xref: /freebsd/sys/kern/kern_exec.c (revision 73577bf01de5c4677dc54d97f93310286c254780)
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/priv.h>
55 #include <sys/proc.h>
56 #include <sys/ptrace.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/shm.h>
63 #include <sys/signalvar.h>
64 #include <sys/smp.h>
65 #include <sys/stat.h>
66 #include <sys/syscallsubr.h>
67 #include <sys/sysctl.h>
68 #include <sys/sysent.h>
69 #include <sys/sysproto.h>
70 #include <sys/vnode.h>
71 #include <sys/wait.h>
72 #ifdef KTRACE
73 #include <sys/ktrace.h>
74 #endif
75 
76 #include <vm/vm.h>
77 #include <vm/vm_param.h>
78 #include <vm/pmap.h>
79 #include <vm/vm_page.h>
80 #include <vm/vm_map.h>
81 #include <vm/vm_kern.h>
82 #include <vm/vm_extern.h>
83 #include <vm/vm_object.h>
84 #include <vm/vm_pager.h>
85 
86 #ifdef	HWPMC_HOOKS
87 #include <sys/pmckern.h>
88 #endif
89 
90 #include <machine/reg.h>
91 
92 #include <security/audit/audit.h>
93 #include <security/mac/mac_framework.h>
94 
95 #ifdef KDTRACE_HOOKS
96 #include <sys/dtrace_bsd.h>
97 dtrace_execexit_func_t	dtrace_fasttrap_exec;
98 #endif
99 
100 SDT_PROVIDER_DECLARE(proc);
101 SDT_PROBE_DEFINE1(proc, , , exec, "char *");
102 SDT_PROBE_DEFINE1(proc, , , exec__failure, "int");
103 SDT_PROBE_DEFINE1(proc, , , exec__success, "char *");
104 
105 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
106 
107 int coredump_pack_fileinfo = 1;
108 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_fileinfo, CTLFLAG_RWTUN,
109     &coredump_pack_fileinfo, 0,
110     "Enable file path packing in 'procstat -f' coredump notes");
111 
112 int coredump_pack_vmmapinfo = 1;
113 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_vmmapinfo, CTLFLAG_RWTUN,
114     &coredump_pack_vmmapinfo, 0,
115     "Enable file path packing in 'procstat -v' coredump notes");
116 
117 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
118 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
119 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
120 static int do_execve(struct thread *td, struct image_args *args,
121     struct mac *mac_p, struct vmspace *oldvmspace);
122 
123 /* XXX This should be vm_size_t. */
124 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD|
125     CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_ps_strings, "LU",
126     "Location of process' ps_strings structure");
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     "Top of process stack");
132 
133 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD|CTLFLAG_MPSAFE,
134     NULL, 0, sysctl_kern_stackprot, "I",
135     "Stack memory permissions");
136 
137 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
138 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
139     &ps_arg_cache_limit, 0,
140     "Process' command line characters cache limit");
141 
142 static int disallow_high_osrel;
143 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW,
144     &disallow_high_osrel, 0,
145     "Disallow execution of binaries built for higher version of the world");
146 
147 static int map_at_zero = 0;
148 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RWTUN, &map_at_zero, 0,
149     "Permit processes to map an object at virtual address 0.");
150 
151 static int
152 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
153 {
154 	struct proc *p;
155 	int error;
156 
157 	p = curproc;
158 #ifdef SCTL_MASK32
159 	if (req->flags & SCTL_MASK32) {
160 		unsigned int val;
161 		val = (unsigned int)p->p_sysent->sv_psstrings;
162 		error = SYSCTL_OUT(req, &val, sizeof(val));
163 	} else
164 #endif
165 		error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
166 		   sizeof(p->p_sysent->sv_psstrings));
167 	return error;
168 }
169 
170 static int
171 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
172 {
173 	struct proc *p;
174 	int error;
175 
176 	p = curproc;
177 #ifdef SCTL_MASK32
178 	if (req->flags & SCTL_MASK32) {
179 		unsigned int val;
180 		val = (unsigned int)p->p_sysent->sv_usrstack;
181 		error = SYSCTL_OUT(req, &val, sizeof(val));
182 	} else
183 #endif
184 		error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
185 		    sizeof(p->p_sysent->sv_usrstack));
186 	return error;
187 }
188 
189 static int
190 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
191 {
192 	struct proc *p;
193 
194 	p = curproc;
195 	return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
196 	    sizeof(p->p_sysent->sv_stackprot)));
197 }
198 
199 /*
200  * Each of the items is a pointer to a `const struct execsw', hence the
201  * double pointer here.
202  */
203 static const struct execsw **execsw;
204 
205 #ifndef _SYS_SYSPROTO_H_
206 struct execve_args {
207 	char    *fname;
208 	char    **argv;
209 	char    **envv;
210 };
211 #endif
212 
213 int
214 sys_execve(struct thread *td, struct execve_args *uap)
215 {
216 	struct image_args args;
217 	struct vmspace *oldvmspace;
218 	int error;
219 
220 	error = pre_execve(td, &oldvmspace);
221 	if (error != 0)
222 		return (error);
223 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
224 	    uap->argv, uap->envv);
225 	if (error == 0)
226 		error = kern_execve(td, &args, NULL, oldvmspace);
227 	post_execve(td, error, oldvmspace);
228 	AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
229 	return (error);
230 }
231 
232 #ifndef _SYS_SYSPROTO_H_
233 struct fexecve_args {
234 	int	fd;
235 	char	**argv;
236 	char	**envv;
237 };
238 #endif
239 int
240 sys_fexecve(struct thread *td, struct fexecve_args *uap)
241 {
242 	struct image_args args;
243 	struct vmspace *oldvmspace;
244 	int error;
245 
246 	error = pre_execve(td, &oldvmspace);
247 	if (error != 0)
248 		return (error);
249 	error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
250 	    uap->argv, uap->envv);
251 	if (error == 0) {
252 		args.fd = uap->fd;
253 		error = kern_execve(td, &args, NULL, oldvmspace);
254 	}
255 	post_execve(td, error, oldvmspace);
256 	AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
257 	return (error);
258 }
259 
260 #ifndef _SYS_SYSPROTO_H_
261 struct __mac_execve_args {
262 	char	*fname;
263 	char	**argv;
264 	char	**envv;
265 	struct mac	*mac_p;
266 };
267 #endif
268 
269 int
270 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap)
271 {
272 #ifdef MAC
273 	struct image_args args;
274 	struct vmspace *oldvmspace;
275 	int error;
276 
277 	error = pre_execve(td, &oldvmspace);
278 	if (error != 0)
279 		return (error);
280 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
281 	    uap->argv, uap->envv);
282 	if (error == 0)
283 		error = kern_execve(td, &args, uap->mac_p, oldvmspace);
284 	post_execve(td, error, oldvmspace);
285 	AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
286 	return (error);
287 #else
288 	return (ENOSYS);
289 #endif
290 }
291 
292 int
293 pre_execve(struct thread *td, struct vmspace **oldvmspace)
294 {
295 	struct proc *p;
296 	int error;
297 
298 	KASSERT(td == curthread, ("non-current thread %p", td));
299 	error = 0;
300 	p = td->td_proc;
301 	if ((p->p_flag & P_HADTHREADS) != 0) {
302 		PROC_LOCK(p);
303 		if (thread_single(p, SINGLE_BOUNDARY) != 0)
304 			error = ERESTART;
305 		PROC_UNLOCK(p);
306 	}
307 	KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0,
308 	    ("nested execve"));
309 	*oldvmspace = p->p_vmspace;
310 	return (error);
311 }
312 
313 void
314 post_execve(struct thread *td, int error, struct vmspace *oldvmspace)
315 {
316 	struct proc *p;
317 
318 	KASSERT(td == curthread, ("non-current thread %p", td));
319 	p = td->td_proc;
320 	if ((p->p_flag & P_HADTHREADS) != 0) {
321 		PROC_LOCK(p);
322 		/*
323 		 * If success, we upgrade to SINGLE_EXIT state to
324 		 * force other threads to suicide.
