xref: /freebsd/sys/kern/kern_exec.c (revision 287472b39c7985d968be84ea145c3e75a3e6b875)
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_DEFINE1(proc, kernel, , exec, exec, "char *");
100 SDT_PROBE_DEFINE1(proc, kernel, , exec_failure, exec-failure, "int");
101 SDT_PROBE_DEFINE1(proc, kernel, , exec_success, exec-success, "char *");
102 
103 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
104 
105 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
106 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
107 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
108 static int do_execve(struct thread *td, struct image_args *args,
109     struct mac *mac_p);
110 
111 /* XXX This should be vm_size_t. */
112 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD,
113     NULL, 0, sysctl_kern_ps_strings, "LU", "");
114 
115 /* XXX This should be vm_size_t. */
116 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD|
117     CTLFLAG_CAPRD, NULL, 0, sysctl_kern_usrstack, "LU", "");
118 
119 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD,
120     NULL, 0, sysctl_kern_stackprot, "I", "");
121 
122 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
123 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
124     &ps_arg_cache_limit, 0, "");
125 
126 static int map_at_zero = 0;
127 TUNABLE_INT("security.bsd.map_at_zero", &map_at_zero);
128 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RW, &map_at_zero, 0,
129     "Permit processes to map an object at virtual address 0.");
130 
131 static int
132 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
133 {
134 	struct proc *p;
135 	int error;
136 
137 	p = curproc;
138 #ifdef SCTL_MASK32
139 	if (req->flags & SCTL_MASK32) {
140 		unsigned int val;
141 		val = (unsigned int)p->p_sysent->sv_psstrings;
142 		error = SYSCTL_OUT(req, &val, sizeof(val));
143 	} else
144 #endif
145 		error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
146 		   sizeof(p->p_sysent->sv_psstrings));
147 	return error;
148 }
149 
150 static int
151 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
152 {
153 	struct proc *p;
154 	int error;
155 
156 	p = curproc;
157 #ifdef SCTL_MASK32
158 	if (req->flags & SCTL_MASK32) {
159 		unsigned int val;
160 		val = (unsigned int)p->p_sysent->sv_usrstack;
161 		error = SYSCTL_OUT(req, &val, sizeof(val));
162 	} else
163 #endif
164 		error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
165 		    sizeof(p->p_sysent->sv_usrstack));
166 	return error;
167 }
168 
169 static int
170 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
171 {
172 	struct proc *p;
173 
174 	p = curproc;
175 	return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
176 	    sizeof(p->p_sysent->sv_stackprot)));
177 }
178 
179 /*
180  * Each of the items is a pointer to a `const struct execsw', hence the
181  * double pointer here.
182  */
183 static const struct execsw **execsw;
184 
185 #ifndef _SYS_SYSPROTO_H_
186 struct execve_args {
187 	char    *fname;
188 	char    **argv;
189 	char    **envv;
190 };
191 #endif
192 
193 int
194 sys_execve(td, uap)
195 	struct thread *td;
196 	struct execve_args /* {
197 		char *fname;
198 		char **argv;
199 		char **envv;
200 	} */ *uap;
201 {
202 	int error;
203 	struct image_args args;
204 
205 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
206 	    uap->argv, uap->envv);
207 	if (error == 0)
208 		error = kern_execve(td, &args, NULL);
209 	return (error);
210 }
211 
212 #ifndef _SYS_SYSPROTO_H_
213 struct fexecve_args {
214 	int	fd;
215 	char	**argv;
216 	char	**envv;
217 }
218 #endif
219 int
220 sys_fexecve(struct thread *td, struct fexecve_args *uap)
221 {
222 	int error;
223 	struct image_args args;
224 
225 	error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
226 	    uap->argv, uap->envv);
227 	if (error == 0) {
228 		args.fd = uap->fd;
229 		error = kern_execve(td, &args, NULL);
230 	}
231 	return (error);
232 }
233 
234 #ifndef _SYS_SYSPROTO_H_
235 struct __mac_execve_args {
236 	char	*fname;
237 	char	**argv;
238 	char	**envv;
239 	struct mac	*mac_p;
240 };
241 #endif
242 
243 int
244 sys___mac_execve(td, uap)
245 	struct thread *td;
246 	struct __mac_execve_args /* {
247 		char *fname;
248 		char **argv;
249 		char **envv;
250 		struct mac *mac_p;
251 	} */ *uap;
252 {
253 #ifdef MAC
254 	int error;
255 	struct image_args args;
256 
257 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
258 	    uap->argv, uap->envv);
259 	if (error == 0)
260 		error = kern_execve(td, &args, uap->mac_p);
261 	return (error);
262 #else
263 	return (ENOSYS);
264 #endif
265 }
266 
267 /*
268  * XXX: kern_execve has the astonishing property of not always returning to
269  * the caller.  If sufficiently bad things happen during the call to
270  * do_execve(), it can end up calling exit1(); as a result, callers must
271  * avoid doing anything which they might need to undo (e.g., allocating
272  * memory).
