xref: /freebsd/sys/kern/kern_exec.c (revision 41840d7587afd6ce27e3725b80481dd4d8f26b1a)
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/rwlock.h>
61 #include <sys/sched.h>
62 #include <sys/sdt.h>
63 #include <sys/sf_buf.h>
64 #include <sys/syscallsubr.h>
65 #include <sys/sysent.h>
66 #include <sys/shm.h>
67 #include <sys/sysctl.h>
68 #include <sys/vnode.h>
69 #include <sys/stat.h>
70 #ifdef KTRACE
71 #include <sys/ktrace.h>
72 #endif
73 
74 #include <vm/vm.h>
75 #include <vm/vm_param.h>
76 #include <vm/pmap.h>
77 #include <vm/vm_page.h>
78 #include <vm/vm_map.h>
79 #include <vm/vm_kern.h>
80 #include <vm/vm_extern.h>
81 #include <vm/vm_object.h>
82 #include <vm/vm_pager.h>
83 
84 #ifdef	HWPMC_HOOKS
85 #include <sys/pmckern.h>
86 #endif
87 
88 #include <machine/reg.h>
89 
90 #include <security/audit/audit.h>
91 #include <security/mac/mac_framework.h>
92 
93 #ifdef KDTRACE_HOOKS
94 #include <sys/dtrace_bsd.h>
95 dtrace_execexit_func_t	dtrace_fasttrap_exec;
96 #endif
97 
98 SDT_PROVIDER_DECLARE(proc);
99 SDT_PROBE_DEFINE(proc, kernel, , exec, exec);
100 SDT_PROBE_ARGTYPE(proc, kernel, , exec, 0, "char *");
101 SDT_PROBE_DEFINE(proc, kernel, , exec_failure, exec-failure);
102 SDT_PROBE_ARGTYPE(proc, kernel, , exec_failure, 0, "int");
103 SDT_PROBE_DEFINE(proc, kernel, , exec_success, exec-success);
104 SDT_PROBE_ARGTYPE(proc, kernel, , exec_success, 0, "char *");
105 
106 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
107 
108 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
109 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
110 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
111 static int do_execve(struct thread *td, struct image_args *args,
112     struct mac *mac_p);
113 
114 /* XXX This should be vm_size_t. */
115 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD,
116     NULL, 0, sysctl_kern_ps_strings, "LU", "");
117 
118 /* XXX This should be vm_size_t. */
119 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD|
120     CTLFLAG_CAPRD, NULL, 0, sysctl_kern_usrstack, "LU", "");
121 
122 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD,
123     NULL, 0, sysctl_kern_stackprot, "I", "");
124 
125 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
126 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
127     &ps_arg_cache_limit, 0, "");
128 
129 static int map_at_zero = 0;
130 TUNABLE_INT("security.bsd.map_at_zero", &map_at_zero);
131 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RW, &map_at_zero, 0,
132     "Permit processes to map an object at virtual address 0.");
133 
134 static int
135 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
136 {
137 	struct proc *p;
138 	int error;
139 
140 	p = curproc;
141 #ifdef SCTL_MASK32
142 	if (req->flags & SCTL_MASK32) {
143 		unsigned int val;
144 		val = (unsigned int)p->p_sysent->sv_psstrings;
145 		error = SYSCTL_OUT(req, &val, sizeof(val));
146 	} else
147 #endif
148 		error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
149 		   sizeof(p->p_sysent->sv_psstrings));
150 	return error;
151 }
152 
153 static int
154 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
155 {
156 	struct proc *p;
157 	int error;
158 
159 	p = curproc;
160 #ifdef SCTL_MASK32
161 	if (req->flags & SCTL_MASK32) {
162 		unsigned int val;
163 		val = (unsigned int)p->p_sysent->sv_usrstack;
164 		error = SYSCTL_OUT(req, &val, sizeof(val));
165 	} else
166 #endif
167 		error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
168 		    sizeof(p->p_sysent->sv_usrstack));
169 	return error;
170 }
171 
172 static int
173 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
174 {
175 	struct proc *p;
176 
177 	p = curproc;
178 	return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
179 	    sizeof(p->p_sysent->sv_stackprot)));
180 }
181 
182 /*
183  * Each of the items is a pointer to a `const struct execsw', hence the
184  * double pointer here.
185  */
186 static const struct execsw **execsw;
187 
188 #ifndef _SYS_SYSPROTO_H_
189 struct execve_args {
190 	char    *fname;
191 	char    **argv;
192 	char    **envv;
193 };
194 #endif
195 
196 int
197 sys_execve(td, uap)
198 	struct thread *td;
199 	struct execve_args /* {
200 		char *fname;
201 		char **argv;
202 		char **envv;
203 	} */ *uap;
204 {
205 	int error;
206 	struct image_args args;
207 
208 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
209 	    uap->argv, uap->envv);
210 	if (error == 0)
211 		error = kern_execve(td, &args, NULL);
212 	return (error);
213 }
214 
215 #ifndef _SYS_SYSPROTO_H_
216 struct fexecve_args {
217 	int	fd;
218 	char	**argv;
219 	char	**envv;
220 }
221 #endif
222 int
223 sys_fexecve(struct thread *td, struct fexecve_args *uap)
224 {
225 	int error;
226 	struct image_args args;
227 
228 	error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
229 	    uap->argv, uap->envv);
230 	if (error == 0) {
231 		args.fd = uap->fd;
232 		error = kern_execve(td, &args, NULL);
233 	}
234 	return (error);
235 }
236 
237 #ifndef _SYS_SYSPROTO_H_
238 struct __mac_execve_args {
239 	char	*fname;
240 	char	**argv;
241 	char	**envv;
242 	struct mac	*mac_p;
243 };
244 #endif
245 
246 int
247 sys___mac_execve(td, uap)
248 	struct thread *td;
249 	struct __mac_execve_args /* {
250 		char *fname;
251 		char **argv;
252 		char **envv;
253 		struct mac *mac_p;
254 	} */ *uap;
255 {
256 #ifdef MAC
257 	int error;
258 	struct image_args args;
259 
260 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
261 	    uap->argv, uap->envv);
262 	if (error == 0)
263 		error = kern_execve(td, &args, uap->mac_p);
264 	return (error);
265 #else
266 	return (ENOSYS);
267 #endif
268 }
269 
270 /*
271  * XXX: kern_execve has the astonishing property of not always returning to
272  * the caller.  If sufficiently bad things happen during the call to
273  * do_execve(), it can end up calling exit1(); as a result, callers must
274  * avoid doing anything which they might need to undo (e.g., allocating
275  * memory).
