xref: /freebsd/sys/kern/kern_exec.c (revision 7431dfd4580e850375fe5478d92ec770344db098)
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;
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 	bzero(imgp, sizeof(*imgp));
383 	imgp->proc = p;
384 	imgp->attr = &attr;
385 	imgp->args = args;
386 
387 #ifdef MAC
388 	error = mac_execve_enter(imgp, mac_p);
389 	if (error)
390 		goto exec_fail;
391 #endif
392 
393 	/*
394 	 * Translate the file name. namei() returns a vnode pointer
395 	 *	in ni_vp amoung other things.
396 	 *
397 	 * XXXAUDIT: It would be desirable to also audit the name of the
398 	 * interpreter if this is an interpreted binary.
399 	 */
400 	if (args->fname != NULL) {
401 		NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME
402 		    | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
403 	}
404 
405 	SDT_PROBE(proc, kernel, , exec, args->fname, 0, 0, 0, 0 );
406 
407 interpret:
408 	if (args->fname != NULL) {
409 #ifdef CAPABILITY_MODE
410 		/*
411 		 * While capability mode can't reach this point via direct
412 		 * path arguments to execve(), we also don't allow
413 		 * interpreters to be used in capability mode (for now).
414 		 * Catch indirect lookups and return a permissions error.
415 		 */
416 		if (IN_CAPABILITY_MODE(td)) {
417 			error = ECAPMODE;
418 			goto exec_fail;
419 		}
420 #endif
421 		error = namei(&nd);
422 		if (error)
423 			goto exec_fail;
424 
425 		binvp = nd.ni_vp;
426 		imgp->vp = binvp;
427 	} else {
428 		AUDIT_ARG_FD(args->fd);
429 		/*
430 		 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
431 		 */
432 		error = fgetvp_exec(td, args->fd,
433 		    cap_rights_init(&rights, CAP_FEXECVE), &binvp);
434 		if (error)
435 			goto exec_fail;
436 		vn_lock(binvp, LK_EXCLUSIVE | LK_RETRY);
437 		AUDIT_ARG_VNODE1(binvp);
438 		imgp->vp = binvp;
439 	}
440 
441 	/*
442 	 * Check file permissions (also 'opens' file)
443 	 */
444 	error = exec_check_permissions(imgp);
445 	if (error)
446 		goto exec_fail_dealloc;
447 
448 	imgp->object = imgp->vp->v_object;
449 	if (imgp->object != NULL)
450 		vm_object_reference(imgp->object);
451 
452 	/*
453 	 * Set VV_TEXT now so no one can write to the executable while we're
454 	 * activating it.
455 	 *
456 	 * Remember if this was set before and unset it in case this is not
457 	 * actually an executable image.
458 	 */
459 	textset = VOP_IS_TEXT(imgp->vp);
460 	VOP_SET_TEXT(imgp->vp);
461 
462 	error = exec_map_first_page(imgp);
463 	if (error)
464 		goto exec_fail_dealloc;
465 
466 	imgp->proc->p_osrel = 0;
467 	/*
468 	 *	If the current process has a special image activator it
469 	 *	wants to try first, call it.   For example, emulating shell
470 	 *	scripts differently.
471 	 */
472 	error = -1;
473 	if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
474 		error = img_first(imgp);
475 
476 	/*
477 	 *	Loop through the list of image activators, calling each one.
478 	 *	An activator returns -1 if there is no match, 0 on success,
479 	 *	and an error otherwise.
480 	 */
481 	for (i = 0; error == -1 && execsw[i]; ++i) {
482 		if (execsw[i]->ex_imgact == NULL ||
483 		    execsw[i]->ex_imgact == img_first) {
484 			continue;
485 		}
486 		error = (*execsw[i]->ex_imgact)(imgp);
487 	}
488 
489 	if (error) {
490 		if (error == -1) {
491 			if (textset == 0)
492 				VOP_UNSET_TEXT(imgp->vp);
493 			error = ENOEXEC;
494 		}
495 		goto exec_fail_dealloc;
496 	}
497 
498 	/*
499 	 * Special interpreter operation, cleanup and loop up to try to
500 	 * activate the interpreter.
501 	 */
502 	if (imgp->interpreted) {
503 		exec_unmap_first_page(imgp);
504 		/*
505 		 * VV_TEXT needs to be unset for scripts.  There is a short
506 		 * period before we determine that something is a script where
507 		 * VV_TEXT will be set. The vnode lock is held over this
508 		 * entire period so nothing should illegitimately be blocked.
