xref: /freebsd/sys/kern/kern_exec.c (revision aa24f48b361effe51163877d84f1b70d32b77e04)
1 /*-
2  * Copyright (c) 1993, David Greenman
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  */
26 
27 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD$");
29 
30 #include "opt_capsicum.h"
31 #include "opt_compat.h"
32 #include "opt_hwpmc_hooks.h"
33 #include "opt_ktrace.h"
34 #include "opt_vm.h"
35 
36 #include <sys/param.h>
37 #include <sys/systm.h>
38 #include <sys/acct.h>
39 #include <sys/capsicum.h>
40 #include <sys/eventhandler.h>
41 #include <sys/exec.h>
42 #include <sys/fcntl.h>
43 #include <sys/filedesc.h>
44 #include <sys/imgact.h>
45 #include <sys/imgact_elf.h>
46 #include <sys/kernel.h>
47 #include <sys/lock.h>
48 #include <sys/malloc.h>
49 #include <sys/mman.h>
50 #include <sys/mount.h>
51 #include <sys/mutex.h>
52 #include <sys/namei.h>
53 #include <sys/pioctl.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);
122 
123 /* XXX This should be vm_size_t. */
124 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD|
125     CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_ps_strings, "LU", "");
126 
127 /* XXX This should be vm_size_t. */
128 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD|
129     CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_usrstack, "LU", "");
130 
131 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD|CTLFLAG_MPSAFE,
132     NULL, 0, sysctl_kern_stackprot, "I", "");
133 
134 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
135 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
136     &ps_arg_cache_limit, 0, "");
137 
138 static int disallow_high_osrel;
139 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW,
140     &disallow_high_osrel, 0,
141     "Disallow execution of binaries built for higher version of the world");
142 
143 static int map_at_zero = 0;
144 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RWTUN, &map_at_zero, 0,
145     "Permit processes to map an object at virtual address 0.");
146 
147 static int
148 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
149 {
150 	struct proc *p;
151 	int error;
152 
153 	p = curproc;
154 #ifdef SCTL_MASK32
155 	if (req->flags & SCTL_MASK32) {
156 		unsigned int val;
157 		val = (unsigned int)p->p_sysent->sv_psstrings;
158 		error = SYSCTL_OUT(req, &val, sizeof(val));
159 	} else
160 #endif
161 		error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
162 		   sizeof(p->p_sysent->sv_psstrings));
163 	return error;
164 }
165 
166 static int
167 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
168 {
169 	struct proc *p;
170 	int error;
171 
172 	p = curproc;
173 #ifdef SCTL_MASK32
174 	if (req->flags & SCTL_MASK32) {
175 		unsigned int val;
176 		val = (unsigned int)p->p_sysent->sv_usrstack;
177 		error = SYSCTL_OUT(req, &val, sizeof(val));
178 	} else
179 #endif
180 		error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
181 		    sizeof(p->p_sysent->sv_usrstack));
182 	return error;
183 }
184 
185 static int
186 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
187 {
188 	struct proc *p;
189 
190 	p = curproc;
191 	return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
192 	    sizeof(p->p_sysent->sv_stackprot)));
193 }
194 
195 /*
196  * Each of the items is a pointer to a `const struct execsw', hence the
197  * double pointer here.
198  */
199 static const struct execsw **execsw;
200 
201 #ifndef _SYS_SYSPROTO_H_
202 struct execve_args {
203 	char    *fname;
204 	char    **argv;
205 	char    **envv;
206 };
207 #endif
208 
209 int
210 sys_execve(struct thread *td, struct execve_args *uap)
211 {
212 	struct image_args args;
213 	struct vmspace *oldvmspace;
214 	int error;
215 
216 	error = pre_execve(td, &oldvmspace);
217 	if (error != 0)
218 		return (error);
219 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
220 	    uap->argv, uap->envv);
221 	if (error == 0)
222 		error = kern_execve(td, &args, NULL);
223 	post_execve(td, error, oldvmspace);
224 	return (error);
225 }
226 
227 #ifndef _SYS_SYSPROTO_H_
228 struct fexecve_args {
229 	int	fd;
230 	char	**argv;
231 	char	**envv;
232 }
233 #endif
234 int
235 sys_fexecve(struct thread *td, struct fexecve_args *uap)
236 {
237 	struct image_args args;
238 	struct vmspace *oldvmspace;
239 	int error;
240 
241 	error = pre_execve(td, &oldvmspace);
242 	if (error != 0)
243 		return (error);
244 	error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
245 	    uap->argv, uap->envv);
246 	if (error == 0) {
247 		args.fd = uap->fd;
248 		error = kern_execve(td, &args, NULL);
249 	}
250 	post_execve(td, error, oldvmspace);
251 	return (error);
252 }
253 
254 #ifndef _SYS_SYSPROTO_H_
255 struct __mac_execve_args {
256 	char	*fname;
257 	char	**argv;
258 	char	**envv;
259 	struct mac	*mac_p;
260 };
261 #endif
262 
263 int
264 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap)
265 {
266 #ifdef MAC
267 	struct image_args args;
268 	struct vmspace *oldvmspace;
269 	int error;
270 
271 	error = pre_execve(td, &oldvmspace);
272 	if (error != 0)
273 		return (error);
274 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
275 	    uap->argv, uap->envv);
276 	if (error == 0)
277 		error = kern_execve(td, &args, uap->mac_p);
278 	post_execve(td, error, oldvmspace);
279 	return (error);
280 #else
281 	return (ENOSYS);
282 #endif
283 }
284 
285 int
286 pre_execve(struct thread *td, struct vmspace **oldvmspace)
287 {
288 	struct proc *p;
289 	int error;
290 
291 	KASSERT(td == curthread, ("non-current thread %p", td));
292 	error = 0;
293 	p = td->td_proc;
294 	if ((p->p_flag & P_HADTHREADS) != 0) {
295 		PROC_LOCK(p);
296 		if (thread_single(p, SINGLE_BOUNDARY) != 0)
297 			error = ERESTART;
298 		PROC_UNLOCK(p);
299 	}
300 	KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0,
301 	    ("nested execve"));
302 	*oldvmspace = p->p_vmspace;
303 	return (error);
304 }
305 
306 void
307 post_execve(struct thread *td, int error, struct vmspace *oldvmspace)
308 {
309 	struct proc *p;
310 
311 	KASSERT(td == curthread, ("non-current thread %p", td));
312 	p = td->td_proc;
313 	if ((p->p_flag & P_HADTHREADS) != 0) {
314 		PROC_LOCK(p);
315 		/*
316 		 * If success, we upgrade to SINGLE_EXIT state to
317 		 * force other threads to suicide.
