xref: /freebsd/sys/kern/kern_exec.c (revision bd81e07d2761cf1c13063eb49a5c0cb4a6951318)
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(struct thread *td, struct execve_args *uap)
197 {
198 	struct image_args args;
199 	struct vmspace *oldvmspace;
200 	int error;
201 
202 	error = pre_execve(td, &oldvmspace);
203 	if (error != 0)
204 		return (error);
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 	post_execve(td, error, oldvmspace);
210 	return (error);
211 }
212 
213 #ifndef _SYS_SYSPROTO_H_
214 struct fexecve_args {
215 	int	fd;
216 	char	**argv;
217 	char	**envv;
218 }
219 #endif
220 int
221 sys_fexecve(struct thread *td, struct fexecve_args *uap)
222 {
223 	struct image_args args;
224 	struct vmspace *oldvmspace;
225 	int error;
226 
227 	error = pre_execve(td, &oldvmspace);
228 	if (error != 0)
229 		return (error);
230 	error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
231 	    uap->argv, uap->envv);
232 	if (error == 0) {
233 		args.fd = uap->fd;
234 		error = kern_execve(td, &args, NULL);
235 	}
236 	post_execve(td, error, oldvmspace);
237 	return (error);
238 }
239 
240 #ifndef _SYS_SYSPROTO_H_
241 struct __mac_execve_args {
242 	char	*fname;
243 	char	**argv;
244 	char	**envv;
245 	struct mac	*mac_p;
246 };
247 #endif
248 
249 int
250 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap)
251 {
252 #ifdef MAC
253 	struct image_args args;
254 	struct vmspace *oldvmspace;
255 	int error;
256 
257 	error = pre_execve(td, &oldvmspace);
258 	if (error != 0)
259 		return (error);
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 	post_execve(td, error, oldvmspace);
265 	return (error);
266 #else
267 	return (ENOSYS);
268 #endif
269 }
270 
271 int
272 pre_execve(struct thread *td, struct vmspace **oldvmspace)
273 {
274 	struct proc *p;
275 	int error;
276 
277 	KASSERT(td == curthread, ("non-current thread %p", td));
278 	error = 0;
279 	p = td->td_proc;
280 	if ((p->p_flag & P_HADTHREADS) != 0) {
281 		PROC_LOCK(p);
282 		if (thread_single(p, SINGLE_BOUNDARY) != 0)
283 			error = ERESTART;
284 		PROC_UNLOCK(p);
285 	}
286 	KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0,
287 	    ("nested execve"));
288 	*oldvmspace = p->p_vmspace;
289 	return (error);
290 }
291 
292 void
293 post_execve(struct thread *td, int error, struct vmspace *oldvmspace)
294 {
295 	struct proc *p;
296 
297 	KASSERT(td == curthread, ("non-current thread %p", td));
298 	p = td->td_proc;
299 	if ((p->p_flag & P_HADTHREADS) != 0) {
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(p, SINGLE_EXIT);
307 		else
308 			thread_single_end(p, SINGLE_BOUNDARY);
309 		PROC_UNLOCK(p);
310 	}
311 	if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
312 		KASSERT(p->p_vmspace != oldvmspace,
313 		    ("oldvmspace still used"));
314 		vmspace_free(oldvmspace);
315 		td->td_pflags &= ~TDP_EXECVMSPC;
316 	}
317 }
318 
319 /*
320  * XXX: kern_execve has the astonishing property of not always returning to
321  * the caller.  If sufficiently bad things happen during the call to
322  * do_execve(), it can end up calling exit1(); as a result, callers must
323  * avoid doing anything which they might need to undo (e.g., allocating
324  * memory).
325  */
326 int
327 kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p)
328 {
329 
330 	AUDIT_ARG_ARGV(args->begin_argv, args->argc,
331 	    args->begin_envv - args->begin_argv);
332 	AUDIT_ARG_ENVV(args->begin_envv, args->envc,
333 	    args->endp - args->begin_envv);
334 	return (do_execve(td, args, mac_p));
335 }
336 
337 /*
338  * In-kernel implementation of execve().  All arguments are assumed to be
339  * userspace pointers from the passed thread.
340  */
341 static int
342 do_execve(td, args, mac_p)
343 	struct thread *td;
344 	struct image_args *args;
345 	struct mac *mac_p;
346 {
347 	struct proc *p = td->td_proc;
348 	struct nameidata nd;
349 	struct ucred *newcred = NULL, *oldcred;
350 	struct uidinfo *euip = NULL;
351 	register_t *stack_base;
352 	int error, i;
353 	struct image_params image_params, *imgp;
354 	struct vattr attr;
355 	int (*img_first)(struct image_params *);
356 	struct pargs *oldargs = NULL, *newargs = NULL;
357 	struct sigacts *oldsigacts, *newsigacts;
358 #ifdef KTRACE
359 	struct vnode *tracevp = NULL;
360 	struct ucred *tracecred = NULL;
361 #endif
362 	struct vnode *oldtextvp = NULL, *newtextvp;
363 	cap_rights_t rights;
364 	int credential_changing;
365 	int textset;
366 #ifdef MAC
367 	struct label *interpvplabel = NULL;
368 	int will_transition;
369 #endif
370 #ifdef HWPMC_HOOKS
371 	struct pmckern_procexec pe;
372 #endif
373 	static const char fexecv_proc_title[] = "(fexecv)";
374 
375 	imgp = &image_params;
376 
377 	/*
378 	 * Lock the process and set the P_INEXEC flag to indicate that
379 	 * it should be left alone until we're done here.  This is
380 	 * necessary to avoid race conditions - e.g. in ptrace() -
381 	 * that might allow a local user to illicitly obtain elevated
382 	 * privileges.