276  */
277 int
278 kern_execve(td, args, mac_p)
279 	struct thread *td;
280 	struct image_args *args;
281 	struct mac *mac_p;
282 {
283 	struct proc *p = td->td_proc;
284 	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 	error = do_execve(td, args, mac_p);
301 
302 	if (p->p_flag & P_HADTHREADS) {
303 		PROC_LOCK(p);
304 		/*
305 		 * If success, we upgrade to SINGLE_EXIT state to
306 		 * force other threads to suicide.
307 		 */
308 		if (error == 0)
309 			thread_single(SINGLE_EXIT);
310 		else
311 			thread_single_end();
312 		PROC_UNLOCK(p);
313 	}
314 
315 	return (error);
316 }
317 
318 /*
319  * In-kernel implementation of execve().  All arguments are assumed to be
320  * userspace pointers from the passed thread.
321  */
322 static int
323 do_execve(td, args, mac_p)
324 	struct thread *td;
325 	struct image_args *args;
326 	struct mac *mac_p;
327 {
328 	struct proc *p = td->td_proc;
329 	struct nameidata nd;
330 	struct ucred *newcred = NULL, *oldcred;
331 	struct uidinfo *euip;
332 	register_t *stack_base;
333 	int error, i;
334 	struct image_params image_params, *imgp;
335 	struct vattr attr;
336 	int (*img_first)(struct image_params *);
337 	struct pargs *oldargs = NULL, *newargs = NULL;
338 	struct sigacts *oldsigacts, *newsigacts;
339 #ifdef KTRACE
340 	struct vnode *tracevp = NULL;
341 	struct ucred *tracecred = NULL;
342 #endif
343 	struct vnode *textvp = NULL, *binvp = NULL;
344 	int credential_changing;
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 	imgp = &image_params;
356 
357 	/*
358 	 * Lock the process and set the P_INEXEC flag to indicate that
359 	 * it should be left alone until we're done here.  This is
360 	 * necessary to avoid race conditions - e.g. in ptrace() -
361 	 * that might allow a local user to illicitly obtain elevated
362 	 * privileges.
363 	 */
364 	PROC_LOCK(p);
365 	KASSERT((p->p_flag & P_INEXEC) == 0,
366 	    ("%s(): process already has P_INEXEC flag", __func__));
367 	p->p_flag |= P_INEXEC;
368 	PROC_UNLOCK(p);
369 
370 	/*
371 	 * Initialize part of the common data
372 	 */
373 	imgp->proc = p;
374 	imgp->execlabel = NULL;
375 	imgp->attr = &attr;
376 	imgp->entry_addr = 0;
377 	imgp->reloc_base = 0;
378 	imgp->vmspace_destroyed = 0;
379 	imgp->interpreted = 0;
380 	imgp->opened = 0;
381 	imgp->interpreter_name = NULL;
382 	imgp->auxargs = NULL;
383 	imgp->vp = NULL;
384 	imgp->object = NULL;
385 	imgp->firstpage = NULL;
386 	imgp->ps_strings = 0;
387 	imgp->auxarg_size = 0;
388 	imgp->args = args;
389 	imgp->execpath = imgp->freepath = NULL;
390 	imgp->execpathp = 0;
391 	imgp->canary = 0;
392 	imgp->canarylen = 0;
393 	imgp->pagesizes = 0;
394 	imgp->pagesizeslen = 0;
395 	imgp->stack_prot = 0;
396 
397 #ifdef MAC
398 	error = mac_execve_enter(imgp, mac_p);
399 	if (error)
400 		goto exec_fail;
401 #endif
402 
403 	imgp->image_header = NULL;
404 
405 	/*
406 	 * Translate the file name. namei() returns a vnode pointer
407 	 *	in ni_vp amoung other things.
408 	 *
409 	 * XXXAUDIT: It would be desirable to also audit the name of the
410 	 * interpreter if this is an interpreted binary.
411 	 */
412 	if (args->fname != NULL) {
413 		NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME
414 		    | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
415 	}
416 
417 	SDT_PROBE(proc, kernel, , exec, args->fname, 0, 0, 0, 0 );
418 
419 interpret:
420 	if (args->fname != NULL) {
421 #ifdef CAPABILITY_MODE
422 		/*
423 		 * While capability mode can't reach this point via direct
424 		 * path arguments to execve(), we also don't allow
425 		 * interpreters to be used in capability mode (for now).
426 		 * Catch indirect lookups and return a permissions error.