273  */
274 int
275 kern_execve(td, args, mac_p)
276 	struct thread *td;
277 	struct image_args *args;
278 	struct mac *mac_p;
279 {
280 	struct proc *p = td->td_proc;
281 	int error;
282 
283 	AUDIT_ARG_ARGV(args->begin_argv, args->argc,
284 	    args->begin_envv - args->begin_argv);
285 	AUDIT_ARG_ENVV(args->begin_envv, args->envc,
286 	    args->endp - args->begin_envv);
287 	if (p->p_flag & P_HADTHREADS) {
288 		PROC_LOCK(p);
289 		if (thread_single(SINGLE_BOUNDARY)) {
290 			PROC_UNLOCK(p);
291 	       		exec_free_args(args);
292 			return (ERESTART);	/* Try again later. */
293 		}
294 		PROC_UNLOCK(p);
295 	}
296 
297 	error = do_execve(td, args, mac_p);
298 
299 	if (p->p_flag & P_HADTHREADS) {
300 		PROC_LOCK(p);
301 		/*
302 		 * If success, we upgrade to SINGLE_EXIT state to
303 		 * force other threads to suicide.
304 		 */
305 		if (error == 0)
306 			thread_single(SINGLE_EXIT);
307 		else
308 			thread_single_end();
309 		PROC_UNLOCK(p);
310 	}
311 
312 	return (error);
313 }
314 
315 /*
316  * In-kernel implementation of execve().  All arguments are assumed to be
317  * userspace pointers from the passed thread.
318  */
319 static int
320 do_execve(td, args, mac_p)
321 	struct thread *td;
322 	struct image_args *args;
323 	struct mac *mac_p;
324 {
325 	struct proc *p = td->td_proc;
326 	struct nameidata nd;
327 	struct ucred *newcred = NULL, *oldcred;
328 	struct uidinfo *euip;
329 	register_t *stack_base;
330 	int error, i;
331 	struct image_params image_params, *imgp;
332 	struct vattr attr;
333 	int (*img_first)(struct image_params *);
334 	struct pargs *oldargs = NULL, *newargs = NULL;
335 	struct sigacts *oldsigacts, *newsigacts;
336 #ifdef KTRACE
337 	struct vnode *tracevp = NULL;
338 	struct ucred *tracecred = NULL;
339 #endif
340 	struct vnode *textvp = NULL, *binvp = NULL;
341 	int credential_changing;
342 	int textset;
343 #ifdef MAC
344 	struct label *interpvplabel = NULL;
345 	int will_transition;
346 #endif
347 #ifdef HWPMC_HOOKS
348 	struct pmckern_procexec pe;
349 #endif
350 	static const char fexecv_proc_title[] = "(fexecv)";
351 
352 	imgp = &image_params;
353 
354 	/*
355 	 * Lock the process and set the P_INEXEC flag to indicate that
356 	 * it should be left alone until we're done here.  This is
357 	 * necessary to avoid race conditions - e.g. in ptrace() -
358 	 * that might allow a local user to illicitly obtain elevated
359 	 * privileges.
360 	 */
361 	PROC_LOCK(p);
362 	KASSERT((p->p_flag & P_INEXEC) == 0,
363 	    ("%s(): process already has P_INEXEC flag", __func__));
364 	p->p_flag |= P_INEXEC;
365 	PROC_UNLOCK(p);
366 
367 	/*
368 	 * Initialize part of the common data
369 	 */
370 	imgp->proc = p;
371 	imgp->execlabel = NULL;
372 	imgp->attr = &attr;
373 	imgp->entry_addr = 0;
374 	imgp->reloc_base = 0;
375 	imgp->vmspace_destroyed = 0;
376 	imgp->interpreted = 0;
377 	imgp->opened = 0;
378 	imgp->interpreter_name = NULL;
379 	imgp->auxargs = NULL;
380 	imgp->vp = NULL;
381 	imgp->object = NULL;
382 	imgp->firstpage = NULL;
383 	imgp->ps_strings = 0;
384 	imgp->auxarg_size = 0;
385 	imgp->args = args;
386 	imgp->execpath = imgp->freepath = NULL;
387 	imgp->execpathp = 0;
388 	imgp->canary = 0;
389 	imgp->canarylen = 0;
390 	imgp->pagesizes = 0;
391 	imgp->pagesizeslen = 0;
392 	imgp->stack_prot = 0;
393 
394 #ifdef MAC
395 	error = mac_execve_enter(imgp, mac_p);
396 	if (error)
397 		goto exec_fail;
398 #endif
399 
400 	imgp->image_header = NULL;
401 
402 	/*
403 	 * Translate the file name. namei() returns a vnode pointer
404 	 *	in ni_vp amoung other things.
405 	 *
406 	 * XXXAUDIT: It would be desirable to also audit the name of the
407 	 * interpreter if this is an interpreted binary.
408 	 */
409 	if (args->fname != NULL) {
410 		NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME
411 		    | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
412 	}
413 
414 	SDT_PROBE(proc, kernel, , exec, args->fname, 0, 0, 0, 0 );
415 
416 interpret:
417 	if (args->fname != NULL) {
418 #ifdef CAPABILITY_MODE
419 		/*
420 		 * While capability mode can't reach this point via direct
421 		 * path arguments to execve(), we also don't allow
422 		 * interpreters to be used in capability mode (for now).
423 		 * Catch indirect lookups and return a permissions error.