383 	 */
384 	PROC_LOCK(p);
385 	KASSERT((p->p_flag & P_INEXEC) == 0,
386 	    ("%s(): process already has P_INEXEC flag", __func__));
387 	p->p_flag |= P_INEXEC;
388 	PROC_UNLOCK(p);
389 
390 	/*
391 	 * Initialize part of the common data
392 	 */
393 	bzero(imgp, sizeof(*imgp));
394 	imgp->proc = p;
395 	imgp->attr = &attr;
396 	imgp->args = args;
397 
398 #ifdef MAC
399 	error = mac_execve_enter(imgp, mac_p);
400 	if (error)
401 		goto exec_fail;
402 #endif
403 
404 	/*
405 	 * Translate the file name. namei() returns a vnode pointer
406 	 *	in ni_vp amoung other things.
407 	 *
408 	 * XXXAUDIT: It would be desirable to also audit the name of the
409 	 * interpreter if this is an interpreted binary.
410 	 */
411 	if (args->fname != NULL) {
412 		NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME
413 		    | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
414 	}
415 
416 	SDT_PROBE(proc, kernel, , exec, args->fname, 0, 0, 0, 0 );
417 
418 interpret:
419 	if (args->fname != NULL) {
420 #ifdef CAPABILITY_MODE
421 		/*
422 		 * While capability mode can't reach this point via direct
423 		 * path arguments to execve(), we also don't allow
424 		 * interpreters to be used in capability mode (for now).
425 		 * Catch indirect lookups and return a permissions error.
426 		 */
427 		if (IN_CAPABILITY_MODE(td)) {
428 			error = ECAPMODE;
429 			goto exec_fail;
430 		}
431 #endif
432 		error = namei(&nd);
433 		if (error)
434 			goto exec_fail;
435 
436 		newtextvp = nd.ni_vp;
437 		imgp->vp = newtextvp;
438 	} else {
439 		AUDIT_ARG_FD(args->fd);
440 		/*
441 		 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
442 		 */
443 		error = fgetvp_exec(td, args->fd,
444 		    cap_rights_init(&rights, CAP_FEXECVE), &newtextvp);
445 		if (error)
446 			goto exec_fail;
447 		vn_lock(newtextvp, LK_EXCLUSIVE | LK_RETRY);
448 		AUDIT_ARG_VNODE1(newtextvp);
449 		imgp->vp = newtextvp;
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(newtextvp, &interpvplabel);
527 #endif
528 		if (imgp->opened) {
529 			VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td);
530 			imgp->opened = 0;
531 		}
532 		vput(newtextvp);
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 	if (disallow_high_osrel &&
557 	    P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
558 		error = ENOEXEC;
559 		uprintf("Osrel %d for image %s too high\n", p->p_osrel,
560 		    imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
561 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
562 		goto exec_fail_dealloc;
563 	}
564 
565 	/* ABI enforces the use of Capsicum. Switch into capabilities mode. */
566 	if (SV_PROC_FLAG(p, SV_CAPSICUM))
567 		sys_cap_enter(td, NULL);
568 
569 	/*
570 	 * Copy out strings (args and env) and initialize stack base
571 	 */
572 	if (p->p_sysent->sv_copyout_strings)
573 		stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
574 	else
575 		stack_base = exec_copyout_strings(imgp);
576 
577 	/*
578 	 * If custom stack fixup routine present for this process
579 	 * let it do the stack setup.
580 	 * Else stuff argument count as first item on stack
581 	 */
582 	if (p->p_sysent->sv_fixup != NULL)
583 		(*p->p_sysent->sv_fixup)(&stack_base, imgp);
584 	else
585 		suword(--stack_base, imgp->args->argc);
586 
587 	if (args->fdp != NULL) {
588 		/* Install a brand new file descriptor table. */
589 		fdinstall_remapped(td, args->fdp);
590 		args->fdp = NULL;
591 	} else {
592 		/*
593 		 * Keep on using the existing file descriptor table. For
594 		 * security and other reasons, the file descriptor table
595 		 * cannot be shared after an exec.
596 		 */
597 		fdunshare(td);
598 		/* close files on exec */
599 		fdcloseexec(td);
600 	}
601 
602 	/*
603 	 * Malloc things before we need locks.
604 	 */
605 	i = imgp->args->begin_envv - imgp->args->begin_argv;
606 	/* Cache arguments if they fit inside our allowance */
607 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
608 		newargs = pargs_alloc(i);
609 		bcopy(imgp->args->begin_argv, newargs->ar_args, i);
610 	}
611 
612 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
613 
614 	/*
615 	 * For security and other reasons, signal handlers cannot
616 	 * be shared after an exec. The new process gets a copy of the old
617 	 * handlers. In execsigs(), the new process will have its signals
618 	 * reset.