318 		 */
319 		if (error == 0)
320 			thread_single(p, SINGLE_EXIT);
321 		else
322 			thread_single_end(p, SINGLE_BOUNDARY);
323 		PROC_UNLOCK(p);
324 	}
325 	if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
326 		KASSERT(p->p_vmspace != oldvmspace,
327 		    ("oldvmspace still used"));
328 		vmspace_free(oldvmspace);
329 		td->td_pflags &= ~TDP_EXECVMSPC;
330 	}
331 }
332 
333 /*
334  * XXX: kern_execve has the astonishing property of not always returning to
335  * the caller.  If sufficiently bad things happen during the call to
336  * do_execve(), it can end up calling exit1(); as a result, callers must
337  * avoid doing anything which they might need to undo (e.g., allocating
338  * memory).
339  */
340 int
341 kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p)
342 {
343 
344 	AUDIT_ARG_ARGV(args->begin_argv, args->argc,
345 	    args->begin_envv - args->begin_argv);
346 	AUDIT_ARG_ENVV(args->begin_envv, args->envc,
347 	    args->endp - args->begin_envv);
348 	return (do_execve(td, args, mac_p));
349 }
350 
351 /*
352  * In-kernel implementation of execve().  All arguments are assumed to be
353  * userspace pointers from the passed thread.
354  */
355 static int
356 do_execve(struct thread *td, struct image_args *args, struct mac *mac_p)
357 {
358 	struct proc *p = td->td_proc;
359 	struct nameidata nd;
360 	struct ucred *oldcred;
361 	struct uidinfo *euip = NULL;
362 	register_t *stack_base;
363 	int error, i;
364 	struct image_params image_params, *imgp;
365 	struct vattr attr;
366 	int (*img_first)(struct image_params *);
367 	struct pargs *oldargs = NULL, *newargs = NULL;
368 	struct sigacts *oldsigacts = NULL, *newsigacts = NULL;
369 #ifdef KTRACE
370 	struct vnode *tracevp = NULL;
371 	struct ucred *tracecred = NULL;
372 #endif
373 	struct vnode *oldtextvp = NULL, *newtextvp;
374 	cap_rights_t rights;
375 	int credential_changing;
376 	int textset;
377 #ifdef MAC
378 	struct label *interpvplabel = NULL;
379 	int will_transition;
380 #endif
381 #ifdef HWPMC_HOOKS
382 	struct pmckern_procexec pe;
383 #endif
384 	static const char fexecv_proc_title[] = "(fexecv)";
385 
386 	imgp = &image_params;
387 
388 	/*
389 	 * Lock the process and set the P_INEXEC flag to indicate that
390 	 * it should be left alone until we're done here.  This is
391 	 * necessary to avoid race conditions - e.g. in ptrace() -
392 	 * that might allow a local user to illicitly obtain elevated
393 	 * privileges.
394 	 */
395 	PROC_LOCK(p);
396 	KASSERT((p->p_flag & P_INEXEC) == 0,
397 	    ("%s(): process already has P_INEXEC flag", __func__));
398 	p->p_flag |= P_INEXEC;
399 	PROC_UNLOCK(p);
400 
401 	/*
402 	 * Initialize part of the common data
403 	 */
404 	bzero(imgp, sizeof(*imgp));
405 	imgp->proc = p;
406 	imgp->attr = &attr;
407 	imgp->args = args;
408 	oldcred = p->p_ucred;
409 
410 #ifdef MAC
411 	error = mac_execve_enter(imgp, mac_p);
412 	if (error)
413 		goto exec_fail;
414 #endif
415 
416 	/*
417 	 * Translate the file name. namei() returns a vnode pointer
418 	 *	in ni_vp among other things.
419 	 *
420 	 * XXXAUDIT: It would be desirable to also audit the name of the
421 	 * interpreter if this is an interpreted binary.
422 	 */
423 	if (args->fname != NULL) {
424 		NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME
425 		    | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
426 	}
427 
428 	SDT_PROBE1(proc, , , exec, args->fname);
429 
430 interpret:
431 	if (args->fname != NULL) {
432 #ifdef CAPABILITY_MODE
433 		/*
434 		 * While capability mode can't reach this point via direct
435 		 * path arguments to execve(), we also don't allow
436 		 * interpreters to be used in capability mode (for now).
437 		 * Catch indirect lookups and return a permissions error.
438 		 */
439 		if (IN_CAPABILITY_MODE(td)) {
440 			error = ECAPMODE;
441 			goto exec_fail;
442 		}
443 #endif
444 		error = namei(&nd);
445 		if (error)
446 			goto exec_fail;
447 
448 		newtextvp = nd.ni_vp;
449 		imgp->vp = newtextvp;
450 	} else {
451 		AUDIT_ARG_FD(args->fd);
452 		/*
453 		 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
454 		 */
455 		error = fgetvp_exec(td, args->fd,
456 		    cap_rights_init(&rights, CAP_FEXECVE), &newtextvp);
457 		if (error)
458 			goto exec_fail;
459 		vn_lock(newtextvp, LK_EXCLUSIVE | LK_RETRY);
460 		AUDIT_ARG_VNODE1(newtextvp);
461 		imgp->vp = newtextvp;
462 	}
463 
464 	/*
465 	 * Check file permissions (also 'opens' file)
466 	 */
467 	error = exec_check_permissions(imgp);
468 	if (error)
469 		goto exec_fail_dealloc;
470 
471 	imgp->object = imgp->vp->v_object;
472 	if (imgp->object != NULL)
473 		vm_object_reference(imgp->object);
474 
475 	/*
476 	 * Set VV_TEXT now so no one can write to the executable while we're
477 	 * activating it.
