xref: /freebsd/sys/kern/kern_exec.c (revision 037479ff5ee18977b1c48e1e59770aad2f200a5a)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 1993, David Greenman
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include "opt_capsicum.h"
33 #include "opt_hwpmc_hooks.h"
34 #include "opt_ktrace.h"
35 #include "opt_vm.h"
36 
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/acct.h>
40 #include <sys/capsicum.h>
41 #include <sys/eventhandler.h>
42 #include <sys/exec.h>
43 #include <sys/fcntl.h>
44 #include <sys/filedesc.h>
45 #include <sys/imgact.h>
46 #include <sys/imgact_elf.h>
47 #include <sys/kernel.h>
48 #include <sys/lock.h>
49 #include <sys/malloc.h>
50 #include <sys/mman.h>
51 #include <sys/mount.h>
52 #include <sys/mutex.h>
53 #include <sys/namei.h>
54 #include <sys/pioctl.h>
55 #include <sys/priv.h>
56 #include <sys/proc.h>
57 #include <sys/ptrace.h>
58 #include <sys/resourcevar.h>
59 #include <sys/rwlock.h>
60 #include <sys/sched.h>
61 #include <sys/sdt.h>
62 #include <sys/sf_buf.h>
63 #include <sys/shm.h>
64 #include <sys/signalvar.h>
65 #include <sys/smp.h>
66 #include <sys/stat.h>
67 #include <sys/syscallsubr.h>
68 #include <sys/sysctl.h>
69 #include <sys/sysent.h>
70 #include <sys/sysproto.h>
71 #include <sys/vnode.h>
72 #include <sys/wait.h>
73 #ifdef KTRACE
74 #include <sys/ktrace.h>
75 #endif
76 
77 #include <vm/vm.h>
78 #include <vm/vm_param.h>
79 #include <vm/pmap.h>
80 #include <vm/vm_page.h>
81 #include <vm/vm_map.h>
82 #include <vm/vm_kern.h>
83 #include <vm/vm_extern.h>
84 #include <vm/vm_object.h>
85 #include <vm/vm_pager.h>
86 
87 #ifdef	HWPMC_HOOKS
88 #include <sys/pmckern.h>
89 #endif
90 
91 #include <machine/reg.h>
92 
93 #include <security/audit/audit.h>
94 #include <security/mac/mac_framework.h>
95 
96 #ifdef KDTRACE_HOOKS
97 #include <sys/dtrace_bsd.h>
98 dtrace_execexit_func_t	dtrace_fasttrap_exec;
99 #endif
100 
101 SDT_PROVIDER_DECLARE(proc);
102 SDT_PROBE_DEFINE1(proc, , , exec, "char *");
103 SDT_PROBE_DEFINE1(proc, , , exec__failure, "int");
104 SDT_PROBE_DEFINE1(proc, , , exec__success, "char *");
105 
106 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
107 
108 int coredump_pack_fileinfo = 1;
109 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_fileinfo, CTLFLAG_RWTUN,
110     &coredump_pack_fileinfo, 0,
111     "Enable file path packing in 'procstat -f' coredump notes");
112 
113 int coredump_pack_vmmapinfo = 1;
114 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_vmmapinfo, CTLFLAG_RWTUN,
115     &coredump_pack_vmmapinfo, 0,
116     "Enable file path packing in 'procstat -v' coredump notes");
117 
118 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
119 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
120 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
121 static int do_execve(struct thread *td, struct image_args *args,
122     struct mac *mac_p);
123 
124 /* XXX This should be vm_size_t. */
125 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD|
126     CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_ps_strings, "LU", "");
127 
128 /* XXX This should be vm_size_t. */
129 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD|
130     CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_usrstack, "LU", "");
131 
132 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD|CTLFLAG_MPSAFE,
133     NULL, 0, sysctl_kern_stackprot, "I", "");
134 
135 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
136 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
137     &ps_arg_cache_limit, 0, "");
138 
139 static int disallow_high_osrel;
140 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW,
141     &disallow_high_osrel, 0,
142     "Disallow execution of binaries built for higher version of the world");
143 
144 static int map_at_zero = 0;
145 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RWTUN, &map_at_zero, 0,
146     "Permit processes to map an object at virtual address 0.");
147 
148 EVENTHANDLER_LIST_DECLARE(process_exec);
149 
150 static int
151 sysctl_kern_ps_strings(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_psstrings;
161 		error = SYSCTL_OUT(req, &val, sizeof(val));
162 	} else
163 #endif
164 		error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
165 		   sizeof(p->p_sysent->sv_psstrings));
166 	return error;
167 }
168 
169 static int
170 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
171 {
172 	struct proc *p;
173 	int error;
174 
175 	p = curproc;
176 #ifdef SCTL_MASK32
177 	if (req->flags & SCTL_MASK32) {
178 		unsigned int val;
179 		val = (unsigned int)p->p_sysent->sv_usrstack;
180 		error = SYSCTL_OUT(req, &val, sizeof(val));
181 	} else
182 #endif
183 		error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
184 		    sizeof(p->p_sysent->sv_usrstack));
185 	return error;
186 }
187 
188 static int
189 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
190 {
191 	struct proc *p;
192 
193 	p = curproc;
194 	return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
195 	    sizeof(p->p_sysent->sv_stackprot)));
196 }
197 
198 /*
199  * Each of the items is a pointer to a `const struct execsw', hence the
200  * double pointer here.
201  */
202 static const struct execsw **execsw;
203 
204 #ifndef _SYS_SYSPROTO_H_
205 struct execve_args {
206 	char    *fname;
207 	char    **argv;
208 	char    **envv;
209 };
210 #endif
211 
212 int
213 sys_execve(struct thread *td, struct execve_args *uap)
214 {
215 	struct image_args args;
216 	struct vmspace *oldvmspace;
217 	int error;
218 
219 	error = pre_execve(td, &oldvmspace);
220 	if (error != 0)
221 		return (error);
222 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
223 	    uap->argv, uap->envv);
224 	if (error == 0)
225 		error = kern_execve(td, &args, NULL);
226 	post_execve(td, error, oldvmspace);
227 	return (error);
228 }
229 
230 #ifndef _SYS_SYSPROTO_H_
231 struct fexecve_args {
232 	int	fd;
233 	char	**argv;
234 	char	**envv;
235 }
236 #endif
237 int
238 sys_fexecve(struct thread *td, struct fexecve_args *uap)
239 {
240 	struct image_args args;
241 	struct vmspace *oldvmspace;
242 	int error;
243 
244 	error = pre_execve(td, &oldvmspace);
245 	if (error != 0)
246 		return (error);
247 	error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
248 	    uap->argv, uap->envv);
249 	if (error == 0) {
250 		args.fd = uap->fd;
251 		error = kern_execve(td, &args, NULL);
252 	}
253 	post_execve(td, error, oldvmspace);
254 	return (error);
255 }
256 
257 #ifndef _SYS_SYSPROTO_H_
258 struct __mac_execve_args {
259 	char	*fname;
260 	char	**argv;
261 	char	**envv;
262 	struct mac	*mac_p;
263 };
264 #endif
265 
266 int
267 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap)
268 {
269 #ifdef MAC
270 	struct image_args args;
271 	struct vmspace *oldvmspace;
272 	int error;
273 
274 	error = pre_execve(td, &oldvmspace);
275 	if (error != 0)
276 		return (error);
277 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
278 	    uap->argv, uap->envv);
279 	if (error == 0)
280 		error = kern_execve(td, &args, uap->mac_p);
281 	post_execve(td, error, oldvmspace);
282 	return (error);
283 #else
284 	return (ENOSYS);
285 #endif
286 }
287 
288 int
289 pre_execve(struct thread *td, struct vmspace **oldvmspace)
290 {
291 	struct proc *p;
292 	int error;
293 
294 	KASSERT(td == curthread, ("non-current thread %p", td));
295 	error = 0;
296 	p = td->td_proc;
297 	if ((p->p_flag & P_HADTHREADS) != 0) {
298 		PROC_LOCK(p);
299 		if (thread_single(p, SINGLE_BOUNDARY) != 0)
300 			error = ERESTART;
301 		PROC_UNLOCK(p);
302 	}
303 	KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0,
304 	    ("nested execve"));
305 	*oldvmspace = p->p_vmspace;
306 	return (error);
307 }
308 
309 void
310 post_execve(struct thread *td, int error, struct vmspace *oldvmspace)
311 {
312 	struct proc *p;
313 
314 	KASSERT(td == curthread, ("non-current thread %p", td));
315 	p = td->td_proc;
316 	if ((p->p_flag & P_HADTHREADS) != 0) {
317 		PROC_LOCK(p);
318 		/*
319 		 * If success, we upgrade to SINGLE_EXIT state to
320 		 * force other threads to suicide.