509 		 */
510 		VOP_UNSET_TEXT(imgp->vp);
511 		/* free name buffer and old vnode */
512 		if (args->fname != NULL)
513 			NDFREE(&nd, NDF_ONLY_PNBUF);
514 #ifdef MAC
515 		mac_execve_interpreter_enter(binvp, &interpvplabel);
516 #endif
517 		if (imgp->opened) {
518 			VOP_CLOSE(binvp, FREAD, td->td_ucred, td);
519 			imgp->opened = 0;
520 		}
521 		vput(binvp);
522 		vm_object_deallocate(imgp->object);
523 		imgp->object = NULL;
524 		/* set new name to that of the interpreter */
525 		NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
526 		    UIO_SYSSPACE, imgp->interpreter_name, td);
527 		args->fname = imgp->interpreter_name;
528 		goto interpret;
529 	}
530 
531 	/*
532 	 * NB: We unlock the vnode here because it is believed that none
533 	 * of the sv_copyout_strings/sv_fixup operations require the vnode.
534 	 */
535 	VOP_UNLOCK(imgp->vp, 0);
536 
537 	/*
538 	 * Do the best to calculate the full path to the image file.
539 	 */
540 	if (imgp->auxargs != NULL &&
541 	    ((args->fname != NULL && args->fname[0] == '/') ||
542 	     vn_fullpath(td, imgp->vp, &imgp->execpath, &imgp->freepath) != 0))
543 		imgp->execpath = args->fname;
544 
545 	if (disallow_high_osrel &&
546 	    P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
547 		error = ENOEXEC;
548 		uprintf("Osrel %d for image %s too high\n", p->p_osrel,
549 		    imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
550 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
551 		goto exec_fail_dealloc;
552 	}
553 
554 	/*
555 	 * Copy out strings (args and env) and initialize stack base
556 	 */
557 	if (p->p_sysent->sv_copyout_strings)
558 		stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
559 	else
560 		stack_base = exec_copyout_strings(imgp);
561 
562 	/*
563 	 * If custom stack fixup routine present for this process
564 	 * let it do the stack setup.
565 	 * Else stuff argument count as first item on stack
566 	 */
567 	if (p->p_sysent->sv_fixup != NULL)
568 		(*p->p_sysent->sv_fixup)(&stack_base, imgp);
569 	else
570 		suword(--stack_base, imgp->args->argc);
571 
572 	/*
573 	 * For security and other reasons, the file descriptor table cannot
574 	 * be shared after an exec.
575 	 */
576 	fdunshare(td);
577 	/* close files on exec */
578 	fdcloseexec(td);
579 
580 	/*
581 	 * Malloc things before we need locks.
582 	 */
583 	i = imgp->args->begin_envv - imgp->args->begin_argv;
584 	/* Cache arguments if they fit inside our allowance */
585 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
586 		newargs = pargs_alloc(i);
587 		bcopy(imgp->args->begin_argv, newargs->ar_args, i);
588 	}
589 
590 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
591 
592 	/* Get a reference to the vnode prior to locking the proc */
593 	VREF(binvp);
594 
595 	/*
596 	 * For security and other reasons, signal handlers cannot
597 	 * be shared after an exec. The new process gets a copy of the old
598 	 * handlers. In execsigs(), the new process will have its signals
599 	 * reset.
600 	 */
601 	if (sigacts_shared(p->p_sigacts)) {
602 		oldsigacts = p->p_sigacts;
603 		newsigacts = sigacts_alloc();
604 		sigacts_copy(newsigacts, oldsigacts);
605 	} else {
606 		oldsigacts = NULL;
607 		newsigacts = NULL; /* satisfy gcc */
608 	}
609 
610 	PROC_LOCK(p);
611 	if (oldsigacts)
612 		p->p_sigacts = newsigacts;
613 	oldcred = p->p_ucred;
614 	/* Stop profiling */
615 	stopprofclock(p);
616 
617 	/* reset caught signals */
618 	execsigs(p);
619 
620 	/* name this process - nameiexec(p, ndp) */
621 	bzero(p->p_comm, sizeof(p->p_comm));
622 	if (args->fname)
623 		bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
624 		    min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
625 	else if (vn_commname(binvp, p->p_comm, sizeof(p->p_comm)) != 0)
626 		bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
627 	bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
628 #ifdef KTR
629 	sched_clear_tdname(td);
630 #endif
631 
632 	/*
633 	 * mark as execed, wakeup the process that vforked (if any) and tell
634 	 * it that it now has its own resources back
635 	 */
636 	p->p_flag |= P_EXEC;
637 	if (p->p_flag & P_PPWAIT) {
638 		p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
639 		cv_broadcast(&p->p_pwait);
640 	}
641 
642 	/*
643 	 * Implement image setuid/setgid.
644 	 *
645 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
646 	 * the process is being traced.
647 	 *
648 	 * We disable setuid/setgid/etc in compatibility mode on the basis
649 	 * that most setugid applications are not written with that
650 	 * environment in mind, and will therefore almost certainly operate
651 	 * incorrectly. In principle there's no reason that setugid
652 	 * applications might not be useful in capability mode, so we may want
653 	 * to reconsider this conservative design choice in the future.