478 	 *
479 	 * Remember if this was set before and unset it in case this is not
480 	 * actually an executable image.
481 	 */
482 	textset = VOP_IS_TEXT(imgp->vp);
483 	VOP_SET_TEXT(imgp->vp);
484 
485 	error = exec_map_first_page(imgp);
486 	if (error)
487 		goto exec_fail_dealloc;
488 
489 	imgp->proc->p_osrel = 0;
490 
491 	/*
492 	 * Implement image setuid/setgid.
493 	 *
494 	 * Determine new credentials before attempting image activators
495 	 * so that it can be used by process_exec handlers to determine
496 	 * credential/setid changes.
497 	 *
498 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
499 	 * the process is being traced.
500 	 *
501 	 * We disable setuid/setgid/etc in capability mode on the basis
502 	 * that most setugid applications are not written with that
503 	 * environment in mind, and will therefore almost certainly operate
504 	 * incorrectly. In principle there's no reason that setugid
505 	 * applications might not be useful in capability mode, so we may want
506 	 * to reconsider this conservative design choice in the future.
507 	 *
508 	 * XXXMAC: For the time being, use NOSUID to also prohibit
509 	 * transitions on the file system.
510 	 */
511 	credential_changing = 0;
512 	credential_changing |= (attr.va_mode & S_ISUID) &&
513 	    oldcred->cr_uid != attr.va_uid;
514 	credential_changing |= (attr.va_mode & S_ISGID) &&
515 	    oldcred->cr_gid != attr.va_gid;
516 #ifdef MAC
517 	will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
518 	    interpvplabel, imgp);
519 	credential_changing |= will_transition;
520 #endif
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, 0);
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_EXCLUSIVE | 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, 0);
563 			imgp->newcred = crdup(oldcred);
564 			vn_lock(imgp->vp, LK_EXCLUSIVE | 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, 0);
578 		if (vn_fullpath(td, imgp->vp, &imgp->execpath,
579 		    &imgp->freepath) != 0)
580 			imgp->execpath = args->fname;
581 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
582 	}
583 
584 	/*
585 	 *	If the current process has a special image activator it
586 	 *	wants to try first, call it.   For example, emulating shell
587 	 *	scripts differently.
588 	 */
589 	error = -1;
590 	if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
591 		error = img_first(imgp);
592 
593 	/*
594 	 *	Loop through the list of image activators, calling each one.
595 	 *	An activator returns -1 if there is no match, 0 on success,
596 	 *	and an error otherwise.
597 	 */
598 	for (i = 0; error == -1 && execsw[i]; ++i) {
599 		if (execsw[i]->ex_imgact == NULL ||
600 		    execsw[i]->ex_imgact == img_first) {
601 			continue;
602 		}
603 		error = (*execsw[i]->ex_imgact)(imgp);
604 	}
605 
606 	if (error) {
607 		if (error == -1) {
608 			if (textset == 0)
609 				VOP_UNSET_TEXT(imgp->vp);
610 			error = ENOEXEC;
611 		}
612 		goto exec_fail_dealloc;
613 	}
614 
615 	/*
616 	 * Special interpreter operation, cleanup and loop up to try to
617 	 * activate the interpreter.
618 	 */
619 	if (imgp->interpreted) {
620 		exec_unmap_first_page(imgp);
621 		/*
622 		 * VV_TEXT needs to be unset for scripts.  There is a short
623 		 * period before we determine that something is a script where
624 		 * VV_TEXT will be set. The vnode lock is held over this
625 		 * entire period so nothing should illegitimately be blocked.
626 		 */
627 		VOP_UNSET_TEXT(imgp->vp);
628 		/* free name buffer and old vnode */
629 		if (args->fname != NULL)
630 			NDFREE(&nd, NDF_ONLY_PNBUF);
631 #ifdef MAC
632 		mac_execve_interpreter_enter(newtextvp, &interpvplabel);
633 #endif
634 		if (imgp->opened) {
635 			VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td);
636 			imgp->opened = 0;
637 		}
638 		vput(newtextvp);
639 		vm_object_deallocate(imgp->object);
640 		imgp->object = NULL;
641 		imgp->credential_setid = false;
642 		if (imgp->newcred != NULL) {
643 			crfree(imgp->newcred);
644 			imgp->newcred = NULL;
645 		}
646 		imgp->execpath = NULL;
647 		free(imgp->freepath, M_TEMP);
648 		imgp->freepath = NULL;
649 		/* set new name to that of the interpreter */
650 		NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
651 		    UIO_SYSSPACE, imgp->interpreter_name, td);
652 		args->fname = imgp->interpreter_name;
653 		goto interpret;
654 	}
655 
656 	/*
657 	 * NB: We unlock the vnode here because it is believed that none
658 	 * of the sv_copyout_strings/sv_fixup operations require the vnode.
659 	 */
660 	VOP_UNLOCK(imgp->vp, 0);
661 
662 	if (disallow_high_osrel &&
663 	    P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
664 		error = ENOEXEC;
665 		uprintf("Osrel %d for image %s too high\n", p->p_osrel,
666 		    imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
667 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
668 		goto exec_fail_dealloc;
669 	}
670 
671 	/* ABI enforces the use of Capsicum. Switch into capabilities mode. */
672 	if (SV_PROC_FLAG(p, SV_CAPSICUM))
673 		sys_cap_enter(td, NULL);
674 
675 	/*
676 	 * Copy out strings (args and env) and initialize stack base
677 	 */
678 	if (p->p_sysent->sv_copyout_strings)
679 		stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
680 	else
681 		stack_base = exec_copyout_strings(imgp);
682 
683 	/*
684 	 * If custom stack fixup routine present for this process
685 	 * let it do the stack setup.