424 		 */
425 		if (IN_CAPABILITY_MODE(td)) {
426 			error = ECAPMODE;
427 			goto exec_fail;
428 		}
429 #endif
430 		error = namei(&nd);
431 		if (error)
432 			goto exec_fail;
433 
434 		binvp  = nd.ni_vp;
435 		imgp->vp = binvp;
436 	} else {
437 		AUDIT_ARG_FD(args->fd);
438 		/*
439 		 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
440 		 */
441 		error = fgetvp_exec(td, args->fd, CAP_FEXECVE, &binvp);
442 		if (error)
443 			goto exec_fail;
444 		vn_lock(binvp, LK_EXCLUSIVE | LK_RETRY);
445 		AUDIT_ARG_VNODE1(binvp);
446 		imgp->vp = binvp;
447 	}
448 
449 	/*
450 	 * Check file permissions (also 'opens' file)
451 	 */
452 	error = exec_check_permissions(imgp);
453 	if (error)
454 		goto exec_fail_dealloc;
455 
456 	imgp->object = imgp->vp->v_object;
457 	if (imgp->object != NULL)
458 		vm_object_reference(imgp->object);
459 
460 	/*
461 	 * Set VV_TEXT now so no one can write to the executable while we're
462 	 * activating it.
463 	 *
464 	 * Remember if this was set before and unset it in case this is not
465 	 * actually an executable image.
466 	 */
467 	textset = VOP_IS_TEXT(imgp->vp);
468 	VOP_SET_TEXT(imgp->vp);
469 
470 	error = exec_map_first_page(imgp);
471 	if (error)
472 		goto exec_fail_dealloc;
473 
474 	imgp->proc->p_osrel = 0;
475 	/*
476 	 *	If the current process has a special image activator it
477 	 *	wants to try first, call it.   For example, emulating shell
478 	 *	scripts differently.
479 	 */
480 	error = -1;
481 	if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
482 		error = img_first(imgp);
483 
484 	/*
485 	 *	Loop through the list of image activators, calling each one.
486 	 *	An activator returns -1 if there is no match, 0 on success,
487 	 *	and an error otherwise.
488 	 */
489 	for (i = 0; error == -1 && execsw[i]; ++i) {
490 		if (execsw[i]->ex_imgact == NULL ||
491 		    execsw[i]->ex_imgact == img_first) {
492 			continue;
493 		}
494 		error = (*execsw[i]->ex_imgact)(imgp);
495 	}
496 
497 	if (error) {
498 		if (error == -1) {
499 			if (textset == 0)
500 				VOP_UNSET_TEXT(imgp->vp);
501 			error = ENOEXEC;
502 		}
503 		goto exec_fail_dealloc;
504 	}
505 
506 	/*
507 	 * Special interpreter operation, cleanup and loop up to try to
508 	 * activate the interpreter.
509 	 */
510 	if (imgp->interpreted) {
511 		exec_unmap_first_page(imgp);
512 		/*
513 		 * VV_TEXT needs to be unset for scripts.  There is a short
514 		 * period before we determine that something is a script where
515 		 * VV_TEXT will be set. The vnode lock is held over this
516 		 * entire period so nothing should illegitimately be blocked.
517 		 */
518 		VOP_UNSET_TEXT(imgp->vp);
519 		/* free name buffer and old vnode */
520 		if (args->fname != NULL)
521 			NDFREE(&nd, NDF_ONLY_PNBUF);
522 #ifdef MAC
523 		mac_execve_interpreter_enter(binvp, &interpvplabel);
524 #endif
525 		if (imgp->opened) {
526 			VOP_CLOSE(binvp, FREAD, td->td_ucred, td);
527 			imgp->opened = 0;
528 		}
529 		vput(binvp);
530 		vm_object_deallocate(imgp->object);
531 		imgp->object = NULL;
532 		/* set new name to that of the interpreter */
533 		NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
534 		    UIO_SYSSPACE, imgp->interpreter_name, td);
535 		args->fname = imgp->interpreter_name;
536 		goto interpret;
537 	}
538 
539 	/*
540 	 * NB: We unlock the vnode here because it is believed that none
541 	 * of the sv_copyout_strings/sv_fixup operations require the vnode.
542 	 */
543 	VOP_UNLOCK(imgp->vp, 0);
544 
545 	/*
546 	 * Do the best to calculate the full path to the image file.
547 	 */
548 	if (imgp->auxargs != NULL &&
549 	    ((args->fname != NULL && args->fname[0] == '/') ||
550 	     vn_fullpath(td, imgp->vp, &imgp->execpath, &imgp->freepath) != 0))
551 		imgp->execpath = args->fname;
552 
553 	/*
554 	 * Copy out strings (args and env) and initialize stack base
555 	 */
556 	if (p->p_sysent->sv_copyout_strings)
557 		stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
558 	else
559 		stack_base = exec_copyout_strings(imgp);
560 
561 	/*
562 	 * If custom stack fixup routine present for this process
563 	 * let it do the stack setup.
564 	 * Else stuff argument count as first item on stack
565 	 */
566 	if (p->p_sysent->sv_fixup != NULL)
567 		(*p->p_sysent->sv_fixup)(&stack_base, imgp);
568 	else
569 		suword(--stack_base, imgp->args->argc);
570 
571 	/*
572 	 * For security and other reasons, the file descriptor table cannot
573 	 * be shared after an exec.
574 	 */
575 	fdunshare(p, td);
576 
577 	/*
578 	 * Malloc things before we need locks.
579 	 */
580 	newcred = crget();
581 	euip = uifind(attr.va_uid);
582 	i = imgp->args->begin_envv - imgp->args->begin_argv;
583 	/* Cache arguments if they fit inside our allowance */
584 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
585 		newargs = pargs_alloc(i);
586 		bcopy(imgp->args->begin_argv, newargs->ar_args, i);
587 	}
588 
589 	/* close files on exec */
590 	fdcloseexec(td);
591 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
592 
593 	/* Get a reference to the vnode prior to locking the proc */
594 	VREF(binvp);
595 
596 	/*
597 	 * For security and other reasons, signal handlers cannot
598 	 * be shared after an exec. The new process gets a copy of the old
599 	 * handlers. In execsigs(), the new process will have its signals
600 	 * reset.