427 		 */
428 		if (IN_CAPABILITY_MODE(td)) {
429 			error = ECAPMODE;
430 			goto exec_fail;
431 		}
432 #endif
433 		error = namei(&nd);
434 		if (error)
435 			goto exec_fail;
436 
437 		binvp  = nd.ni_vp;
438 		imgp->vp = binvp;
439 	} else {
440 		AUDIT_ARG_FD(args->fd);
441 		/*
442 		 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
443 		 */
444 		error = fgetvp_exec(td, args->fd, CAP_FEXECVE, &binvp);
445 		if (error)
446 			goto exec_fail;
447 		vn_lock(binvp, LK_EXCLUSIVE | LK_RETRY);
448 		AUDIT_ARG_VNODE1(binvp);
449 		imgp->vp = binvp;
450 	}
451 
452 	/*
453 	 * Check file permissions (also 'opens' file)
454 	 */
455 	error = exec_check_permissions(imgp);
456 	if (error)
457 		goto exec_fail_dealloc;
458 
459 	imgp->object = imgp->vp->v_object;
460 	if (imgp->object != NULL)
461 		vm_object_reference(imgp->object);
462 
463 	/*
464 	 * Set VV_TEXT now so no one can write to the executable while we're
465 	 * activating it.
466 	 *
467 	 * Remember if this was set before and unset it in case this is not
468 	 * actually an executable image.
469 	 */
470 	textset = VOP_IS_TEXT(imgp->vp);
471 	VOP_SET_TEXT(imgp->vp);
472 
473 	error = exec_map_first_page(imgp);
474 	if (error)
475 		goto exec_fail_dealloc;
476 
477 	imgp->proc->p_osrel = 0;
478 	/*
479 	 *	If the current process has a special image activator it
480 	 *	wants to try first, call it.   For example, emulating shell
481 	 *	scripts differently.
482 	 */
483 	error = -1;
484 	if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
485 		error = img_first(imgp);
486 
487 	/*
488 	 *	Loop through the list of image activators, calling each one.
489 	 *	An activator returns -1 if there is no match, 0 on success,
490 	 *	and an error otherwise.
491 	 */
492 	for (i = 0; error == -1 && execsw[i]; ++i) {
493 		if (execsw[i]->ex_imgact == NULL ||
494 		    execsw[i]->ex_imgact == img_first) {
495 			continue;
496 		}
497 		error = (*execsw[i]->ex_imgact)(imgp);
498 	}
499 
500 	if (error) {
501 		if (error == -1) {
502 			if (textset == 0)
503 				VOP_UNSET_TEXT(imgp->vp);
504 			error = ENOEXEC;
505 		}
506 		goto exec_fail_dealloc;
507 	}
508 
509 	/*
510 	 * Special interpreter operation, cleanup and loop up to try to
511 	 * activate the interpreter.
512 	 */
513 	if (imgp->interpreted) {
514 		exec_unmap_first_page(imgp);
515 		/*
516 		 * VV_TEXT needs to be unset for scripts.  There is a short
517 		 * period before we determine that something is a script where
518 		 * VV_TEXT will be set. The vnode lock is held over this
519 		 * entire period so nothing should illegitimately be blocked.
520 		 */
521 		VOP_UNSET_TEXT(imgp->vp);
522 		/* free name buffer and old vnode */
523 		if (args->fname != NULL)
524 			NDFREE(&nd, NDF_ONLY_PNBUF);
525 #ifdef MAC
526 		mac_execve_interpreter_enter(binvp, &interpvplabel);
527 #endif
528 		if (imgp->opened) {
529 			VOP_CLOSE(binvp, FREAD, td->td_ucred, td);
530 			imgp->opened = 0;
531 		}
532 		vput(binvp);
533 		vm_object_deallocate(imgp->object);
534 		imgp->object = NULL;
535 		/* set new name to that of the interpreter */
536 		NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
537 		    UIO_SYSSPACE, imgp->interpreter_name, td);
538 		args->fname = imgp->interpreter_name;
539 		goto interpret;
540 	}
541 
542 	/*
543 	 * NB: We unlock the vnode here because it is believed that none
544 	 * of the sv_copyout_strings/sv_fixup operations require the vnode.
545 	 */
546 	VOP_UNLOCK(imgp->vp, 0);
547 
548 	/*
549 	 * Do the best to calculate the full path to the image file.
550 	 */
551 	if (imgp->auxargs != NULL &&
552 	    ((args->fname != NULL && args->fname[0] == '/') ||
553 	     vn_fullpath(td, imgp->vp, &imgp->execpath, &imgp->freepath) != 0))
554 		imgp->execpath = args->fname;
555 
556 	/*
557 	 * Copy out strings (args and env) and initialize stack base
558 	 */
559 	if (p->p_sysent->sv_copyout_strings)
560 		stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
561 	else
562 		stack_base = exec_copyout_strings(imgp);
563 
564 	/*
565 	 * If custom stack fixup routine present for this process
566 	 * let it do the stack setup.
567 	 * Else stuff argument count as first item on stack
568 	 */
569 	if (p->p_sysent->sv_fixup != NULL)
570 		(*p->p_sysent->sv_fixup)(&stack_base, imgp);
571 	else
572 		suword(--stack_base, imgp->args->argc);
573 
574 	/*
575 	 * For security and other reasons, the file descriptor table cannot
576 	 * be shared after an exec.
577 	 */
578 	fdunshare(p, td);
579 
580 	/*
581 	 * Malloc things before we need locks.
582 	 */
583 	newcred = crget();
584 	euip = uifind(attr.va_uid);
585 	i = imgp->args->begin_envv - imgp->args->begin_argv;
586 	/* Cache arguments if they fit inside our allowance */
587 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
588 		newargs = pargs_alloc(i);
589 		bcopy(imgp->args->begin_argv, newargs->ar_args, i);
590 	}
591 
592 	/* close files on exec */
593 	fdcloseexec(td);
594 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
595 
596 	/* Get a reference to the vnode prior to locking the proc */
597 	VREF(binvp);
598 
599 	/*
600 	 * For security and other reasons, signal handlers cannot
601 	 * be shared after an exec. The new process gets a copy of the old
602 	 * handlers. In execsigs(), the new process will have its signals
603 	 * reset.