619 	 */
620 	if (sigacts_shared(p->p_sigacts)) {
621 		oldsigacts = p->p_sigacts;
622 		newsigacts = sigacts_alloc();
623 		sigacts_copy(newsigacts, oldsigacts);
624 	} else {
625 		oldsigacts = NULL;
626 		newsigacts = NULL; /* satisfy gcc */
627 	}
628 
629 	PROC_LOCK(p);
630 	if (oldsigacts)
631 		p->p_sigacts = newsigacts;
632 	oldcred = p->p_ucred;
633 	/* Stop profiling */
634 	stopprofclock(p);
635 
636 	/* reset caught signals */
637 	execsigs(p);
638 
639 	/* name this process - nameiexec(p, ndp) */
640 	bzero(p->p_comm, sizeof(p->p_comm));
641 	if (args->fname)
642 		bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
643 		    min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
644 	else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0)
645 		bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
646 	bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
647 #ifdef KTR
648 	sched_clear_tdname(td);
649 #endif
650 
651 	/*
652 	 * mark as execed, wakeup the process that vforked (if any) and tell
653 	 * it that it now has its own resources back
654 	 */
655 	p->p_flag |= P_EXEC;
656 	if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0)
657 		p->p_flag2 &= ~P2_NOTRACE;
658 	if (p->p_flag & P_PPWAIT) {
659 		p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
660 		cv_broadcast(&p->p_pwait);
661 	}
662 
663 	/*
664 	 * Implement image setuid/setgid.
665 	 *
666 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
667 	 * the process is being traced.
668 	 *
669 	 * We disable setuid/setgid/etc in compatibility mode on the basis
670 	 * that most setugid applications are not written with that
671 	 * environment in mind, and will therefore almost certainly operate
672 	 * incorrectly. In principle there's no reason that setugid
673 	 * applications might not be useful in capability mode, so we may want
674 	 * to reconsider this conservative design choice in the future.
675 	 *
676 	 * XXXMAC: For the time being, use NOSUID to also prohibit
677 	 * transitions on the file system.
678 	 */
679 	credential_changing = 0;
680 	credential_changing |= (attr.va_mode & S_ISUID) && oldcred->cr_uid !=
681 	    attr.va_uid;
682 	credential_changing |= (attr.va_mode & S_ISGID) && oldcred->cr_gid !=
683 	    attr.va_gid;
684 #ifdef MAC
685 	will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
686 	    interpvplabel, imgp);
687 	credential_changing |= will_transition;
688 #endif
689 
690 	if (credential_changing &&
691 #ifdef CAPABILITY_MODE
692 	    ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
693 #endif
694 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
695 	    (p->p_flag & P_TRACED) == 0) {
696 		/*
697 		 * Turn off syscall tracing for set-id programs, except for
698 		 * root.  Record any set-id flags first to make sure that
699 		 * we do not regain any tracing during a possible block.
700 		 */
701 		setsugid(p);
702 
703 #ifdef KTRACE
704 		if (p->p_tracecred != NULL &&
705 		    priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED, 0))
706 			ktrprocexec(p, &tracecred, &tracevp);
707 #endif
708 		/*
709 		 * Close any file descriptors 0..2 that reference procfs,
710 		 * then make sure file descriptors 0..2 are in use.
711 		 *
712 		 * Both fdsetugidsafety() and fdcheckstd() may call functions
713 		 * taking sleepable locks, so temporarily drop our locks.
714 		 */
715 		PROC_UNLOCK(p);
716 		VOP_UNLOCK(imgp->vp, 0);
717 		fdsetugidsafety(td);
718 		error = fdcheckstd(td);
719 		if (error != 0)
720 			goto done1;
721 		newcred = crdup(oldcred);
722 		euip = uifind(attr.va_uid);
723 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
724 		PROC_LOCK(p);
725 		/*
726 		 * Set the new credentials.
727 		 */
728 		if (attr.va_mode & S_ISUID)
729 			change_euid(newcred, euip);
730 		if (attr.va_mode & S_ISGID)
731 			change_egid(newcred, attr.va_gid);
732 #ifdef MAC
733 		if (will_transition) {
734 			mac_vnode_execve_transition(oldcred, newcred, imgp->vp,
735 			    interpvplabel, imgp);
736 		}
737 #endif
738 		/*
739 		 * Implement correct POSIX saved-id behavior.
740 		 *
741 		 * XXXMAC: Note that the current logic will save the
742 		 * uid and gid if a MAC domain transition occurs, even
743 		 * though maybe it shouldn't.
744 		 */
745 		change_svuid(newcred, newcred->cr_uid);
746 		change_svgid(newcred, newcred->cr_gid);
747 		proc_set_cred(p, newcred);
748 	} else {
749 		if (oldcred->cr_uid == oldcred->cr_ruid &&
750 		    oldcred->cr_gid == oldcred->cr_rgid)
751 			p->p_flag &= ~P_SUGID;
752 		/*
753 		 * Implement correct POSIX saved-id behavior.
754 		 *
755 		 * XXX: It's not clear that the existing behavior is
756 		 * POSIX-compliant.  A number of sources indicate that the
757 		 * saved uid/gid should only be updated if the new ruid is
758 		 * not equal to the old ruid, or the new euid is not equal
759 		 * to the old euid and the new euid is not equal to the old
760 		 * ruid.  The FreeBSD code always updates the saved uid/gid.
761 		 * Also, this code uses the new (replaced) euid and egid as
762 		 * the source, which may or may not be the right ones to use.
763 		 */
764 		if (oldcred->cr_svuid != oldcred->cr_uid ||
765 		    oldcred->cr_svgid != oldcred->cr_gid) {
766 			PROC_UNLOCK(p);
767 			VOP_UNLOCK(imgp->vp, 0);
768 			newcred = crdup(oldcred);
769 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
770 			PROC_LOCK(p);
771 			change_svuid(newcred, newcred->cr_uid);
772 			change_svgid(newcred, newcred->cr_gid);
773 			proc_set_cred(p, newcred);
774 		}
775 	}
776 
777 	/*
778 	 * Store the vp for use in procfs.  This vnode was referenced by namei
779 	 * or fgetvp_exec.