325 		 */
326 		if (error == EJUSTRETURN)
327 			thread_single(p, SINGLE_EXIT);
328 		else
329 			thread_single_end(p, SINGLE_BOUNDARY);
330 		PROC_UNLOCK(p);
331 	}
332 	exec_cleanup(td, oldvmspace);
333 }
334 
335 /*
336  * kern_execve() has the astonishing property of not always returning to
337  * the caller.  If sufficiently bad things happen during the call to
338  * do_execve(), it can end up calling exit1(); as a result, callers must
339  * avoid doing anything which they might need to undo (e.g., allocating
340  * memory).
341  */
342 int
343 kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p,
344     struct vmspace *oldvmspace)
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, oldvmspace));
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     struct vmspace *oldvmspace)
361 {
362 	struct proc *p = td->td_proc;
363 	struct nameidata nd;
364 	struct ucred *oldcred;
365 	struct uidinfo *euip = NULL;
366 	uintptr_t stack_base;
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 #ifdef MAC
379 	struct label *interpvplabel = NULL;
380 	int will_transition;
381 #endif
382 #ifdef HWPMC_HOOKS
383 	struct pmckern_procexec pe;
384 #endif
385 	int error, i, orig_osrel;
386 	uint32_t orig_fctl0;
387 	static const char fexecv_proc_title[] = "(fexecv)";
388 
389 	imgp = &image_params;
390 
391 	/*
392 	 * Lock the process and set the P_INEXEC flag to indicate that
393 	 * it should be left alone until we're done here.  This is
394 	 * necessary to avoid race conditions - e.g. in ptrace() -
395 	 * that might allow a local user to illicitly obtain elevated
396 	 * privileges.
397 	 */
398 	PROC_LOCK(p);
399 	KASSERT((p->p_flag & P_INEXEC) == 0,
400 	    ("%s(): process already has P_INEXEC flag", __func__));
401 	p->p_flag |= P_INEXEC;
402 	PROC_UNLOCK(p);
403 
404 	/*
405 	 * Initialize part of the common data
406 	 */
407 	bzero(imgp, sizeof(*imgp));
408 	imgp->proc = p;
409 	imgp->attr = &attr;
410 	imgp->args = args;
411 	oldcred = p->p_ucred;
412 	orig_osrel = p->p_osrel;
413 	orig_fctl0 = p->p_fctl0;
414 
415 #ifdef MAC
416 	error = mac_execve_enter(imgp, mac_p);
417 	if (error)
418 		goto exec_fail;
419 #endif
420 
421 	/*
422 	 * Translate the file name. namei() returns a vnode pointer
423 	 *	in ni_vp among other things.
424 	 *
425 	 * XXXAUDIT: It would be desirable to also audit the name of the
426 	 * interpreter if this is an interpreted binary.
427 	 */
428 	if (args->fname != NULL) {
429 		NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW |
430 		    SAVENAME | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
431 	}
432 
433 	SDT_PROBE1(proc, , , exec, args->fname);
434 
435 interpret:
436 	if (args->fname != NULL) {
437 #ifdef CAPABILITY_MODE
438 		/*
439 		 * While capability mode can't reach this point via direct
440 		 * path arguments to execve(), we also don't allow
441 		 * interpreters to be used in capability mode (for now).
442 		 * Catch indirect lookups and return a permissions error.
443 		 */
444 		if (IN_CAPABILITY_MODE(td)) {
445 			error = ECAPMODE;
446 			goto exec_fail;
447 		}
448 #endif
449 		error = namei(&nd);
450 		if (error)
451 			goto exec_fail;
452 
453 		newtextvp = nd.ni_vp;
454 		imgp->vp = newtextvp;
455 	} else {
456 		AUDIT_ARG_FD(args->fd);
457 		/*
458 		 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
459 		 */
460 		error = fgetvp_exec(td, args->fd, &cap_fexecve_rights, &newtextvp);
461 		if (error)
462 			goto exec_fail;
463 		vn_lock(newtextvp, LK_SHARED | LK_RETRY);
464 		AUDIT_ARG_VNODE1(newtextvp);
465 		imgp->vp = newtextvp;
466 	}
467 
468 	/*
469 	 * Check file permissions.  Also 'opens' file and sets its vnode to
470 	 * text mode.
471 	 */
472 	error = exec_check_permissions(imgp);
473 	if (error)
474 		goto exec_fail_dealloc;
475 
476 	imgp->object = imgp->vp->v_object;
477 	if (imgp->object != NULL)
478 		vm_object_reference(imgp->object);
479 
480 	error = exec_map_first_page(imgp);
481 	if (error)
482 		goto exec_fail_dealloc;
483 
484 	imgp->proc->p_osrel = 0;
485 	imgp->proc->p_fctl0 = 0;
486 
487 	/*
488 	 * Implement image setuid/setgid.
489 	 *
490 	 * Determine new credentials before attempting image activators
491 	 * so that it can be used by process_exec handlers to determine
492 	 * credential/setid changes.
493 	 *
494 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
495 	 * the process is being traced.
496 	 *
497 	 * We disable setuid/setgid/etc in capability mode on the basis
498 	 * that most setugid applications are not written with that
499 	 * environment in mind, and will therefore almost certainly operate
500 	 * incorrectly. In principle there's no reason that setugid
501 	 * applications might not be useful in capability mode, so we may want
502 	 * to reconsider this conservative design choice in the future.
503 	 *
504 	 * XXXMAC: For the time being, use NOSUID to also prohibit
505 	 * transitions on the file system.
506 	 */
507 	credential_changing = 0;
508 	credential_changing |= (attr.va_mode & S_ISUID) &&
509 	    oldcred->cr_uid != attr.va_uid;
510 	credential_changing |= (attr.va_mode & S_ISGID) &&
511 	    oldcred->cr_gid != attr.va_gid;
512 #ifdef MAC
513 	will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
514 	    interpvplabel, imgp);
515 	credential_changing |= will_transition;
516 #endif
517 
518 	/* Don't inherit PROC_PDEATHSIG_CTL value if setuid/setgid. */
519 	if (credential_changing)
520 		imgp->proc->p_pdeathsig = 0;
521 
522 	if (credential_changing &&
523 #ifdef CAPABILITY_MODE
524 	    ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
525 #endif
526 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
527 	    (p->p_flag & P_TRACED) == 0) {
528 		imgp->credential_setid = true;
529 		VOP_UNLOCK(imgp->vp);
530 		imgp->newcred = crdup(oldcred);
531 		if (attr.va_mode & S_ISUID) {
532 			euip = uifind(attr.va_uid);
533 			change_euid(imgp->newcred, euip);
534 		}
535 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
536 		if (attr.va_mode & S_ISGID)
537 			change_egid(imgp->newcred, attr.va_gid);
538 		/*
539 		 * Implement correct POSIX saved-id behavior.