604 	 */
605 	PROC_LOCK(p);
606 	oldcred = crcopysafe(p, newcred);
607 	if (sigacts_shared(p->p_sigacts)) {
608 		oldsigacts = p->p_sigacts;
609 		PROC_UNLOCK(p);
610 		newsigacts = sigacts_alloc();
611 		sigacts_copy(newsigacts, oldsigacts);
612 		PROC_LOCK(p);
613 		p->p_sigacts = newsigacts;
614 	} else
615 		oldsigacts = NULL;
616 
617 	/* Stop profiling */
618 	stopprofclock(p);
619 
620 	/* reset caught signals */
621 	execsigs(p);
622 
623 	/* name this process - nameiexec(p, ndp) */
624 	bzero(p->p_comm, sizeof(p->p_comm));
625 	if (args->fname)
626 		bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
627 		    min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
628 	else if (vn_commname(binvp, p->p_comm, sizeof(p->p_comm)) != 0)
629 		bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
630 	bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
631 #ifdef KTR
632 	sched_clear_tdname(td);
633 #endif
634 
635 	/*
636 	 * mark as execed, wakeup the process that vforked (if any) and tell
637 	 * it that it now has its own resources back
638 	 */
639 	p->p_flag |= P_EXEC;
640 	if (p->p_pptr && (p->p_flag & P_PPWAIT)) {
641 		p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
642 		cv_broadcast(&p->p_pwait);
643 	}
644 
645 	/*
646 	 * Implement image setuid/setgid.
647 	 *
648 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
649 	 * the process is being traced.
650 	 *
651 	 * We disable setuid/setgid/etc in compatibility mode on the basis
652 	 * that most setugid applications are not written with that
653 	 * environment in mind, and will therefore almost certainly operate
654 	 * incorrectly. In principle there's no reason that setugid
655 	 * applications might not be useful in capability mode, so we may want
656 	 * to reconsider this conservative design choice in the future.
657 	 *
658 	 * XXXMAC: For the time being, use NOSUID to also prohibit
659 	 * transitions on the file system.
660 	 */
661 	credential_changing = 0;
662 	credential_changing |= (attr.va_mode & S_ISUID) && oldcred->cr_uid !=
663 	    attr.va_uid;
664 	credential_changing |= (attr.va_mode & S_ISGID) && oldcred->cr_gid !=
665 	    attr.va_gid;
666 #ifdef MAC
667 	will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
668 	    interpvplabel, imgp);
669 	credential_changing |= will_transition;
670 #endif
671 
672 	if (credential_changing &&
673 #ifdef CAPABILITY_MODE
674 	    ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
675 #endif
676 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
677 	    (p->p_flag & P_TRACED) == 0) {
678 		/*
679 		 * Turn off syscall tracing for set-id programs, except for
680 		 * root.  Record any set-id flags first to make sure that
681 		 * we do not regain any tracing during a possible block.
682 		 */
683 		setsugid(p);
684 
685 #ifdef KTRACE
686 		if (p->p_tracecred != NULL &&
687 		    priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED, 0))
688 			ktrprocexec(p, &tracecred, &tracevp);
689 #endif
690 		/*
691 		 * Close any file descriptors 0..2 that reference procfs,
692 		 * then make sure file descriptors 0..2 are in use.
693 		 *
694 		 * setugidsafety() may call closef() and then pfind()
695 		 * which may grab the process lock.
696 		 * fdcheckstd() may call falloc() which may block to
697 		 * allocate memory, so temporarily drop the process lock.
698 		 */
699 		PROC_UNLOCK(p);
700 		VOP_UNLOCK(imgp->vp, 0);
701 		setugidsafety(td);
702 		error = fdcheckstd(td);
703 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
704 		if (error != 0)
705 			goto done1;
706 		PROC_LOCK(p);
707 		/*
708 		 * Set the new credentials.
709 		 */
710 		if (attr.va_mode & S_ISUID)
711 			change_euid(newcred, euip);
712 		if (attr.va_mode & S_ISGID)
713 			change_egid(newcred, attr.va_gid);
714 #ifdef MAC
715 		if (will_transition) {
716 			mac_vnode_execve_transition(oldcred, newcred, imgp->vp,
717 			    interpvplabel, imgp);
718 		}
719 #endif
720 		/*
721 		 * Implement correct POSIX saved-id behavior.
722 		 *
723 		 * XXXMAC: Note that the current logic will save the
724 		 * uid and gid if a MAC domain transition occurs, even
725 		 * though maybe it shouldn't.
726 		 */
727 		change_svuid(newcred, newcred->cr_uid);
728 		change_svgid(newcred, newcred->cr_gid);
729 		p->p_ucred = newcred;
730 		newcred = NULL;
731 	} else {
732 		if (oldcred->cr_uid == oldcred->cr_ruid &&
733 		    oldcred->cr_gid == oldcred->cr_rgid)
734 			p->p_flag &= ~P_SUGID;
735 		/*
736 		 * Implement correct POSIX saved-id behavior.
737 		 *
738 		 * XXX: It's not clear that the existing behavior is
739 		 * POSIX-compliant.  A number of sources indicate that the
740 		 * saved uid/gid should only be updated if the new ruid is
741 		 * not equal to the old ruid, or the new euid is not equal
742 		 * to the old euid and the new euid is not equal to the old
743 		 * ruid.  The FreeBSD code always updates the saved uid/gid.
744 		 * Also, this code uses the new (replaced) euid and egid as
745 		 * the source, which may or may not be the right ones to use.