654 	 *
655 	 * XXXMAC: For the time being, use NOSUID to also prohibit
656 	 * transitions on the file system.
657 	 */
658 	credential_changing = 0;
659 	credential_changing |= (attr.va_mode & S_ISUID) && oldcred->cr_uid !=
660 	    attr.va_uid;
661 	credential_changing |= (attr.va_mode & S_ISGID) && oldcred->cr_gid !=
662 	    attr.va_gid;
663 #ifdef MAC
664 	will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
665 	    interpvplabel, imgp);
666 	credential_changing |= will_transition;
667 #endif
668 
669 	if (credential_changing &&
670 #ifdef CAPABILITY_MODE
671 	    ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
672 #endif
673 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
674 	    (p->p_flag & P_TRACED) == 0) {
675 		/*
676 		 * Turn off syscall tracing for set-id programs, except for
677 		 * root.  Record any set-id flags first to make sure that
678 		 * we do not regain any tracing during a possible block.
679 		 */
680 		setsugid(p);
681 
682 #ifdef KTRACE
683 		if (p->p_tracecred != NULL &&
684 		    priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED, 0))
685 			ktrprocexec(p, &tracecred, &tracevp);
686 #endif
687 		/*
688 		 * Close any file descriptors 0..2 that reference procfs,
689 		 * then make sure file descriptors 0..2 are in use.
690 		 *
691 		 * Both fdsetugidsafety() and fdcheckstd() may call functions
692 		 * taking sleepable locks, so temporarily drop our locks.
693 		 */
694 		PROC_UNLOCK(p);
695 		VOP_UNLOCK(imgp->vp, 0);
696 		fdsetugidsafety(td);
697 		error = fdcheckstd(td);
698 		if (error != 0)
699 			goto done1;
700 		newcred = crdup(oldcred);
701 		euip = uifind(attr.va_uid);
702 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
703 		PROC_LOCK(p);
704 		/*
705 		 * Set the new credentials.
706 		 */
707 		if (attr.va_mode & S_ISUID)
708 			change_euid(newcred, euip);
709 		if (attr.va_mode & S_ISGID)
710 			change_egid(newcred, attr.va_gid);
711 #ifdef MAC
712 		if (will_transition) {
713 			mac_vnode_execve_transition(oldcred, newcred, imgp->vp,
714 			    interpvplabel, imgp);
715 		}
716 #endif
717 		/*
718 		 * Implement correct POSIX saved-id behavior.
719 		 *
720 		 * XXXMAC: Note that the current logic will save the
721 		 * uid and gid if a MAC domain transition occurs, even
722 		 * though maybe it shouldn't.
723 		 */
724 		change_svuid(newcred, newcred->cr_uid);
725 		change_svgid(newcred, newcred->cr_gid);
726 		p->p_ucred = newcred;
727 	} else {
728 		if (oldcred->cr_uid == oldcred->cr_ruid &&
729 		    oldcred->cr_gid == oldcred->cr_rgid)
730 			p->p_flag &= ~P_SUGID;
731 		/*
732 		 * Implement correct POSIX saved-id behavior.
733 		 *
734 		 * XXX: It's not clear that the existing behavior is
735 		 * POSIX-compliant.  A number of sources indicate that the
736 		 * saved uid/gid should only be updated if the new ruid is
737 		 * not equal to the old ruid, or the new euid is not equal
738 		 * to the old euid and the new euid is not equal to the old
739 		 * ruid.  The FreeBSD code always updates the saved uid/gid.
740 		 * Also, this code uses the new (replaced) euid and egid as
741 		 * the source, which may or may not be the right ones to use.
742 		 */
743 		if (oldcred->cr_svuid != oldcred->cr_uid ||
744 		    oldcred->cr_svgid != oldcred->cr_gid) {
745 			PROC_UNLOCK(p);
746 			VOP_UNLOCK(imgp->vp, 0);
747 			newcred = crdup(oldcred);
748 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
749 			PROC_LOCK(p);
750 			change_svuid(newcred, newcred->cr_uid);
751 			change_svgid(newcred, newcred->cr_gid);
752 			p->p_ucred = newcred;
753 		}
754 	}
755 
756 	/*
757 	 * Store the vp for use in procfs.  This vnode was referenced prior
758 	 * to locking the proc lock.
759 	 */
760 	textvp = p->p_textvp;
761 	p->p_textvp = binvp;
762 
763 #ifdef KDTRACE_HOOKS
764 	/*
765 	 * Tell the DTrace fasttrap provider about the exec if it
766 	 * has declared an interest.
767 	 */
768 	if (dtrace_fasttrap_exec)
769 		dtrace_fasttrap_exec(p);
770 #endif
771 
772 	/*
773 	 * Notify others that we exec'd, and clear the P_INEXEC flag
774 	 * as we're now a bona fide freshly-execed process.
