag & P_HADTHREADS) {
291 		PROC_LOCK(p);
292 		if (thread_single(SINGLE_BOUNDARY)) {
293 			PROC_UNLOCK(p);
294 	       		exec_free_args(args);
295 			return (ERESTART);	/* Try again later. */
296 		}
297 		PROC_UNLOCK(p);
298 	}
299 
300 	KASSERT((td->td_pflags & TDP_EXECVMSPC) == 0, ("nested execve"));
301 	oldvmspace = td->td_proc->p_vmspace;
302 	error = do_execve(td, args, mac_p);
303 
304 	if (p->p_flag & P_HADTHREADS) {
305 		PROC_LOCK(p);
306 		/*
307 		 * If success, we upgrade to SINGLE_EXIT state to
308 		 * force other threads to suicide.
309 		 */
310 		if (error == 0)
311 			thread_single(SINGLE_EXIT);
312 		else
313 			thread_single_end();
314 		PROC_UNLOCK(p);
315 	}
316 	if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
317 		KASSERT(td->td_proc->p_vmspace != oldvmspace,
318 		    ("oldvmspace still used"));
319 		vmspace_free(oldvmspace);
320 		td->td_pflags &= ~TDP_EXECVMSPC;
321 	}
322 
323 	return (error);
324 }
325 
326 /*
327  * In-kernel implementation of execve().  All arguments are assumed to be
328  * userspace pointers from the passed thread.
329  */
330 static int
331 do_execve(td, args, mac_p)
332 	struct thread *td;
333 	struct image_args *args;
334 	struct mac *mac_p;
335 {
336 	struct proc *p = td->td_proc;
337 	struct nameidata nd;
338 	struct ucred *newcred = NULL, *oldcred;
339 	struct uidinfo *euip = NULL;
340 	register_t *stack_base;
341 	int error, i;
342 	struct image_params image_params, *imgp;
343 	struct vattr attr;
344 	int (*img_first)(struct image_params *);
345 	struct pargs *oldargs = NULL, *newargs = NULL;
346 	struct sigacts *oldsigacts, *newsigacts;
347 #ifdef KTRACE
348 	struct vnode *tracevp = NULL;
349 	struct ucred *tracecred = NULL;
350 #endif
351 	struct vnode *textvp = NULL, *binvp = NULL;
352 	cap_rights_t rights;
353 	int credential_changing;
354 	int textset;
355 #ifdef MAC
356 	struct label *interpvplabel = NULL;
357 	int will_transition;
358 #endif
359 #ifdef HWPMC_HOOKS
360 	struct pmckern_procexec pe;
361 #endif
362 	static const char fexecv_proc_title[] = "(fexecv)";
363 
364 	imgp = &image_params;
365 
366 	/*
367 	 * Lock the process and set the P_INEXEC flag to indicate that
368 	 * it should be left alone until we're done here.  This is
369 	 * necessary to avoid race conditions - e.g. in ptrace() -
370 	 * that might allow a local user to illicitly obtain elevated
371 	 * privileges.
372 	 */
373 	PROC_LOCK(p);
374 	KASSERT((p->p_flag & P_INEXEC) == 0,
375 	    ("%s(): process already has P_INEXEC flag", __func__));
376 	p->p_flag |= P_INEXEC;
377 	PROC_UNLOCK(p);
378 
379 	/*
380 	 * Initialize part of the common data
381 	 */
382 	imgp->proc = p;
383 	imgp->execlabel = NULL;
384 	imgp->attr = &attr;
385 	imgp->entry_addr = 0;
386 	imgp->reloc_base = 0;
387 	imgp->vmspace_destroyed = 0;
388 	imgp->interpreted = 0;
389 	imgp->opened = 0;
390 	imgp->interpreter_name = NULL;
391 	imgp->auxargs = NULL;
392 	imgp->vp = NULL;
393 	imgp->object = NULL;
394 	imgp->firstpage = NULL;
395 	imgp->ps_strings = 0;
396 	imgp->auxarg_size = 0;
397 	imgp->args = args;
398 	imgp->execpath = imgp->freepath = NULL;
399 	imgp->execpathp = 0;
400 	imgp->canary = 0;
401 	imgp->canarylen = 0;
402 	imgp->pagesizes = 0;
403 	imgp->pagesizeslen = 0;
404 	imgp->stack_prot = 0;
405 
406 #ifdef MAC
407 	error = mac_execve_enter(imgp, mac_p);
408 	if (error)
409 		goto exec_fail;
410 #endif
411 
412 	imgp->image_header = NULL;
413 
414 	/*
415 	 * Translate the file name. namei() returns a vnode pointer
416 	 *	in ni_vp amoung other things.
417 	 *
418 	 * XXXAUDIT: It would be desirable to also audit the name of the
419 	 * interpreter if this is an interpreted binary.