746 		 */
747 		if (oldcred->cr_svuid != oldcred->cr_uid ||
748 		    oldcred->cr_svgid != oldcred->cr_gid) {
749 			change_svuid(newcred, newcred->cr_uid);
750 			change_svgid(newcred, newcred->cr_gid);
751 			p->p_ucred = newcred;
752 			newcred = NULL;
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 done1:
826 	/*
827 	 * Free any resources malloc'd earlier that we didn't use.
828 	 */
829 	uifree(euip);
830 	if (newcred == NULL)
831 		crfree(oldcred);
832 	else
833 		crfree(newcred);
834 	VOP_UNLOCK(imgp->vp, 0);
835 
836 	/*
837 	 * Handle deferred decrement of ref counts.
838 	 */
839 	if (textvp != NULL)
840 		vrele(textvp);
841 	if (binvp && error != 0)
842 		vrele(binvp);
843 #ifdef KTRACE
844 	if (tracevp != NULL)
845 		vrele(tracevp);
846 	if (tracecred != NULL)
847 		crfree(tracecred);
848 #endif
849 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
850 	pargs_drop(oldargs);
851 	pargs_drop(newargs);
852 	if (oldsigacts != NULL)
853 		sigacts_free(oldsigacts);
854 
855 exec_fail_dealloc:
856 
857 	/*
858 	 * free various allocated resources
859 	 */
860 	if (imgp->firstpage != NULL)
861 		exec_unmap_first_page(imgp);
862 
863 	if (imgp->vp != NULL) {
864 		if (args->fname)
865 			NDFREE(&nd, NDF_ONLY_PNBUF);
866 		if (imgp->opened)
867 			VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
868 		vput(imgp->vp);
869 	}
870 
871 	if (imgp->object != NULL)
872 		vm_object_deallocate(imgp->object);
873 
874 	free(imgp->freepath, M_TEMP);
875 
876 	if (error == 0) {
877 		PROC_LOCK(p);
878 		td->td_dbgflags |= TDB_EXEC;
879 		PROC_UNLOCK(p);
880 
881 		/*
882 		 * Stop the process here if its stop event mask has
883 		 * the S_EXEC bit set.
884 		 */
885 		STOPEVENT(p, S_EXEC, 0);
886 		goto done2;
887 	}
888 
889 exec_fail:
890 	/* we're done here, clear P_INEXEC */
891 	PROC_LOCK(p);
892 	p->p_flag &= ~P_INEXEC;
893 	PROC_UNLOCK(p);
894 
895 	SDT_PROBE(proc, kernel, , exec_failure, error, 0, 0, 0, 0);
896 
897 done2:
898 #ifdef MAC
899 	mac_execve_exit(imgp);
900 	mac_execve_interpreter_exit(interpvplabel);
901 #endif
902 	exec_free_args(args);
903 
904 	if (error && imgp->vmspace_destroyed) {
905 		/* sorry, no more process anymore. exit gracefully */
906 		exit1(td, W_EXITCODE(0, SIGABRT));
907 		/* NOT REACHED */
908 	}
909 
910 #ifdef KTRACE
911 	if (error == 0)
912 		ktrprocctor(p);
913 #endif
914 
915 	return (error);
916 }
917 
918 int
919 exec_map_first_page(imgp)
920 	struct image_params *imgp;
921 {
922 	int rv, i;
923 	int initial_pagein;
924 	vm_page_t ma[VM_INITIAL_PAGEIN];
925 	vm_object_t object;
926 
927 	if (imgp->firstpage != NULL)
928 		exec_unmap_first_page(imgp);
929 
930 	object = imgp->vp->v_object;
931 	if (object == NULL)
932 		return (EACCES);
933 	VM_OBJECT_WLOCK(object);
934 #if VM_NRESERVLEVEL > 0
935 	if ((object->flags & OBJ_COLORED) == 0) {
936 		object->flags |= OBJ_COLORED;
937 		object->pg_color = 0;
938 	}
939 #endif
940 	ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY |
941 	    VM_ALLOC_RETRY);
942 	if (ma[0]->valid != VM_PAGE_BITS_ALL) {
943 		vm_page_busy(ma[0]);
944 		initial_pagein = VM_INITIAL_PAGEIN;
945 		if (initial_pagein > object->size)
946 			initial_pagein = object->size;
947 		for (i = 1; i < initial_pagein; i++) {
948 			if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) {
949 				if (ma[i]->valid)
950 					break;
951 				if ((ma[i]->oflags & VPO_BUSY) || ma[i]->busy)
952 					break;
953 				vm_page_busy(ma[i]);
954 			} else {
955 				ma[i] = vm_page_alloc(object, i,
956 				    VM_ALLOC_NORMAL | VM_ALLOC_IFNOTCACHED);
957 				if (ma[i] == NULL)
958 					break;
959 			}
960 		}
961 		initial_pagein = i;
962 		rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
963 		ma[0] = vm_page_lookup(object, 0);
964 		if ((rv != VM_PAGER_OK) || (ma[0] == NULL)) {
965 			if (ma[0] != NULL) {
966 				vm_page_lock(ma[0]);
967 				vm_page_free(ma[0]);
968 				vm_page_unlock(ma[0]);
969 			}
970 			VM_OBJECT_WUNLOCK(object);
971 			return (EIO);
972 		}
973 		vm_page_wakeup(ma[0]);
974 	}
975 	vm_page_lock(ma[0]);
976 	vm_page_hold(ma[0]);
977 	vm_page_unlock(ma[0]);
978 	VM_OBJECT_WUNLOCK(object);
979 
980 	imgp->firstpage = sf_buf_alloc(ma[0], 0);
981 	imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
982 
983 	return (0);
984 }
985 
986 void
987 exec_unmap_first_page(imgp)
988 	struct image_params *imgp;
989 {
990 	vm_page_t m;
991 
992 	if (imgp->firstpage != NULL) {
993 		m = sf_buf_page(imgp->firstpage);
994 		sf_buf_free(imgp->firstpage);
995 		imgp->firstpage = NULL;
996 		vm_page_lock(m);
997 		vm_page_unhold(m);
998 		vm_page_unlock(m);
999 	}
1000 }
1001 
1002 /*
1003  * Destroy old address space, and allocate a new stack
1004  *	The new stack is only SGROWSIZ large because it is grown
1005  *	automatically in trap.c.