775 	 */
776 	KNOTE_LOCKED(&p->p_klist, NOTE_EXEC);
777 	p->p_flag &= ~P_INEXEC;
778 
779 	/* clear "fork but no exec" flag, as we _are_ execing */
780 	p->p_acflag &= ~AFORK;
781 
782 	/*
783 	 * Free any previous argument cache and replace it with
784 	 * the new argument cache, if any.
785 	 */
786 	oldargs = p->p_args;
787 	p->p_args = newargs;
788 	newargs = NULL;
789 
790 #ifdef	HWPMC_HOOKS
791 	/*
792 	 * Check if system-wide sampling is in effect or if the
793 	 * current process is using PMCs.  If so, do exec() time
794 	 * processing.  This processing needs to happen AFTER the
795 	 * P_INEXEC flag is cleared.
796 	 *
797 	 * The proc lock needs to be released before taking the PMC
798 	 * SX.
799 	 */
800 	if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
801 		PROC_UNLOCK(p);
802 		VOP_UNLOCK(imgp->vp, 0);
803 		pe.pm_credentialschanged = credential_changing;
804 		pe.pm_entryaddr = imgp->entry_addr;
805 
806 		PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
807 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
808 	} else
809 		PROC_UNLOCK(p);
810 #else  /* !HWPMC_HOOKS */
811 	PROC_UNLOCK(p);
812 #endif
813 
814 	/* Set values passed into the program in registers. */
815 	if (p->p_sysent->sv_setregs)
816 		(*p->p_sysent->sv_setregs)(td, imgp,
817 		    (u_long)(uintptr_t)stack_base);
818 	else
819 		exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base);
820 
821 	vfs_mark_atime(imgp->vp, td->td_ucred);
822 
823 	SDT_PROBE(proc, kernel, , exec__success, args->fname, 0, 0, 0, 0);
824 
825 	VOP_UNLOCK(imgp->vp, 0);
826 done1:
827 	/*
828 	 * Free any resources malloc'd earlier that we didn't use.
829 	 */
830 	if (euip != NULL)
831 		uifree(euip);
832 	if (newcred != NULL)
833 		crfree(oldcred);
834 
835 	/*
836 	 * Handle deferred decrement of ref counts.
837 	 */
838 	if (textvp != NULL)
839 		vrele(textvp);
840 	if (error != 0)
841 		vrele(binvp);
842 #ifdef KTRACE
843 	if (tracevp != NULL)
844 		vrele(tracevp);
845 	if (tracecred != NULL)
846 		crfree(tracecred);
847 #endif
848 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
849 	pargs_drop(oldargs);
850 	pargs_drop(newargs);
851 	if (oldsigacts != NULL)
852 		sigacts_free(oldsigacts);
853 
854 exec_fail_dealloc:
855 
856 	/*
857 	 * free various allocated resources
858 	 */
859 	if (imgp->firstpage != NULL)
860 		exec_unmap_first_page(imgp);
861 
862 	if (imgp->vp != NULL) {
863 		if (args->fname)
864 			NDFREE(&nd, NDF_ONLY_PNBUF);
865 		if (imgp->opened)
866 			VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
867 		vput(imgp->vp);
868 	}
869 
870 	if (imgp->object != NULL)
871 		vm_object_deallocate(imgp->object);
872 
873 	free(imgp->freepath, M_TEMP);
874 
875 	if (error == 0) {
876 		PROC_LOCK(p);
877 		td->td_dbgflags |= TDB_EXEC;
878 		PROC_UNLOCK(p);
879 
880 		/*
881 		 * Stop the process here if its stop event mask has
882 		 * the S_EXEC bit set.