420 	 */
421 	if (args->fname != NULL) {
422 		NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME
423 		    | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
424 	}
425 
426 	SDT_PROBE(proc, kernel, , exec, args->fname, 0, 0, 0, 0 );
427 
428 interpret:
429 	if (args->fname != NULL) {
430 #ifdef CAPABILITY_MODE
431 		/*
432 		 * While capability mode can't reach this point via direct
433 		 * path arguments to execve(), we also don't allow
434 		 * interpreters to be used in capability mode (for now).
435 		 * Catch indirect lookups and return a permissions error.
436 		 */
437 		if (IN_CAPABILITY_MODE(td)) {
438 			error = ECAPMODE;
439 			goto exec_fail;
440 		}
441 #endif
442 		error = namei(&nd);
443 		if (error)
444 			goto exec_fail;
445 
446 		binvp  = nd.ni_vp;
447 		imgp->vp = binvp;
448 	} else {
449 		AUDIT_ARG_FD(args->fd);
450 		/*
451 		 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
452 		 */
453 		error = fgetvp_exec(td, args->fd,
454 		    cap_rights_init(&rights, CAP_FEXECVE), &binvp);
455 		if (error)
456 			goto exec_fail;
457 		vn_lock(binvp, LK_EXCLUSIVE | LK_RETRY);
458 		AUDIT_ARG_VNODE1(binvp);
459 		imgp->vp = binvp;
460 	}
461 
462 	/*
463 	 * Check file permissions (also 'opens' file)
464 	 */
465 	error = exec_check_permissions(imgp);
466 	if (error)
467 		goto exec_fail_dealloc;
468 
469 	imgp->object = imgp->vp->v_object;
470 	if (imgp->object != NULL)
471 		vm_object_reference(imgp->object);
472 
473 	/*
474 	 * Set VV_TEXT now so no one can write to the executable while we're
475 	 * activating it.
476 	 *
477 	 * Remember if this was set before and unset it in case this is not
478 	 * actually an executable image.
479 	 */
480 	textset = VOP_IS_TEXT(imgp->vp);
481 	VOP_SET_TEXT(imgp->vp);
482 
483 	error = exec_map_first_page(imgp);
484 	if (error)
485 		goto exec_fail_dealloc;
486 
487 	imgp->proc->p_osrel = 0;
488 	/*
489 	 *	If the current process has a special image activator it
490 	 *	wants to try first, call it.   For example, emulating shell
491 	 *	scripts differently.
492 	 */
493 	error = -1;
494 	if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
495 		error = img_first(imgp);
496 
497 	/*
498 	 *	Loop through the list of image activators, calling each one.
499 	 *	An activator returns -1 if there is no match, 0 on success,
500 	 *	and an error otherwise.
501 	 */
502 	for (i = 0; error == -1 && execsw[i]; ++i) {
503 		if (execsw[i]->ex_imgact == NULL ||
504 		    execsw[i]->ex_imgact == img_first) {
505 			continue;
506 		}
507 		error = (*execsw[i]->ex_imgact)(imgp);
508 	}
509 
510 	if (error) {
511 		if (error == -1) {
512 			if (textset == 0)
513 				VOP_UNSET_TEXT(imgp->vp);
514 			error = ENOEXEC;
515 		}
516 		goto exec_fail_dealloc;
517 	}
518 
519 	/*
520 	 * Special interpreter operation, cleanup and loop up to try to
521 	 * activate the interpreter.
522 	 */
523 	if (imgp->interpreted) {
524 		exec_unmap_first_page(imgp);
525 		/*
526 		 * VV_TEXT needs to be unset for scripts.  There is a short
527 		 * period before we determine that something is a script where
528 		 * VV_TEXT will be set. The vnode lock is held over this
529 		 * entire period so nothing should illegitimately be blocked.
530 		 */
531 		VOP_UNSET_TEXT(imgp->vp);
532 		/* free name buffer and old vnode */
533 		if (args->fname != NULL)
534 			NDFREE(&nd, NDF_ONLY_PNBUF);
535 #ifdef MAC
536 		mac_execve_interpreter_enter(binvp, &interpvplabel);
537 #endif
538 		if (imgp->opened) {
539 			VOP_CLOSE(binvp, FREAD, td->td_ucred, td);
540 			imgp->opened = 0;
541 		}
542 		vput(binvp);
543 		vm_object_deallocate(imgp->object);
544 		imgp->object = NULL;
545 		/* set new name to that of the interpreter */
546 		NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
547 		    UIO_SYSSPACE, imgp->interpreter_name, td);
548 		args->fname = imgp->interpreter_name;
549 		goto interpret;
550 	}
551 
552 	/*
553 	 * NB: We unlock the vnode here because it is believed that none
554 	 * of the sv_copyout_strings/sv_fixup operations require the vnode.