1006  */
1007 int
1008 exec_new_vmspace(imgp, sv)
1009 	struct image_params *imgp;
1010 	struct sysentvec *sv;
1011 {
1012 	int error;
1013 	struct proc *p = imgp->proc;
1014 	struct vmspace *vmspace = p->p_vmspace;
1015 	vm_object_t obj;
1016 	vm_offset_t sv_minuser, stack_addr;
1017 	vm_map_t map;
1018 	u_long ssiz;
1019 
1020 	imgp->vmspace_destroyed = 1;
1021 	imgp->sysent = sv;
1022 
1023 	/* May be called with Giant held */
1024 	EVENTHANDLER_INVOKE(process_exec, p, imgp);
1025 
1026 	/*
1027 	 * Blow away entire process VM, if address space not shared,
1028 	 * otherwise, create a new VM space so that other threads are
1029 	 * not disrupted
1030 	 */
1031 	map = &vmspace->vm_map;
1032 	if (map_at_zero)
1033 		sv_minuser = sv->sv_minuser;
1034 	else
1035 		sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1036 	if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser &&
1037 	    vm_map_max(map) == sv->sv_maxuser) {
1038 		shmexit(vmspace);
1039 		pmap_remove_pages(vmspace_pmap(vmspace));
1040 		vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1041 	} else {
1042 		error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1043 		if (error)
1044 			return (error);
1045 		vmspace = p->p_vmspace;
1046 		map = &vmspace->vm_map;
1047 	}
1048 
1049 	/* Map a shared page */
1050 	obj = sv->sv_shared_page_obj;
1051 	if (obj != NULL) {
1052 		vm_object_reference(obj);
1053 		error = vm_map_fixed(map, obj, 0,
1054 		    sv->sv_shared_page_base, sv->sv_shared_page_len,
1055 		    VM_PROT_READ | VM_PROT_EXECUTE,
1056 		    VM_PROT_READ | VM_PROT_EXECUTE,
1057 		    MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1058 		if (error) {
1059 			vm_object_deallocate(obj);
1060 			return (error);
1061 		}
1062 	}
1063 
1064 	/* Allocate a new stack */
1065 	if (sv->sv_maxssiz != NULL)
1066 		ssiz = *sv->sv_maxssiz;
1067 	else
1068 		ssiz = maxssiz;
1069 	stack_addr = sv->sv_usrstack - ssiz;
1070 	error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1071 	    obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1072 		sv->sv_stackprot,
1073 	    VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1074 	if (error)
1075 		return (error);
1076 
1077 #ifdef __ia64__
1078 	/* Allocate a new register stack */
1079 	stack_addr = IA64_BACKINGSTORE;
1080 	error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1081 	    sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_UP);
1082 	if (error)
1083 		return (error);
1084 #endif
1085 
1086 	/* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
1087 	 * VM_STACK case, but they are still used to monitor the size of the
1088 	 * process stack so we can check the stack rlimit.
1089 	 */
1090 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1091 	vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - ssiz;
1092 
1093 	return (0);
1094 }
1095 
1096 /*
1097  * Copy out argument and environment strings from the old process address
1098  * space into the temporary string buffer.
1099  */
1100 int
1101 exec_copyin_args(struct image_args *args, char *fname,
1102     enum uio_seg segflg, char **argv, char **envv)
1103 {
1104 	char *argp, *envp;
1105 	int error;
1106 	size_t length;
1107 
1108 	bzero(args, sizeof(*args));
1109 	if (argv == NULL)
1110 		return (EFAULT);
1111 
1112 	/*
1113 	 * Allocate demand-paged memory for the file name, argument, and
1114 	 * environment strings.
1115 	 */
1116 	error = exec_alloc_args(args);
1117 	if (error != 0)
1118 		return (error);
1119 
1120 	/*
1121 	 * Copy the file name.
1122 	 */
1123 	if (fname != NULL) {
1124 		args->fname = args->buf;
1125 		error = (segflg == UIO_SYSSPACE) ?
1126 		    copystr(fname, args->fname, PATH_MAX, &length) :
1127 		    copyinstr(fname, args->fname, PATH_MAX, &length);
1128 		if (error != 0)
1129 			goto err_exit;
1130 	} else
1131 		length = 0;
1132 
1133 	args->begin_argv = args->buf + length;
1134 	args->endp = args->begin_argv;
1135 	args->stringspace = ARG_MAX;
1136 
1137 	/*
1138 	 * extract arguments first
1139 	 */
1140 	while ((argp = (caddr_t) (intptr_t) fuword(argv++))) {
1141 		if (argp == (caddr_t) -1) {
1142 			error = EFAULT;
1143 			goto err_exit;
1144 		}
1145 		if ((error = copyinstr(argp, args->endp,
1146 		    args->stringspace, &length))) {
1147 			if (error == ENAMETOOLONG)
1148 				error = E2BIG;
1149 			goto err_exit;
1150 		}
1151 		args->stringspace -= length;
1152 		args->endp += length;
1153 		args->argc++;
1154 	}
1155 
1156 	args->begin_envv = args->endp;
1157 
1158 	/*
1159 	 * extract environment strings
1160 	 */
1161 	if (envv) {
1162 		while ((envp = (caddr_t)(intptr_t)fuword(envv++))) {
1163 			if (envp == (caddr_t)-1) {
1164 				error = EFAULT;
1165 				goto err_exit;
1166 			}
1167 			if ((error = copyinstr(envp, args->endp,
1168 			    args->stringspace, &length))) {
1169 				if (error == ENAMETOOLONG)
1170 					error = E2BIG;
1171 				goto err_exit;
1172 			}
1173 			args->stringspace -= length;
1174 			args->endp += length;
1175 			args->envc++;
1176 		}
1177 	}
1178 
1179 	return (0);
1180 
1181 err_exit:
1182 	exec_free_args(args);
1183 	return (error);
1184 }
1185 
1186 /*
1187  * Allocate temporary demand-paged, zero-filled memory for the file name,
1188  * argument, and environment strings.  Returns zero if the allocation succeeds
1189  * and ENOMEM otherwise.