883 		 */
884 		STOPEVENT(p, S_EXEC, 0);
885 		goto done2;
886 	}
887 
888 exec_fail:
889 	/* we're done here, clear P_INEXEC */
890 	PROC_LOCK(p);
891 	p->p_flag &= ~P_INEXEC;
892 	PROC_UNLOCK(p);
893 
894 	SDT_PROBE(proc, kernel, , exec__failure, error, 0, 0, 0, 0);
895 
896 done2:
897 #ifdef MAC
898 	mac_execve_exit(imgp);
899 	mac_execve_interpreter_exit(interpvplabel);
900 #endif
901 	exec_free_args(args);
902 
903 	if (error && imgp->vmspace_destroyed) {
904 		/* sorry, no more process anymore. exit gracefully */
905 		exit1(td, W_EXITCODE(0, SIGABRT));
906 		/* NOT REACHED */
907 	}
908 
909 #ifdef KTRACE
910 	if (error == 0)
911 		ktrprocctor(p);
912 #endif
913 
914 	return (error);
915 }
916 
917 int
918 exec_map_first_page(imgp)
919 	struct image_params *imgp;
920 {
921 	int rv, i;
922 	int initial_pagein;
923 	vm_page_t ma[VM_INITIAL_PAGEIN];
924 	vm_object_t object;
925 
926 	if (imgp->firstpage != NULL)
927 		exec_unmap_first_page(imgp);
928 
929 	object = imgp->vp->v_object;
930 	if (object == NULL)
931 		return (EACCES);
932 	VM_OBJECT_WLOCK(object);
933 #if VM_NRESERVLEVEL > 0
934 	if ((object->flags & OBJ_COLORED) == 0) {
935 		object->flags |= OBJ_COLORED;
936 		object->pg_color = 0;
937 	}
938 #endif
939 	ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL);
940 	if (ma[0]->valid != VM_PAGE_BITS_ALL) {
941 		initial_pagein = VM_INITIAL_PAGEIN;
942 		if (initial_pagein > object->size)
943 			initial_pagein = object->size;
944 		for (i = 1; i < initial_pagein; i++) {
945 			if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) {
946 				if (ma[i]->valid)
947 					break;
948 				if (vm_page_tryxbusy(ma[i]))
949 					break;
950 			} else {
951 				ma[i] = vm_page_alloc(object, i,
952 				    VM_ALLOC_NORMAL | VM_ALLOC_IFNOTCACHED);
953 				if (ma[i] == NULL)
954 					break;
955 			}
956 		}
957 		initial_pagein = i;
958 		rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
959 		ma[0] = vm_page_lookup(object, 0);
960 		if ((rv != VM_PAGER_OK) || (ma[0] == NULL)) {
961 			if (ma[0] != NULL) {
962 				vm_page_lock(ma[0]);
963 				vm_page_free(ma[0]);
964 				vm_page_unlock(ma[0]);
965 			}
966 			VM_OBJECT_WUNLOCK(object);
967 			return (EIO);
968 		}
969 	}
970 	vm_page_xunbusy(ma[0]);
971 	vm_page_lock(ma[0]);
972 	vm_page_hold(ma[0]);
973 	vm_page_activate(ma[0]);
974 	vm_page_unlock(ma[0]);
975 	VM_OBJECT_WUNLOCK(object);
976 
977 	imgp->firstpage = sf_buf_alloc(ma[0], 0);
978 	imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
979 
980 	return (0);
981 }
982 
983 void
984 exec_unmap_first_page(imgp)
985 	struct image_params *imgp;
986 {
987 	vm_page_t m;
988 
989 	if (imgp->firstpage != NULL) {
990 		m = sf_buf_page(imgp->firstpage);
991 		sf_buf_free(imgp->firstpage);
992 		imgp->firstpage = NULL;
993 		vm_page_lock(m);
994 		vm_page_unhold(m);
995 		vm_page_unlock(m);
996 	}
997 }
998 
999 /*
1000  * Destroy old address space, and allocate a new stack
1001  *	The new stack is only SGROWSIZ large because it is grown
1002  *	automatically in trap.c.
1003  */
1004 int
1005 exec_new_vmspace(imgp, sv)
1006 	struct image_params *imgp;
1007 	struct sysentvec *sv;
1008 {
1009 	int error;
1010 	struct proc *p = imgp->proc;
1011 	struct vmspace *vmspace = p->p_vmspace;
1012 	vm_object_t obj;
1013 	vm_offset_t sv_minuser, stack_addr;
1014 	vm_map_t map;
1015 	u_long ssiz;
1016 
1017 	imgp->vmspace_destroyed = 1;
1018 	imgp->sysent = sv;
1019 
1020 	/* May be called with Giant held */
1021 	EVENTHANDLER_INVOKE(process_exec, p, imgp);
1022 
1023 	/*
1024 	 * Blow away entire process VM, if address space not shared,
1025 	 * otherwise, create a new VM space so that other threads are
1026 	 * not disrupted
1027 	 */
1028 	map = &vmspace->vm_map;
1029 	if (map_at_zero)
1030 		sv_minuser = sv->sv_minuser;
1031 	else
1032 		sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1033 	if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser &&
1034 	    vm_map_max(map) == sv->sv_maxuser) {
1035 		shmexit(vmspace);
1036 		pmap_remove_pages(vmspace_pmap(vmspace));
1037 		vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1038 	} else {
1039 		error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1040 		if (error)
1041 			return (error);
1042 		vmspace = p->p_vmspace;
1043 		map = &vmspace->vm_map;
1044 	}
1045 
1046 	/* Map a shared page */
1047 	obj = sv->sv_shared_page_obj;
1048 	if (obj != NULL) {
1049 		vm_object_reference(obj);
1050 		error = vm_map_fixed(map, obj, 0,
1051 		    sv->sv_shared_page_base, sv->sv_shared_page_len,
1052 		    VM_PROT_READ | VM_PROT_EXECUTE,
1053 		    VM_PROT_READ | VM_PROT_EXECUTE,
1054 		    MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1055 		if (error) {
1056 			vm_object_deallocate(obj);
1057 			return (error);
1058 		}
1059 	}
1060 
1061 	/* Allocate a new stack */
1062 	if (sv->sv_maxssiz != NULL)
1063 		ssiz = *sv->sv_maxssiz;
1064 	else
1065 		ssiz = maxssiz;
1066 	stack_addr = sv->sv_usrstack - ssiz;
1067 	error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1068 	    obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1069 		sv->sv_stackprot,
1070 	    VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1071 	if (error)
1072 		return (error);
1073 
1074 	/*
1075 	 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they
1076 	 * are still used to enforce the stack rlimit on the process stack.