321 		 */
322 		if (error == EJUSTRETURN)
323 			thread_single(p, SINGLE_EXIT);
324 		else
325 			thread_single_end(p, SINGLE_BOUNDARY);
326 		PROC_UNLOCK(p);
327 	}
328 	if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
329 		KASSERT(p->p_vmspace != oldvmspace,
330 		    ("oldvmspace still used"));
331 		vmspace_free(oldvmspace);
332 		td->td_pflags &= ~TDP_EXECVMSPC;
333 	}
334 }
335 
336 /*
337  * XXX: kern_execve has the astonishing property of not always returning to
338  * the caller.  If sufficiently bad things happen during the call to
339  * do_execve(), it can end up calling exit1(); as a result, callers must
340  * avoid doing anything which they might need to undo (e.g., allocating
341  * memory).
342  */
343 int
344 kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p)
345 {
346 
347 	AUDIT_ARG_ARGV(args->begin_argv, args->argc,
348 	    exec_args_get_begin_envv(args) - args->begin_argv);
349 	AUDIT_ARG_ENVV(exec_args_get_begin_envv(args), args->envc,
350 	    args->endp - exec_args_get_begin_envv(args));
351 	return (do_execve(td, args, mac_p));
352 }
353 
354 /*
355  * In-kernel implementation of execve().  All arguments are assumed to be
356  * userspace pointers from the passed thread.
357  */
358 static int
359 do_execve(struct thread *td, struct image_args *args, struct mac *mac_p)
360 {
361 	struct proc *p = td->td_proc;
362 	struct nameidata nd;
363 	struct ucred *oldcred;
364 	struct uidinfo *euip = NULL;
365 	register_t *stack_base;
366 	int error, i;
367 	struct image_params image_params, *imgp;
368 	struct vattr attr;
369 	int (*img_first)(struct image_params *);
370 	struct pargs *oldargs = NULL, *newargs = NULL;
371 	struct sigacts *oldsigacts = NULL, *newsigacts = NULL;
372 #ifdef KTRACE
373 	struct vnode *tracevp = NULL;
374 	struct ucred *tracecred = NULL;
375 #endif
376 	struct vnode *oldtextvp = NULL, *newtextvp;
377 	int credential_changing;
378 	int textset;
379 #ifdef MAC
380 	struct label *interpvplabel = NULL;
381 	int will_transition;
382 #endif
383 #ifdef HWPMC_HOOKS
384 	struct pmckern_procexec pe;
385 #endif
386 	static const char fexecv_proc_title[] = "(fexecv)";
387 
388 	imgp = &image_params;
389 
390 	/*
391 	 * Lock the process and set the P_INEXEC flag to indicate that
392 	 * it should be left alone until we're done here.  This is
393 	 * necessary to avoid race conditions - e.g. in ptrace() -
394 	 * that might allow a local user to illicitly obtain elevated
395 	 * privileges.
396 	 */
397 	PROC_LOCK(p);
398 	KASSERT((p->p_flag & P_INEXEC) == 0,
399 	    ("%s(): process already has P_INEXEC flag", __func__));
400 	p->p_flag |= P_INEXEC;
401 	PROC_UNLOCK(p);
402 
403 	/*
404 	 * Initialize part of the common data
405 	 */
406 	bzero(imgp, sizeof(*imgp));
407 	imgp->proc = p;
408 	imgp->attr = &attr;
409 	imgp->args = args;
410 	oldcred = p->p_ucred;
411 
412 #ifdef MAC
413 	error = mac_execve_enter(imgp, mac_p);
414 	if (error)
415 		goto exec_fail;
416 #endif
417 
418 	/*
419 	 * Translate the file name. namei() returns a vnode pointer
420 	 *	in ni_vp among other things.
421 	 *
422 	 * XXXAUDIT: It would be desirable to also audit the name of the
423 	 * interpreter if this is an interpreted binary.
424 	 */
425 	if (args->fname != NULL) {
426 		NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME
427 		    | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
428 	}
429 
430 	SDT_PROBE1(proc, , , exec, args->fname);
431 
432 interpret:
433 	if (args->fname != NULL) {
434 #ifdef CAPABILITY_MODE
435 		/*
436 		 * While capability mode can't reach this point via direct
437 		 * path arguments to execve(), we also don't allow
438 		 * interpreters to be used in capability mode (for now).
439 		 * Catch indirect lookups and return a permissions error.
440 		 */
441 		if (IN_CAPABILITY_MODE(td)) {
442 			error = ECAPMODE;
443 			goto exec_fail;
444 		}
445 #endif
446 		error = namei(&nd);
447 		if (error)
448 			goto exec_fail;
449 
450 		newtextvp = nd.ni_vp;
451 		imgp->vp = newtextvp;
452 	} else {
453 		AUDIT_ARG_FD(args->fd);
454 		/*
455 		 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
456 		 */
457 		error = fgetvp_exec(td, args->fd, &cap_fexecve_rights, &newtextvp);
458 		if (error)
459 			goto exec_fail;
460 		vn_lock(newtextvp, LK_EXCLUSIVE | LK_RETRY);
461 		AUDIT_ARG_VNODE1(newtextvp);
462 		imgp->vp = newtextvp;
463 	}
464 
465 	/*
466 	 * Check file permissions (also 'opens' file)
467 	 */
468 	error = exec_check_permissions(imgp);
469 	if (error)
470 		goto exec_fail_dealloc;
471 
472 	imgp->object = imgp->vp->v_object;
473 	if (imgp->object != NULL)
474 		vm_object_reference(imgp->object);
475 
476 	/*
477 	 * Set VV_TEXT now so no one can write to the executable while we're
478 	 * activating it.
479 	 *
480 	 * Remember if this was set before and unset it in case this is not
481 	 * actually an executable image.
482 	 */
483 	textset = VOP_IS_TEXT(imgp->vp);
484 	VOP_SET_TEXT(imgp->vp);
485 
486 	error = exec_map_first_page(imgp);
487 	if (error)
488 		goto exec_fail_dealloc;
489 
490 	imgp->proc->p_osrel = 0;
491 	imgp->proc->p_fctl0 = 0;
492 
493 	/*
494 	 * Implement image setuid/setgid.
495 	 *
496 	 * Determine new credentials before attempting image activators
497 	 * so that it can be used by process_exec handlers to determine
498 	 * credential/setid changes.
499 	 *
500 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
501 	 * the process is being traced.
502 	 *
503 	 * We disable setuid/setgid/etc in capability mode on the basis
504 	 * that most setugid applications are not written with that
505 	 * environment in mind, and will therefore almost certainly operate
506 	 * incorrectly. In principle there's no reason that setugid
507 	 * applications might not be useful in capability mode, so we may want
508 	 * to reconsider this conservative design choice in the future.
509 	 *
510 	 * XXXMAC: For the time being, use NOSUID to also prohibit
511 	 * transitions on the file system.