601 	 */
602 	PROC_LOCK(p);
603 	oldcred = crcopysafe(p, newcred);
604 	if (sigacts_shared(p->p_sigacts)) {
605 		oldsigacts = p->p_sigacts;
606 		PROC_UNLOCK(p);
607 		newsigacts = sigacts_alloc();
608 		sigacts_copy(newsigacts, oldsigacts);
609 		PROC_LOCK(p);
610 		p->p_sigacts = newsigacts;
611 	} else
612 		oldsigacts = NULL;
613 
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_pptr && (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 		 * setugidsafety() may call closef() and then pfind()
692 		 * which may grab the process lock.
693 		 * fdcheckstd() may call falloc() which may block to
694 		 * allocate memory, so temporarily drop the process lock.
695 		 */
696 		PROC_UNLOCK(p);
697 		VOP_UNLOCK(imgp->vp, 0);
698 		setugidsafety(td);
699 		error = fdcheckstd(td);
700 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
701 		if (error != 0)
702 			goto done1;
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 		newcred = NULL;
728 	} else {
729 		if (oldcred->cr_uid == oldcred->cr_ruid &&
730 		    oldcred->cr_gid == oldcred->cr_rgid)
731 			p->p_flag &= ~P_SUGID;
732 		/*
733 		 * Implement correct POSIX saved-id behavior.
734 		 *
735 		 * XXX: It's not clear that the existing behavior is
736 		 * POSIX-compliant.  A number of sources indicate that the
737 		 * saved uid/gid should only be updated if the new ruid is
738 		 * not equal to the old ruid, or the new euid is not equal
739 		 * to the old euid and the new euid is not equal to the old
740 		 * ruid.  The FreeBSD code always updates the saved uid/gid.
741 		 * Also, this code uses the new (replaced) euid and egid as
742 		 * the source, which may or may not be the right ones to use.
743 		 */
744 		if (oldcred->cr_svuid != oldcred->cr_uid ||
745 		    oldcred->cr_svgid != oldcred->cr_gid) {
746 			change_svuid(newcred, newcred->cr_uid);
747 			change_svgid(newcred, newcred->cr_gid);
748 			p->p_ucred = newcred;
749 			newcred = NULL;
750 		}
751 	}
752 
753 	/*
754 	 * Store the vp for use in procfs.  This vnode was referenced prior
755 	 * to locking the proc lock.
756 	 */
757 	textvp = p->p_textvp;
758 	p->p_textvp = binvp;
759 
760 #ifdef KDTRACE_HOOKS
761 	/*
762 	 * Tell the DTrace fasttrap provider about the exec if it
763 	 * has declared an interest.
764 	 */
765 	if (dtrace_fasttrap_exec)
766 		dtrace_fasttrap_exec(p);
767 #endif
768 
769 	/*
770 	 * Notify others that we exec'd, and clear the P_INEXEC flag
771 	 * as we're now a bona fide freshly-execed process.
772 	 */
773 	KNOTE_LOCKED(&p->p_klist, NOTE_EXEC);
774 	p->p_flag &= ~P_INEXEC;
775 
776 	/* clear "fork but no exec" flag, as we _are_ execing */
777 	p->p_acflag &= ~AFORK;
778 
779 	/*
780 	 * Free any previous argument cache and replace it with
781 	 * the new argument cache, if any.
782 	 */
783 	oldargs = p->p_args;
784 	p->p_args = newargs;
785 	newargs = NULL;
786 
787 #ifdef	HWPMC_HOOKS
788 	/*
789 	 * Check if system-wide sampling is in effect or if the
790 	 * current process is using PMCs.  If so, do exec() time
791 	 * processing.  This processing needs to happen AFTER the
792 	 * P_INEXEC flag is cleared.
793 	 *
794 	 * The proc lock needs to be released before taking the PMC
795 	 * SX.
796 	 */
797 	if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
798 		PROC_UNLOCK(p);
799 		VOP_UNLOCK(imgp->vp, 0);
800 		pe.pm_credentialschanged = credential_changing;
801 		pe.pm_entryaddr = imgp->entry_addr;
802 
803 		PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
804 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
805 	} else
806 		PROC_UNLOCK(p);
807 #else  /* !HWPMC_HOOKS */
808 	PROC_UNLOCK(p);
809 #endif
810 
811 	/* Set values passed into the program in registers. */
812 	if (p->p_sysent->sv_setregs)
813 		(*p->p_sysent->sv_setregs)(td, imgp,
814 		    (u_long)(uintptr_t)stack_base);
815 	else
816 		exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base);
817 
818 	vfs_mark_atime(imgp->vp, td->td_ucred);
819 
820 	SDT_PROBE(proc, kernel, , exec_success, args->fname, 0, 0, 0, 0);
821 
822 done1:
823 	/*
824 	 * Free any resources malloc'd earlier that we didn't use.
825 	 */
826 	uifree(euip);
827 	if (newcred == NULL)
828 		crfree(oldcred);
829 	else
830 		crfree(newcred);
831 	VOP_UNLOCK(imgp->vp, 0);
832 
833 	/*
834 	 * Handle deferred decrement of ref counts.