1077 	 */
1078 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1079 	vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - ssiz;
1080 
1081 	return (0);
1082 }
1083 
1084 /*
1085  * Copy out argument and environment strings from the old process address
1086  * space into the temporary string buffer.
1087  */
1088 int
1089 exec_copyin_args(struct image_args *args, char *fname,
1090     enum uio_seg segflg, char **argv, char **envv)
1091 {
1092 	u_long argp, envp;
1093 	int error;
1094 	size_t length;
1095 
1096 	bzero(args, sizeof(*args));
1097 	if (argv == NULL)
1098 		return (EFAULT);
1099 
1100 	/*
1101 	 * Allocate demand-paged memory for the file name, argument, and
1102 	 * environment strings.
1103 	 */
1104 	error = exec_alloc_args(args);
1105 	if (error != 0)
1106 		return (error);
1107 
1108 	/*
1109 	 * Copy the file name.
1110 	 */
1111 	if (fname != NULL) {
1112 		args->fname = args->buf;
1113 		error = (segflg == UIO_SYSSPACE) ?
1114 		    copystr(fname, args->fname, PATH_MAX, &length) :
1115 		    copyinstr(fname, args->fname, PATH_MAX, &length);
1116 		if (error != 0)
1117 			goto err_exit;
1118 	} else
1119 		length = 0;
1120 
1121 	args->begin_argv = args->buf + length;
1122 	args->endp = args->begin_argv;
1123 	args->stringspace = ARG_MAX;
1124 
1125 	/*
1126 	 * extract arguments first
1127 	 */
1128 	for (;;) {
1129 		error = fueword(argv++, &argp);
1130 		if (error == -1) {
1131 			error = EFAULT;
1132 			goto err_exit;
1133 		}
1134 		if (argp == 0)
1135 			break;
1136 		error = copyinstr((void *)(uintptr_t)argp, args->endp,
1137 		    args->stringspace, &length);
1138 		if (error != 0) {
1139 			if (error == ENAMETOOLONG)
1140 				error = E2BIG;
1141 			goto err_exit;
1142 		}
1143 		args->stringspace -= length;
1144 		args->endp += length;
1145 		args->argc++;
1146 	}
1147 
1148 	args->begin_envv = args->endp;
1149 
1150 	/*
1151 	 * extract environment strings
1152 	 */
1153 	if (envv) {
1154 		for (;;) {
1155 			error = fueword(envv++, &envp);
1156 			if (error == -1) {
1157 				error = EFAULT;
1158 				goto err_exit;
1159 			}
1160 			if (envp == 0)
1161 				break;
1162 			error = copyinstr((void *)(uintptr_t)envp,
1163 			    args->endp, args->stringspace, &length);
1164 			if (error != 0) {
1165 				if (error == ENAMETOOLONG)
1166 					error = E2BIG;
1167 				goto err_exit;
1168 			}
1169 			args->stringspace -= length;
1170 			args->endp += length;
1171 			args->envc++;
1172 		}
1173 	}
1174 
1175 	return (0);
1176 
1177 err_exit:
1178 	exec_free_args(args);
1179 	return (error);
1180 }
1181 
1182 /*
1183  * Allocate temporary demand-paged, zero-filled memory for the file name,
1184  * argument, and environment strings.  Returns zero if the allocation succeeds
1185  * and ENOMEM otherwise.
1186  */
1187 int
1188 exec_alloc_args(struct image_args *args)
1189 {
1190 
1191 	args->buf = (char *)kmap_alloc_wait(exec_map, PATH_MAX + ARG_MAX);
1192 	return (args->buf != NULL ? 0 : ENOMEM);
1193 }
1194 
1195 void
1196 exec_free_args(struct image_args *args)
1197 {
1198 
1199 	if (args->buf != NULL) {
1200 		kmap_free_wakeup(exec_map, (vm_offset_t)args->buf,
1201 		    PATH_MAX + ARG_MAX);
1202 		args->buf = NULL;
1203 	}
1204 	if (args->fname_buf != NULL) {
1205 		free(args->fname_buf, M_TEMP);
1206 		args->fname_buf = NULL;
1207 	}
1208 }
1209 
1210 /*
1211  * Copy strings out to the new process address space, constructing new arg
1212  * and env vector tables. Return a pointer to the base so that it can be used
1213  * as the initial stack pointer.