555 	 */
556 	VOP_UNLOCK(imgp->vp, 0);
557 
558 	/*
559 	 * Do the best to calculate the full path to the image file.
560 	 */
561 	if (imgp->auxargs != NULL &&
562 	    ((args->fname != NULL && args->fname[0] == '/') ||
563 	     vn_fullpath(td, imgp->vp, &imgp->execpath, &imgp->freepath) != 0))
564 		imgp->execpath = args->fname;
565 
566 	if (disallow_high_osrel &&
567 	    P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
568 		error = ENOEXEC;
569 		uprintf("Osrel %d for image %s too high\n", p->p_osrel,
570 		    imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
571 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
572 		goto exec_fail_dealloc;
573 	}
574 
575 	/*
576 	 * Copy out strings (args and env) and initialize stack base
577 	 */
578 	if (p->p_sysent->sv_copyout_strings)
579 		stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
580 	else
581 		stack_base = exec_copyout_strings(imgp);
582 
583 	/*
584 	 * If custom stack fixup routine present for this process
585 	 * let it do the stack setup.
586 	 * Else stuff argument count as first item on stack
587 	 */
588 	if (p->p_sysent->sv_fixup != NULL)
589 		(*p->p_sysent->sv_fixup)(&stack_base, imgp);
590 	else
591 		suword(--stack_base, imgp->args->argc);
592 
593 	/*
594 	 * For security and other reasons, the file descriptor table cannot
595 	 * be shared after an exec.
596 	 */
597 	fdunshare(td);
598 	/* close files on exec */
599 	fdcloseexec(td);
600 
601 	/*
602 	 * Malloc things before we need locks.
603 	 */
604 	i = imgp->args->begin_envv - imgp->args->begin_argv;
605 	/* Cache arguments if they fit inside our allowance */
606 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
607 		newargs = pargs_alloc(i);
608 		bcopy(imgp->args->begin_argv, newargs->ar_args, i);
609 	}
610 
611 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
612 
613 	/* Get a reference to the vnode prior to locking the proc */
614 	VREF(binvp);
615 
616 	/*
617 	 * For security and other reasons, signal handlers cannot
618 	 * be shared after an exec. The new process gets a copy of the old
619 	 * handlers. In execsigs(), the new process will have its signals
620 	 * reset.
621 	 */
622 	if (sigacts_shared(p->p_sigacts)) {
623 		oldsigacts = p->p_sigacts;
624 		newsigacts = sigacts_alloc();
625 		sigacts_copy(newsigacts, oldsigacts);
626 	} else {
627 		oldsigacts = NULL;
628 		newsigacts = NULL; /* satisfy gcc */
629 	}
630 
631 	PROC_LOCK(p);
632 	if (oldsigacts)
633 		p->p_sigacts = newsigacts;
634 	oldcred = p->p_ucred;
635 	/* Stop profiling */
636 	stopprofclock(p);
637 
638 	/* reset caught signals */
639 	execsigs(p);
640 
641 	/* name this process - nameiexec(p, ndp) */
642 	bzero(p->p_comm, sizeof(p->p_comm));
643 	if (args->fname)
644 		bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
645 		    min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
646 	else if (vn_commname(binvp, p->p_comm, sizeof(p->p_comm)) != 0)
647 		bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
648 	bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
649 #ifdef KTR
650 	sched_clear_tdname(td);
651 #endif
652 
653 	/*
654 	 * mark as execed, wakeup the process that vforked (if any) and tell
655 	 * it that it now has its own resources back
656 	 */
657 	p->p_flag |= P_EXEC;
658 	if (p->p_flag & P_PPWAIT) {
659 		p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
660 		cv_broadcast(&p->p_pwait);
661 	}
662 
663 	/*
664 	 * Implement image setuid/setgid.
665 	 *
666 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
667 	 * the process is being traced.
668 	 *
669 	 * We disable setuid/setgid/etc in compatibility mode on the basis
670 	 * that most setugid applications are not written with that
671 	 * environment in mind, and will therefore almost certainly operate
672 	 * incorrectly. In principle there's no reason that setugid
673 	 * applications might not be useful in capability mode, so we may want
674 	 * to reconsider this conservative design choice in the future.
675 	 *
676 	 * XXXMAC: For the time being, use NOSUID to also prohibit
677 	 * transitions on the file system.
678 	 */
679 	credential_changing = 0;
680 	credential_changing |= (attr.va_mode & S_ISUID) && oldcred->cr_uid !=
681 	    attr.va_uid;
682 	credential_changing |= (attr.va_mode & S_ISGID) && oldcred->cr_gid !=
683 	    attr.va_gid;
684 #ifdef MAC
685 	will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
686 	    interpvplabel, imgp);
687 	credential_changing |= will_transition;
688 #endif
689 
690 	if (credential_changing &&
691 #ifdef CAPABILITY_MODE
692 	    ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
693 #endif
694 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
695 	    (p->p_flag & P_TRACED) == 0) {
696 		/*
697 		 * Turn off syscall tracing for set-id programs, except for
698 		 * root.  Record any set-id flags first to make sure that
699 		 * we do not regain any tracing during a possible block.