780 	 */
781 	oldtextvp = p->p_textvp;
782 	p->p_textvp = newtextvp;
783 
784 #ifdef KDTRACE_HOOKS
785 	/*
786 	 * Tell the DTrace fasttrap provider about the exec if it
787 	 * has declared an interest.
788 	 */
789 	if (dtrace_fasttrap_exec)
790 		dtrace_fasttrap_exec(p);
791 #endif
792 
793 	/*
794 	 * Notify others that we exec'd, and clear the P_INEXEC flag
795 	 * as we're now a bona fide freshly-execed process.
796 	 */
797 	KNOTE_LOCKED(&p->p_klist, NOTE_EXEC);
798 	p->p_flag &= ~P_INEXEC;
799 
800 	/* clear "fork but no exec" flag, as we _are_ execing */
801 	p->p_acflag &= ~AFORK;
802 
803 	/*
804 	 * Free any previous argument cache and replace it with
805 	 * the new argument cache, if any.
806 	 */
807 	oldargs = p->p_args;
808 	p->p_args = newargs;
809 	newargs = NULL;
810 
811 #ifdef	HWPMC_HOOKS
812 	/*
813 	 * Check if system-wide sampling is in effect or if the
814 	 * current process is using PMCs.  If so, do exec() time
815 	 * processing.  This processing needs to happen AFTER the
816 	 * P_INEXEC flag is cleared.
817 	 *
818 	 * The proc lock needs to be released before taking the PMC
819 	 * SX.
820 	 */
821 	if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
822 		PROC_UNLOCK(p);
823 		VOP_UNLOCK(imgp->vp, 0);
824 		pe.pm_credentialschanged = credential_changing;
825 		pe.pm_entryaddr = imgp->entry_addr;
826 
827 		PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
828 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
829 	} else
830 		PROC_UNLOCK(p);
831 #else  /* !HWPMC_HOOKS */
832 	PROC_UNLOCK(p);
833 #endif
834 
835 	/* Set values passed into the program in registers. */
836 	if (p->p_sysent->sv_setregs)
837 		(*p->p_sysent->sv_setregs)(td, imgp,
838 		    (u_long)(uintptr_t)stack_base);
839 	else
840 		exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base);
841 
842 	vfs_mark_atime(imgp->vp, td->td_ucred);
843 
844 	SDT_PROBE(proc, kernel, , exec__success, args->fname, 0, 0, 0, 0);
845 
846 	VOP_UNLOCK(imgp->vp, 0);
847 done1:
848 	/*
849 	 * Free any resources malloc'd earlier that we didn't use.
850 	 */
851 	if (euip != NULL)
852 		uifree(euip);
853 	if (newcred != NULL)
854 		crfree(oldcred);
855 
856 	/*
857 	 * Handle deferred decrement of ref counts.
858 	 */
859 	if (oldtextvp != NULL)
860 		vrele(oldtextvp);
861 #ifdef KTRACE
862 	if (tracevp != NULL)
863 		vrele(tracevp);
864 	if (tracecred != NULL)
865 		crfree(tracecred);
866 #endif
867 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
868 	pargs_drop(oldargs);
869 	pargs_drop(newargs);
870 	if (oldsigacts != NULL)
871 		sigacts_free(oldsigacts);
872 
873 exec_fail_dealloc:
874 
875 	/*
876 	 * free various allocated resources
877 	 */
878 	if (imgp->firstpage != NULL)
879 		exec_unmap_first_page(imgp);
880 
881 	if (imgp->vp != NULL) {
882 		if (args->fname)
883 			NDFREE(&nd, NDF_ONLY_PNBUF);
884 		if (imgp->opened)
885 			VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
886 		if (error != 0)
887 			vput(imgp->vp);
888 		else
889 			VOP_UNLOCK(imgp->vp, 0);
890 	}
891 
892 	if (imgp->object != NULL)
893 		vm_object_deallocate(imgp->object);
894 
895 	free(imgp->freepath, M_TEMP);
896 
897 	if (error == 0) {
898 		PROC_LOCK(p);
899 		td->td_dbgflags |= TDB_EXEC;
900 		PROC_UNLOCK(p);
901 
902 		/*
903 		 * Stop the process here if its stop event mask has
904 		 * the S_EXEC bit set.