1190  */
1191 int
1192 exec_alloc_args(struct image_args *args)
1193 {
1194 
1195 	args->buf = (char *)kmem_alloc_wait(exec_map, PATH_MAX + ARG_MAX);
1196 	return (args->buf != NULL ? 0 : ENOMEM);
1197 }
1198 
1199 void
1200 exec_free_args(struct image_args *args)
1201 {
1202 
1203 	if (args->buf != NULL) {
1204 		kmem_free_wakeup(exec_map, (vm_offset_t)args->buf,
1205 		    PATH_MAX + ARG_MAX);
1206 		args->buf = NULL;
1207 	}
1208 	if (args->fname_buf != NULL) {
1209 		free(args->fname_buf, M_TEMP);
1210 		args->fname_buf = NULL;
1211 	}
1212 }
1213 
1214 /*
1215  * Copy strings out to the new process address space, constructing new arg
1216  * and env vector tables. Return a pointer to the base so that it can be used
1217  * as the initial stack pointer.
1218  */
1219 register_t *
1220 exec_copyout_strings(imgp)
1221 	struct image_params *imgp;
1222 {
1223 	int argc, envc;
1224 	char **vectp;
1225 	char *stringp, *destp;
1226 	register_t *stack_base;
1227 	struct ps_strings *arginfo;
1228 	struct proc *p;
1229 	size_t execpath_len;
1230 	int szsigcode, szps;
1231 	char canary[sizeof(long) * 8];
1232 
1233 	szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1234 	/*
1235 	 * Calculate string base and vector table pointers.
1236 	 * Also deal with signal trampoline code for this exec type.
1237 	 */
1238 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
1239 		execpath_len = strlen(imgp->execpath) + 1;
1240 	else
1241 		execpath_len = 0;
1242 	p = imgp->proc;
1243 	szsigcode = 0;
1244 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1245 	if (p->p_sysent->sv_sigcode_base == 0) {
1246 		if (p->p_sysent->sv_szsigcode != NULL)
1247 			szsigcode = *(p->p_sysent->sv_szsigcode);
1248 	}
1249 	destp =	(caddr_t)arginfo - szsigcode - SPARE_USRSPACE -
1250 	    roundup(execpath_len, sizeof(char *)) -
1251 	    roundup(sizeof(canary), sizeof(char *)) -
1252 	    roundup(szps, sizeof(char *)) -
1253 	    roundup((ARG_MAX - imgp->args->stringspace), sizeof(char *));
1254 
1255 	/*
1256 	 * install sigcode
1257 	 */
1258 	if (szsigcode != 0)
1259 		copyout(p->p_sysent->sv_sigcode, ((caddr_t)arginfo -
1260 		    szsigcode), szsigcode);
1261 
1262 	/*
1263 	 * Copy the image path for the rtld.
1264 	 */
1265 	if (execpath_len != 0) {
1266 		imgp->execpathp = (uintptr_t)arginfo - szsigcode - execpath_len;
1267 		copyout(imgp->execpath, (void *)imgp->execpathp,
1268 		    execpath_len);
1269 	}
1270 
1271 	/*
1272 	 * Prepare the canary for SSP.
1273 	 */
1274 	arc4rand(canary, sizeof(canary), 0);
1275 	imgp->canary = (uintptr_t)arginfo - szsigcode - execpath_len -
1276 	    sizeof(canary);
1277 	copyout(canary, (void *)imgp->canary, sizeof(canary));
1278 	imgp->canarylen = sizeof(canary);
1279 
1280 	/*
1281 	 * Prepare the pagesizes array.
1282 	 */
1283 	imgp->pagesizes = (uintptr_t)arginfo - szsigcode - execpath_len -
1284 	    roundup(sizeof(canary), sizeof(char *)) - szps;
1285 	copyout(pagesizes, (void *)imgp->pagesizes, szps);
1286 	imgp->pagesizeslen = szps;
1287 
1288 	/*
1289 	 * If we have a valid auxargs ptr, prepare some room
1290 	 * on the stack.
1291 	 */
1292 	if (imgp->auxargs) {
1293 		/*
1294 		 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1295 		 * lower compatibility.
1296 		 */
1297 		imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1298 		    (AT_COUNT * 2);
1299 		/*
1300 		 * The '+ 2' is for the null pointers at the end of each of
1301 		 * the arg and env vector sets,and imgp->auxarg_size is room
1302 		 * for argument of Runtime loader.
1303 		 */
1304 		vectp = (char **)(destp - (imgp->args->argc +
1305 		    imgp->args->envc + 2 + imgp->auxarg_size)
1306 		    * sizeof(char *));
1307 	} else {
1308 		/*
1309 		 * The '+ 2' is for the null pointers at the end of each of
1310 		 * the arg and env vector sets
1311 		 */
1312 		vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc + 2) *
1313 		    sizeof(char *));
1314 	}
1315 
1316 	/*
1317 	 * vectp also becomes our initial stack base
1318 	 */
1319 	stack_base = (register_t *)vectp;
1320 
1321 	stringp = imgp->args->begin_argv;
1322 	argc = imgp->args->argc;
1323 	envc = imgp->args->envc;
1324 
1325 	/*
1326 	 * Copy out strings - arguments and environment.