512 	 */
513 	credential_changing = 0;
514 	credential_changing |= (attr.va_mode & S_ISUID) &&
515 	    oldcred->cr_uid != attr.va_uid;
516 	credential_changing |= (attr.va_mode & S_ISGID) &&
517 	    oldcred->cr_gid != attr.va_gid;
518 #ifdef MAC
519 	will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
520 	    interpvplabel, imgp);
521 	credential_changing |= will_transition;
522 #endif
523 
524 	/* Don't inherit PROC_PDEATHSIG_CTL value if setuid/setgid. */
525 	if (credential_changing)
526 		imgp->proc->p_pdeathsig = 0;
527 
528 	if (credential_changing &&
529 #ifdef CAPABILITY_MODE
530 	    ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
531 #endif
532 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
533 	    (p->p_flag & P_TRACED) == 0) {
534 		imgp->credential_setid = true;
535 		VOP_UNLOCK(imgp->vp, 0);
536 		imgp->newcred = crdup(oldcred);
537 		if (attr.va_mode & S_ISUID) {
538 			euip = uifind(attr.va_uid);
539 			change_euid(imgp->newcred, euip);
540 		}
541 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
542 		if (attr.va_mode & S_ISGID)
543 			change_egid(imgp->newcred, attr.va_gid);
544 		/*
545 		 * Implement correct POSIX saved-id behavior.
546 		 *
547 		 * XXXMAC: Note that the current logic will save the
548 		 * uid and gid if a MAC domain transition occurs, even
549 		 * though maybe it shouldn't.
550 		 */
551 		change_svuid(imgp->newcred, imgp->newcred->cr_uid);
552 		change_svgid(imgp->newcred, imgp->newcred->cr_gid);
553 	} else {
554 		/*
555 		 * Implement correct POSIX saved-id behavior.
556 		 *
557 		 * XXX: It's not clear that the existing behavior is
558 		 * POSIX-compliant.  A number of sources indicate that the
559 		 * saved uid/gid should only be updated if the new ruid is
560 		 * not equal to the old ruid, or the new euid is not equal
561 		 * to the old euid and the new euid is not equal to the old
562 		 * ruid.  The FreeBSD code always updates the saved uid/gid.
563 		 * Also, this code uses the new (replaced) euid and egid as
564 		 * the source, which may or may not be the right ones to use.
565 		 */
566 		if (oldcred->cr_svuid != oldcred->cr_uid ||
567 		    oldcred->cr_svgid != oldcred->cr_gid) {
568 			VOP_UNLOCK(imgp->vp, 0);
569 			imgp->newcred = crdup(oldcred);
570 			vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
571 			change_svuid(imgp->newcred, imgp->newcred->cr_uid);
572 			change_svgid(imgp->newcred, imgp->newcred->cr_gid);
573 		}
574 	}
575 	/* The new credentials are installed into the process later. */
576 
577 	/*
578 	 * Do the best to calculate the full path to the image file.
579 	 */
580 	if (args->fname != NULL && args->fname[0] == '/')
581 		imgp->execpath = args->fname;
582 	else {
583 		VOP_UNLOCK(imgp->vp, 0);
584 		if (vn_fullpath(td, imgp->vp, &imgp->execpath,
585 		    &imgp->freepath) != 0)
586 			imgp->execpath = args->fname;
587 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
588 	}
589 
590 	/*
591 	 *	If the current process has a special image activator it
592 	 *	wants to try first, call it.   For example, emulating shell
593 	 *	scripts differently.
594 	 */
595 	error = -1;
596 	if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
597 		error = img_first(imgp);
598 
599 	/*
600 	 *	Loop through the list of image activators, calling each one.
601 	 *	An activator returns -1 if there is no match, 0 on success,
602 	 *	and an error otherwise.
603 	 */
604 	for (i = 0; error == -1 && execsw[i]; ++i) {
605 		if (execsw[i]->ex_imgact == NULL ||
606 		    execsw[i]->ex_imgact == img_first) {
607 			continue;
608 		}
609 		error = (*execsw[i]->ex_imgact)(imgp);
610 	}
611 
612 	if (error) {
613 		if (error == -1) {
614 			if (textset == 0)
615 				VOP_UNSET_TEXT(imgp->vp);
616 			error = ENOEXEC;
617 		}
618 		goto exec_fail_dealloc;
619 	}
620 
621 	/*
622 	 * Special interpreter operation, cleanup and loop up to try to
623 	 * activate the interpreter.
624 	 */
625 	if (imgp->interpreted) {
626 		exec_unmap_first_page(imgp);
627 		/*
628 		 * VV_TEXT needs to be unset for scripts.  There is a short
629 		 * period before we determine that something is a script where
630 		 * VV_TEXT will be set. The vnode lock is held over this
631 		 * entire period so nothing should illegitimately be blocked.
632 		 */
633 		VOP_UNSET_TEXT(imgp->vp);
634 		/* free name buffer and old vnode */
635 		if (args->fname != NULL)
636 			NDFREE(&nd, NDF_ONLY_PNBUF);
637 #ifdef MAC
638 		mac_execve_interpreter_enter(newtextvp, &interpvplabel);
639 #endif
640 		if (imgp->opened) {
641 			VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td);
642 			imgp->opened = 0;
643 		}
644 		vput(newtextvp);
645 		vm_object_deallocate(imgp->object);
646 		imgp->object = NULL;
647 		imgp->credential_setid = false;
648 		if (imgp->newcred != NULL) {
649 			crfree(imgp->newcred);
650 			imgp->newcred = NULL;
651 		}
652 		imgp->execpath = NULL;
653 		free(imgp->freepath, M_TEMP);
654 		imgp->freepath = NULL;
655 		/* set new name to that of the interpreter */
656 		NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
657 		    UIO_SYSSPACE, imgp->interpreter_name, td);
658 		args->fname = imgp->interpreter_name;
659 		goto interpret;
660 	}
661 
662 	/*
663 	 * NB: We unlock the vnode here because it is believed that none
664 	 * of the sv_copyout_strings/sv_fixup operations require the vnode.
665 	 */
666 	VOP_UNLOCK(imgp->vp, 0);
667 
668 	if (disallow_high_osrel &&
669 	    P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
670 		error = ENOEXEC;
671 		uprintf("Osrel %d for image %s too high\n", p->p_osrel,
672 		    imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
673 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
674 		goto exec_fail_dealloc;
675 	}
676 
677 	/* ABI enforces the use of Capsicum. Switch into capabilities mode. */
678 	if (SV_PROC_FLAG(p, SV_CAPSICUM))
679 		sys_cap_enter(td, NULL);
680 
681 	/*
682 	 * Copy out strings (args and env) and initialize stack base
683 	 */
684 	if (p->p_sysent->sv_copyout_strings)
685 		stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
686 	else
687 		stack_base = exec_copyout_strings(imgp);
688 
689 	/*
690 	 * If custom stack fixup routine present for this process
691 	 * let it do the stack setup.
692 	 * Else stuff argument count as first item on stack
693 	 */
694 	if (p->p_sysent->sv_fixup != NULL)
695 		error = (*p->p_sysent->sv_fixup)(&stack_base, imgp);
696 	else
697 		error = suword(--stack_base, imgp->args->argc) == 0 ?
698 		    0 : EFAULT;
699 	if (error != 0)
700 		goto exec_fail_dealloc;
701 
702 	if (args->fdp != NULL) {
703 		/* Install a brand new file descriptor table. */
704 		fdinstall_remapped(td, args->fdp);
705 		args->fdp = NULL;
706 	} else {
707 		/*
708 		 * Keep on using the existing file descriptor table. For
709 		 * security and other reasons, the file descriptor table
710 		 * cannot be shared after an exec.
711 		 */
712 		fdunshare(td);
713 		/* close files on exec */
714 		fdcloseexec(td);
715 	}
716 
717 	/*
718 	 * Malloc things before we need locks.