540 		 *
541 		 * XXXMAC: Note that the current logic will save the
542 		 * uid and gid if a MAC domain transition occurs, even
543 		 * though maybe it shouldn't.
544 		 */
545 		change_svuid(imgp->newcred, imgp->newcred->cr_uid);
546 		change_svgid(imgp->newcred, imgp->newcred->cr_gid);
547 	} else {
548 		/*
549 		 * Implement correct POSIX saved-id behavior.
550 		 *
551 		 * XXX: It's not clear that the existing behavior is
552 		 * POSIX-compliant.  A number of sources indicate that the
553 		 * saved uid/gid should only be updated if the new ruid is
554 		 * not equal to the old ruid, or the new euid is not equal
555 		 * to the old euid and the new euid is not equal to the old
556 		 * ruid.  The FreeBSD code always updates the saved uid/gid.
557 		 * Also, this code uses the new (replaced) euid and egid as
558 		 * the source, which may or may not be the right ones to use.
559 		 */
560 		if (oldcred->cr_svuid != oldcred->cr_uid ||
561 		    oldcred->cr_svgid != oldcred->cr_gid) {
562 			VOP_UNLOCK(imgp->vp);
563 			imgp->newcred = crdup(oldcred);
564 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
565 			change_svuid(imgp->newcred, imgp->newcred->cr_uid);
566 			change_svgid(imgp->newcred, imgp->newcred->cr_gid);
567 		}
568 	}
569 	/* The new credentials are installed into the process later. */
570 
571 	/*
572 	 * Do the best to calculate the full path to the image file.
573 	 */
574 	if (args->fname != NULL && args->fname[0] == '/')
575 		imgp->execpath = args->fname;
576 	else {
577 		VOP_UNLOCK(imgp->vp);
578 		if (vn_fullpath(imgp->vp, &imgp->execpath, &imgp->freepath) != 0)
579 			imgp->execpath = args->fname;
580 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
581 	}
582 
583 	/*
584 	 *	If the current process has a special image activator it
585 	 *	wants to try first, call it.   For example, emulating shell
586 	 *	scripts differently.
587 	 */
588 	error = -1;
589 	if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
590 		error = img_first(imgp);
591 
592 	/*
593 	 *	Loop through the list of image activators, calling each one.
594 	 *	An activator returns -1 if there is no match, 0 on success,
595 	 *	and an error otherwise.
596 	 */
597 	for (i = 0; error == -1 && execsw[i]; ++i) {
598 		if (execsw[i]->ex_imgact == NULL ||
599 		    execsw[i]->ex_imgact == img_first) {
600 			continue;
601 		}
602 		error = (*execsw[i]->ex_imgact)(imgp);
603 	}
604 
605 	if (error) {
606 		if (error == -1)
607 			error = ENOEXEC;
608 		goto exec_fail_dealloc;
609 	}
610 
611 	/*
612 	 * Special interpreter operation, cleanup and loop up to try to
613 	 * activate the interpreter.
614 	 */
615 	if (imgp->interpreted) {
616 		exec_unmap_first_page(imgp);
617 		/*
618 		 * The text reference needs to be removed for scripts.
619 		 * There is a short period before we determine that
620 		 * something is a script where text reference is active.
621 		 * The vnode lock is held over this entire period
622 		 * so nothing should illegitimately be blocked.
623 		 */
624 		MPASS(imgp->textset);
625 		VOP_UNSET_TEXT_CHECKED(newtextvp);
626 		imgp->textset = false;
627 		/* free name buffer and old vnode */
628 		if (args->fname != NULL)
629 			NDFREE(&nd, NDF_ONLY_PNBUF);
630 #ifdef MAC
631 		mac_execve_interpreter_enter(newtextvp, &interpvplabel);
632 #endif
633 		if (imgp->opened) {
634 			VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td);
635 			imgp->opened = 0;
636 		}
637 		vput(newtextvp);
638 		vm_object_deallocate(imgp->object);
639 		imgp->object = NULL;
640 		imgp->credential_setid = false;
641 		if (imgp->newcred != NULL) {
642 			crfree(imgp->newcred);
643 			imgp->newcred = NULL;
644 		}
645 		imgp->execpath = NULL;
646 		free(imgp->freepath, M_TEMP);
647 		imgp->freepath = NULL;
648 		/* set new name to that of the interpreter */
649 		NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW |
650 		    SAVENAME, UIO_SYSSPACE, imgp->interpreter_name, td);
651 		args->fname = imgp->interpreter_name;
652 		goto interpret;
653 	}
654 
655 	/*
656 	 * NB: We unlock the vnode here because it is believed that none
657 	 * of the sv_copyout_strings/sv_fixup operations require the vnode.
658 	 */
659 	VOP_UNLOCK(imgp->vp);
660 
661 	if (disallow_high_osrel &&
662 	    P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
663 		error = ENOEXEC;
664 		uprintf("Osrel %d for image %s too high\n", p->p_osrel,
665 		    imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
666 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
667 		goto exec_fail_dealloc;
668 	}
669 
670 	/* ABI enforces the use of Capsicum. Switch into capabilities mode. */
671 	if (SV_PROC_FLAG(p, SV_CAPSICUM))
672 		sys_cap_enter(td, NULL);
673 
674 	/*
675 	 * Copy out strings (args and env) and initialize stack base.
676 	 */
677 	error = (*p->p_sysent->sv_copyout_strings)(imgp, &stack_base);
678 	if (error != 0) {
679 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
680 		goto exec_fail_dealloc;
681 	}
682 
683 	/*
684 	 * Stack setup.
685 	 */
686 	error = (*p->p_sysent->sv_fixup)(&stack_base, imgp);
687 	if (error != 0) {
688 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
689 		goto exec_fail_dealloc;
690 	}
691 
692 	if (args->fdp != NULL) {
693 		/* Install a brand new file descriptor table. */
694 		fdinstall_remapped(td, args->fdp);
695 		args->fdp = NULL;
696 	} else {
697 		/*
698 		 * Keep on using the existing file descriptor table. For
699 		 * security and other reasons, the file descriptor table
700 		 * cannot be shared after an exec.
701 		 */
702 		fdunshare(td);
703 		/* close files on exec */
704 		fdcloseexec(td);
705 	}
706 
707 	/*
708 	 * Malloc things before we need locks.
709 	 */
710 	i = exec_args_get_begin_envv(imgp->args) - imgp->args->begin_argv;
711 	/* Cache arguments if they fit inside our allowance */
712 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
713 		newargs = pargs_alloc(i);
714 		bcopy(imgp->args->begin_argv, newargs->ar_args, i);
715 	}
716 
717 	/*
718 	 * For security and other reasons, signal handlers cannot
719 	 * be shared after an exec. The new process gets a copy of the old
720 	 * handlers. In execsigs(), the new process will have its signals
721 	 * reset.