686 	 * Else stuff argument count as first item on stack
687 	 */
688 	if (p->p_sysent->sv_fixup != NULL)
689 		(*p->p_sysent->sv_fixup)(&stack_base, imgp);
690 	else
691 		suword(--stack_base, imgp->args->argc);
692 
693 	if (args->fdp != NULL) {
694 		/* Install a brand new file descriptor table. */
695 		fdinstall_remapped(td, args->fdp);
696 		args->fdp = NULL;
697 	} else {
698 		/*
699 		 * Keep on using the existing file descriptor table. For
700 		 * security and other reasons, the file descriptor table
701 		 * cannot be shared after an exec.
702 		 */
703 		fdunshare(td);
704 		/* close files on exec */
705 		fdcloseexec(td);
706 	}
707 
708 	/*
709 	 * Malloc things before we need locks.
710 	 */
711 	i = imgp->args->begin_envv - imgp->args->begin_argv;
712 	/* Cache arguments if they fit inside our allowance */
713 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
714 		newargs = pargs_alloc(i);
715 		bcopy(imgp->args->begin_argv, newargs->ar_args, i);
716 	}
717 
718 	/*
719 	 * For security and other reasons, signal handlers cannot
720 	 * be shared after an exec. The new process gets a copy of the old
721 	 * handlers. In execsigs(), the new process will have its signals
722 	 * reset.
723 	 */
724 	if (sigacts_shared(p->p_sigacts)) {
725 		oldsigacts = p->p_sigacts;
726 		newsigacts = sigacts_alloc();
727 		sigacts_copy(newsigacts, oldsigacts);
728 	}
729 
730 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
731 
732 	PROC_LOCK(p);
733 	if (oldsigacts)
734 		p->p_sigacts = newsigacts;
735 	/* Stop profiling */
736 	stopprofclock(p);
737 
738 	/* reset caught signals */
739 	execsigs(p);
740 
741 	/* name this process - nameiexec(p, ndp) */
742 	bzero(p->p_comm, sizeof(p->p_comm));
743 	if (args->fname)
744 		bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
745 		    min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
746 	else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0)
747 		bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
748 	bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
749 #ifdef KTR
750 	sched_clear_tdname(td);
751 #endif
752 
753 	/*
754 	 * mark as execed, wakeup the process that vforked (if any) and tell
755 	 * it that it now has its own resources back
756 	 */
757 	p->p_flag |= P_EXEC;
758 	if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0)
759 		p->p_flag2 &= ~P2_NOTRACE;
760 	if (p->p_flag & P_PPWAIT) {
761 		p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
762 		cv_broadcast(&p->p_pwait);
763 		/* STOPs are no longer ignored, arrange for AST */
764 		signotify(td);
765 	}
766 
767 	/*
768 	 * Implement image setuid/setgid installation.
769 	 */
770 	if (imgp->credential_setid) {
771 		/*
772 		 * Turn off syscall tracing for set-id programs, except for
773 		 * root.  Record any set-id flags first to make sure that
774 		 * we do not regain any tracing during a possible block.
775 		 */
776 		setsugid(p);
777 
778 #ifdef KTRACE
779 		if (p->p_tracecred != NULL &&
780 		    priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED, 0))
781 			ktrprocexec(p, &tracecred, &tracevp);
782 #endif
783 		/*
784 		 * Close any file descriptors 0..2 that reference procfs,
785 		 * then make sure file descriptors 0..2 are in use.
786 		 *
787 		 * Both fdsetugidsafety() and fdcheckstd() may call functions
788 		 * taking sleepable locks, so temporarily drop our locks.
789 		 */
790 		PROC_UNLOCK(p);
791 		VOP_UNLOCK(imgp->vp, 0);
792 		fdsetugidsafety(td);
793 		error = fdcheckstd(td);
794 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
795 		if (error != 0)
796 			goto exec_fail_dealloc;
797 		PROC_LOCK(p);
798 #ifdef MAC
799 		if (will_transition) {
800 			mac_vnode_execve_transition(oldcred, imgp->newcred,
801 			    imgp->vp, interpvplabel, imgp);
802 		}
803 #endif
804 	} else {
805 		if (oldcred->cr_uid == oldcred->cr_ruid &&
806 		    oldcred->cr_gid == oldcred->cr_rgid)
807 			p->p_flag &= ~P_SUGID;
808 	}
809 	/*
810 	 * Set the new credentials.
811 	 */
812 	if (imgp->newcred != NULL) {
813 		proc_set_cred(p, imgp->newcred);
814 		crfree(oldcred);
815 		oldcred = NULL;
816 	}
817 
818 	/*
819 	 * Store the vp for use in procfs.  This vnode was referenced by namei
820 	 * or fgetvp_exec.
821 	 */
822 	oldtextvp = p->p_textvp;
823 	p->p_textvp = newtextvp;
824 
825 #ifdef KDTRACE_HOOKS
826 	/*
827 	 * Tell the DTrace fasttrap provider about the exec if it
828 	 * has declared an interest.
829 	 */
830 	if (dtrace_fasttrap_exec)
831 		dtrace_fasttrap_exec(p);
832 #endif
833 
834 	/*
835 	 * Notify others that we exec'd, and clear the P_INEXEC flag
836 	 * as we're now a bona fide freshly-execed process.
837 	 */
838 	KNOTE_LOCKED(p->p_klist, NOTE_EXEC);
839 	p->p_flag &= ~P_INEXEC;
840 
841 	/* clear "fork but no exec" flag, as we _are_ execing */
842 	p->p_acflag &= ~AFORK;
843 
844 	/*
845 	 * Free any previous argument cache and replace it with
846 	 * the new argument cache, if any.
847 	 */
848 	oldargs = p->p_args;
849 	p->p_args = newargs;
850 	newargs = NULL;
851 
852 #ifdef	HWPMC_HOOKS
853 	/*
854 	 * Check if system-wide sampling is in effect or if the
855 	 * current process is using PMCs.  If so, do exec() time
856 	 * processing.  This processing needs to happen AFTER the
857 	 * P_INEXEC flag is cleared.
858 	 *
859 	 * The proc lock needs to be released before taking the PMC
860 	 * SX.