719 	 */
720 	i = exec_args_get_begin_envv(imgp->args) - imgp->args->begin_argv;
721 	/* Cache arguments if they fit inside our allowance */
722 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
723 		newargs = pargs_alloc(i);
724 		bcopy(imgp->args->begin_argv, newargs->ar_args, i);
725 	}
726 
727 	/*
728 	 * For security and other reasons, signal handlers cannot
729 	 * be shared after an exec. The new process gets a copy of the old
730 	 * handlers. In execsigs(), the new process will have its signals
731 	 * reset.
732 	 */
733 	if (sigacts_shared(p->p_sigacts)) {
734 		oldsigacts = p->p_sigacts;
735 		newsigacts = sigacts_alloc();
736 		sigacts_copy(newsigacts, oldsigacts);
737 	}
738 
739 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
740 
741 	PROC_LOCK(p);
742 	if (oldsigacts)
743 		p->p_sigacts = newsigacts;
744 	/* Stop profiling */
745 	stopprofclock(p);
746 
747 	/* reset caught signals */
748 	execsigs(p);
749 
750 	/* name this process - nameiexec(p, ndp) */
751 	bzero(p->p_comm, sizeof(p->p_comm));
752 	if (args->fname)
753 		bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
754 		    min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
755 	else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0)
756 		bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
757 	bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
758 #ifdef KTR
759 	sched_clear_tdname(td);
760 #endif
761 
762 	/*
763 	 * mark as execed, wakeup the process that vforked (if any) and tell
764 	 * it that it now has its own resources back
765 	 */
766 	p->p_flag |= P_EXEC;
767 	if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0)
768 		p->p_flag2 &= ~P2_NOTRACE;
769 	if (p->p_flag & P_PPWAIT) {
770 		p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
771 		cv_broadcast(&p->p_pwait);
772 		/* STOPs are no longer ignored, arrange for AST */
773 		signotify(td);
774 	}
775 
776 	/*
777 	 * Implement image setuid/setgid installation.
778 	 */
779 	if (imgp->credential_setid) {
780 		/*
781 		 * Turn off syscall tracing for set-id programs, except for
782 		 * root.  Record any set-id flags first to make sure that
783 		 * we do not regain any tracing during a possible block.
784 		 */
785 		setsugid(p);
786 
787 #ifdef KTRACE
788 		if (p->p_tracecred != NULL &&
789 		    priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED, 0))
790 			ktrprocexec(p, &tracecred, &tracevp);
791 #endif
792 		/*
793 		 * Close any file descriptors 0..2 that reference procfs,
794 		 * then make sure file descriptors 0..2 are in use.
795 		 *
796 		 * Both fdsetugidsafety() and fdcheckstd() may call functions
797 		 * taking sleepable locks, so temporarily drop our locks.
798 		 */
799 		PROC_UNLOCK(p);
800 		VOP_UNLOCK(imgp->vp, 0);
801 		fdsetugidsafety(td);
802 		error = fdcheckstd(td);
803 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
804 		if (error != 0)
805 			goto exec_fail_dealloc;
806 		PROC_LOCK(p);
807 #ifdef MAC
808 		if (will_transition) {
809 			mac_vnode_execve_transition(oldcred, imgp->newcred,
810 			    imgp->vp, interpvplabel, imgp);
811 		}
812 #endif
813 	} else {
814 		if (oldcred->cr_uid == oldcred->cr_ruid &&
815 		    oldcred->cr_gid == oldcred->cr_rgid)
816 			p->p_flag &= ~P_SUGID;
817 	}
818 	/*
819 	 * Set the new credentials.
820 	 */
821 	if (imgp->newcred != NULL) {
822 		proc_set_cred(p, imgp->newcred);
823 		crfree(oldcred);
824 		oldcred = NULL;
825 	}
826 
827 	/*
828 	 * Store the vp for use in procfs.  This vnode was referenced by namei
829 	 * or fgetvp_exec.
830 	 */
831 	oldtextvp = p->p_textvp;
832 	p->p_textvp = newtextvp;
833 
834 #ifdef KDTRACE_HOOKS
835 	/*
836 	 * Tell the DTrace fasttrap provider about the exec if it
837 	 * has declared an interest.
838 	 */
839 	if (dtrace_fasttrap_exec)
840 		dtrace_fasttrap_exec(p);
841 #endif
842 
843 	/*
844 	 * Notify others that we exec'd, and clear the P_INEXEC flag
845 	 * as we're now a bona fide freshly-execed process.
846 	 */
847 	KNOTE_LOCKED(p->p_klist, NOTE_EXEC);
848 	p->p_flag &= ~P_INEXEC;
849 
850 	/* clear "fork but no exec" flag, as we _are_ execing */
851 	p->p_acflag &= ~AFORK;
852 
853 	/*
854 	 * Free any previous argument cache and replace it with
855 	 * the new argument cache, if any.
856 	 */
857 	oldargs = p->p_args;
858 	p->p_args = newargs;
859 	newargs = NULL;
860 
861 	PROC_UNLOCK(p);
862 
863 #ifdef	HWPMC_HOOKS
864 	/*
865 	 * Check if system-wide sampling is in effect or if the
866 	 * current process is using PMCs.  If so, do exec() time
867 	 * processing.  This processing needs to happen AFTER the
868 	 * P_INEXEC flag is cleared.
869 	 */
870 	if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
871 		VOP_UNLOCK(imgp->vp, 0);
872 		pe.pm_credentialschanged = credential_changing;
873 		pe.pm_entryaddr = imgp->entry_addr;
874 
875 		PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
876 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
877 	}
878 #endif
879 
880 	/* Set values passed into the program in registers. */
881 	if (p->p_sysent->sv_setregs)
882 		(*p->p_sysent->sv_setregs)(td, imgp,
883 		    (u_long)(uintptr_t)stack_base);
884 	else
885 		exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base);
886 
887 	vfs_mark_atime(imgp->vp, td->td_ucred);
888 
889 	SDT_PROBE1(proc, , , exec__success, args->fname);
890 
891 exec_fail_dealloc:
892 	if (imgp->firstpage != NULL)
893 		exec_unmap_first_page(imgp);
894 
895 	if (imgp->vp != NULL) {
896 		if (args->fname)
897 			NDFREE(&nd, NDF_ONLY_PNBUF);
898 		if (imgp->opened)
899 			VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
900 		if (error != 0)
901 			vput(imgp->vp);
902 		else
903 			VOP_UNLOCK(imgp->vp, 0);
904 	}
905 
906 	if (imgp->object != NULL)
907 		vm_object_deallocate(imgp->object);
908 
909 	free(imgp->freepath, M_TEMP);
910 
911 	if (error == 0) {
912 		if (p->p_ptevents & PTRACE_EXEC) {
913 			PROC_LOCK(p);
914 			if (p->p_ptevents & PTRACE_EXEC)
915 				td->td_dbgflags |= TDB_EXEC;
916 			PROC_UNLOCK(p);
917 		}
918 
919 		/*
920 		 * Stop the process here if its stop event mask has
921 		 * the S_EXEC bit set.
922 		 */
923 		STOPEVENT(p, S_EXEC, 0);
924 	} else {
925 exec_fail:
926 		/* we're done here, clear P_INEXEC */
927 		PROC_LOCK(p);
928 		p->p_flag &= ~P_INEXEC;
929 		PROC_UNLOCK(p);
930 
931 		SDT_PROBE1(proc, , , exec__failure, error);
932 	}
933 
934 	if (imgp->newcred != NULL && oldcred != NULL)
935 		crfree(imgp->newcred);
936 
937 #ifdef MAC
938 	mac_execve_exit(imgp);
939 	mac_execve_interpreter_exit(interpvplabel);
940 #endif
941 	exec_free_args(args);
942 
943 	/*
944 	 * Handle deferred decrement of ref counts.