700 		 */
701 		setsugid(p);
702 
703 #ifdef KTRACE
704 		if (p->p_tracecred != NULL &&
705 		    priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED, 0))
706 			ktrprocexec(p, &tracecred, &tracevp);
707 #endif
708 		/*
709 		 * Close any file descriptors 0..2 that reference procfs,
710 		 * then make sure file descriptors 0..2 are in use.
711 		 *
712 		 * setugidsafety() may call closef() and then pfind()
713 		 * which may grab the process lock.
714 		 * fdcheckstd() may call falloc() which may block to
715 		 * allocate memory, so temporarily drop the process lock.
716 		 */
717 		PROC_UNLOCK(p);
718 		VOP_UNLOCK(imgp->vp, 0);
719 		setugidsafety(td);
720 		error = fdcheckstd(td);
721 		if (error != 0)
722 			goto done1;
723 		newcred = crdup(oldcred);
724 		euip = uifind(attr.va_uid);
725 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
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 	} else {
751 		if (oldcred->cr_uid == oldcred->cr_ruid &&
752 		    oldcred->cr_gid == oldcred->cr_rgid)
753 			p->p_flag &= ~P_SUGID;
754 		/*
755 		 * Implement correct POSIX saved-id behavior.
756 		 *
757 		 * XXX: It's not clear that the existing behavior is
758 		 * POSIX-compliant.  A number of sources indicate that the
759 		 * saved uid/gid should only be updated if the new ruid is
760 		 * not equal to the old ruid, or the new euid is not equal
761 		 * to the old euid and the new euid is not equal to the old
762 		 * ruid.  The FreeBSD code always updates the saved uid/gid.
763 		 * Also, this code uses the new (replaced) euid and egid as
764 		 * the source, which may or may not be the right ones to use.
765 		 */
766 		if (oldcred->cr_svuid != oldcred->cr_uid ||
767 		    oldcred->cr_svgid != oldcred->cr_gid) {
768 			PROC_UNLOCK(p);
769 			VOP_UNLOCK(imgp->vp, 0);
770 			newcred = crdup(oldcred);
771 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
772 			PROC_LOCK(p);
773 			change_svuid(newcred, newcred->cr_uid);
774 			change_svgid(newcred, newcred->cr_gid);
775 			p->p_ucred = newcred;
776 		}
777 	}
778 
779 	/*
780 	 * Store the vp for use in procfs.  This vnode was referenced prior
781 	 * to locking the proc lock.
782 	 */
783 	textvp = p->p_textvp;
784 	p->p_textvp = binvp;
785 
786 #ifdef KDTRACE_HOOKS
787 	/*
788 	 * Tell the DTrace fasttrap provider about the exec if it
789 	 * has declared an interest.
790 	 */
791 	if (dtrace_fasttrap_exec)
792 		dtrace_fasttrap_exec(p);
793 #endif
794 
795 	/*
796 	 * Notify others that we exec'd, and clear the P_INEXEC flag
797 	 * as we're now a bona fide freshly-execed process.
798 	 */
799 	KNOTE_LOCKED(&p->p_klist, NOTE_EXEC);
800 	p->p_flag &= ~P_INEXEC;
801 
802 	/* clear "fork but no exec" flag, as we _are_ execing */
803 	p->p_acflag &= ~AFORK;
804 
805 	/*
806 	 * Free any previous argument cache and replace it with
807 	 * the new argument cache, if any.
808 	 */
809 	oldargs = p->p_args;
810 	p->p_args = newargs;
811 	newargs = NULL;
812 
813 #ifdef	HWPMC_HOOKS
814 	/*
815 	 * Check if system-wide sampling is in effect or if the
816 	 * current process is using PMCs.  If so, do exec() time
817 	 * processing.  This processing needs to happen AFTER the
818 	 * P_INEXEC flag is cleared.
819 	 *
820 	 * The proc lock needs to be released before taking the PMC
821 	 * SX.