861 	 */
862 	if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
863 		PROC_UNLOCK(p);
864 		VOP_UNLOCK(imgp->vp, 0);
865 		pe.pm_credentialschanged = credential_changing;
866 		pe.pm_entryaddr = imgp->entry_addr;
867 
868 		PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
869 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
870 	} else
871 		PROC_UNLOCK(p);
872 #else  /* !HWPMC_HOOKS */
873 	PROC_UNLOCK(p);
874 #endif
875 
876 	/* Set values passed into the program in registers. */
877 	if (p->p_sysent->sv_setregs)
878 		(*p->p_sysent->sv_setregs)(td, imgp,
879 		    (u_long)(uintptr_t)stack_base);
880 	else
881 		exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base);
882 
883 	vfs_mark_atime(imgp->vp, td->td_ucred);
884 
885 	SDT_PROBE1(proc, , , exec__success, args->fname);
886 
887 exec_fail_dealloc:
888 	if (imgp->firstpage != NULL)
889 		exec_unmap_first_page(imgp);
890 
891 	if (imgp->vp != NULL) {
892 		if (args->fname)
893 			NDFREE(&nd, NDF_ONLY_PNBUF);
894 		if (imgp->opened)
895 			VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
896 		if (error != 0)
897 			vput(imgp->vp);
898 		else
899 			VOP_UNLOCK(imgp->vp, 0);
900 	}
901 
902 	if (imgp->object != NULL)
903 		vm_object_deallocate(imgp->object);
904 
905 	free(imgp->freepath, M_TEMP);
906 
907 	if (error == 0) {
908 		PROC_LOCK(p);
909 		if (p->p_ptevents & PTRACE_EXEC)
910 			td->td_dbgflags |= TDB_EXEC;
911 		PROC_UNLOCK(p);
912 
913 		/*
914 		 * Stop the process here if its stop event mask has
915 		 * the S_EXEC bit set.
916 		 */
917 		STOPEVENT(p, S_EXEC, 0);
918 	} else {
919 exec_fail:
920 		/* we're done here, clear P_INEXEC */
921 		PROC_LOCK(p);
922 		p->p_flag &= ~P_INEXEC;
923 		PROC_UNLOCK(p);
924 
925 		SDT_PROBE1(proc, , , exec__failure, error);
926 	}
927 
928 	if (imgp->newcred != NULL && oldcred != NULL)
929 		crfree(imgp->newcred);
930 
931 #ifdef MAC
932 	mac_execve_exit(imgp);
933 	mac_execve_interpreter_exit(interpvplabel);
934 #endif
935 	exec_free_args(args);
936 
937 	/*
938 	 * Handle deferred decrement of ref counts.
939 	 */
940 	if (oldtextvp != NULL)
941 		vrele(oldtextvp);
942 #ifdef KTRACE
943 	if (tracevp != NULL)
944 		vrele(tracevp);
945 	if (tracecred != NULL)
946 		crfree(tracecred);
947 #endif
948 	pargs_drop(oldargs);
949 	pargs_drop(newargs);
950 	if (oldsigacts != NULL)
951 		sigacts_free(oldsigacts);
952 	if (euip != NULL)
953 		uifree(euip);
954 
955 	if (error && imgp->vmspace_destroyed) {
956 		/* sorry, no more process anymore. exit gracefully */
957 		exit1(td, 0, SIGABRT);
958 		/* NOT REACHED */
959 	}
960 
961 #ifdef KTRACE
962 	if (error == 0)
963 		ktrprocctor(p);
964 #endif
965 
966 	return (error);
967 }
968 
969 int
970 exec_map_first_page(imgp)
971 	struct image_params *imgp;
972 {
973 	int rv, i, after, initial_pagein;
974 	vm_page_t ma[VM_INITIAL_PAGEIN];
975 	vm_object_t object;
976 
977 	if (imgp->firstpage != NULL)
978 		exec_unmap_first_page(imgp);
979 
980 	object = imgp->vp->v_object;
981 	if (object == NULL)
982 		return (EACCES);
983 	VM_OBJECT_WLOCK(object);
984 #if VM_NRESERVLEVEL > 0
985 	vm_object_color(object, 0);
986 #endif
987 	ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY);
988 	if (ma[0]->valid != VM_PAGE_BITS_ALL) {
989 		vm_page_xbusy(ma[0]);
990 		if (!vm_pager_has_page(object, 0, NULL, &after)) {
991 			vm_page_lock(ma[0]);
992 			vm_page_free(ma[0]);
993 			vm_page_unlock(ma[0]);
994 			VM_OBJECT_WUNLOCK(object);
995 			return (EIO);
996 		}
997 		initial_pagein = min(after, VM_INITIAL_PAGEIN);
998 		KASSERT(initial_pagein <= object->size,
999 		    ("%s: initial_pagein %d object->size %ju",
1000 		    __func__, initial_pagein, (uintmax_t )object->size));
1001 		for (i = 1; i < initial_pagein; i++) {
1002 			if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) {
1003 				if (ma[i]->valid)
1004 					break;
1005 				if (vm_page_tryxbusy(ma[i]))
1006 					break;
1007 			} else {
1008 				ma[i] = vm_page_alloc(object, i,
1009 				    VM_ALLOC_NORMAL);
1010 				if (ma[i] == NULL)
1011 					break;
1012 			}
1013 		}
1014 		initial_pagein = i;
1015 		rv = vm_pager_get_pages(object, ma, initial_pagein, NULL, NULL);
1016 		if (rv != VM_PAGER_OK) {
1017 			for (i = 0; i < initial_pagein; i++) {
1018 				vm_page_lock(ma[i]);
1019 				vm_page_free(ma[i]);
1020 				vm_page_unlock(ma[i]);
1021 			}
1022 			VM_OBJECT_WUNLOCK(object);
1023 			return (EIO);
1024 		}
1025 		vm_page_xunbusy(ma[0]);
1026 		for (i = 1; i < initial_pagein; i++)
1027 			vm_page_readahead_finish(ma[i]);
1028 	}
1029 	vm_page_lock(ma[0]);
1030 	vm_page_hold(ma[0]);
1031 	vm_page_activate(ma[0]);
1032 	vm_page_unlock(ma[0]);
1033 	VM_OBJECT_WUNLOCK(object);
1034 
1035 	imgp->firstpage = sf_buf_alloc(ma[0], 0);
1036 	imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
1037 
1038 	return (0);
1039 }
1040 
1041 void
1042 exec_unmap_first_page(struct image_params *imgp)
1043 {
1044 	vm_page_t m;
1045 
1046 	if (imgp->firstpage != NULL) {
1047 		m = sf_buf_page(imgp->firstpage);
1048 		sf_buf_free(imgp->firstpage);
1049 		imgp->firstpage = NULL;
1050 		vm_page_lock(m);
1051 		vm_page_unhold(m);
1052 		vm_page_unlock(m);
1053 	}
1054 }
1055 
1056 /*
1057  * Destroy old address space, and allocate a new stack.