905 		 */
906 		STOPEVENT(p, S_EXEC, 0);
907 		goto done2;
908 	}
909 
910 exec_fail:
911 	/* we're done here, clear P_INEXEC */
912 	PROC_LOCK(p);
913 	p->p_flag &= ~P_INEXEC;
914 	PROC_UNLOCK(p);
915 
916 	SDT_PROBE(proc, kernel, , exec__failure, error, 0, 0, 0, 0);
917 
918 done2:
919 #ifdef MAC
920 	mac_execve_exit(imgp);
921 	mac_execve_interpreter_exit(interpvplabel);
922 #endif
923 	exec_free_args(args);
924 
925 	if (error && imgp->vmspace_destroyed) {
926 		/* sorry, no more process anymore. exit gracefully */
927 		exit1(td, 0, SIGABRT);
928 		/* NOT REACHED */
929 	}
930 
931 #ifdef KTRACE
932 	if (error == 0)
933 		ktrprocctor(p);
934 #endif
935 
936 	return (error);
937 }
938 
939 int
940 exec_map_first_page(imgp)
941 	struct image_params *imgp;
942 {
943 	int rv, i;
944 	int initial_pagein;
945 	vm_page_t ma[VM_INITIAL_PAGEIN];
946 	vm_object_t object;
947 
948 	if (imgp->firstpage != NULL)
949 		exec_unmap_first_page(imgp);
950 
951 	object = imgp->vp->v_object;
952 	if (object == NULL)
953 		return (EACCES);
954 	VM_OBJECT_WLOCK(object);
955 #if VM_NRESERVLEVEL > 0
956 	vm_object_color(object, 0);
957 #endif
958 	ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL);
959 	if (ma[0]->valid != VM_PAGE_BITS_ALL) {
960 		initial_pagein = VM_INITIAL_PAGEIN;
961 		if (initial_pagein > object->size)
962 			initial_pagein = object->size;
963 		for (i = 1; i < initial_pagein; i++) {
964 			if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) {
965 				if (ma[i]->valid)
966 					break;
967 				if (vm_page_tryxbusy(ma[i]))
968 					break;
969 			} else {
970 				ma[i] = vm_page_alloc(object, i,
971 				    VM_ALLOC_NORMAL | VM_ALLOC_IFNOTCACHED);
972 				if (ma[i] == NULL)
973 					break;
974 			}
975 		}
976 		initial_pagein = i;
977 		rv = vm_pager_get_pages(object, ma, initial_pagein, 0);
978 		if (rv != VM_PAGER_OK) {
979 			vm_page_lock(ma[0]);
980 			vm_page_free(ma[0]);
981 			vm_page_unlock(ma[0]);
982 			VM_OBJECT_WUNLOCK(object);
983 			return (EIO);
984 		}
985 	}
986 	vm_page_xunbusy(ma[0]);
987 	vm_page_lock(ma[0]);
988 	vm_page_hold(ma[0]);
989 	vm_page_activate(ma[0]);
990 	vm_page_unlock(ma[0]);
991 	VM_OBJECT_WUNLOCK(object);
992 
993 	imgp->firstpage = sf_buf_alloc(ma[0], 0);
994 	imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
995 
996 	return (0);
997 }
998 
999 void
1000 exec_unmap_first_page(imgp)
1001 	struct image_params *imgp;
1002 {
1003 	vm_page_t m;
1004 
1005 	if (imgp->firstpage != NULL) {
1006 		m = sf_buf_page(imgp->firstpage);
1007 		sf_buf_free(imgp->firstpage);
1008 		imgp->firstpage = NULL;
1009 		vm_page_lock(m);
1010 		vm_page_unhold(m);
1011 		vm_page_unlock(m);
1012 	}
1013 }
1014 
1015 /*
1016  * Destroy old address space, and allocate a new stack
1017  *	The new stack is only SGROWSIZ large because it is grown
1018  *	automatically in trap.c.
1019  */
1020 int
1021 exec_new_vmspace(imgp, sv)
1022 	struct image_params *imgp;
1023 	struct sysentvec *sv;
1024 {
1025 	int error;
1026 	struct proc *p = imgp->proc;
1027 	struct vmspace *vmspace = p->p_vmspace;
1028 	vm_object_t obj;
1029 	struct rlimit rlim_stack;
1030 	vm_offset_t sv_minuser, stack_addr;
1031 	vm_map_t map;
1032 	u_long ssiz;
1033 
1034 	imgp->vmspace_destroyed = 1;
1035 	imgp->sysent = sv;
1036 
1037 	/* May be called with Giant held */
1038 	EVENTHANDLER_INVOKE(process_exec, p, imgp);
1039 
1040 	/*
1041 	 * Blow away entire process VM, if address space not shared,
1042 	 * otherwise, create a new VM space so that other threads are
1043 	 * not disrupted
1044 	 */
1045 	map = &vmspace->vm_map;
1046 	if (map_at_zero)
1047 		sv_minuser = sv->sv_minuser;
1048 	else
1049 		sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1050 	if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser &&
1051 	    vm_map_max(map) == sv->sv_maxuser) {
1052 		shmexit(vmspace);
1053 		pmap_remove_pages(vmspace_pmap(vmspace));
1054 		vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1055 	} else {
1056 		error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1057 		if (error)
1058 			return (error);
1059 		vmspace = p->p_vmspace;
1060 		map = &vmspace->vm_map;
1061 	}
1062 
1063 	/* Map a shared page */
1064 	obj = sv->sv_shared_page_obj;
1065 	if (obj != NULL) {
1066 		vm_object_reference(obj);
1067 		error = vm_map_fixed(map, obj, 0,
1068 		    sv->sv_shared_page_base, sv->sv_shared_page_len,
1069 		    VM_PROT_READ | VM_PROT_EXECUTE,
1070 		    VM_PROT_READ | VM_PROT_EXECUTE,
1071 		    MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1072 		if (error) {
1073 			vm_object_deallocate(obj);
1074 			return (error);
1075 		}
1076 	}
1077 
1078 	/* Allocate a new stack */
1079 	if (imgp->stack_sz != 0) {
1080 		ssiz = trunc_page(imgp->stack_sz);
1081 		PROC_LOCK(p);
1082 		lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack);
1083 		PROC_UNLOCK(p);
1084 		if (ssiz > rlim_stack.rlim_max)
1085 			ssiz = rlim_stack.rlim_max;
1086 		if (ssiz > rlim_stack.rlim_cur) {
1087 			rlim_stack.rlim_cur = ssiz;
1088 			kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack);
1089 		}
1090 	} else if (sv->sv_maxssiz != NULL) {
1091 		ssiz = *sv->sv_maxssiz;
1092 	} else {
1093 		ssiz = maxssiz;
1094 	}
1095 	stack_addr = sv->sv_usrstack - ssiz;
1096 	error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1097 	    obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1098 		sv->sv_stackprot,
1099 	    VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1100 	if (error)
1101 		return (error);
1102 
1103 	/*
1104 	 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they
1105 	 * are still used to enforce the stack rlimit on the process stack.