835 	 */
836 	if (textvp != NULL)
837 		vrele(textvp);
838 	if (binvp && error != 0)
839 		vrele(binvp);
840 #ifdef KTRACE
841 	if (tracevp != NULL)
842 		vrele(tracevp);
843 	if (tracecred != NULL)
844 		crfree(tracecred);
845 #endif
846 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
847 	pargs_drop(oldargs);
848 	pargs_drop(newargs);
849 	if (oldsigacts != NULL)
850 		sigacts_free(oldsigacts);
851 
852 exec_fail_dealloc:
853 
854 	/*
855 	 * free various allocated resources
856 	 */
857 	if (imgp->firstpage != NULL)
858 		exec_unmap_first_page(imgp);
859 
860 	if (imgp->vp != NULL) {
861 		if (args->fname)
862 			NDFREE(&nd, NDF_ONLY_PNBUF);
863 		if (imgp->opened)
864 			VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
865 		vput(imgp->vp);
866 	}
867 
868 	if (imgp->object != NULL)
869 		vm_object_deallocate(imgp->object);
870 
871 	free(imgp->freepath, M_TEMP);
872 
873 	if (error == 0) {
874 		PROC_LOCK(p);
875 		td->td_dbgflags |= TDB_EXEC;
876 		PROC_UNLOCK(p);
877 
878 		/*
879 		 * Stop the process here if its stop event mask has
880 		 * the S_EXEC bit set.
881 		 */
882 		STOPEVENT(p, S_EXEC, 0);
883 		goto done2;
884 	}
885 
886 exec_fail:
887 	/* we're done here, clear P_INEXEC */
888 	PROC_LOCK(p);
889 	p->p_flag &= ~P_INEXEC;
890 	PROC_UNLOCK(p);
891 
892 	SDT_PROBE(proc, kernel, , exec_failure, error, 0, 0, 0, 0);
893 
894 done2:
895 #ifdef MAC
896 	mac_execve_exit(imgp);
897 	mac_execve_interpreter_exit(interpvplabel);
898 #endif
899 	exec_free_args(args);
900 
901 	if (error && imgp->vmspace_destroyed) {
902 		/* sorry, no more process anymore. exit gracefully */
903 		exit1(td, W_EXITCODE(0, SIGABRT));
904 		/* NOT REACHED */
905 	}
906 
907 #ifdef KTRACE
908 	if (error == 0)
909 		ktrprocctor(p);
910 #endif
911 
912 	return (error);
913 }
914 
915 int
916 exec_map_first_page(imgp)
917 	struct image_params *imgp;
918 {
919 	int rv, i;
920 	int initial_pagein;
921 	vm_page_t ma[VM_INITIAL_PAGEIN];
922 	vm_object_t object;
923 
924 	if (imgp->firstpage != NULL)
925 		exec_unmap_first_page(imgp);
926 
927 	object = imgp->vp->v_object;
928 	if (object == NULL)
929 		return (EACCES);
930 	VM_OBJECT_WLOCK(object);
931 #if VM_NRESERVLEVEL > 0
932 	if ((object->flags & OBJ_COLORED) == 0) {
933 		object->flags |= OBJ_COLORED;
934 		object->pg_color = 0;
935 	}
936 #endif
937 	ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL);
938 	if (ma[0]->valid != VM_PAGE_BITS_ALL) {
939 		initial_pagein = VM_INITIAL_PAGEIN;
940 		if (initial_pagein > object->size)
941 			initial_pagein = object->size;
942 		for (i = 1; i < initial_pagein; i++) {
943 			if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) {
944 				if (ma[i]->valid)
945 					break;
946 				if (vm_page_tryxbusy(ma[i]))
947 					break;
948 			} else {
949 				ma[i] = vm_page_alloc(object, i,
950 				    VM_ALLOC_NORMAL | VM_ALLOC_IFNOTCACHED);
951 				if (ma[i] == NULL)
952 					break;
953 			}
954 		}
955 		initial_pagein = i;
956 		rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
957 		ma[0] = vm_page_lookup(object, 0);
958 		if ((rv != VM_PAGER_OK) || (ma[0] == NULL)) {
959 			if (ma[0] != NULL) {
960 				vm_page_lock(ma[0]);
961 				vm_page_free(ma[0]);
962 				vm_page_unlock(ma[0]);
963 			}
964 			VM_OBJECT_WUNLOCK(object);
965 			return (EIO);
966 		}
967 	}
968 	vm_page_xunbusy(ma[0]);
969 	vm_page_lock(ma[0]);
970 	vm_page_hold(ma[0]);
971 	vm_page_unlock(ma[0]);
972 	VM_OBJECT_WUNLOCK(object);
973 
974 	imgp->firstpage = sf_buf_alloc(ma[0], 0);
975 	imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
976 
977 	return (0);
978 }
979 
980 void
981 exec_unmap_first_page(imgp)
982 	struct image_params *imgp;
983 {
984 	vm_page_t m;
985 
986 	if (imgp->firstpage != NULL) {
987 		m = sf_buf_page(imgp->firstpage);
988 		sf_buf_free(imgp->firstpage);
989 		imgp->firstpage = NULL;
990 		vm_page_lock(m);
991 		vm_page_unhold(m);
992 		vm_page_unlock(m);
993 	}
994 }
995 
996 /*
997  * Destroy old address space, and allocate a new stack
998  *	The new stack is only SGROWSIZ large because it is grown
999  *	automatically in trap.c.