822 	 */
823 	if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
824 		PROC_UNLOCK(p);
825 		VOP_UNLOCK(imgp->vp, 0);
826 		pe.pm_credentialschanged = credential_changing;
827 		pe.pm_entryaddr = imgp->entry_addr;
828 
829 		PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
830 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
831 	} else
832 		PROC_UNLOCK(p);
833 #else  /* !HWPMC_HOOKS */
834 	PROC_UNLOCK(p);
835 #endif
836 
837 	/* Set values passed into the program in registers. */
838 	if (p->p_sysent->sv_setregs)
839 		(*p->p_sysent->sv_setregs)(td, imgp,
840 		    (u_long)(uintptr_t)stack_base);
841 	else
842 		exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base);
843 
844 	vfs_mark_atime(imgp->vp, td->td_ucred);
845 
846 	SDT_PROBE(proc, kernel, , exec__success, args->fname, 0, 0, 0, 0);
847 
848 	VOP_UNLOCK(imgp->vp, 0);
849 done1:
850 	/*
851 	 * Free any resources malloc'd earlier that we didn't use.
852 	 */
853 	if (euip != NULL)
854 		uifree(euip);
855 	if (newcred != NULL)
856 		crfree(oldcred);
857 
858 	/*
859 	 * Handle deferred decrement of ref counts.
860 	 */
861 	if (textvp != NULL)
862 		vrele(textvp);
863 	if (binvp && error != 0)
864 		vrele(binvp);
865 #ifdef KTRACE
866 	if (tracevp != NULL)
867 		vrele(tracevp);
868 	if (tracecred != NULL)
869 		crfree(tracecred);
870 #endif
871 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
872 	pargs_drop(oldargs);
873 	pargs_drop(newargs);
874 	if (oldsigacts != NULL)
875 		sigacts_free(oldsigacts);
876 
877 exec_fail_dealloc:
878 
879 	/*
880 	 * free various allocated resources
881 	 */
882 	if (imgp->firstpage != NULL)
883 		exec_unmap_first_page(imgp);
884 
885 	if (imgp->vp != NULL) {
886 		if (args->fname)
887 			NDFREE(&nd, NDF_ONLY_PNBUF);
888 		if (imgp->opened)
889 			VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
890 		vput(imgp->vp);
891 	}
892 
893 	if (imgp->object != NULL)
894 		vm_object_deallocate(imgp->object);
895 
896 	free(imgp->freepath, M_TEMP);
897 
898 	if (error == 0) {
899 		PROC_LOCK(p);
900 		td->td_dbgflags |= TDB_EXEC;
901 		PROC_UNLOCK(p);
902 
903 		/*
904 		 * Stop the process here if its stop event mask has
905 		 * the S_EXEC bit set.
906 		 */
907 		STOPEVENT(p, S_EXEC, 0);
908 		goto done2;
909 	}
910 
911 exec_fail:
912 	/* we're done here, clear P_INEXEC */
913 	PROC_LOCK(p);
914 	p->p_flag &= ~P_INEXEC;
915 	PROC_UNLOCK(p);
916 
917 	SDT_PROBE(proc, kernel, , exec__failure, error, 0, 0, 0, 0);
918 
919 done2:
920 #ifdef MAC
921 	mac_execve_exit(imgp);
922 	mac_execve_interpreter_exit(interpvplabel);
923 #endif
924 	exec_free_args(args);
925 
926 	if (error && imgp->vmspace_destroyed) {
927 		/* sorry, no more process anymore. exit gracefully */
928 		exit1(td, W_EXITCODE(0, SIGABRT));
929 		/* NOT REACHED */
930 	}
931 
932 #ifdef KTRACE
933 	if (error == 0)
934 		ktrprocctor(p);
935 #endif
936 
937 	return (error);
938 }
939 
940 int
941 exec_map_first_page(imgp)
942 	struct image_params *imgp;
943 {
944 	int rv, i;
945 	int initial_pagein;
946 	vm_page_t ma[VM_INITIAL_PAGEIN];
947 	vm_object_t object;
948 
949 	if (imgp->firstpage != NULL)
950 		exec_unmap_first_page(imgp);
951 
952 	object = imgp->vp->v_object;
953 	if (object == NULL)
954 		return (EACCES);
955 	VM_OBJECT_WLOCK(object);
956 #if VM_NRESERVLEVEL > 0
957 	if ((object->flags & OBJ_COLORED) == 0) {
958 		object->flags |= OBJ_COLORED;
959 		object->pg_color = 0;
960 	}
961 #endif
962 	ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL);
963 	if (ma[0]->valid != VM_PAGE_BITS_ALL) {
964 		initial_pagein = VM_INITIAL_PAGEIN;
965 		if (initial_pagein > object->size)
966 			initial_pagein = object->size;
967 		for (i = 1; i < initial_pagein; i++) {
968 			if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) {
969 				if (ma[i]->valid)
970 					break;
971 				if (vm_page_tryxbusy(ma[i]))
972 					break;
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_WUNLOCK(object);
990 			return (EIO);
991 		}
992 	}
993 	vm_page_xunbusy(ma[0]);
994 	vm_page_lock(ma[0]);
995 	vm_page_hold(ma[0]);
996 	vm_page_unlock(ma[0]);
997 	VM_OBJECT_WUNLOCK(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,
1075 		    VM_PROT_READ | VM_PROT_EXECUTE,
1076 		    MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1077 		if (error) {
1078 			vm_object_deallocate(obj);
1079 			return (error);
1080 		}
1081 	}
1082 
1083 	/* Allocate a new stack */
1084 	if (sv->sv_maxssiz != NULL)
1085 		ssiz = *sv->sv_maxssiz;
1086 	else
1087 		ssiz = maxssiz;
1088 	stack_addr = sv->sv_usrstack - ssiz;
1089 	error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1090 	    obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1091 		sv->sv_stackprot,
1092 	    VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1093 	if (error)
1094 		return (error);
1095 
1096 	/*
1097 	 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they
1098 	 * are still used to enforce the stack rlimit on the process stack.