722 	 */
723 	if (sigacts_shared(p->p_sigacts)) {
724 		oldsigacts = p->p_sigacts;
725 		newsigacts = sigacts_alloc();
726 		sigacts_copy(newsigacts, oldsigacts);
727 	}
728 
729 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
730 
731 	PROC_LOCK(p);
732 	if (oldsigacts)
733 		p->p_sigacts = newsigacts;
734 	/* Stop profiling */
735 	stopprofclock(p);
736 
737 	/* reset caught signals */
738 	execsigs(p);
739 
740 	/* name this process - nameiexec(p, ndp) */
741 	bzero(p->p_comm, sizeof(p->p_comm));
742 	if (args->fname)
743 		bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
744 		    min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
745 	else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0)
746 		bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
747 	bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
748 #ifdef KTR
749 	sched_clear_tdname(td);
750 #endif
751 
752 	/*
753 	 * mark as execed, wakeup the process that vforked (if any) and tell
754 	 * it that it now has its own resources back
755 	 */
756 	p->p_flag |= P_EXEC;
757 	if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0)
758 		p->p_flag2 &= ~P2_NOTRACE;
759 	if ((p->p_flag2 & P2_STKGAP_DISABLE_EXEC) == 0)
760 		p->p_flag2 &= ~P2_STKGAP_DISABLE;
761 	if (p->p_flag & P_PPWAIT) {
762 		p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
763 		cv_broadcast(&p->p_pwait);
764 		/* STOPs are no longer ignored, arrange for AST */
765 		signotify(td);
766 	}
767 
768 	/*
769 	 * Implement image setuid/setgid installation.
770 	 */
771 	if (imgp->credential_setid) {
772 		/*
773 		 * Turn off syscall tracing for set-id programs, except for
774 		 * root.  Record any set-id flags first to make sure that
775 		 * we do not regain any tracing during a possible block.
776 		 */
777 		setsugid(p);
778 
779 #ifdef KTRACE
780 		if (p->p_tracecred != NULL &&
781 		    priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED))
782 			ktrprocexec(p, &tracecred, &tracevp);
783 #endif
784 		/*
785 		 * Close any file descriptors 0..2 that reference procfs,
786 		 * then make sure file descriptors 0..2 are in use.
787 		 *
788 		 * Both fdsetugidsafety() and fdcheckstd() may call functions
789 		 * taking sleepable locks, so temporarily drop our locks.
790 		 */
791 		PROC_UNLOCK(p);
792 		VOP_UNLOCK(imgp->vp);
793 		fdsetugidsafety(td);
794 		error = fdcheckstd(td);
795 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
796 		if (error != 0)
797 			goto exec_fail_dealloc;
798 		PROC_LOCK(p);
799 #ifdef MAC
800 		if (will_transition) {
801 			mac_vnode_execve_transition(oldcred, imgp->newcred,
802 			    imgp->vp, interpvplabel, imgp);
803 		}
804 #endif
805 	} else {
806 		if (oldcred->cr_uid == oldcred->cr_ruid &&
807 		    oldcred->cr_gid == oldcred->cr_rgid)
808 			p->p_flag &= ~P_SUGID;
809 	}
810 	/*
811 	 * Set the new credentials.
812 	 */
813 	if (imgp->newcred != NULL) {
814 		proc_set_cred(p, imgp->newcred);
815 		crfree(oldcred);
816 		oldcred = NULL;
817 	}
818 
819 	/*
820 	 * Store the vp for use in procfs.  This vnode was referenced by namei
821 	 * or fgetvp_exec.
822 	 */
823 	oldtextvp = p->p_textvp;
824 	p->p_textvp = newtextvp;
825 
826 #ifdef KDTRACE_HOOKS
827 	/*
828 	 * Tell the DTrace fasttrap provider about the exec if it
829 	 * has declared an interest.
830 	 */
831 	if (dtrace_fasttrap_exec)
832 		dtrace_fasttrap_exec(p);
833 #endif
834 
835 	/*
836 	 * Notify others that we exec'd, and clear the P_INEXEC flag
837 	 * as we're now a bona fide freshly-execed process.
838 	 */
839 	KNOTE_LOCKED(p->p_klist, NOTE_EXEC);
840 	p->p_flag &= ~P_INEXEC;
841 
842 	/* clear "fork but no exec" flag, as we _are_ execing */
843 	p->p_acflag &= ~AFORK;
844 
845 	/*
846 	 * Free any previous argument cache and replace it with
847 	 * the new argument cache, if any.
848 	 */
849 	oldargs = p->p_args;
850 	p->p_args = newargs;
851 	newargs = NULL;
852 
853 	PROC_UNLOCK(p);
854 
855 #ifdef	HWPMC_HOOKS
856 	/*
857 	 * Check if system-wide sampling is in effect or if the
858 	 * current process is using PMCs.  If so, do exec() time
859 	 * processing.  This processing needs to happen AFTER the
860 	 * P_INEXEC flag is cleared.
861 	 */
862 	if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
863 		VOP_UNLOCK(imgp->vp);
864 		pe.pm_credentialschanged = credential_changing;
865 		pe.pm_entryaddr = imgp->entry_addr;
866 
867 		PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
868 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
869 	}
870 #endif
871 
872 	/* Set values passed into the program in registers. */
873 	(*p->p_sysent->sv_setregs)(td, imgp, stack_base);
874 
875 	VOP_MMAPPED(imgp->vp);
876 
877 	SDT_PROBE1(proc, , , exec__success, args->fname);
878 
879 exec_fail_dealloc:
880 	if (error != 0) {
881 		p->p_osrel = orig_osrel;
882 		p->p_fctl0 = orig_fctl0;
883 	}
884 
885 	if (imgp->firstpage != NULL)
886 		exec_unmap_first_page(imgp);
887 
888 	if (imgp->vp != NULL) {
889 		if (args->fname)
890 			NDFREE(&nd, NDF_ONLY_PNBUF);
891 		if (imgp->opened)
892 			VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
893 		if (imgp->textset)
894 			VOP_UNSET_TEXT_CHECKED(imgp->vp);
895 		if (error != 0)
896 			vput(imgp->vp);
897 		else
898 			VOP_UNLOCK(imgp->vp);
899 	}
900 
901 	if (imgp->object != NULL)
902 		vm_object_deallocate(imgp->object);
903 
904 	free(imgp->freepath, M_TEMP);
905 
906 	if (error == 0) {
907 		if (p->p_ptevents & PTRACE_EXEC) {
908 			PROC_LOCK(p);
909 			if (p->p_ptevents & PTRACE_EXEC)
910 				td->td_dbgflags |= TDB_EXEC;
911 			PROC_UNLOCK(p);
912 		}
913 	} else {
914 exec_fail:
915 		/* we're done here, clear P_INEXEC */
916 		PROC_LOCK(p);
917 		p->p_flag &= ~P_INEXEC;
918 		PROC_UNLOCK(p);
919 
920 		SDT_PROBE1(proc, , , exec__failure, error);
921 	}
922 
923 	if (imgp->newcred != NULL && oldcred != NULL)
924 		crfree(imgp->newcred);
925 
926 #ifdef MAC
927 	mac_execve_exit(imgp);
928 	mac_execve_interpreter_exit(interpvplabel);
929 #endif
930 	exec_free_args(args);
931 
932 	/*
933 	 * Handle deferred decrement of ref counts.
934 	 */
935 	if (oldtextvp != NULL)
936 		vrele(oldtextvp);
937 #ifdef KTRACE
938 	if (tracevp != NULL)
939 		vrele(tracevp);
940 	if (tracecred != NULL)
941 		crfree(tracecred);
942 #endif
943 	pargs_drop(oldargs);
944 	pargs_drop(newargs);
945 	if (oldsigacts != NULL)
946 		sigacts_free(oldsigacts);
947 	if (euip != NULL)
948 		uifree(euip);
949 
950 	if (error && imgp->vmspace_destroyed) {
951 		/* sorry, no more process anymore. exit gracefully */
952 		exec_cleanup(td, oldvmspace);
953 		exit1(td, 0, SIGABRT);
954 		/* NOT REACHED */
955 	}
956 
957 #ifdef KTRACE
958 	if (error == 0)
959 		ktrprocctor(p);
960 #endif
961 
962 	/*
963 	 * We don't want cpu_set_syscall_retval() to overwrite any of
964 	 * the register values put in place by exec_setregs().