1058  *	The new stack is only sgrowsiz large because it is grown
1059  *	automatically on a page fault.
1060  */
1061 int
1062 exec_new_vmspace(struct image_params *imgp, struct sysentvec *sv)
1063 {
1064 	int error;
1065 	struct proc *p = imgp->proc;
1066 	struct vmspace *vmspace = p->p_vmspace;
1067 	vm_object_t obj;
1068 	struct rlimit rlim_stack;
1069 	vm_offset_t sv_minuser, stack_addr;
1070 	vm_map_t map;
1071 	u_long ssiz;
1072 
1073 	imgp->vmspace_destroyed = 1;
1074 	imgp->sysent = sv;
1075 
1076 	/* May be called with Giant held */
1077 	EVENTHANDLER_INVOKE(process_exec, p, imgp);
1078 
1079 	/*
1080 	 * Blow away entire process VM, if address space not shared,
1081 	 * otherwise, create a new VM space so that other threads are
1082 	 * not disrupted
1083 	 */
1084 	map = &vmspace->vm_map;
1085 	if (map_at_zero)
1086 		sv_minuser = sv->sv_minuser;
1087 	else
1088 		sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1089 	if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser &&
1090 	    vm_map_max(map) == sv->sv_maxuser) {
1091 		shmexit(vmspace);
1092 		pmap_remove_pages(vmspace_pmap(vmspace));
1093 		vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1094 		/* An exec terminates mlockall(MCL_FUTURE). */
1095 		vm_map_lock(map);
1096 		vm_map_modflags(map, 0, MAP_WIREFUTURE);
1097 		vm_map_unlock(map);
1098 	} else {
1099 		error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1100 		if (error)
1101 			return (error);
1102 		vmspace = p->p_vmspace;
1103 		map = &vmspace->vm_map;
1104 	}
1105 
1106 	/* Map a shared page */
1107 	obj = sv->sv_shared_page_obj;
1108 	if (obj != NULL) {
1109 		vm_object_reference(obj);
1110 		error = vm_map_fixed(map, obj, 0,
1111 		    sv->sv_shared_page_base, sv->sv_shared_page_len,
1112 		    VM_PROT_READ | VM_PROT_EXECUTE,
1113 		    VM_PROT_READ | VM_PROT_EXECUTE,
1114 		    MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1115 		if (error != KERN_SUCCESS) {
1116 			vm_object_deallocate(obj);
1117 			return (vm_mmap_to_errno(error));
1118 		}
1119 	}
1120 
1121 	/* Allocate a new stack */
1122 	if (imgp->stack_sz != 0) {
1123 		ssiz = trunc_page(imgp->stack_sz);
1124 		PROC_LOCK(p);
1125 		lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack);
1126 		PROC_UNLOCK(p);
1127 		if (ssiz > rlim_stack.rlim_max)
1128 			ssiz = rlim_stack.rlim_max;
1129 		if (ssiz > rlim_stack.rlim_cur) {
1130 			rlim_stack.rlim_cur = ssiz;
1131 			kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack);
1132 		}
1133 	} else if (sv->sv_maxssiz != NULL) {
1134 		ssiz = *sv->sv_maxssiz;
1135 	} else {
1136 		ssiz = maxssiz;
1137 	}
1138 	stack_addr = sv->sv_usrstack - ssiz;
1139 	error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1140 	    obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1141 	    sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1142 	if (error != KERN_SUCCESS)
1143 		return (vm_mmap_to_errno(error));
1144 
1145 	/*
1146 	 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they
1147 	 * are still used to enforce the stack rlimit on the process stack.
1148 	 */
1149 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1150 	vmspace->vm_maxsaddr = (char *)stack_addr;
1151 
1152 	return (0);
1153 }
1154 
1155 /*
1156  * Copy out argument and environment strings from the old process address
1157  * space into the temporary string buffer.
1158  */
1159 int
1160 exec_copyin_args(struct image_args *args, char *fname,
1161     enum uio_seg segflg, char **argv, char **envv)
1162 {
1163 	u_long argp, envp;
1164 	int error;
1165 	size_t length;
1166 
1167 	bzero(args, sizeof(*args));
1168 	if (argv == NULL)
1169 		return (EFAULT);
1170 
1171 	/*
1172 	 * Allocate demand-paged memory for the file name, argument, and
1173 	 * environment strings.
1174 	 */
1175 	error = exec_alloc_args(args);
1176 	if (error != 0)
1177 		return (error);
1178 
1179 	/*
1180 	 * Copy the file name.
1181 	 */
1182 	if (fname != NULL) {
1183 		args->fname = args->buf;
1184 		error = (segflg == UIO_SYSSPACE) ?