1327 	 */
1328 	copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
1329 
1330 	/*
1331 	 * Fill in "ps_strings" struct for ps, w, etc.
1332 	 */
1333 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1334 	suword32(&arginfo->ps_nargvstr, argc);
1335 
1336 	/*
1337 	 * Fill in argument portion of vector table.
1338 	 */
1339 	for (; argc > 0; --argc) {
1340 		suword(vectp++, (long)(intptr_t)destp);
1341 		while (*stringp++ != 0)
1342 			destp++;
1343 		destp++;
1344 	}
1345 
1346 	/* a null vector table pointer separates the argp's from the envp's */
1347 	suword(vectp++, 0);
1348 
1349 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1350 	suword32(&arginfo->ps_nenvstr, envc);
1351 
1352 	/*
1353 	 * Fill in environment portion of vector table.
1354 	 */
1355 	for (; envc > 0; --envc) {
1356 		suword(vectp++, (long)(intptr_t)destp);
1357 		while (*stringp++ != 0)
1358 			destp++;
1359 		destp++;
1360 	}
1361 
1362 	/* end of vector table is a null pointer */
1363 	suword(vectp, 0);
1364 
1365 	return (stack_base);
1366 }
1367 
1368 /*
1369  * Check permissions of file to execute.
1370  *	Called with imgp->vp locked.
1371  *	Return 0 for success or error code on failure.
1372  */
1373 int
1374 exec_check_permissions(imgp)
1375 	struct image_params *imgp;
1376 {
1377 	struct vnode *vp = imgp->vp;
1378 	struct vattr *attr = imgp->attr;
1379 	struct thread *td;
1380 	int error, writecount;
1381 
1382 	td = curthread;
1383 
1384 	/* Get file attributes */
1385 	error = VOP_GETATTR(vp, attr, td->td_ucred);
1386 	if (error)
1387 		return (error);
1388 
1389 #ifdef MAC
1390 	error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1391 	if (error)
1392 		return (error);
1393 #endif
1394 
1395 	/*
1396 	 * 1) Check if file execution is disabled for the filesystem that
1397 	 *    this file resides on.
1398 	 * 2) Ensure that at least one execute bit is on. Otherwise, a
1399 	 *    privileged user will always succeed, and we don't want this
1400 	 *    to happen unless the file really is executable.
1401 	 * 3) Ensure that the file is a regular file.
1402 	 */
1403 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1404 	    (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1405 	    (attr->va_type != VREG))
1406 		return (EACCES);
1407 
1408 	/*
1409 	 * Zero length files can't be exec'd
1410 	 */
1411 	if (attr->va_size == 0)
1412 		return (ENOEXEC);
1413 
1414 	/*
1415 	 *  Check for execute permission to file based on current credentials.
1416 	 */
1417 	error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1418 	if (error)
1419 		return (error);
1420 
1421 	/*
1422 	 * Check number of open-for-writes on the file and deny execution
1423 	 * if there are any.
1424 	 */
1425 	error = VOP_GET_WRITECOUNT(vp, &writecount);
1426 	if (error != 0)
1427 		return (error);
1428 	if (writecount != 0)
1429 		return (ETXTBSY);
1430 
1431 	/*
1432 	 * Call filesystem specific open routine (which does nothing in the
1433 	 * general case).
1434 	 */
1435 	error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1436 	if (error == 0)
1437 		imgp->opened = 1;
1438 	return (error);
1439 }
1440 
1441 /*
1442  * Exec handler registration
1443  */
1444 int
1445 exec_register(execsw_arg)
1446 	const struct execsw *execsw_arg;
1447 {
1448 	const struct execsw **es, **xs, **newexecsw;
1449 	int count = 2;	/* New slot and trailing NULL */
1450 
1451 	if (execsw)
1452 		for (es = execsw; *es; es++)
1453 			count++;
1454 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1455 	if (newexecsw == NULL)
1456 		return (ENOMEM);
1457 	xs = newexecsw;
1458 	if (execsw)
1459 		for (es = execsw; *es; es++)
1460 			*xs++ = *es;
1461 	*xs++ = execsw_arg;
1462 	*xs = NULL;
1463 	if (execsw)
1464 		free(execsw, M_TEMP);
1465 	execsw = newexecsw;
1466 	return (0);
1467 }
1468 
1469 int
1470 exec_unregister(execsw_arg)
1471 	const struct execsw *execsw_arg;
1472 {
1473 	const struct execsw **es, **xs, **newexecsw;
1474 	int count = 1;
1475 
1476 	if (execsw == NULL)
1477 		panic("unregister with no handlers left?\n");
1478 
1479 	for (es = execsw; *es; es++) {
1480 		if (*es == execsw_arg)
1481 			break;
1482 	}
1483 	if (*es == NULL)
1484 		return (ENOENT);
1485 	for (es = execsw; *es; es++)
1486 		if (*es != execsw_arg)
1487 			count++;
1488 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1489 	if (newexecsw == NULL)
1490 		return (ENOMEM);
1491 	xs = newexecsw;
1492 	for (es = execsw; *es; es++)
1493 		if (*es != execsw_arg)
1494 			*xs++ = *es;
1495 	*xs = NULL;
1496 	if (execsw)
1497 		free(execsw, M_TEMP);
1498 	execsw = newexecsw;
1499 	return (0);
1500 }
1501