965 	 * Implementations of cpu_set_syscall_retval() will leave
966 	 * registers unmodified when returning EJUSTRETURN.
967 	 */
968 	return (error == 0 ? EJUSTRETURN : error);
969 }
970 
971 void
972 exec_cleanup(struct thread *td, struct vmspace *oldvmspace)
973 {
974 	if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
975 		KASSERT(td->td_proc->p_vmspace != oldvmspace,
976 		    ("oldvmspace still used"));
977 		vmspace_free(oldvmspace);
978 		td->td_pflags &= ~TDP_EXECVMSPC;
979 	}
980 }
981 
982 int
983 exec_map_first_page(struct image_params *imgp)
984 {
985 	vm_object_t object;
986 	vm_page_t m;
987 	int error;
988 
989 	if (imgp->firstpage != NULL)
990 		exec_unmap_first_page(imgp);
991 
992 	object = imgp->vp->v_object;
993 	if (object == NULL)
994 		return (EACCES);
995 #if VM_NRESERVLEVEL > 0
996 	if ((object->flags & OBJ_COLORED) == 0) {
997 		VM_OBJECT_WLOCK(object);
998 		vm_object_color(object, 0);
999 		VM_OBJECT_WUNLOCK(object);
1000 	}
1001 #endif
1002 	error = vm_page_grab_valid_unlocked(&m, object, 0,
1003 	    VM_ALLOC_COUNT(VM_INITIAL_PAGEIN) |
1004             VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED);
1005 
1006 	if (error != VM_PAGER_OK)
1007 		return (EIO);
1008 	imgp->firstpage = sf_buf_alloc(m, 0);
1009 	imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
1010 
1011 	return (0);
1012 }
1013 
1014 void
1015 exec_unmap_first_page(struct image_params *imgp)
1016 {
1017 	vm_page_t m;
1018 
1019 	if (imgp->firstpage != NULL) {
1020 		m = sf_buf_page(imgp->firstpage);
1021 		sf_buf_free(imgp->firstpage);
1022 		imgp->firstpage = NULL;
1023 		vm_page_unwire(m, PQ_ACTIVE);
1024 	}
1025 }
1026 
1027 /*
1028  * Destroy old address space, and allocate a new stack.
1029  *	The new stack is only sgrowsiz large because it is grown
1030  *	automatically on a page fault.
1031  */
1032 int
1033 exec_new_vmspace(struct image_params *imgp, struct sysentvec *sv)
1034 {
1035 	int error;
1036 	struct proc *p = imgp->proc;
1037 	struct vmspace *vmspace = p->p_vmspace;
1038 	struct thread *td = curthread;
1039 	vm_object_t obj;
1040 	struct rlimit rlim_stack;
1041 	vm_offset_t sv_minuser, stack_addr;
1042 	vm_map_t map;
1043 	u_long ssiz;
1044 
1045 	imgp->vmspace_destroyed = 1;
1046 	imgp->sysent = sv;
1047 
1048 	sigfastblock_clear(td);
1049 
1050 	/* May be called with Giant held */
1051 	EVENTHANDLER_DIRECT_INVOKE(process_exec, p, imgp);
1052 
1053 	/*
1054 	 * Blow away entire process VM, if address space not shared,
1055 	 * otherwise, create a new VM space so that other threads are
1056 	 * not disrupted
1057 	 */
1058 	map = &vmspace->vm_map;
1059 	if (map_at_zero)
1060 		sv_minuser = sv->sv_minuser;
1061 	else
1062 		sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1063 	if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser &&
1064 	    vm_map_max(map) == sv->sv_maxuser &&
1065 	    cpu_exec_vmspace_reuse(p, map)) {
1066 		shmexit(vmspace);
1067 		pmap_remove_pages(vmspace_pmap(vmspace));
1068 		vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1069 		/*
1070 		 * An exec terminates mlockall(MCL_FUTURE), ASLR state
1071 		 * must be re-evaluated.
1072 		 */
1073 		vm_map_lock(map);
1074 		vm_map_modflags(map, 0, MAP_WIREFUTURE | MAP_ASLR |
1075 		    MAP_ASLR_IGNSTART);
1076 		vm_map_unlock(map);
1077 	} else {
1078 		error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1079 		if (error)
1080 			return (error);
1081 		vmspace = p->p_vmspace;
1082 		map = &vmspace->vm_map;
1083 	}
1084 	map->flags |= imgp->map_flags;
1085 
1086 	/* Map a shared page */
1087 	obj = sv->sv_shared_page_obj;
1088 	if (obj != NULL) {
1089 		vm_object_reference(obj);
1090 		error = vm_map_fixed(map, obj, 0,
1091 		    sv->sv_shared_page_base, sv->sv_shared_page_len,
1092 		    VM_PROT_READ | VM_PROT_EXECUTE,
1093 		    VM_PROT_READ | VM_PROT_EXECUTE,
1094 		    MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1095 		if (error != KERN_SUCCESS) {
1096 			vm_object_deallocate(obj);
1097 			return (vm_mmap_to_errno(error));
1098 		}
1099 	}
1100 
1101 	/* Allocate a new stack */
1102 	if (imgp->stack_sz != 0) {
1103 		ssiz = trunc_page(imgp->stack_sz);
1104 		PROC_LOCK(p);
1105 		lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack);
1106 		PROC_UNLOCK(p);
1107 		if (ssiz > rlim_stack.rlim_max)
1108 			ssiz = rlim_stack.rlim_max;
1109 		if (ssiz > rlim_stack.rlim_cur) {
1110 			rlim_stack.rlim_cur = ssiz;
1111 			kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack);
1112 		}
1113 	} else if (sv->sv_maxssiz != NULL) {
1114 		ssiz = *sv->sv_maxssiz;
1115 	} else {
1116 		ssiz = maxssiz;
1117 	}
1118 	imgp->eff_stack_sz = lim_cur(curthread, RLIMIT_STACK);
1119 	if (ssiz < imgp->eff_stack_sz)
1120 		imgp->eff_stack_sz = ssiz;
1121 	stack_addr = sv->sv_usrstack - ssiz;
1122 	error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1123 	    obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1124 	    sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1125 	if (error != KERN_SUCCESS)
1126 		return (vm_mmap_to_errno(error));
1127 
1128 	/*
1129 	 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they
1130 	 * are still used to enforce the stack rlimit on the process stack.
1131 	 */
1132 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1133 	vmspace->vm_maxsaddr = (char *)stack_addr;
1134 
1135 	return (0);
1136 }
1137 
1138 /*
1139  * Copy out argument and environment strings from the old process address
1140  * space into the temporary string buffer.
1141  */
1142 int
1143 exec_copyin_args(struct image_args *args, const char *fname,
1144     enum uio_seg segflg, char **argv, char **envv)
1145 {
1146 	u_long arg, env;
1147 	int error;
1148 
1149 	bzero(args, sizeof(*args));
1150 	if (argv == NULL)
1151 		return (EFAULT);
1152 
1153 	/*
1154 	 * Allocate demand-paged memory for the file name, argument, and
1155 	 * environment strings.
1156 	 */
1157 	error = exec_alloc_args(args);
1158 	if (error != 0)
1159 		return (error);
1160 
1161 	/*
1162 	 * Copy the file name.