1106 	 */
1107 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1108 	vmspace->vm_maxsaddr = (char *)stack_addr;
1109 
1110 	return (0);
1111 }
1112 
1113 /*
1114  * Copy out argument and environment strings from the old process address
1115  * space into the temporary string buffer.
1116  */
1117 int
1118 exec_copyin_args(struct image_args *args, char *fname,
1119     enum uio_seg segflg, char **argv, char **envv)
1120 {
1121 	u_long argp, envp;
1122 	int error;
1123 	size_t length;
1124 
1125 	bzero(args, sizeof(*args));
1126 	if (argv == NULL)
1127 		return (EFAULT);
1128 
1129 	/*
1130 	 * Allocate demand-paged memory for the file name, argument, and
1131 	 * environment strings.
1132 	 */
1133 	error = exec_alloc_args(args);
1134 	if (error != 0)
1135 		return (error);
1136 
1137 	/*
1138 	 * Copy the file name.
1139 	 */
1140 	if (fname != NULL) {
1141 		args->fname = args->buf;
1142 		error = (segflg == UIO_SYSSPACE) ?
1143 		    copystr(fname, args->fname, PATH_MAX, &length) :
1144 		    copyinstr(fname, args->fname, PATH_MAX, &length);
1145 		if (error != 0)
1146 			goto err_exit;
1147 	} else
1148 		length = 0;
1149 
1150 	args->begin_argv = args->buf + length;
1151 	args->endp = args->begin_argv;
1152 	args->stringspace = ARG_MAX;
1153 
1154 	/*
1155 	 * extract arguments first
1156 	 */
1157 	for (;;) {
1158 		error = fueword(argv++, &argp);
1159 		if (error == -1) {
1160 			error = EFAULT;
1161 			goto err_exit;
1162 		}
1163 		if (argp == 0)
1164 			break;
1165 		error = copyinstr((void *)(uintptr_t)argp, args->endp,
1166 		    args->stringspace, &length);
1167 		if (error != 0) {
1168 			if (error == ENAMETOOLONG)
1169 				error = E2BIG;
1170 			goto err_exit;
1171 		}
1172 		args->stringspace -= length;
1173 		args->endp += length;
1174 		args->argc++;
1175 	}
1176 
1177 	args->begin_envv = args->endp;
1178 
1179 	/*
1180 	 * extract environment strings
1181 	 */
1182 	if (envv) {
1183 		for (;;) {
1184 			error = fueword(envv++, &envp);
1185 			if (error == -1) {
1186 				error = EFAULT;
1187 				goto err_exit;
1188 			}
1189 			if (envp == 0)
1190 				break;
1191 			error = copyinstr((void *)(uintptr_t)envp,
1192 			    args->endp, args->stringspace, &length);
1193 			if (error != 0) {
1194 				if (error == ENAMETOOLONG)
1195 					error = E2BIG;
1196 				goto err_exit;
1197 			}
1198 			args->stringspace -= length;
1199 			args->endp += length;
1200 			args->envc++;
1201 		}
1202 	}
1203 
1204 	return (0);
1205 
1206 err_exit:
1207 	exec_free_args(args);
1208 	return (error);
1209 }
1210 
1211 int
1212 exec_copyin_data_fds(struct thread *td, struct image_args *args,
1213     const void *data, size_t datalen, const int *fds, size_t fdslen)
1214 {
1215 	struct filedesc *ofdp;
1216 	const char *p;
1217 	int *kfds;
1218 	int error;
1219 
1220 	memset(args, '\0', sizeof(*args));
1221 	ofdp = td->td_proc->p_fd;
1222 	if (datalen >= ARG_MAX || fdslen > ofdp->fd_lastfile + 1)
1223 		return (E2BIG);
1224 	error = exec_alloc_args(args);
1225 	if (error != 0)
1226 		return (error);
1227 
1228 	args->begin_argv = args->buf;
1229 	args->stringspace = ARG_MAX;
1230 
1231 	if (datalen > 0) {
1232 		/*
1233 		 * Argument buffer has been provided. Copy it into the
1234 		 * kernel as a single string and add a terminating null
1235 		 * byte.
1236 		 */
1237 		error = copyin(data, args->begin_argv, datalen);
1238 		if (error != 0)
1239 			goto err_exit;
1240 		args->begin_argv[datalen] = '\0';
1241 		args->endp = args->begin_argv + datalen + 1;
1242 		args->stringspace -= datalen + 1;
1243 
1244 		/*
1245 		 * Traditional argument counting. Count the number of
1246 		 * null bytes.
1247 		 */
1248 		for (p = args->begin_argv; p < args->endp; ++p)
1249 			if (*p == '\0')
1250 				++args->argc;
1251 	} else {
1252 		/* No argument buffer provided. */
1253 		args->endp = args->begin_argv;
1254 	}
1255 	/* There are no environment variables. */
1256 	args->begin_envv = args->endp;
1257 
1258 	/* Create new file descriptor table. */
1259 	kfds = malloc(fdslen * sizeof(int), M_TEMP, M_WAITOK);
1260 	error = copyin(fds, kfds, fdslen * sizeof(int));
1261 	if (error != 0) {
1262 		free(kfds, M_TEMP);
1263 		goto err_exit;
1264 	}
1265 	error = fdcopy_remapped(ofdp, kfds, fdslen, &args->fdp);
1266 	free(kfds, M_TEMP);
1267 	if (error != 0)
1268 		goto err_exit;
1269 
1270 	return (0);
1271 err_exit:
1272 	exec_free_args(args);
1273 	return (error);
1274 }
1275 
1276 /*
1277  * Allocate temporary demand-paged, zero-filled memory for the file name,
1278  * argument, and environment strings.  Returns zero if the allocation succeeds
1279  * and ENOMEM otherwise.