1000  */
1001 int
1002 exec_new_vmspace(imgp, sv)
1003 	struct image_params *imgp;
1004 	struct sysentvec *sv;
1005 {
1006 	int error;
1007 	struct proc *p = imgp->proc;
1008 	struct vmspace *vmspace = p->p_vmspace;
1009 	vm_object_t obj;
1010 	vm_offset_t sv_minuser, stack_addr;
1011 	vm_map_t map;
1012 	u_long ssiz;
1013 
1014 	imgp->vmspace_destroyed = 1;
1015 	imgp->sysent = sv;
1016 
1017 	/* May be called with Giant held */
1018 	EVENTHANDLER_INVOKE(process_exec, p, imgp);
1019 
1020 	/*
1021 	 * Blow away entire process VM, if address space not shared,
1022 	 * otherwise, create a new VM space so that other threads are
1023 	 * not disrupted
1024 	 */
1025 	map = &vmspace->vm_map;
1026 	if (map_at_zero)
1027 		sv_minuser = sv->sv_minuser;
1028 	else
1029 		sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1030 	if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser &&
1031 	    vm_map_max(map) == sv->sv_maxuser) {
1032 		shmexit(vmspace);
1033 		pmap_remove_pages(vmspace_pmap(vmspace));
1034 		vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1035 	} else {
1036 		error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1037 		if (error)
1038 			return (error);
1039 		vmspace = p->p_vmspace;
1040 		map = &vmspace->vm_map;
1041 	}
1042 
1043 	/* Map a shared page */
1044 	obj = sv->sv_shared_page_obj;
1045 	if (obj != NULL) {
1046 		vm_object_reference(obj);
1047 		error = vm_map_fixed(map, obj, 0,
1048 		    sv->sv_shared_page_base, sv->sv_shared_page_len,
1049 		    VM_PROT_READ | VM_PROT_EXECUTE,
1050 		    VM_PROT_READ | VM_PROT_EXECUTE,
1051 		    MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1052 		if (error) {
1053 			vm_object_deallocate(obj);
1054 			return (error);
1055 		}
1056 	}
1057 
1058 	/* Allocate a new stack */
1059 	if (sv->sv_maxssiz != NULL)
1060 		ssiz = *sv->sv_maxssiz;
1061 	else
1062 		ssiz = maxssiz;
1063 	stack_addr = sv->sv_usrstack - ssiz;
1064 	error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1065 	    obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1066 		sv->sv_stackprot,
1067 	    VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1068 	if (error)
1069 		return (error);
1070 
1071 #ifdef __ia64__
1072 	/* Allocate a new register stack */
1073 	stack_addr = IA64_BACKINGSTORE;
1074 	error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1075 	    sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_UP);
1076 	if (error)
1077 		return (error);
1078 #endif
1079 
1080 	/* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the
1081 	 * VM_STACK case, but they are still used to monitor the size of the
1082 	 * process stack so we can check the stack rlimit.
1083 	 */
1084 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1085 	vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - ssiz;
1086 
1087 	return (0);
1088 }
1089 
1090 /*
1091  * Copy out argument and environment strings from the old process address
1092  * space into the temporary string buffer.
1093  */
1094 int
1095 exec_copyin_args(struct image_args *args, char *fname,
1096     enum uio_seg segflg, char **argv, char **envv)
1097 {
1098 	char *argp, *envp;
1099 	int error;
1100 	size_t length;
1101 
1102 	bzero(args, sizeof(*args));
1103 	if (argv == NULL)
1104 		return (EFAULT);
1105 
1106 	/*
1107 	 * Allocate demand-paged memory for the file name, argument, and
1108 	 * environment strings.
1109 	 */
1110 	error = exec_alloc_args(args);
1111 	if (error != 0)
1112 		return (error);
1113 
1114 	/*
1115 	 * Copy the file name.
1116 	 */
1117 	if (fname != NULL) {
1118 		args->fname = args->buf;
1119 		error = (segflg == UIO_SYSSPACE) ?
1120 		    copystr(fname, args->fname, PATH_MAX, &length) :
1121 		    copyinstr(fname, args->fname, PATH_MAX, &length);
1122 		if (error != 0)
1123 			goto err_exit;
1124 	} else
1125 		length = 0;
1126 
1127 	args->begin_argv = args->buf + length;
1128 	args->endp = args->begin_argv;
1129 	args->stringspace = ARG_MAX;
1130 
1131 	/*
1132 	 * extract arguments first
1133 	 */
1134 	while ((argp = (caddr_t) (intptr_t) fuword(argv++))) {
1135 		if (argp == (caddr_t) -1) {
1136 			error = EFAULT;
1137 			goto err_exit;
1138 		}
1139 		if ((error = copyinstr(argp, args->endp,
1140 		    args->stringspace, &length))) {
1141 			if (error == ENAMETOOLONG)
1142 				error = E2BIG;
1143 			goto err_exit;
1144 		}
1145 		args->stringspace -= length;
1146 		args->endp += length;
1147 		args->argc++;
1148 	}
1149 
1150 	args->begin_envv = args->endp;
1151 
1152 	/*
1153 	 * extract environment strings
1154 	 */
1155 	if (envv) {
1156 		while ((envp = (caddr_t)(intptr_t)fuword(envv++))) {
1157 			if (envp == (caddr_t)-1) {
1158 				error = EFAULT;
1159 				goto err_exit;
1160 			}
1161 			if ((error = copyinstr(envp, args->endp,
1162 			    args->stringspace, &length))) {
1163 				if (error == ENAMETOOLONG)
1164 					error = E2BIG;
1165 				goto err_exit;
1166 			}
1167 			args->stringspace -= length;
1168 			args->endp += length;
1169 			args->envc++;
1170 		}
1171 	}
1172 
1173 	return (0);
1174 
1175 err_exit:
1176 	exec_free_args(args);
1177 	return (error);
1178 }
1179 
1180 /*
1181  * Allocate temporary demand-paged, zero-filled memory for the file name,
1182  * argument, and environment strings.  Returns zero if the allocation succeeds
1183  * and ENOMEM otherwise.