945 	 */
946 	if (oldtextvp != NULL)
947 		vrele(oldtextvp);
948 #ifdef KTRACE
949 	if (tracevp != NULL)
950 		vrele(tracevp);
951 	if (tracecred != NULL)
952 		crfree(tracecred);
953 #endif
954 	pargs_drop(oldargs);
955 	pargs_drop(newargs);
956 	if (oldsigacts != NULL)
957 		sigacts_free(oldsigacts);
958 	if (euip != NULL)
959 		uifree(euip);
960 
961 	if (error && imgp->vmspace_destroyed) {
962 		/* sorry, no more process anymore. exit gracefully */
963 		exit1(td, 0, SIGABRT);
964 		/* NOT REACHED */
965 	}
966 
967 #ifdef KTRACE
968 	if (error == 0)
969 		ktrprocctor(p);
970 #endif
971 
972 	/*
973 	 * We don't want cpu_set_syscall_retval() to overwrite any of
974 	 * the register values put in place by exec_setregs().
975 	 * Implementations of cpu_set_syscall_retval() will leave
976 	 * registers unmodified when returning EJUSTRETURN.
977 	 */
978 	return (error == 0 ? EJUSTRETURN : error);
979 }
980 
981 int
982 exec_map_first_page(struct image_params *imgp)
983 {
984 	int rv, i, after, initial_pagein;
985 	vm_page_t ma[VM_INITIAL_PAGEIN];
986 	vm_object_t object;
987 
988 	if (imgp->firstpage != NULL)
989 		exec_unmap_first_page(imgp);
990 
991 	object = imgp->vp->v_object;
992 	if (object == NULL)
993 		return (EACCES);
994 	VM_OBJECT_WLOCK(object);
995 #if VM_NRESERVLEVEL > 0
996 	vm_object_color(object, 0);
997 #endif
998 	ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY);
999 	if (ma[0]->valid != VM_PAGE_BITS_ALL) {
1000 		vm_page_xbusy(ma[0]);
1001 		if (!vm_pager_has_page(object, 0, NULL, &after)) {
1002 			vm_page_lock(ma[0]);
1003 			vm_page_free(ma[0]);
1004 			vm_page_unlock(ma[0]);
1005 			VM_OBJECT_WUNLOCK(object);
1006 			return (EIO);
1007 		}
1008 		initial_pagein = min(after, VM_INITIAL_PAGEIN);
1009 		KASSERT(initial_pagein <= object->size,
1010 		    ("%s: initial_pagein %d object->size %ju",
1011 		    __func__, initial_pagein, (uintmax_t )object->size));
1012 		for (i = 1; i < initial_pagein; i++) {
1013 			if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) {
1014 				if (ma[i]->valid)
1015 					break;
1016 				if (!vm_page_tryxbusy(ma[i]))
1017 					break;
1018 			} else {
1019 				ma[i] = vm_page_alloc(object, i,
1020 				    VM_ALLOC_NORMAL);
1021 				if (ma[i] == NULL)
1022 					break;
1023 			}
1024 		}
1025 		initial_pagein = i;
1026 		rv = vm_pager_get_pages(object, ma, initial_pagein, NULL, NULL);
1027 		if (rv != VM_PAGER_OK) {
1028 			for (i = 0; i < initial_pagein; i++) {
1029 				vm_page_lock(ma[i]);
1030 				vm_page_free(ma[i]);
1031 				vm_page_unlock(ma[i]);
1032 			}
1033 			VM_OBJECT_WUNLOCK(object);
1034 			return (EIO);
1035 		}
1036 		vm_page_xunbusy(ma[0]);
1037 		for (i = 1; i < initial_pagein; i++)
1038 			vm_page_readahead_finish(ma[i]);
1039 	}
1040 	vm_page_lock(ma[0]);
1041 	vm_page_hold(ma[0]);
1042 	vm_page_activate(ma[0]);
1043 	vm_page_unlock(ma[0]);
1044 	VM_OBJECT_WUNLOCK(object);
1045 
1046 	imgp->firstpage = sf_buf_alloc(ma[0], 0);
1047 	imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
1048 
1049 	return (0);
1050 }
1051 
1052 void
1053 exec_unmap_first_page(struct image_params *imgp)
1054 {
1055 	vm_page_t m;
1056 
1057 	if (imgp->firstpage != NULL) {
1058 		m = sf_buf_page(imgp->firstpage);
1059 		sf_buf_free(imgp->firstpage);
1060 		imgp->firstpage = NULL;
1061 		vm_page_lock(m);
1062 		vm_page_unhold(m);
1063 		vm_page_unlock(m);
1064 	}
1065 }
1066 
1067 /*
1068  * Destroy old address space, and allocate a new stack.
1069  *	The new stack is only sgrowsiz large because it is grown
1070  *	automatically on a page fault.
1071  */
1072 int
1073 exec_new_vmspace(struct image_params *imgp, struct sysentvec *sv)
1074 {
1075 	int error;
1076 	struct proc *p = imgp->proc;
1077 	struct vmspace *vmspace = p->p_vmspace;
1078 	vm_object_t obj;
1079 	struct rlimit rlim_stack;
1080 	vm_offset_t sv_minuser, stack_addr;
1081 	vm_map_t map;
1082 	u_long ssiz;
1083 
1084 	imgp->vmspace_destroyed = 1;
1085 	imgp->sysent = sv;
1086 
1087 	/* May be called with Giant held */
1088 	EVENTHANDLER_DIRECT_INVOKE(process_exec, p, imgp);
1089 
1090 	/*
1091 	 * Blow away entire process VM, if address space not shared,
1092 	 * otherwise, create a new VM space so that other threads are
1093 	 * not disrupted
1094 	 */
1095 	map = &vmspace->vm_map;
1096 	if (map_at_zero)
1097 		sv_minuser = sv->sv_minuser;
1098 	else
1099 		sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1100 	if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser &&
1101 	    vm_map_max(map) == sv->sv_maxuser) {
1102 		shmexit(vmspace);
1103 		pmap_remove_pages(vmspace_pmap(vmspace));
1104 		vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1105 		/* An exec terminates mlockall(MCL_FUTURE). */
1106 		vm_map_lock(map);
1107 		vm_map_modflags(map, 0, MAP_WIREFUTURE);
1108 		vm_map_unlock(map);
1109 	} else {
1110 		error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1111 		if (error)
1112 			return (error);
1113 		vmspace = p->p_vmspace;
1114 		map = &vmspace->vm_map;
1115 	}
1116 
1117 	/* Map a shared page */
1118 	obj = sv->sv_shared_page_obj;
1119 	if (obj != NULL) {
1120 		vm_object_reference(obj);
1121 		error = vm_map_fixed(map, obj, 0,
1122 		    sv->sv_shared_page_base, sv->sv_shared_page_len,
1123 		    VM_PROT_READ | VM_PROT_EXECUTE,
1124 		    VM_PROT_READ | VM_PROT_EXECUTE,
1125 		    MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1126 		if (error != KERN_SUCCESS) {
1127 			vm_object_deallocate(obj);
1128 			return (vm_mmap_to_errno(error));
1129 		}
1130 	}
1131 
1132 	/* Allocate a new stack */
1133 	if (imgp->stack_sz != 0) {
1134 		ssiz = trunc_page(imgp->stack_sz);
1135 		PROC_LOCK(p);
1136 		lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack);
1137 		PROC_UNLOCK(p);
1138 		if (ssiz > rlim_stack.rlim_max)
1139 			ssiz = rlim_stack.rlim_max;
1140 		if (ssiz > rlim_stack.rlim_cur) {
1141 			rlim_stack.rlim_cur = ssiz;
1142 			kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack);
1143 		}
1144 	} else if (sv->sv_maxssiz != NULL) {
1145 		ssiz = *sv->sv_maxssiz;
1146 	} else {
1147 		ssiz = maxssiz;
1148 	}
1149 	stack_addr = sv->sv_usrstack - ssiz;
1150 	error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1151 	    obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1152 	    sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1153 	if (error != KERN_SUCCESS)
1154 		return (vm_mmap_to_errno(error));
1155 
1156 	/*
1157 	 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they
1158 	 * are still used to enforce the stack rlimit on the process stack.