1099 	 */
1100 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1101 	vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - ssiz;
1102 
1103 	return (0);
1104 }
1105 
1106 /*
1107  * Copy out argument and environment strings from the old process address
1108  * space into the temporary string buffer.
1109  */
1110 int
1111 exec_copyin_args(struct image_args *args, char *fname,
1112     enum uio_seg segflg, char **argv, char **envv)
1113 {
1114 	char *argp, *envp;
1115 	int error;
1116 	size_t length;
1117 
1118 	bzero(args, sizeof(*args));
1119 	if (argv == NULL)
1120 		return (EFAULT);
1121 
1122 	/*
1123 	 * Allocate demand-paged memory for the file name, argument, and
1124 	 * environment strings.
1125 	 */
1126 	error = exec_alloc_args(args);
1127 	if (error != 0)
1128 		return (error);
1129 
1130 	/*
1131 	 * Copy the file name.
1132 	 */
1133 	if (fname != NULL) {
1134 		args->fname = args->buf;
1135 		error = (segflg == UIO_SYSSPACE) ?
1136 		    copystr(fname, args->fname, PATH_MAX, &length) :
1137 		    copyinstr(fname, args->fname, PATH_MAX, &length);
1138 		if (error != 0)
1139 			goto err_exit;
1140 	} else
1141 		length = 0;
1142 
1143 	args->begin_argv = args->buf + length;
1144 	args->endp = args->begin_argv;
1145 	args->stringspace = ARG_MAX;
1146 
1147 	/*
1148 	 * extract arguments first
1149 	 */
1150 	while ((argp = (caddr_t) (intptr_t) fuword(argv++))) {
1151 		if (argp == (caddr_t) -1) {
1152 			error = EFAULT;
1153 			goto err_exit;
1154 		}
1155 		if ((error = copyinstr(argp, args->endp,
1156 		    args->stringspace, &length))) {
1157 			if (error == ENAMETOOLONG)
1158 				error = E2BIG;
1159 			goto err_exit;
1160 		}
1161 		args->stringspace -= length;
1162 		args->endp += length;
1163 		args->argc++;
1164 	}
1165 
1166 	args->begin_envv = args->endp;
1167 
1168 	/*
1169 	 * extract environment strings
1170 	 */
1171 	if (envv) {
1172 		while ((envp = (caddr_t)(intptr_t)fuword(envv++))) {
1173 			if (envp == (caddr_t)-1) {
1174 				error = EFAULT;
1175 				goto err_exit;
1176 			}
1177 			if ((error = copyinstr(envp, args->endp,
1178 			    args->stringspace, &length))) {
1179 				if (error == ENAMETOOLONG)
1180 					error = E2BIG;
1181 				goto err_exit;
1182 			}
1183 			args->stringspace -= length;
1184 			args->endp += length;
1185 			args->envc++;
1186 		}
1187 	}
1188 
1189 	return (0);
1190 
1191 err_exit:
1192 	exec_free_args(args);
1193 	return (error);
1194 }
1195 
1196 /*
1197  * Allocate temporary demand-paged, zero-filled memory for the file name,
1198  * argument, and environment strings.  Returns zero if the allocation succeeds
1199  * and ENOMEM otherwise.
1200  */
1201 int
1202 exec_alloc_args(struct image_args *args)
1203 {
1204 
1205 	args->buf = (char *)kmap_alloc_wait(exec_map, PATH_MAX + ARG_MAX);
1206 	return (args->buf != NULL ? 0 : ENOMEM);
1207 }
1208 
1209 void
1210 exec_free_args(struct image_args *args)
1211 {
1212 
1213 	if (args->buf != NULL) {
1214 		kmap_free_wakeup(exec_map, (vm_offset_t)args->buf,
1215 		    PATH_MAX + ARG_MAX);
1216 		args->buf = NULL;
1217 	}
1218 	if (args->fname_buf != NULL) {
1219 		free(args->fname_buf, M_TEMP);
1220 		args->fname_buf = NULL;
1221 	}
1222 }
1223 
1224 /*
1225  * Copy strings out to the new process address space, constructing new arg
1226  * and env vector tables. Return a pointer to the base so that it can be used
1227  * as the initial stack pointer.