1184  */
1185 int
1186 exec_alloc_args(struct image_args *args)
1187 {
1188 
1189 	args->buf = (char *)kmap_alloc_wait(exec_map, PATH_MAX + ARG_MAX);
1190 	return (args->buf != NULL ? 0 : ENOMEM);
1191 }
1192 
1193 void
1194 exec_free_args(struct image_args *args)
1195 {
1196 
1197 	if (args->buf != NULL) {
1198 		kmap_free_wakeup(exec_map, (vm_offset_t)args->buf,
1199 		    PATH_MAX + ARG_MAX);
1200 		args->buf = NULL;
1201 	}
1202 	if (args->fname_buf != NULL) {
1203 		free(args->fname_buf, M_TEMP);
1204 		args->fname_buf = NULL;
1205 	}
1206 }
1207 
1208 /*
1209  * Copy strings out to the new process address space, constructing new arg
1210  * and env vector tables. Return a pointer to the base so that it can be used
1211  * as the initial stack pointer.
1212  */
1213 register_t *
1214 exec_copyout_strings(imgp)
1215 	struct image_params *imgp;
1216 {
1217 	int argc, envc;
1218 	char **vectp;
1219 	char *stringp, *destp;
1220 	register_t *stack_base;
1221 	struct ps_strings *arginfo;
1222 	struct proc *p;
1223 	size_t execpath_len;
1224 	int szsigcode, szps;
1225 	char canary[sizeof(long) * 8];
1226 
1227 	szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1228 	/*
1229 	 * Calculate string base and vector table pointers.
1230 	 * Also deal with signal trampoline code for this exec type.
1231 	 */
1232 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
1233 		execpath_len = strlen(imgp->execpath) + 1;
1234 	else
1235 		execpath_len = 0;
1236 	p = imgp->proc;
1237 	szsigcode = 0;
1238 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1239 	if (p->p_sysent->sv_sigcode_base == 0) {
1240 		if (p->p_sysent->sv_szsigcode != NULL)
1241 			szsigcode = *(p->p_sysent->sv_szsigcode);
1242 	}
1243 	destp =	(caddr_t)arginfo - szsigcode - SPARE_USRSPACE -
1244 	    roundup(execpath_len, sizeof(char *)) -
1245 	    roundup(sizeof(canary), sizeof(char *)) -
1246 	    roundup(szps, sizeof(char *)) -
1247 	    roundup((ARG_MAX - imgp->args->stringspace), sizeof(char *));
1248 
1249 	/*
1250 	 * install sigcode
1251 	 */
1252 	if (szsigcode != 0)
1253 		copyout(p->p_sysent->sv_sigcode, ((caddr_t)arginfo -
1254 		    szsigcode), szsigcode);
1255 
1256 	/*
1257 	 * Copy the image path for the rtld.
1258 	 */
1259 	if (execpath_len != 0) {
1260 		imgp->execpathp = (uintptr_t)arginfo - szsigcode - execpath_len;
1261 		copyout(imgp->execpath, (void *)imgp->execpathp,
1262 		    execpath_len);
1263 	}
1264 
1265 	/*
1266 	 * Prepare the canary for SSP.
1267 	 */
1268 	arc4rand(canary, sizeof(canary), 0);
1269 	imgp->canary = (uintptr_t)arginfo - szsigcode - execpath_len -
1270 	    sizeof(canary);
1271 	copyout(canary, (void *)imgp->canary, sizeof(canary));
1272 	imgp->canarylen = sizeof(canary);
1273 
1274 	/*
1275 	 * Prepare the pagesizes array.
1276 	 */
1277 	imgp->pagesizes = (uintptr_t)arginfo - szsigcode - execpath_len -
1278 	    roundup(sizeof(canary), sizeof(char *)) - szps;
1279 	copyout(pagesizes, (void *)imgp->pagesizes, szps);
1280 	imgp->pagesizeslen = szps;
1281 
1282 	/*
1283 	 * If we have a valid auxargs ptr, prepare some room
1284 	 * on the stack.
1285 	 */
1286 	if (imgp->auxargs) {
1287 		/*
1288 		 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1289 		 * lower compatibility.
1290 		 */
1291 		imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1292 		    (AT_COUNT * 2);
1293 		/*
1294 		 * The '+ 2' is for the null pointers at the end of each of
1295 		 * the arg and env vector sets,and imgp->auxarg_size is room
1296 		 * for argument of Runtime loader.
1297 		 */
1298 		vectp = (char **)(destp - (imgp->args->argc +
1299 		    imgp->args->envc + 2 + imgp->auxarg_size)
1300 		    * sizeof(char *));
1301 	} else {
1302 		/*
1303 		 * The '+ 2' is for the null pointers at the end of each of
1304 		 * the arg and env vector sets
1305 		 */
1306 		vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc + 2) *
1307 		    sizeof(char *));
1308 	}
1309 
1310 	/*
1311 	 * vectp also becomes our initial stack base
1312 	 */
1313 	stack_base = (register_t *)vectp;
1314 
1315 	stringp = imgp->args->begin_argv;
1316 	argc = imgp->args->argc;
1317 	envc = imgp->args->envc;
1318 
1319 	/*
1320 	 * Copy out strings - arguments and environment.