1163 	 */
1164 	error = exec_args_add_fname(args, fname, segflg);
1165 	if (error != 0)
1166 		goto err_exit;
1167 
1168 	/*
1169 	 * extract arguments first
1170 	 */
1171 	for (;;) {
1172 		error = fueword(argv++, &arg);
1173 		if (error == -1) {
1174 			error = EFAULT;
1175 			goto err_exit;
1176 		}
1177 		if (arg == 0)
1178 			break;
1179 		error = exec_args_add_arg(args, (char *)(uintptr_t)arg,
1180 		    UIO_USERSPACE);
1181 		if (error != 0)
1182 			goto err_exit;
1183 	}
1184 
1185 	/*
1186 	 * extract environment strings
1187 	 */
1188 	if (envv) {
1189 		for (;;) {
1190 			error = fueword(envv++, &env);
1191 			if (error == -1) {
1192 				error = EFAULT;
1193 				goto err_exit;
1194 			}
1195 			if (env == 0)
1196 				break;
1197 			error = exec_args_add_env(args,
1198 			    (char *)(uintptr_t)env, UIO_USERSPACE);
1199 			if (error != 0)
1200 				goto err_exit;
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_nfiles)
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 
1256 	/* Create new file descriptor table. */
1257 	kfds = malloc(fdslen * sizeof(int), M_TEMP, M_WAITOK);
1258 	error = copyin(fds, kfds, fdslen * sizeof(int));
1259 	if (error != 0) {
1260 		free(kfds, M_TEMP);
1261 		goto err_exit;
1262 	}
1263 	error = fdcopy_remapped(ofdp, kfds, fdslen, &args->fdp);
1264 	free(kfds, M_TEMP);
1265 	if (error != 0)
1266 		goto err_exit;
1267 
1268 	return (0);
1269 err_exit:
1270 	exec_free_args(args);
1271 	return (error);
1272 }
1273 
1274 struct exec_args_kva {
1275 	vm_offset_t addr;
1276 	u_int gen;
1277 	SLIST_ENTRY(exec_args_kva) next;
1278 };
1279 
1280 DPCPU_DEFINE_STATIC(struct exec_args_kva *, exec_args_kva);
1281 
1282 static SLIST_HEAD(, exec_args_kva) exec_args_kva_freelist;
1283 static struct mtx exec_args_kva_mtx;
1284 static u_int exec_args_gen;
1285 
1286 static void
1287 exec_prealloc_args_kva(void *arg __unused)
1288 {
1289 	struct exec_args_kva *argkva;
1290 	u_int i;
1291 
1292 	SLIST_INIT(&exec_args_kva_freelist);
1293 	mtx_init(&exec_args_kva_mtx, "exec args kva", NULL, MTX_DEF);
1294 	for (i = 0; i < exec_map_entries; i++) {
1295 		argkva = malloc(sizeof(*argkva), M_PARGS, M_WAITOK);
1296 		argkva->addr = kmap_alloc_wait(exec_map, exec_map_entry_size);
1297 		argkva->gen = exec_args_gen;
1298 		SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1299 	}
1300 }
1301 SYSINIT(exec_args_kva, SI_SUB_EXEC, SI_ORDER_ANY, exec_prealloc_args_kva, NULL);
1302 
1303 static vm_offset_t
1304 exec_alloc_args_kva(void **cookie)
1305 {
1306 	struct exec_args_kva *argkva;
1307 
1308 	argkva = (void *)atomic_readandclear_ptr(
1309 	    (uintptr_t *)DPCPU_PTR(exec_args_kva));
1310 	if (argkva == NULL) {
1311 		mtx_lock(&exec_args_kva_mtx);
1312 		while ((argkva = SLIST_FIRST(&exec_args_kva_freelist)) == NULL)
1313 			(void)mtx_sleep(&exec_args_kva_freelist,
1314 			    &exec_args_kva_mtx, 0, "execkva", 0);
1315 		SLIST_REMOVE_HEAD(&exec_args_kva_freelist, next);
1316 		mtx_unlock(&exec_args_kva_mtx);
1317 	}
1318 	*(struct exec_args_kva **)cookie = argkva;
1319 	return (argkva->addr);
1320 }
1321 
1322 static void
1323 exec_release_args_kva(struct exec_args_kva *argkva, u_int gen)
1324 {
1325 	vm_offset_t base;
1326 
1327 	base = argkva->addr;
1328 	if (argkva->gen != gen) {
1329 		(void)vm_map_madvise(exec_map, base, base + exec_map_entry_size,
1330 		    MADV_FREE);
1331 		argkva->gen = gen;
1332 	}
1333 	if (!atomic_cmpset_ptr((uintptr_t *)DPCPU_PTR(exec_args_kva),
1334 	    (uintptr_t)NULL, (uintptr_t)argkva)) {
1335 		mtx_lock(&exec_args_kva_mtx);
1336 		SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1337 		wakeup_one(&exec_args_kva_freelist);
1338 		mtx_unlock(&exec_args_kva_mtx);
1339 	}
1340 }
1341 
1342 static void
1343 exec_free_args_kva(void *cookie)
1344 {
1345 
1346 	exec_release_args_kva(cookie, exec_args_gen);
1347 }
1348 
1349 static void
1350 exec_args_kva_lowmem(void *arg __unused)
1351 {
1352 	SLIST_HEAD(, exec_args_kva) head;
1353 	struct exec_args_kva *argkva;
1354 	u_int gen;
1355 	int i;
1356 
1357 	gen = atomic_fetchadd_int(&exec_args_gen, 1) + 1;
1358 
1359 	/*
1360 	 * Force an madvise of each KVA range. Any currently allocated ranges
1361 	 * will have MADV_FREE applied once they are freed.
1362 	 */
1363 	SLIST_INIT(&head);
1364 	mtx_lock(&exec_args_kva_mtx);
1365 	SLIST_SWAP(&head, &exec_args_kva_freelist, exec_args_kva);
1366 	mtx_unlock(&exec_args_kva_mtx);
1367 	while ((argkva = SLIST_FIRST(&head)) != NULL) {
1368 		SLIST_REMOVE_HEAD(&head, next);
1369 		exec_release_args_kva(argkva, gen);
1370 	}
1371 
1372 	CPU_FOREACH(i) {
1373 		argkva = (void *)atomic_readandclear_ptr(
1374 		    (uintptr_t *)DPCPU_ID_PTR(i, exec_args_kva));
1375 		if (argkva != NULL)
1376 			exec_release_args_kva(argkva, gen);
1377 	}
1378 }
1379 EVENTHANDLER_DEFINE(vm_lowmem, exec_args_kva_lowmem, NULL,
1380     EVENTHANDLER_PRI_ANY);
1381 
1382 /*
1383  * Allocate temporary demand-paged, zero-filled memory for the file name,
1384  * argument, and environment strings.
1385  */
1386 int
1387 exec_alloc_args(struct image_args *args)
1388 {
1389 
1390 	args->buf = (char *)exec_alloc_args_kva(&args->bufkva);
1391 	return (0);
1392 }
1393 
1394 void
1395 exec_free_args(struct image_args *args)
1396 {
1397 
1398 	if (args->buf != NULL) {
1399 		exec_free_args_kva(args->bufkva);
1400 		args->buf = NULL;
1401 	}
1402 	if (args->fname_buf != NULL) {
1403 		free(args->fname_buf, M_TEMP);
1404 		args->fname_buf = NULL;
1405 	}
1406 	if (args->fdp != NULL)
1407 		fdescfree_remapped(args->fdp);
1408 }
1409 
1410 /*
1411  * A set to functions to fill struct image args.