1214  */
1215 register_t *
1216 exec_copyout_strings(imgp)
1217 	struct image_params *imgp;
1218 {
1219 	int argc, envc;
1220 	char **vectp;
1221 	char *stringp;
1222 	uintptr_t destp;
1223 	register_t *stack_base;
1224 	struct ps_strings *arginfo;
1225 	struct proc *p;
1226 	size_t execpath_len;
1227 	int szsigcode, szps;
1228 	char canary[sizeof(long) * 8];
1229 
1230 	szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1231 	/*
1232 	 * Calculate string base and vector table pointers.
1233 	 * Also deal with signal trampoline code for this exec type.
1234 	 */
1235 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
1236 		execpath_len = strlen(imgp->execpath) + 1;
1237 	else
1238 		execpath_len = 0;
1239 	p = imgp->proc;
1240 	szsigcode = 0;
1241 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1242 	if (p->p_sysent->sv_sigcode_base == 0) {
1243 		if (p->p_sysent->sv_szsigcode != NULL)
1244 			szsigcode = *(p->p_sysent->sv_szsigcode);
1245 	}
1246 	destp =	(uintptr_t)arginfo;
1247 
1248 	/*
1249 	 * install sigcode
1250 	 */
1251 	if (szsigcode != 0) {
1252 		destp -= szsigcode;
1253 		destp = rounddown2(destp, sizeof(void *));
1254 		copyout(p->p_sysent->sv_sigcode, (void *)destp, szsigcode);
1255 	}
1256 
1257 	/*
1258 	 * Copy the image path for the rtld.
1259 	 */
1260 	if (execpath_len != 0) {
1261 		destp -= execpath_len;
1262 		imgp->execpathp = destp;
1263 		copyout(imgp->execpath, (void *)destp, execpath_len);
1264 	}
1265 
1266 	/*
1267 	 * Prepare the canary for SSP.
1268 	 */
1269 	arc4rand(canary, sizeof(canary), 0);
1270 	destp -= sizeof(canary);
1271 	imgp->canary = destp;
1272 	copyout(canary, (void *)destp, sizeof(canary));
1273 	imgp->canarylen = sizeof(canary);
1274 
1275 	/*
1276 	 * Prepare the pagesizes array.
1277 	 */
1278 	destp -= szps;
1279 	destp = rounddown2(destp, sizeof(void *));
1280 	imgp->pagesizes = destp;
1281 	copyout(pagesizes, (void *)destp, szps);
1282 	imgp->pagesizeslen = szps;
1283 
1284 	destp -= ARG_MAX - imgp->args->stringspace;
1285 	destp = rounddown2(destp, sizeof(void *));
1286 
1287 	/*
1288 	 * If we have a valid auxargs ptr, prepare some room
1289 	 * on the stack.
1290 	 */
1291 	if (imgp->auxargs) {
1292 		/*
1293 		 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1294 		 * lower compatibility.
1295 		 */
1296 		imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1297 		    (AT_COUNT * 2);
1298 		/*
1299 		 * The '+ 2' is for the null pointers at the end of each of
1300 		 * the arg and env vector sets,and imgp->auxarg_size is room
1301 		 * for argument of Runtime loader.
1302 		 */
1303 		vectp = (char **)(destp - (imgp->args->argc +
1304 		    imgp->args->envc + 2 + imgp->auxarg_size)
1305 		    * sizeof(char *));
1306 	} else {
1307 		/*
1308 		 * The '+ 2' is for the null pointers at the end of each of
1309 		 * the arg and env vector sets
1310 		 */
1311 		vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc
1312 		    + 2) * sizeof(char *));
1313 	}
1314 
1315 	/*
1316 	 * vectp also becomes our initial stack base
1317 	 */
1318 	stack_base = (register_t *)vectp;
1319 
1320 	stringp = imgp->args->begin_argv;
1321 	argc = imgp->args->argc;
1322 	envc = imgp->args->envc;
1323 
1324 	/*
1325 	 * Copy out strings - arguments and environment.
1326 	 */
1327 	copyout(stringp, (void *)destp, ARG_MAX - imgp->args->stringspace);
1328 
1329 	/*
1330 	 * Fill in "ps_strings" struct for ps, w, etc.
1331 	 */
1332 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1333 	suword32(&arginfo->ps_nargvstr, argc);
1334 
1335 	/*
1336 	 * Fill in argument portion of vector table.