1159 	 */
1160 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1161 	vmspace->vm_maxsaddr = (char *)stack_addr;
1162 
1163 	return (0);
1164 }
1165 
1166 /*
1167  * Copy out argument and environment strings from the old process address
1168  * space into the temporary string buffer.
1169  */
1170 int
1171 exec_copyin_args(struct image_args *args, const char *fname,
1172     enum uio_seg segflg, char **argv, char **envv)
1173 {
1174 	u_long arg, env;
1175 	int error;
1176 
1177 	bzero(args, sizeof(*args));
1178 	if (argv == NULL)
1179 		return (EFAULT);
1180 
1181 	/*
1182 	 * Allocate demand-paged memory for the file name, argument, and
1183 	 * environment strings.
1184 	 */
1185 	error = exec_alloc_args(args);
1186 	if (error != 0)
1187 		return (error);
1188 
1189 	/*
1190 	 * Copy the file name.
1191 	 */
1192 	error = exec_args_add_fname(args, fname, segflg);
1193 	if (error != 0)
1194 		goto err_exit;
1195 
1196 	/*
1197 	 * extract arguments first
1198 	 */
1199 	for (;;) {
1200 		error = fueword(argv++, &arg);
1201 		if (error == -1) {
1202 			error = EFAULT;
1203 			goto err_exit;
1204 		}
1205 		if (arg == 0)
1206 			break;
1207 		error = exec_args_add_arg(args, (char *)(uintptr_t)arg,
1208 		    UIO_USERSPACE);
1209 		if (error != 0)
1210 			goto err_exit;
1211 	}
1212 
1213 	/*
1214 	 * extract environment strings
1215 	 */
1216 	if (envv) {
1217 		for (;;) {
1218 			error = fueword(envv++, &env);
1219 			if (error == -1) {
1220 				error = EFAULT;
1221 				goto err_exit;
1222 			}
1223 			if (env == 0)
1224 				break;
1225 			error = exec_args_add_env(args,
1226 			    (char *)(uintptr_t)env, UIO_USERSPACE);
1227 			if (error != 0)
1228 				goto err_exit;
1229 		}
1230 	}
1231 
1232 	return (0);
1233 
1234 err_exit:
1235 	exec_free_args(args);
1236 	return (error);
1237 }
1238 
1239 int
1240 exec_copyin_data_fds(struct thread *td, struct image_args *args,
1241     const void *data, size_t datalen, const int *fds, size_t fdslen)
1242 {
1243 	struct filedesc *ofdp;
1244 	const char *p;
1245 	int *kfds;
1246 	int error;
1247 
1248 	memset(args, '\0', sizeof(*args));
1249 	ofdp = td->td_proc->p_fd;
1250 	if (datalen >= ARG_MAX || fdslen > ofdp->fd_lastfile + 1)
1251 		return (E2BIG);
1252 	error = exec_alloc_args(args);
1253 	if (error != 0)
1254 		return (error);
1255 
1256 	args->begin_argv = args->buf;
1257 	args->stringspace = ARG_MAX;
1258 
1259 	if (datalen > 0) {
1260 		/*
1261 		 * Argument buffer has been provided. Copy it into the
1262 		 * kernel as a single string and add a terminating null
1263 		 * byte.
1264 		 */
1265 		error = copyin(data, args->begin_argv, datalen);
1266 		if (error != 0)
1267 			goto err_exit;
1268 		args->begin_argv[datalen] = '\0';
1269 		args->endp = args->begin_argv + datalen + 1;
1270 		args->stringspace -= datalen + 1;
1271 
1272 		/*
1273 		 * Traditional argument counting. Count the number of
1274 		 * null bytes.
1275 		 */
1276 		for (p = args->begin_argv; p < args->endp; ++p)
1277 			if (*p == '\0')
1278 				++args->argc;
1279 	} else {
1280 		/* No argument buffer provided. */
1281 		args->endp = args->begin_argv;
1282 	}
1283 
1284 	/* Create new file descriptor table. */
1285 	kfds = malloc(fdslen * sizeof(int), M_TEMP, M_WAITOK);
1286 	error = copyin(fds, kfds, fdslen * sizeof(int));
1287 	if (error != 0) {
1288 		free(kfds, M_TEMP);
1289 		goto err_exit;
1290 	}
1291 	error = fdcopy_remapped(ofdp, kfds, fdslen, &args->fdp);
1292 	free(kfds, M_TEMP);
1293 	if (error != 0)
1294 		goto err_exit;
1295 
1296 	return (0);
1297 err_exit:
1298 	exec_free_args(args);
1299 	return (error);
1300 }
1301 
1302 struct exec_args_kva {
1303 	vm_offset_t addr;
1304 	u_int gen;
1305 	SLIST_ENTRY(exec_args_kva) next;
1306 };
1307 
1308 DPCPU_DEFINE_STATIC(struct exec_args_kva *, exec_args_kva);
1309 
1310 static SLIST_HEAD(, exec_args_kva) exec_args_kva_freelist;
1311 static struct mtx exec_args_kva_mtx;
1312 static u_int exec_args_gen;
1313 
1314 static void
1315 exec_prealloc_args_kva(void *arg __unused)
1316 {
1317 	struct exec_args_kva *argkva;
1318 	u_int i;
1319 
1320 	SLIST_INIT(&exec_args_kva_freelist);
1321 	mtx_init(&exec_args_kva_mtx, "exec args kva", NULL, MTX_DEF);
1322 	for (i = 0; i < exec_map_entries; i++) {
1323 		argkva = malloc(sizeof(*argkva), M_PARGS, M_WAITOK);
1324 		argkva->addr = kmap_alloc_wait(exec_map, exec_map_entry_size);
1325 		argkva->gen = exec_args_gen;
1326 		SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1327 	}
1328 }
1329 SYSINIT(exec_args_kva, SI_SUB_EXEC, SI_ORDER_ANY, exec_prealloc_args_kva, NULL);
1330 
1331 static vm_offset_t
1332 exec_alloc_args_kva(void **cookie)
1333 {
1334 	struct exec_args_kva *argkva;
1335 
1336 	argkva = (void *)atomic_readandclear_ptr(
1337 	    (uintptr_t *)DPCPU_PTR(exec_args_kva));
1338 	if (argkva == NULL) {
1339 		mtx_lock(&exec_args_kva_mtx);
1340 		while ((argkva = SLIST_FIRST(&exec_args_kva_freelist)) == NULL)
1341 			(void)mtx_sleep(&exec_args_kva_freelist,
1342 			    &exec_args_kva_mtx, 0, "execkva", 0);
1343 		SLIST_REMOVE_HEAD(&exec_args_kva_freelist, next);
1344 		mtx_unlock(&exec_args_kva_mtx);
1345 	}
1346 	*(struct exec_args_kva **)cookie = argkva;
1347 	return (argkva->addr);
1348 }
1349 
1350 static void
1351 exec_release_args_kva(struct exec_args_kva *argkva, u_int gen)
1352 {
1353 	vm_offset_t base;
1354 
1355 	base = argkva->addr;
1356 	if (argkva->gen != gen) {
1357 		(void)vm_map_madvise(exec_map, base, base + exec_map_entry_size,
1358 		    MADV_FREE);
1359 		argkva->gen = gen;
1360 	}
1361 	if (!atomic_cmpset_ptr((uintptr_t *)DPCPU_PTR(exec_args_kva),
1362 	    (uintptr_t)NULL, (uintptr_t)argkva)) {
1363 		mtx_lock(&exec_args_kva_mtx);
1364 		SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1365 		wakeup_one(&exec_args_kva_freelist);
1366 		mtx_unlock(&exec_args_kva_mtx);
1367 	}
1368 }
1369 
1370 static void
1371 exec_free_args_kva(void *cookie)
1372 {
1373 
1374 	exec_release_args_kva(cookie, exec_args_gen);
1375 }
1376 
1377 static void
1378 exec_args_kva_lowmem(void *arg __unused)
1379 {
1380 	SLIST_HEAD(, exec_args_kva) head;
1381 	struct exec_args_kva *argkva;
1382 	u_int gen;
1383 	int i;
1384 
1385 	gen = atomic_fetchadd_int(&exec_args_gen, 1) + 1;
1386 
1387 	/*
1388 	 * Force an madvise of each KVA range. Any currently allocated ranges
1389 	 * will have MADV_FREE applied once they are freed.