1412  *
1413  * NOTE: exec_args_add_fname() must be called (possibly with a NULL
1414  * fname) before the other functions.  All exec_args_add_arg() calls must
1415  * be made before any exec_args_add_env() calls.  exec_args_adjust_args()
1416  * may be called any time after exec_args_add_fname().
1417  *
1418  * exec_args_add_fname() - install path to be executed
1419  * exec_args_add_arg() - append an argument string
1420  * exec_args_add_env() - append an env string
1421  * exec_args_adjust_args() - adjust location of the argument list to
1422  *                           allow new arguments to be prepended
1423  */
1424 int
1425 exec_args_add_fname(struct image_args *args, const char *fname,
1426     enum uio_seg segflg)
1427 {
1428 	int error;
1429 	size_t length;
1430 
1431 	KASSERT(args->fname == NULL, ("fname already appended"));
1432 	KASSERT(args->endp == NULL, ("already appending to args"));
1433 
1434 	if (fname != NULL) {
1435 		args->fname = args->buf;
1436 		error = segflg == UIO_SYSSPACE ?
1437 		    copystr(fname, args->fname, PATH_MAX, &length) :
1438 		    copyinstr(fname, args->fname, PATH_MAX, &length);
1439 		if (error != 0)
1440 			return (error == ENAMETOOLONG ? E2BIG : error);
1441 	} else
1442 		length = 0;
1443 
1444 	/* Set up for _arg_*()/_env_*() */
1445 	args->endp = args->buf + length;
1446 	/* begin_argv must be set and kept updated */
1447 	args->begin_argv = args->endp;
1448 	KASSERT(exec_map_entry_size - length >= ARG_MAX,
1449 	    ("too little space remaining for arguments %zu < %zu",
1450 	    exec_map_entry_size - length, (size_t)ARG_MAX));
1451 	args->stringspace = ARG_MAX;
1452 
1453 	return (0);
1454 }
1455 
1456 static int
1457 exec_args_add_str(struct image_args *args, const char *str,
1458     enum uio_seg segflg, int *countp)
1459 {
1460 	int error;
1461 	size_t length;
1462 
1463 	KASSERT(args->endp != NULL, ("endp not initialized"));
1464 	KASSERT(args->begin_argv != NULL, ("begin_argp not initialized"));
1465 
1466 	error = (segflg == UIO_SYSSPACE) ?
1467 	    copystr(str, args->endp, args->stringspace, &length) :
1468 	    copyinstr(str, args->endp, args->stringspace, &length);
1469 	if (error != 0)
1470 		return (error == ENAMETOOLONG ? E2BIG : error);
1471 	args->stringspace -= length;
1472 	args->endp += length;
1473 	(*countp)++;
1474 
1475 	return (0);
1476 }
1477 
1478 int
1479 exec_args_add_arg(struct image_args *args, const char *argp,
1480     enum uio_seg segflg)
1481 {
1482 
1483 	KASSERT(args->envc == 0, ("appending args after env"));
1484 
1485 	return (exec_args_add_str(args, argp, segflg, &args->argc));
1486 }
1487 
1488 int
1489 exec_args_add_env(struct image_args *args, const char *envp,
1490     enum uio_seg segflg)
1491 {
1492 
1493 	if (args->envc == 0)
1494 		args->begin_envv = args->endp;
1495 
1496 	return (exec_args_add_str(args, envp, segflg, &args->envc));
1497 }
1498 
1499 int
1500 exec_args_adjust_args(struct image_args *args, size_t consume, ssize_t extend)
1501 {
1502 	ssize_t offset;
1503 
1504 	KASSERT(args->endp != NULL, ("endp not initialized"));
1505 	KASSERT(args->begin_argv != NULL, ("begin_argp not initialized"));
1506 
1507 	offset = extend - consume;
1508 	if (args->stringspace < offset)
1509 		return (E2BIG);
1510 	memmove(args->begin_argv + extend, args->begin_argv + consume,
1511 	    args->endp - args->begin_argv + consume);
1512 	if (args->envc > 0)
1513 		args->begin_envv += offset;
1514 	args->endp += offset;
1515 	args->stringspace -= offset;
1516 	return (0);
1517 }
1518 
1519 char *
1520 exec_args_get_begin_envv(struct image_args *args)
1521 {
1522 
1523 	KASSERT(args->endp != NULL, ("endp not initialized"));
1524 
1525 	if (args->envc > 0)
1526 		return (args->begin_envv);
1527 	return (args->endp);
1528 }
1529 
1530 /*
1531  * Copy strings out to the new process address space, constructing new arg
1532  * and env vector tables. Return a pointer to the base so that it can be used
1533  * as the initial stack pointer.
1534  */
1535 int
1536 exec_copyout_strings(struct image_params *imgp, uintptr_t *stack_base)
1537 {
1538 	int argc, envc;
1539 	char **vectp;
1540 	char *stringp;
1541 	uintptr_t destp, ustringp;
1542 	struct ps_strings *arginfo;
1543 	struct proc *p;
1544 	size_t execpath_len;
1545 	int error, szsigcode, szps;
1546 	char canary[sizeof(long) * 8];
1547 
1548 	szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1549 	/*
1550 	 * Calculate string base and vector table pointers.
1551 	 * Also deal with signal trampoline code for this exec type.
1552 	 */
1553 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
1554 		execpath_len = strlen(imgp->execpath) + 1;
1555 	else
1556 		execpath_len = 0;
1557 	p = imgp->proc;
1558 	szsigcode = 0;
1559 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1560 	imgp->ps_strings = arginfo;
1561 	if (p->p_sysent->sv_sigcode_base == 0) {
1562 		if (p->p_sysent->sv_szsigcode != NULL)
1563 			szsigcode = *(p->p_sysent->sv_szsigcode);
1564 	}
1565 	destp =	(uintptr_t)arginfo;
1566 
1567 	/*
1568 	 * install sigcode
1569 	 */
1570 	if (szsigcode != 0) {
1571 		destp -= szsigcode;
1572 		destp = rounddown2(destp, sizeof(void *));
1573 		error = copyout(p->p_sysent->sv_sigcode, (void *)destp,
1574 		    szsigcode);
1575 		if (error != 0)
1576 			return (error);
1577 	}
1578 
1579 	/*
1580 	 * Copy the image path for the rtld.
1581 	 */
1582 	if (execpath_len != 0) {
1583 		destp -= execpath_len;
1584 		destp = rounddown2(destp, sizeof(void *));
1585 		imgp->execpathp = (void *)destp;
1586 		error = copyout(imgp->execpath, imgp->execpathp, execpath_len);
1587 		if (error != 0)
1588 			return (error);
1589 	}
1590 
1591 	/*
1592 	 * Prepare the canary for SSP.
1593 	 */
1594 	arc4rand(canary, sizeof(canary), 0);
1595 	destp -= sizeof(canary);
1596 	imgp->canary = (void *)destp;
1597 	error = copyout(canary, imgp->canary, sizeof(canary));
1598 	if (error != 0)
1599 		return (error);
1600 	imgp->canarylen = sizeof(canary);
1601 
1602 	/*
1603 	 * Prepare the pagesizes array.
1604 	 */
1605 	destp -= szps;
1606 	destp = rounddown2(destp, sizeof(void *));
1607 	imgp->pagesizes = (void *)destp;
1608 	error = copyout(pagesizes, imgp->pagesizes, szps);
1609 	if (error != 0)
1610 		return (error);
1611 	imgp->pagesizeslen = szps;
1612 
1613 	/*
1614 	 * Allocate room for the argument and environment strings.