1280  */
1281 int
1282 exec_alloc_args(struct image_args *args)
1283 {
1284 
1285 	args->buf = (char *)kmap_alloc_wait(exec_map, PATH_MAX + ARG_MAX);
1286 	return (args->buf != NULL ? 0 : ENOMEM);
1287 }
1288 
1289 void
1290 exec_free_args(struct image_args *args)
1291 {
1292 
1293 	if (args->buf != NULL) {
1294 		kmap_free_wakeup(exec_map, (vm_offset_t)args->buf,
1295 		    PATH_MAX + ARG_MAX);
1296 		args->buf = NULL;
1297 	}
1298 	if (args->fname_buf != NULL) {
1299 		free(args->fname_buf, M_TEMP);
1300 		args->fname_buf = NULL;
1301 	}
1302 	if (args->fdp != NULL)
1303 		fdescfree_remapped(args->fdp);
1304 }
1305 
1306 /*
1307  * Copy strings out to the new process address space, constructing new arg
1308  * and env vector tables. Return a pointer to the base so that it can be used
1309  * as the initial stack pointer.
1310  */
1311 register_t *
1312 exec_copyout_strings(imgp)
1313 	struct image_params *imgp;
1314 {
1315 	int argc, envc;
1316 	char **vectp;
1317 	char *stringp;
1318 	uintptr_t destp;
1319 	register_t *stack_base;
1320 	struct ps_strings *arginfo;
1321 	struct proc *p;
1322 	size_t execpath_len;
1323 	int szsigcode, szps;
1324 	char canary[sizeof(long) * 8];
1325 
1326 	szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1327 	/*
1328 	 * Calculate string base and vector table pointers.
1329 	 * Also deal with signal trampoline code for this exec type.
1330 	 */
1331 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
1332 		execpath_len = strlen(imgp->execpath) + 1;
1333 	else
1334 		execpath_len = 0;
1335 	p = imgp->proc;
1336 	szsigcode = 0;
1337 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1338 	if (p->p_sysent->sv_sigcode_base == 0) {
1339 		if (p->p_sysent->sv_szsigcode != NULL)
1340 			szsigcode = *(p->p_sysent->sv_szsigcode);
1341 	}
1342 	destp =	(uintptr_t)arginfo;
1343 
1344 	/*
1345 	 * install sigcode
1346 	 */
1347 	if (szsigcode != 0) {
1348 		destp -= szsigcode;
1349 		destp = rounddown2(destp, sizeof(void *));
1350 		copyout(p->p_sysent->sv_sigcode, (void *)destp, szsigcode);
1351 	}
1352 
1353 	/*
1354 	 * Copy the image path for the rtld.
1355 	 */
1356 	if (execpath_len != 0) {
1357 		destp -= execpath_len;
1358 		imgp->execpathp = destp;
1359 		copyout(imgp->execpath, (void *)destp, execpath_len);
1360 	}
1361 
1362 	/*
1363 	 * Prepare the canary for SSP.
1364 	 */
1365 	arc4rand(canary, sizeof(canary), 0);
1366 	destp -= sizeof(canary);
1367 	imgp->canary = destp;
1368 	copyout(canary, (void *)destp, sizeof(canary));
1369 	imgp->canarylen = sizeof(canary);
1370 
1371 	/*
1372 	 * Prepare the pagesizes array.
1373 	 */
1374 	destp -= szps;
1375 	destp = rounddown2(destp, sizeof(void *));
1376 	imgp->pagesizes = destp;
1377 	copyout(pagesizes, (void *)destp, szps);
1378 	imgp->pagesizeslen = szps;
1379 
1380 	destp -= ARG_MAX - imgp->args->stringspace;
1381 	destp = rounddown2(destp, sizeof(void *));
1382 
1383 	/*
1384 	 * If we have a valid auxargs ptr, prepare some room
1385 	 * on the stack.
1386 	 */
1387 	if (imgp->auxargs) {
1388 		/*
1389 		 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1390 		 * lower compatibility.
1391 		 */
1392 		imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1393 		    (AT_COUNT * 2);
1394 		/*
1395 		 * The '+ 2' is for the null pointers at the end of each of
1396 		 * the arg and env vector sets,and imgp->auxarg_size is room
1397 		 * for argument of Runtime loader.
1398 		 */
1399 		vectp = (char **)(destp - (imgp->args->argc +
1400 		    imgp->args->envc + 2 + imgp->auxarg_size)
1401 		    * sizeof(char *));
1402 	} else {
1403 		/*
1404 		 * The '+ 2' is for the null pointers at the end of each of
1405 		 * the arg and env vector sets
1406 		 */
1407 		vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc
1408 		    + 2) * sizeof(char *));
1409 	}
1410 
1411 	/*
1412 	 * vectp also becomes our initial stack base
1413 	 */
1414 	stack_base = (register_t *)vectp;
1415 
1416 	stringp = imgp->args->begin_argv;
1417 	argc = imgp->args->argc;
1418 	envc = imgp->args->envc;
1419 
1420 	/*
1421 	 * Copy out strings - arguments and environment.