1185 		    copystr(fname, args->fname, PATH_MAX, &length) :
1186 		    copyinstr(fname, args->fname, PATH_MAX, &length);
1187 		if (error != 0)
1188 			goto err_exit;
1189 	} else
1190 		length = 0;
1191 
1192 	args->begin_argv = args->buf + length;
1193 	args->endp = args->begin_argv;
1194 	args->stringspace = ARG_MAX;
1195 
1196 	/*
1197 	 * extract arguments first
1198 	 */
1199 	for (;;) {
1200 		error = fueword(argv++, &argp);
1201 		if (error == -1) {
1202 			error = EFAULT;
1203 			goto err_exit;
1204 		}
1205 		if (argp == 0)
1206 			break;
1207 		error = copyinstr((void *)(uintptr_t)argp, args->endp,
1208 		    args->stringspace, &length);
1209 		if (error != 0) {
1210 			if (error == ENAMETOOLONG)
1211 				error = E2BIG;
1212 			goto err_exit;
1213 		}
1214 		args->stringspace -= length;
1215 		args->endp += length;
1216 		args->argc++;
1217 	}
1218 
1219 	args->begin_envv = args->endp;
1220 
1221 	/*
1222 	 * extract environment strings
1223 	 */
1224 	if (envv) {
1225 		for (;;) {
1226 			error = fueword(envv++, &envp);
1227 			if (error == -1) {
1228 				error = EFAULT;
1229 				goto err_exit;
1230 			}
1231 			if (envp == 0)
1232 				break;
1233 			error = copyinstr((void *)(uintptr_t)envp,
1234 			    args->endp, args->stringspace, &length);
1235 			if (error != 0) {
1236 				if (error == ENAMETOOLONG)
1237 					error = E2BIG;
1238 				goto err_exit;
1239 			}
1240 			args->stringspace -= length;
1241 			args->endp += length;
1242 			args->envc++;
1243 		}
1244 	}
1245 
1246 	return (0);
1247 
1248 err_exit:
1249 	exec_free_args(args);
1250 	return (error);
1251 }
1252 
1253 int
1254 exec_copyin_data_fds(struct thread *td, struct image_args *args,
1255     const void *data, size_t datalen, const int *fds, size_t fdslen)
1256 {
1257 	struct filedesc *ofdp;
1258 	const char *p;
1259 	int *kfds;
1260 	int error;
1261 
1262 	memset(args, '\0', sizeof(*args));
1263 	ofdp = td->td_proc->p_fd;
1264 	if (datalen >= ARG_MAX || fdslen > ofdp->fd_lastfile + 1)
1265 		return (E2BIG);
1266 	error = exec_alloc_args(args);
1267 	if (error != 0)
1268 		return (error);
1269 
1270 	args->begin_argv = args->buf;
1271 	args->stringspace = ARG_MAX;
1272 
1273 	if (datalen > 0) {
1274 		/*
1275 		 * Argument buffer has been provided. Copy it into the
1276 		 * kernel as a single string and add a terminating null
1277 		 * byte.
1278 		 */
1279 		error = copyin(data, args->begin_argv, datalen);
1280 		if (error != 0)
1281 			goto err_exit;
1282 		args->begin_argv[datalen] = '\0';
1283 		args->endp = args->begin_argv + datalen + 1;
1284 		args->stringspace -= datalen + 1;
1285 
1286 		/*
1287 		 * Traditional argument counting. Count the number of
1288 		 * null bytes.
1289 		 */
1290 		for (p = args->begin_argv; p < args->endp; ++p)
1291 			if (*p == '\0')
1292 				++args->argc;
1293 	} else {
1294 		/* No argument buffer provided. */
1295 		args->endp = args->begin_argv;
1296 	}
1297 	/* There are no environment variables. */
1298 	args->begin_envv = args->endp;
1299 
1300 	/* Create new file descriptor table. */
1301 	kfds = malloc(fdslen * sizeof(int), M_TEMP, M_WAITOK);
1302 	error = copyin(fds, kfds, fdslen * sizeof(int));
1303 	if (error != 0) {
1304 		free(kfds, M_TEMP);
1305 		goto err_exit;
1306 	}
1307 	error = fdcopy_remapped(ofdp, kfds, fdslen, &args->fdp);
1308 	free(kfds, M_TEMP);
1309 	if (error != 0)
1310 		goto err_exit;
1311 
1312 	return (0);
1313 err_exit:
1314 	exec_free_args(args);
1315 	return (error);
1316 }
1317 
1318 struct exec_args_kva {
1319 	vm_offset_t addr;
1320 	u_int gen;
1321 	SLIST_ENTRY(exec_args_kva) next;
1322 };
1323 
1324 static DPCPU_DEFINE(struct exec_args_kva *, exec_args_kva);
1325 
1326 static SLIST_HEAD(, exec_args_kva) exec_args_kva_freelist;
1327 static struct mtx exec_args_kva_mtx;
1328 static u_int exec_args_gen;
1329 
1330 static void
1331 exec_prealloc_args_kva(void *arg __unused)
1332 {
1333 	struct exec_args_kva *argkva;
1334 	u_int i;
1335 
1336 	SLIST_INIT(&exec_args_kva_freelist);
1337 	mtx_init(&exec_args_kva_mtx, "exec args kva", NULL, MTX_DEF);
1338 	for (i = 0; i < exec_map_entries; i++) {
1339 		argkva = malloc(sizeof(*argkva), M_PARGS, M_WAITOK);
1340 		argkva->addr = kmap_alloc_wait(exec_map, exec_map_entry_size);
1341 		argkva->gen = exec_args_gen;
1342 		SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1343 	}
1344 }
1345 SYSINIT(exec_args_kva, SI_SUB_EXEC, SI_ORDER_ANY, exec_prealloc_args_kva, NULL);
1346 
1347 static vm_offset_t
1348 exec_alloc_args_kva(void **cookie)
1349 {
1350 	struct exec_args_kva *argkva;
1351 
1352 	argkva = (void *)atomic_readandclear_ptr(
1353 	    (uintptr_t *)DPCPU_PTR(exec_args_kva));
1354 	if (argkva == NULL) {
1355 		mtx_lock(&exec_args_kva_mtx);
1356 		while ((argkva = SLIST_FIRST(&exec_args_kva_freelist)) == NULL)
1357 			(void)mtx_sleep(&exec_args_kva_freelist,
1358 			    &exec_args_kva_mtx, 0, "execkva", 0);
1359 		SLIST_REMOVE_HEAD(&exec_args_kva_freelist, next);
1360 		mtx_unlock(&exec_args_kva_mtx);
1361 	}
1362 	*(struct exec_args_kva **)cookie = argkva;
1363 	return (argkva->addr);
1364 }
1365 
1366 static void
1367 exec_release_args_kva(struct exec_args_kva *argkva, u_int gen)
1368 {
1369 	vm_offset_t base;
1370 
1371 	base = argkva->addr;
1372 	if (argkva->gen != gen) {
1373 		vm_map_madvise(exec_map, base, base + exec_map_entry_size,
1374 		    MADV_FREE);
1375 		argkva->gen = gen;
1376 	}
1377 	if (!atomic_cmpset_ptr((uintptr_t *)DPCPU_PTR(exec_args_kva),
1378 	    (uintptr_t)NULL, (uintptr_t)argkva)) {
1379 		mtx_lock(&exec_args_kva_mtx);
1380 		SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1381 		wakeup_one(&exec_args_kva_freelist);
1382 		mtx_unlock(&exec_args_kva_mtx);
1383 	}
1384 }
1385 
1386 static void
1387 exec_free_args_kva(void *cookie)
1388 {
1389 
1390 	exec_release_args_kva(cookie, exec_args_gen);
1391 }
1392 
1393 static void
1394 exec_args_kva_lowmem(void *arg __unused)
1395 {
1396 	SLIST_HEAD(, exec_args_kva) head;
1397 	struct exec_args_kva *argkva;
1398 	u_int gen;
1399 	int i;
1400 
1401 	gen = atomic_fetchadd_int(&exec_args_gen, 1) + 1;
1402 
1403 	/*
1404 	 * Force an madvise of each KVA range. Any currently allocated ranges
1405 	 * will have MADV_FREE applied once they are freed.