1337 	 */
1338 	for (; argc > 0; --argc) {
1339 		suword(vectp++, (long)(intptr_t)destp);
1340 		while (*stringp++ != 0)
1341 			destp++;
1342 		destp++;
1343 	}
1344 
1345 	/* a null vector table pointer separates the argp's from the envp's */
1346 	suword(vectp++, 0);
1347 
1348 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1349 	suword32(&arginfo->ps_nenvstr, envc);
1350 
1351 	/*
1352 	 * Fill in environment portion of vector table.
1353 	 */
1354 	for (; envc > 0; --envc) {
1355 		suword(vectp++, (long)(intptr_t)destp);
1356 		while (*stringp++ != 0)
1357 			destp++;
1358 		destp++;
1359 	}
1360 
1361 	/* end of vector table is a null pointer */
1362 	suword(vectp, 0);
1363 
1364 	return (stack_base);
1365 }
1366 
1367 /*
1368  * Check permissions of file to execute.
1369  *	Called with imgp->vp locked.
1370  *	Return 0 for success or error code on failure.
1371  */
1372 int
1373 exec_check_permissions(imgp)
1374 	struct image_params *imgp;
1375 {
1376 	struct vnode *vp = imgp->vp;
1377 	struct vattr *attr = imgp->attr;
1378 	struct thread *td;
1379 	int error, writecount;
1380 
1381 	td = curthread;
1382 
1383 	/* Get file attributes */
1384 	error = VOP_GETATTR(vp, attr, td->td_ucred);
1385 	if (error)
1386 		return (error);
1387 
1388 #ifdef MAC
1389 	error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1390 	if (error)
1391 		return (error);
1392 #endif
1393 
1394 	/*
1395 	 * 1) Check if file execution is disabled for the filesystem that
1396 	 *    this file resides on.
1397 	 * 2) Ensure that at least one execute bit is on. Otherwise, a
1398 	 *    privileged user will always succeed, and we don't want this
1399 	 *    to happen unless the file really is executable.
1400 	 * 3) Ensure that the file is a regular file.
1401 	 */
1402 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1403 	    (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1404 	    (attr->va_type != VREG))
1405 		return (EACCES);
1406 
1407 	/*
1408 	 * Zero length files can't be exec'd
1409 	 */
1410 	if (attr->va_size == 0)
1411 		return (ENOEXEC);
1412 
1413 	/*
1414 	 *  Check for execute permission to file based on current credentials.
1415 	 */
1416 	error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1417 	if (error)
1418 		return (error);
1419 
1420 	/*
1421 	 * Check number of open-for-writes on the file and deny execution
1422 	 * if there are any.
1423 	 */
1424 	error = VOP_GET_WRITECOUNT(vp, &writecount);
1425 	if (error != 0)
1426 		return (error);
1427 	if (writecount != 0)
1428 		return (ETXTBSY);
1429 
1430 	/*
1431 	 * Call filesystem specific open routine (which does nothing in the
1432 	 * general case).
1433 	 */
1434 	error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1435 	if (error == 0)
1436 		imgp->opened = 1;
1437 	return (error);
1438 }
1439 
1440 /*
1441  * Exec handler registration
1442  */
1443 int
1444 exec_register(execsw_arg)
1445 	const struct execsw *execsw_arg;
1446 {
1447 	const struct execsw **es, **xs, **newexecsw;
1448 	int count = 2;	/* New slot and trailing NULL */
1449 
1450 	if (execsw)
1451 		for (es = execsw; *es; es++)
1452 			count++;
1453 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1454 	if (newexecsw == NULL)
1455 		return (ENOMEM);
1456 	xs = newexecsw;
1457 	if (execsw)
1458 		for (es = execsw; *es; es++)
1459 			*xs++ = *es;
1460 	*xs++ = execsw_arg;
1461 	*xs = NULL;
1462 	if (execsw)
1463 		free(execsw, M_TEMP);
1464 	execsw = newexecsw;
1465 	return (0);
1466 }
1467 
1468 int
1469 exec_unregister(execsw_arg)
1470 	const struct execsw *execsw_arg;
1471 {
1472 	const struct execsw **es, **xs, **newexecsw;
1473 	int count = 1;
1474 
1475 	if (execsw == NULL)
1476 		panic("unregister with no handlers left?\n");
1477 
1478 	for (es = execsw; *es; es++) {
1479 		if (*es == execsw_arg)
1480 			break;
1481 	}
1482 	if (*es == NULL)
1483 		return (ENOENT);
1484 	for (es = execsw; *es; es++)
1485 		if (*es != execsw_arg)
1486 			count++;
1487 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1488 	if (newexecsw == NULL)
1489 		return (ENOMEM);
1490 	xs = newexecsw;
1491 	for (es = execsw; *es; es++)
1492 		if (*es != execsw_arg)
1493 			*xs++ = *es;
1494 	*xs = NULL;
1495 	if (execsw)
1496 		free(execsw, M_TEMP);
1497 	execsw = newexecsw;
1498 	return (0);
1499 }
1500