1228  */
1229 register_t *
1230 exec_copyout_strings(imgp)
1231 	struct image_params *imgp;
1232 {
1233 	int argc, envc;
1234 	char **vectp;
1235 	char *stringp;
1236 	uintptr_t destp;
1237 	register_t *stack_base;
1238 	struct ps_strings *arginfo;
1239 	struct proc *p;
1240 	size_t execpath_len;
1241 	int szsigcode, szps;
1242 	char canary[sizeof(long) * 8];
1243 
1244 	szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1245 	/*
1246 	 * Calculate string base and vector table pointers.
1247 	 * Also deal with signal trampoline code for this exec type.
1248 	 */
1249 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
1250 		execpath_len = strlen(imgp->execpath) + 1;
1251 	else
1252 		execpath_len = 0;
1253 	p = imgp->proc;
1254 	szsigcode = 0;
1255 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1256 	if (p->p_sysent->sv_sigcode_base == 0) {
1257 		if (p->p_sysent->sv_szsigcode != NULL)
1258 			szsigcode = *(p->p_sysent->sv_szsigcode);
1259 	}
1260 	destp =	(uintptr_t)arginfo;
1261 
1262 	/*
1263 	 * install sigcode
1264 	 */
1265 	if (szsigcode != 0) {
1266 		destp -= szsigcode;
1267 		destp = rounddown2(destp, sizeof(void *));
1268 		copyout(p->p_sysent->sv_sigcode, (void *)destp, szsigcode);
1269 	}
1270 
1271 	/*
1272 	 * Copy the image path for the rtld.
1273 	 */
1274 	if (execpath_len != 0) {
1275 		destp -= execpath_len;
1276 		imgp->execpathp = destp;
1277 		copyout(imgp->execpath, (void *)destp, execpath_len);
1278 	}
1279 
1280 	/*
1281 	 * Prepare the canary for SSP.
1282 	 */
1283 	arc4rand(canary, sizeof(canary), 0);
1284 	destp -= sizeof(canary);
1285 	imgp->canary = destp;
1286 	copyout(canary, (void *)destp, sizeof(canary));
1287 	imgp->canarylen = sizeof(canary);
1288 
1289 	/*
1290 	 * Prepare the pagesizes array.
1291 	 */
1292 	destp -= szps;
1293 	destp = rounddown2(destp, sizeof(void *));
1294 	imgp->pagesizes = destp;
1295 	copyout(pagesizes, (void *)destp, szps);
1296 	imgp->pagesizeslen = szps;
1297 
1298 	destp -= ARG_MAX - imgp->args->stringspace;
1299 	destp = rounddown2(destp, sizeof(void *));
1300 
1301 	/*
1302 	 * If we have a valid auxargs ptr, prepare some room
1303 	 * on the stack.
1304 	 */
1305 	if (imgp->auxargs) {
1306 		/*
1307 		 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1308 		 * lower compatibility.
1309 		 */
1310 		imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1311 		    (AT_COUNT * 2);
1312 		/*
1313 		 * The '+ 2' is for the null pointers at the end of each of
1314 		 * the arg and env vector sets,and imgp->auxarg_size is room
1315 		 * for argument of Runtime loader.
1316 		 */
1317 		vectp = (char **)(destp - (imgp->args->argc +
1318 		    imgp->args->envc + 2 + imgp->auxarg_size)
1319 		    * sizeof(char *));
1320 	} else {
1321 		/*
1322 		 * The '+ 2' is for the null pointers at the end of each of
1323 		 * the arg and env vector sets
1324 		 */
1325 		vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc
1326 		    + 2) * sizeof(char *));
1327 	}
1328 
1329 	/*
1330 	 * vectp also becomes our initial stack base
1331 	 */
1332 	stack_base = (register_t *)vectp;
1333 
1334 	stringp = imgp->args->begin_argv;
1335 	argc = imgp->args->argc;
1336 	envc = imgp->args->envc;
1337 
1338 	/*
1339 	 * Copy out strings - arguments and environment.
1340 	 */
1341 	copyout(stringp, (void *)destp, ARG_MAX - imgp->args->stringspace);
1342 
1343 	/*
1344 	 * Fill in "ps_strings" struct for ps, w, etc.
1345 	 */
1346 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1347 	suword32(&arginfo->ps_nargvstr, argc);
1348 
1349 	/*
1350 	 * Fill in argument portion of vector table.