1390 	 */
1391 	SLIST_INIT(&head);
1392 	mtx_lock(&exec_args_kva_mtx);
1393 	SLIST_SWAP(&head, &exec_args_kva_freelist, exec_args_kva);
1394 	mtx_unlock(&exec_args_kva_mtx);
1395 	while ((argkva = SLIST_FIRST(&head)) != NULL) {
1396 		SLIST_REMOVE_HEAD(&head, next);
1397 		exec_release_args_kva(argkva, gen);
1398 	}
1399 
1400 	CPU_FOREACH(i) {
1401 		argkva = (void *)atomic_readandclear_ptr(
1402 		    (uintptr_t *)DPCPU_ID_PTR(i, exec_args_kva));
1403 		if (argkva != NULL)
1404 			exec_release_args_kva(argkva, gen);
1405 	}
1406 }
1407 EVENTHANDLER_DEFINE(vm_lowmem, exec_args_kva_lowmem, NULL,
1408     EVENTHANDLER_PRI_ANY);
1409 
1410 /*
1411  * Allocate temporary demand-paged, zero-filled memory for the file name,
1412  * argument, and environment strings.
1413  */
1414 int
1415 exec_alloc_args(struct image_args *args)
1416 {
1417 
1418 	args->buf = (char *)exec_alloc_args_kva(&args->bufkva);
1419 	return (0);
1420 }
1421 
1422 void
1423 exec_free_args(struct image_args *args)
1424 {
1425 
1426 	if (args->buf != NULL) {
1427 		exec_free_args_kva(args->bufkva);
1428 		args->buf = NULL;
1429 	}
1430 	if (args->fname_buf != NULL) {
1431 		free(args->fname_buf, M_TEMP);
1432 		args->fname_buf = NULL;
1433 	}
1434 	if (args->fdp != NULL)
1435 		fdescfree_remapped(args->fdp);
1436 }
1437 
1438 /*
1439  * A set to functions to fill struct image args.
1440  *
1441  * NOTE: exec_args_add_fname() must be called (possibly with a NULL
1442  * fname) before the other functions.  All exec_args_add_arg() calls must
1443  * be made before any exec_args_add_env() calls.  exec_args_adjust_args()
1444  * may be called any time after exec_args_add_fname().
1445  *
1446  * exec_args_add_fname() - install path to be executed
1447  * exec_args_add_arg() - append an argument string
1448  * exec_args_add_env() - append an env string
1449  * exec_args_adjust_args() - adjust location of the argument list to
1450  *                           allow new arguments to be prepended
1451  */
1452 int
1453 exec_args_add_fname(struct image_args *args, const char *fname,
1454     enum uio_seg segflg)
1455 {
1456 	int error;
1457 	size_t length;
1458 
1459 	KASSERT(args->fname == NULL, ("fname already appended"));
1460 	KASSERT(args->endp == NULL, ("already appending to args"));
1461 
1462 	if (fname != NULL) {
1463 		args->fname = args->buf;
1464 		error = segflg == UIO_SYSSPACE ?
1465 		    copystr(fname, args->fname, PATH_MAX, &length) :
1466 		    copyinstr(fname, args->fname, PATH_MAX, &length);
1467 		if (error != 0)
1468 			return (error == ENAMETOOLONG ? E2BIG : error);
1469 	} else
1470 		length = 0;
1471 
1472 	/* Set up for _arg_*()/_env_*() */
1473 	args->endp = args->buf + length;
1474 	/* begin_argv must be set and kept updated */
1475 	args->begin_argv = args->endp;
1476 	KASSERT(exec_map_entry_size - length >= ARG_MAX,
1477 	    ("too little space remaining for arguments %zu < %zu",
1478 	    exec_map_entry_size - length, (size_t)ARG_MAX));
1479 	args->stringspace = ARG_MAX;
1480 
1481 	return (0);
1482 }
1483 
1484 static int
1485 exec_args_add_str(struct image_args *args, const char *str,
1486     enum uio_seg segflg, int *countp)
1487 {
1488 	int error;
1489 	size_t length;
1490 
1491 	KASSERT(args->endp != NULL, ("endp not initialized"));
1492 	KASSERT(args->begin_argv != NULL, ("begin_argp not initialized"));
1493 
1494 	error = (segflg == UIO_SYSSPACE) ?
1495 	    copystr(str, args->endp, args->stringspace, &length) :
1496 	    copyinstr(str, args->endp, args->stringspace, &length);
1497 	if (error != 0)
1498 		return (error == ENAMETOOLONG ? E2BIG : error);
1499 	args->stringspace -= length;
1500 	args->endp += length;
1501 	(*countp)++;
1502 
1503 	return (0);
1504 }
1505 
1506 int
1507 exec_args_add_arg(struct image_args *args, const char *argp,
1508     enum uio_seg segflg)
1509 {
1510 
1511 	KASSERT(args->envc == 0, ("appending args after env"));
1512 
1513 	return (exec_args_add_str(args, argp, segflg, &args->argc));
1514 }
1515 
1516 int
1517 exec_args_add_env(struct image_args *args, const char *envp,
1518     enum uio_seg segflg)
1519 {
1520 
1521 	if (args->envc == 0)
1522 		args->begin_envv = args->endp;
1523 
1524 	return (exec_args_add_str(args, envp, segflg, &args->envc));
1525 }
1526 
1527 int
1528 exec_args_adjust_args(struct image_args *args, size_t consume, ssize_t extend)
1529 {
1530 	ssize_t offset;
1531 
1532 	KASSERT(args->endp != NULL, ("endp not initialized"));
1533 	KASSERT(args->begin_argv != NULL, ("begin_argp not initialized"));
1534 
1535 	offset = extend - consume;
1536 	if (args->stringspace < offset)
1537 		return (E2BIG);
1538 	memmove(args->begin_argv + extend, args->begin_argv + consume,
1539 	    args->endp - args->begin_argv + consume);
1540 	if (args->envc > 0)
1541 		args->begin_envv += offset;
1542 	args->endp += offset;
1543 	args->stringspace -= offset;
1544 	return (0);
1545 }
1546 
1547 char *
1548 exec_args_get_begin_envv(struct image_args *args)
1549 {
1550 
1551 	KASSERT(args->endp != NULL, ("endp not initialized"));
1552 
1553 	if (args->envc > 0)
1554 		return (args->begin_envv);
1555 	return (args->endp);
1556 }
1557 
1558 /*
1559  * Copy strings out to the new process address space, constructing new arg
1560  * and env vector tables. Return a pointer to the base so that it can be used
1561  * as the initial stack pointer.
1562  */
1563 register_t *
1564 exec_copyout_strings(struct image_params *imgp)
1565 {
1566 	int argc, envc;
1567 	char **vectp;
1568 	char *stringp;
1569 	uintptr_t destp;
1570 	register_t *stack_base;
1571 	struct ps_strings *arginfo;
1572 	struct proc *p;
1573 	size_t execpath_len;
1574 	int szsigcode, szps;
1575 	char canary[sizeof(long) * 8];
1576 
1577 	szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1578 	/*
1579 	 * Calculate string base and vector table pointers.
1580 	 * Also deal with signal trampoline code for this exec type.