1422 	 */
1423 	copyout(stringp, (void *)destp, ARG_MAX - imgp->args->stringspace);
1424 
1425 	/*
1426 	 * Fill in "ps_strings" struct for ps, w, etc.
1427 	 */
1428 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1429 	suword32(&arginfo->ps_nargvstr, argc);
1430 
1431 	/*
1432 	 * Fill in argument portion of vector table.
1433 	 */
1434 	for (; argc > 0; --argc) {
1435 		suword(vectp++, (long)(intptr_t)destp);
1436 		while (*stringp++ != 0)
1437 			destp++;
1438 		destp++;
1439 	}
1440 
1441 	/* a null vector table pointer separates the argp's from the envp's */
1442 	suword(vectp++, 0);
1443 
1444 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1445 	suword32(&arginfo->ps_nenvstr, envc);
1446 
1447 	/*
1448 	 * Fill in environment portion of vector table.
1449 	 */
1450 	for (; envc > 0; --envc) {
1451 		suword(vectp++, (long)(intptr_t)destp);
1452 		while (*stringp++ != 0)
1453 			destp++;
1454 		destp++;
1455 	}
1456 
1457 	/* end of vector table is a null pointer */
1458 	suword(vectp, 0);
1459 
1460 	return (stack_base);
1461 }
1462 
1463 /*
1464  * Check permissions of file to execute.
1465  *	Called with imgp->vp locked.
1466  *	Return 0 for success or error code on failure.
1467  */
1468 int
1469 exec_check_permissions(imgp)
1470 	struct image_params *imgp;
1471 {
1472 	struct vnode *vp = imgp->vp;
1473 	struct vattr *attr = imgp->attr;
1474 	struct thread *td;
1475 	int error, writecount;
1476 
1477 	td = curthread;
1478 
1479 	/* Get file attributes */
1480 	error = VOP_GETATTR(vp, attr, td->td_ucred);
1481 	if (error)
1482 		return (error);
1483 
1484 #ifdef MAC
1485 	error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1486 	if (error)
1487 		return (error);
1488 #endif
1489 
1490 	/*
1491 	 * 1) Check if file execution is disabled for the filesystem that
1492 	 *    this file resides on.
1493 	 * 2) Ensure that at least one execute bit is on. Otherwise, a
1494 	 *    privileged user will always succeed, and we don't want this
1495 	 *    to happen unless the file really is executable.
1496 	 * 3) Ensure that the file is a regular file.
1497 	 */
1498 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1499 	    (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1500 	    (attr->va_type != VREG))
1501 		return (EACCES);
1502 
1503 	/*
1504 	 * Zero length files can't be exec'd
1505 	 */
1506 	if (attr->va_size == 0)
1507 		return (ENOEXEC);
1508 
1509 	/*
1510 	 *  Check for execute permission to file based on current credentials.
1511 	 */
1512 	error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1513 	if (error)
1514 		return (error);
1515 
1516 	/*
1517 	 * Check number of open-for-writes on the file and deny execution
1518 	 * if there are any.
1519 	 */
1520 	error = VOP_GET_WRITECOUNT(vp, &writecount);
1521 	if (error != 0)
1522 		return (error);
1523 	if (writecount != 0)
1524 		return (ETXTBSY);
1525 
1526 	/*
1527 	 * Call filesystem specific open routine (which does nothing in the
1528 	 * general case).
1529 	 */
1530 	error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1531 	if (error == 0)
1532 		imgp->opened = 1;
1533 	return (error);
1534 }
1535 
1536 /*
1537  * Exec handler registration
1538  */
1539 int
1540 exec_register(execsw_arg)
1541 	const struct execsw *execsw_arg;
1542 {
1543 	const struct execsw **es, **xs, **newexecsw;
1544 	int count = 2;	/* New slot and trailing NULL */
1545 
1546 	if (execsw)
1547 		for (es = execsw; *es; es++)
1548 			count++;
1549 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1550 	if (newexecsw == NULL)
1551 		return (ENOMEM);
1552 	xs = newexecsw;
1553 	if (execsw)
1554 		for (es = execsw; *es; es++)
1555 			*xs++ = *es;
1556 	*xs++ = execsw_arg;
1557 	*xs = NULL;
1558 	if (execsw)
1559 		free(execsw, M_TEMP);
1560 	execsw = newexecsw;
1561 	return (0);
1562 }
1563 
1564 int
1565 exec_unregister(execsw_arg)
1566 	const struct execsw *execsw_arg;
1567 {
1568 	const struct execsw **es, **xs, **newexecsw;
1569 	int count = 1;
1570 
1571 	if (execsw == NULL)
1572 		panic("unregister with no handlers left?\n");
1573 
1574 	for (es = execsw; *es; es++) {
1575 		if (*es == execsw_arg)
1576 			break;
1577 	}
1578 	if (*es == NULL)
1579 		return (ENOENT);
1580 	for (es = execsw; *es; es++)
1581 		if (*es != execsw_arg)
1582 			count++;
1583 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1584 	if (newexecsw == NULL)
1585 		return (ENOMEM);
1586 	xs = newexecsw;
1587 	for (es = execsw; *es; es++)
1588 		if (*es != execsw_arg)
1589 			*xs++ = *es;
1590 	*xs = NULL;
1591 	if (execsw)
1592 		free(execsw, M_TEMP);
1593 	execsw = newexecsw;
1594 	return (0);
1595 }
1596