1615 	 */
1616 	destp -= ARG_MAX - imgp->args->stringspace;
1617 	destp = rounddown2(destp, sizeof(void *));
1618 	ustringp = destp;
1619 
1620 	if (imgp->sysent->sv_stackgap != NULL)
1621 		imgp->sysent->sv_stackgap(imgp, &destp);
1622 
1623 	if (imgp->auxargs) {
1624 		/*
1625 		 * Allocate room on the stack for the ELF auxargs
1626 		 * array.  It has up to AT_COUNT entries.
1627 		 */
1628 		destp -= AT_COUNT * sizeof(Elf_Auxinfo);
1629 		destp = rounddown2(destp, sizeof(void *));
1630 	}
1631 
1632 	vectp = (char **)destp;
1633 
1634 	/*
1635 	 * Allocate room for the argv[] and env vectors including the
1636 	 * terminating NULL pointers.
1637 	 */
1638 	vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
1639 
1640 	/*
1641 	 * vectp also becomes our initial stack base
1642 	 */
1643 	*stack_base = (uintptr_t)vectp;
1644 
1645 	stringp = imgp->args->begin_argv;
1646 	argc = imgp->args->argc;
1647 	envc = imgp->args->envc;
1648 
1649 	/*
1650 	 * Copy out strings - arguments and environment.
1651 	 */
1652 	error = copyout(stringp, (void *)ustringp,
1653 	    ARG_MAX - imgp->args->stringspace);
1654 	if (error != 0)
1655 		return (error);
1656 
1657 	/*
1658 	 * Fill in "ps_strings" struct for ps, w, etc.
1659 	 */
1660 	imgp->argv = vectp;
1661 	if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 ||
1662 	    suword32(&arginfo->ps_nargvstr, argc) != 0)
1663 		return (EFAULT);
1664 
1665 	/*
1666 	 * Fill in argument portion of vector table.
1667 	 */
1668 	for (; argc > 0; --argc) {
1669 		if (suword(vectp++, ustringp) != 0)
1670 			return (EFAULT);
1671 		while (*stringp++ != 0)
1672 			ustringp++;
1673 		ustringp++;
1674 	}
1675 
1676 	/* a null vector table pointer separates the argp's from the envp's */
1677 	if (suword(vectp++, 0) != 0)
1678 		return (EFAULT);
1679 
1680 	imgp->envv = vectp;
1681 	if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 ||
1682 	    suword32(&arginfo->ps_nenvstr, envc) != 0)
1683 		return (EFAULT);
1684 
1685 	/*
1686 	 * Fill in environment portion of vector table.
1687 	 */
1688 	for (; envc > 0; --envc) {
1689 		if (suword(vectp++, ustringp) != 0)
1690 			return (EFAULT);
1691 		while (*stringp++ != 0)
1692 			ustringp++;
1693 		ustringp++;
1694 	}
1695 
1696 	/* end of vector table is a null pointer */
1697 	if (suword(vectp, 0) != 0)
1698 		return (EFAULT);
1699 
1700 	if (imgp->auxargs) {
1701 		vectp++;
1702 		error = imgp->sysent->sv_copyout_auxargs(imgp,
1703 		    (uintptr_t)vectp);
1704 		if (error != 0)
1705 			return (error);
1706 	}
1707 
1708 	return (0);
1709 }
1710 
1711 /*
1712  * Check permissions of file to execute.
1713  *	Called with imgp->vp locked.
1714  *	Return 0 for success or error code on failure.
1715  */
1716 int
1717 exec_check_permissions(struct image_params *imgp)
1718 {
1719 	struct vnode *vp = imgp->vp;
1720 	struct vattr *attr = imgp->attr;
1721 	struct thread *td;
1722 	int error;
1723 
1724 	td = curthread;
1725 
1726 	/* Get file attributes */
1727 	error = VOP_GETATTR(vp, attr, td->td_ucred);
1728 	if (error)
1729 		return (error);
1730 
1731 #ifdef MAC
1732 	error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1733 	if (error)
1734 		return (error);
1735 #endif
1736 
1737 	/*
1738 	 * 1) Check if file execution is disabled for the filesystem that
1739 	 *    this file resides on.
1740 	 * 2) Ensure that at least one execute bit is on. Otherwise, a
1741 	 *    privileged user will always succeed, and we don't want this
1742 	 *    to happen unless the file really is executable.
1743 	 * 3) Ensure that the file is a regular file.
1744 	 */
1745 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1746 	    (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1747 	    (attr->va_type != VREG))
1748 		return (EACCES);
1749 
1750 	/*
1751 	 * Zero length files can't be exec'd
1752 	 */
1753 	if (attr->va_size == 0)
1754 		return (ENOEXEC);
1755 
1756 	/*
1757 	 *  Check for execute permission to file based on current credentials.
1758 	 */
1759 	error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1760 	if (error)
1761 		return (error);
1762 
1763 	/*
1764 	 * Check number of open-for-writes on the file and deny execution
1765 	 * if there are any.
1766 	 *
1767 	 * Add a text reference now so no one can write to the
1768 	 * executable while we're activating it.
1769 	 *
1770 	 * Remember if this was set before and unset it in case this is not
1771 	 * actually an executable image.
1772 	 */
1773 	error = VOP_SET_TEXT(vp);
1774 	if (error != 0)
1775 		return (error);
1776 	imgp->textset = true;
1777 
1778 	/*
1779 	 * Call filesystem specific open routine (which does nothing in the
1780 	 * general case).
1781 	 */
1782 	error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1783 	if (error == 0)
1784 		imgp->opened = 1;
1785 	return (error);
1786 }
1787 
1788 /*
1789  * Exec handler registration
1790  */
1791 int
1792 exec_register(const struct execsw *execsw_arg)
1793 {
1794 	const struct execsw **es, **xs, **newexecsw;
1795 	u_int count = 2;	/* New slot and trailing NULL */
1796 
1797 	if (execsw)
1798 		for (es = execsw; *es; es++)
1799 			count++;
1800 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1801 	xs = newexecsw;
1802 	if (execsw)
1803 		for (es = execsw; *es; es++)
1804 			*xs++ = *es;
1805 	*xs++ = execsw_arg;
1806 	*xs = NULL;
1807 	if (execsw)
1808 		free(execsw, M_TEMP);
1809 	execsw = newexecsw;
1810 	return (0);
1811 }
1812 
1813 int
1814 exec_unregister(const struct execsw *execsw_arg)
1815 {
1816 	const struct execsw **es, **xs, **newexecsw;
1817 	int count = 1;
1818 
1819 	if (execsw == NULL)
1820 		panic("unregister with no handlers left?\n");
1821 
1822 	for (es = execsw; *es; es++) {
1823 		if (*es == execsw_arg)
1824 			break;
1825 	}
1826 	if (*es == NULL)
1827 		return (ENOENT);
1828 	for (es = execsw; *es; es++)
1829 		if (*es != execsw_arg)
1830 			count++;
1831 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1832 	xs = newexecsw;
1833 	for (es = execsw; *es; es++)
1834 		if (*es != execsw_arg)
1835 			*xs++ = *es;
1836 	*xs = NULL;
1837 	if (execsw)
1838 		free(execsw, M_TEMP);
1839 	execsw = newexecsw;
1840 	return (0);
1841 }
1842