1581 	 */
1582 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
1583 		execpath_len = strlen(imgp->execpath) + 1;
1584 	else
1585 		execpath_len = 0;
1586 	p = imgp->proc;
1587 	szsigcode = 0;
1588 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1589 	if (p->p_sysent->sv_sigcode_base == 0) {
1590 		if (p->p_sysent->sv_szsigcode != NULL)
1591 			szsigcode = *(p->p_sysent->sv_szsigcode);
1592 	}
1593 	destp =	(uintptr_t)arginfo;
1594 
1595 	/*
1596 	 * install sigcode
1597 	 */
1598 	if (szsigcode != 0) {
1599 		destp -= szsigcode;
1600 		destp = rounddown2(destp, sizeof(void *));
1601 		copyout(p->p_sysent->sv_sigcode, (void *)destp, szsigcode);
1602 	}
1603 
1604 	/*
1605 	 * Copy the image path for the rtld.
1606 	 */
1607 	if (execpath_len != 0) {
1608 		destp -= execpath_len;
1609 		destp = rounddown2(destp, sizeof(void *));
1610 		imgp->execpathp = destp;
1611 		copyout(imgp->execpath, (void *)destp, execpath_len);
1612 	}
1613 
1614 	/*
1615 	 * Prepare the canary for SSP.
1616 	 */
1617 	arc4rand(canary, sizeof(canary), 0);
1618 	destp -= sizeof(canary);
1619 	imgp->canary = destp;
1620 	copyout(canary, (void *)destp, sizeof(canary));
1621 	imgp->canarylen = sizeof(canary);
1622 
1623 	/*
1624 	 * Prepare the pagesizes array.
1625 	 */
1626 	destp -= szps;
1627 	destp = rounddown2(destp, sizeof(void *));
1628 	imgp->pagesizes = destp;
1629 	copyout(pagesizes, (void *)destp, szps);
1630 	imgp->pagesizeslen = szps;
1631 
1632 	destp -= ARG_MAX - imgp->args->stringspace;
1633 	destp = rounddown2(destp, sizeof(void *));
1634 
1635 	vectp = (char **)destp;
1636 	if (imgp->auxargs) {
1637 		/*
1638 		 * Allocate room on the stack for the ELF auxargs
1639 		 * array.  It has up to AT_COUNT entries.
1640 		 */
1641 		vectp -= howmany(AT_COUNT * sizeof(Elf_Auxinfo),
1642 		    sizeof(*vectp));
1643 	}
1644 
1645 	/*
1646 	 * Allocate room for the argv[] and env vectors including the
1647 	 * terminating NULL pointers.
1648 	 */
1649 	vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
1650 
1651 	/*
1652 	 * vectp also becomes our initial stack base
1653 	 */
1654 	stack_base = (register_t *)vectp;
1655 
1656 	stringp = imgp->args->begin_argv;
1657 	argc = imgp->args->argc;
1658 	envc = imgp->args->envc;
1659 
1660 	/*
1661 	 * Copy out strings - arguments and environment.
1662 	 */
1663 	copyout(stringp, (void *)destp, ARG_MAX - imgp->args->stringspace);
1664 
1665 	/*
1666 	 * Fill in "ps_strings" struct for ps, w, etc.
1667 	 */
1668 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1669 	suword32(&arginfo->ps_nargvstr, argc);
1670 
1671 	/*
1672 	 * Fill in argument portion of vector table.
1673 	 */
1674 	for (; argc > 0; --argc) {
1675 		suword(vectp++, (long)(intptr_t)destp);
1676 		while (*stringp++ != 0)
1677 			destp++;
1678 		destp++;
1679 	}
1680 
1681 	/* a null vector table pointer separates the argp's from the envp's */
1682 	suword(vectp++, 0);
1683 
1684 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1685 	suword32(&arginfo->ps_nenvstr, envc);
1686 
1687 	/*
1688 	 * Fill in environment portion of vector table.
1689 	 */
1690 	for (; envc > 0; --envc) {
1691 		suword(vectp++, (long)(intptr_t)destp);
1692 		while (*stringp++ != 0)
1693 			destp++;
1694 		destp++;
1695 	}
1696 
1697 	/* end of vector table is a null pointer */
1698 	suword(vectp, 0);
1699 
1700 	return (stack_base);
1701 }
1702 
1703 /*
1704  * Check permissions of file to execute.
1705  *	Called with imgp->vp locked.
1706  *	Return 0 for success or error code on failure.
1707  */
1708 int
1709 exec_check_permissions(struct image_params *imgp)
1710 {
1711 	struct vnode *vp = imgp->vp;
1712 	struct vattr *attr = imgp->attr;
1713 	struct thread *td;
1714 	int error, writecount;
1715 
1716 	td = curthread;
1717 
1718 	/* Get file attributes */
1719 	error = VOP_GETATTR(vp, attr, td->td_ucred);
1720 	if (error)
1721 		return (error);
1722 
1723 #ifdef MAC
1724 	error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1725 	if (error)
1726 		return (error);
1727 #endif
1728 
1729 	/*
1730 	 * 1) Check if file execution is disabled for the filesystem that
1731 	 *    this file resides on.
1732 	 * 2) Ensure that at least one execute bit is on. Otherwise, a
1733 	 *    privileged user will always succeed, and we don't want this
1734 	 *    to happen unless the file really is executable.
1735 	 * 3) Ensure that the file is a regular file.
1736 	 */
1737 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1738 	    (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1739 	    (attr->va_type != VREG))
1740 		return (EACCES);
1741 
1742 	/*
1743 	 * Zero length files can't be exec'd
1744 	 */
1745 	if (attr->va_size == 0)
1746 		return (ENOEXEC);
1747 
1748 	/*
1749 	 *  Check for execute permission to file based on current credentials.
1750 	 */
1751 	error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1752 	if (error)
1753 		return (error);
1754 
1755 	/*
1756 	 * Check number of open-for-writes on the file and deny execution
1757 	 * if there are any.
1758 	 */
1759 	error = VOP_GET_WRITECOUNT(vp, &writecount);
1760 	if (error != 0)
1761 		return (error);
1762 	if (writecount != 0)
1763 		return (ETXTBSY);
1764 
1765 	/*
1766 	 * Call filesystem specific open routine (which does nothing in the
1767 	 * general case).
1768 	 */
1769 	error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1770 	if (error == 0)
1771 		imgp->opened = 1;
1772 	return (error);
1773 }
1774 
1775 /*
1776  * Exec handler registration
1777  */
1778 int
1779 exec_register(const struct execsw *execsw_arg)
1780 {
1781 	const struct execsw **es, **xs, **newexecsw;
1782 	u_int count = 2;	/* New slot and trailing NULL */
1783 
1784 	if (execsw)
1785 		for (es = execsw; *es; es++)
1786 			count++;
1787 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1788 	xs = newexecsw;
1789 	if (execsw)
1790 		for (es = execsw; *es; es++)
1791 			*xs++ = *es;
1792 	*xs++ = execsw_arg;
1793 	*xs = NULL;
1794 	if (execsw)
1795 		free(execsw, M_TEMP);
1796 	execsw = newexecsw;
1797 	return (0);
1798 }
1799 
1800 int
1801 exec_unregister(const struct execsw *execsw_arg)
1802 {
1803 	const struct execsw **es, **xs, **newexecsw;
1804 	int count = 1;
1805 
1806 	if (execsw == NULL)
1807 		panic("unregister with no handlers left?\n");
1808 
1809 	for (es = execsw; *es; es++) {
1810 		if (*es == execsw_arg)
1811 			break;
1812 	}
1813 	if (*es == NULL)
1814 		return (ENOENT);
1815 	for (es = execsw; *es; es++)
1816 		if (*es != execsw_arg)
1817 			count++;
1818 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1819 	xs = newexecsw;
1820 	for (es = execsw; *es; es++)
1821 		if (*es != execsw_arg)
1822 			*xs++ = *es;
1823 	*xs = NULL;
1824 	if (execsw)
1825 		free(execsw, M_TEMP);
1826 	execsw = newexecsw;
1827 	return (0);
1828 }
1829