1321 	 */
1322 	copyout(stringp, destp, ARG_MAX - imgp->args->stringspace);
1323 
1324 	/*
1325 	 * Fill in "ps_strings" struct for ps, w, etc.
1326 	 */
1327 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1328 	suword32(&arginfo->ps_nargvstr, argc);
1329 
1330 	/*
1331 	 * Fill in argument portion of vector table.
1332 	 */
1333 	for (; argc > 0; --argc) {
1334 		suword(vectp++, (long)(intptr_t)destp);
1335 		while (*stringp++ != 0)
1336 			destp++;
1337 		destp++;
1338 	}
1339 
1340 	/* a null vector table pointer separates the argp's from the envp's */
1341 	suword(vectp++, 0);
1342 
1343 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1344 	suword32(&arginfo->ps_nenvstr, envc);
1345 
1346 	/*
1347 	 * Fill in environment portion of vector table.
1348 	 */
1349 	for (; envc > 0; --envc) {
1350 		suword(vectp++, (long)(intptr_t)destp);
1351 		while (*stringp++ != 0)
1352 			destp++;
1353 		destp++;
1354 	}
1355 
1356 	/* end of vector table is a null pointer */
1357 	suword(vectp, 0);
1358 
1359 	return (stack_base);
1360 }
1361 
1362 /*
1363  * Check permissions of file to execute.
1364  *	Called with imgp->vp locked.
1365  *	Return 0 for success or error code on failure.
1366  */
1367 int
1368 exec_check_permissions(imgp)
1369 	struct image_params *imgp;
1370 {
1371 	struct vnode *vp = imgp->vp;
1372 	struct vattr *attr = imgp->attr;
1373 	struct thread *td;
1374 	int error, writecount;
1375 
1376 	td = curthread;
1377 
1378 	/* Get file attributes */
1379 	error = VOP_GETATTR(vp, attr, td->td_ucred);
1380 	if (error)
1381 		return (error);
1382 
1383 #ifdef MAC
1384 	error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1385 	if (error)
1386 		return (error);
1387 #endif
1388 
1389 	/*
1390 	 * 1) Check if file execution is disabled for the filesystem that
1391 	 *    this file resides on.
1392 	 * 2) Ensure that at least one execute bit is on. Otherwise, a
1393 	 *    privileged user will always succeed, and we don't want this
1394 	 *    to happen unless the file really is executable.
1395 	 * 3) Ensure that the file is a regular file.
1396 	 */
1397 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1398 	    (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1399 	    (attr->va_type != VREG))
1400 		return (EACCES);
1401 
1402 	/*
1403 	 * Zero length files can't be exec'd
1404 	 */
1405 	if (attr->va_size == 0)
1406 		return (ENOEXEC);
1407 
1408 	/*
1409 	 *  Check for execute permission to file based on current credentials.
1410 	 */
1411 	error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1412 	if (error)
1413 		return (error);
1414 
1415 	/*
1416 	 * Check number of open-for-writes on the file and deny execution
1417 	 * if there are any.
1418 	 */
1419 	error = VOP_GET_WRITECOUNT(vp, &writecount);
1420 	if (error != 0)
1421 		return (error);
1422 	if (writecount != 0)
1423 		return (ETXTBSY);
1424 
1425 	/*
1426 	 * Call filesystem specific open routine (which does nothing in the
1427 	 * general case).
1428 	 */
1429 	error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1430 	if (error == 0)
1431 		imgp->opened = 1;
1432 	return (error);
1433 }
1434 
1435 /*
1436  * Exec handler registration
1437  */
1438 int
1439 exec_register(execsw_arg)
1440 	const struct execsw *execsw_arg;
1441 {
1442 	const struct execsw **es, **xs, **newexecsw;
1443 	int count = 2;	/* New slot and trailing NULL */
1444 
1445 	if (execsw)
1446 		for (es = execsw; *es; es++)
1447 			count++;
1448 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1449 	if (newexecsw == NULL)
1450 		return (ENOMEM);
1451 	xs = newexecsw;
1452 	if (execsw)
1453 		for (es = execsw; *es; es++)
1454 			*xs++ = *es;
1455 	*xs++ = execsw_arg;
1456 	*xs = NULL;
1457 	if (execsw)
1458 		free(execsw, M_TEMP);
1459 	execsw = newexecsw;
1460 	return (0);
1461 }
1462 
1463 int
1464 exec_unregister(execsw_arg)
1465 	const struct execsw *execsw_arg;
1466 {
1467 	const struct execsw **es, **xs, **newexecsw;
1468 	int count = 1;
1469 
1470 	if (execsw == NULL)
1471 		panic("unregister with no handlers left?\n");
1472 
1473 	for (es = execsw; *es; es++) {
1474 		if (*es == execsw_arg)
1475 			break;
1476 	}
1477 	if (*es == NULL)
1478 		return (ENOENT);
1479 	for (es = execsw; *es; es++)
1480 		if (*es != execsw_arg)
1481 			count++;
1482 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1483 	if (newexecsw == NULL)
1484 		return (ENOMEM);
1485 	xs = newexecsw;
1486 	for (es = execsw; *es; es++)
1487 		if (*es != execsw_arg)
1488 			*xs++ = *es;
1489 	*xs = NULL;
1490 	if (execsw)
1491 		free(execsw, M_TEMP);
1492 	execsw = newexecsw;
1493 	return (0);
1494 }
1495