1351 	 */
1352 	for (; argc > 0; --argc) {
1353 		suword(vectp++, (long)(intptr_t)destp);
1354 		while (*stringp++ != 0)
1355 			destp++;
1356 		destp++;
1357 	}
1358 
1359 	/* a null vector table pointer separates the argp's from the envp's */
1360 	suword(vectp++, 0);
1361 
1362 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1363 	suword32(&arginfo->ps_nenvstr, envc);
1364 
1365 	/*
1366 	 * Fill in environment portion of vector table.
1367 	 */
1368 	for (; envc > 0; --envc) {
1369 		suword(vectp++, (long)(intptr_t)destp);
1370 		while (*stringp++ != 0)
1371 			destp++;
1372 		destp++;
1373 	}
1374 
1375 	/* end of vector table is a null pointer */
1376 	suword(vectp, 0);
1377 
1378 	return (stack_base);
1379 }
1380 
1381 /*
1382  * Check permissions of file to execute.
1383  *	Called with imgp->vp locked.
1384  *	Return 0 for success or error code on failure.
1385  */
1386 int
1387 exec_check_permissions(imgp)
1388 	struct image_params *imgp;
1389 {
1390 	struct vnode *vp = imgp->vp;
1391 	struct vattr *attr = imgp->attr;
1392 	struct thread *td;
1393 	int error, writecount;
1394 
1395 	td = curthread;
1396 
1397 	/* Get file attributes */
1398 	error = VOP_GETATTR(vp, attr, td->td_ucred);
1399 	if (error)
1400 		return (error);
1401 
1402 #ifdef MAC
1403 	error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1404 	if (error)
1405 		return (error);
1406 #endif
1407 
1408 	/*
1409 	 * 1) Check if file execution is disabled for the filesystem that
1410 	 *    this file resides on.
1411 	 * 2) Ensure that at least one execute bit is on. Otherwise, a
1412 	 *    privileged user will always succeed, and we don't want this
1413 	 *    to happen unless the file really is executable.
1414 	 * 3) Ensure that the file is a regular file.
1415 	 */
1416 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1417 	    (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1418 	    (attr->va_type != VREG))
1419 		return (EACCES);
1420 
1421 	/*
1422 	 * Zero length files can't be exec'd
1423 	 */
1424 	if (attr->va_size == 0)
1425 		return (ENOEXEC);
1426 
1427 	/*
1428 	 *  Check for execute permission to file based on current credentials.
1429 	 */
1430 	error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1431 	if (error)
1432 		return (error);
1433 
1434 	/*
1435 	 * Check number of open-for-writes on the file and deny execution
1436 	 * if there are any.
1437 	 */
1438 	error = VOP_GET_WRITECOUNT(vp, &writecount);
1439 	if (error != 0)
1440 		return (error);
1441 	if (writecount != 0)
1442 		return (ETXTBSY);
1443 
1444 	/*
1445 	 * Call filesystem specific open routine (which does nothing in the
1446 	 * general case).
1447 	 */
1448 	error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1449 	if (error == 0)
1450 		imgp->opened = 1;
1451 	return (error);
1452 }
1453 
1454 /*
1455  * Exec handler registration
1456  */
1457 int
1458 exec_register(execsw_arg)
1459 	const struct execsw *execsw_arg;
1460 {
1461 	const struct execsw **es, **xs, **newexecsw;
1462 	int count = 2;	/* New slot and trailing NULL */
1463 
1464 	if (execsw)
1465 		for (es = execsw; *es; es++)
1466 			count++;
1467 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1468 	if (newexecsw == NULL)
1469 		return (ENOMEM);
1470 	xs = newexecsw;
1471 	if (execsw)
1472 		for (es = execsw; *es; es++)
1473 			*xs++ = *es;
1474 	*xs++ = execsw_arg;
1475 	*xs = NULL;
1476 	if (execsw)
1477 		free(execsw, M_TEMP);
1478 	execsw = newexecsw;
1479 	return (0);
1480 }
1481 
1482 int
1483 exec_unregister(execsw_arg)
1484 	const struct execsw *execsw_arg;
1485 {
1486 	const struct execsw **es, **xs, **newexecsw;
1487 	int count = 1;
1488 
1489 	if (execsw == NULL)
1490 		panic("unregister with no handlers left?\n");
1491 
1492 	for (es = execsw; *es; es++) {
1493 		if (*es == execsw_arg)
1494 			break;
1495 	}
1496 	if (*es == NULL)
1497 		return (ENOENT);
1498 	for (es = execsw; *es; es++)
1499 		if (*es != execsw_arg)
1500 			count++;
1501 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1502 	if (newexecsw == NULL)
1503 		return (ENOMEM);
1504 	xs = newexecsw;
1505 	for (es = execsw; *es; es++)
1506 		if (*es != execsw_arg)
1507 			*xs++ = *es;
1508 	*xs = NULL;
1509 	if (execsw)
1510 		free(execsw, M_TEMP);
1511 	execsw = newexecsw;
1512 	return (0);
1513 }
1514