1406 	 */
1407 	SLIST_INIT(&head);
1408 	mtx_lock(&exec_args_kva_mtx);
1409 	SLIST_SWAP(&head, &exec_args_kva_freelist, exec_args_kva);
1410 	mtx_unlock(&exec_args_kva_mtx);
1411 	while ((argkva = SLIST_FIRST(&head)) != NULL) {
1412 		SLIST_REMOVE_HEAD(&head, next);
1413 		exec_release_args_kva(argkva, gen);
1414 	}
1415 
1416 	CPU_FOREACH(i) {
1417 		argkva = (void *)atomic_readandclear_ptr(
1418 		    (uintptr_t *)DPCPU_ID_PTR(i, exec_args_kva));
1419 		if (argkva != NULL)
1420 			exec_release_args_kva(argkva, gen);
1421 	}
1422 }
1423 EVENTHANDLER_DEFINE(vm_lowmem, exec_args_kva_lowmem, NULL,
1424     EVENTHANDLER_PRI_ANY);
1425 
1426 /*
1427  * Allocate temporary demand-paged, zero-filled memory for the file name,
1428  * argument, and environment strings.
1429  */
1430 int
1431 exec_alloc_args(struct image_args *args)
1432 {
1433 
1434 	args->buf = (char *)exec_alloc_args_kva(&args->bufkva);
1435 	return (0);
1436 }
1437 
1438 void
1439 exec_free_args(struct image_args *args)
1440 {
1441 
1442 	if (args->buf != NULL) {
1443 		exec_free_args_kva(args->bufkva);
1444 		args->buf = NULL;
1445 	}
1446 	if (args->fname_buf != NULL) {
1447 		free(args->fname_buf, M_TEMP);
1448 		args->fname_buf = NULL;
1449 	}
1450 	if (args->fdp != NULL)
1451 		fdescfree_remapped(args->fdp);
1452 }
1453 
1454 /*
1455  * Copy strings out to the new process address space, constructing new arg
1456  * and env vector tables. Return a pointer to the base so that it can be used
1457  * as the initial stack pointer.
1458  */
1459 register_t *
1460 exec_copyout_strings(struct image_params *imgp)
1461 {
1462 	int argc, envc;
1463 	char **vectp;
1464 	char *stringp;
1465 	uintptr_t destp;
1466 	register_t *stack_base;
1467 	struct ps_strings *arginfo;
1468 	struct proc *p;
1469 	size_t execpath_len;
1470 	int szsigcode, szps;
1471 	char canary[sizeof(long) * 8];
1472 
1473 	szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1474 	/*
1475 	 * Calculate string base and vector table pointers.
1476 	 * Also deal with signal trampoline code for this exec type.
1477 	 */
1478 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
1479 		execpath_len = strlen(imgp->execpath) + 1;
1480 	else
1481 		execpath_len = 0;
1482 	p = imgp->proc;
1483 	szsigcode = 0;
1484 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1485 	if (p->p_sysent->sv_sigcode_base == 0) {
1486 		if (p->p_sysent->sv_szsigcode != NULL)
1487 			szsigcode = *(p->p_sysent->sv_szsigcode);
1488 	}
1489 	destp =	(uintptr_t)arginfo;
1490 
1491 	/*
1492 	 * install sigcode
1493 	 */
1494 	if (szsigcode != 0) {
1495 		destp -= szsigcode;
1496 		destp = rounddown2(destp, sizeof(void *));
1497 		copyout(p->p_sysent->sv_sigcode, (void *)destp, szsigcode);
1498 	}
1499 
1500 	/*
1501 	 * Copy the image path for the rtld.
1502 	 */
1503 	if (execpath_len != 0) {
1504 		destp -= execpath_len;
1505 		imgp->execpathp = destp;
1506 		copyout(imgp->execpath, (void *)destp, execpath_len);
1507 	}
1508 
1509 	/*
1510 	 * Prepare the canary for SSP.
1511 	 */
1512 	arc4rand(canary, sizeof(canary), 0);
1513 	destp -= sizeof(canary);
1514 	imgp->canary = destp;
1515 	copyout(canary, (void *)destp, sizeof(canary));
1516 	imgp->canarylen = sizeof(canary);
1517 
1518 	/*
1519 	 * Prepare the pagesizes array.
1520 	 */
1521 	destp -= szps;
1522 	destp = rounddown2(destp, sizeof(void *));
1523 	imgp->pagesizes = destp;
1524 	copyout(pagesizes, (void *)destp, szps);
1525 	imgp->pagesizeslen = szps;
1526 
1527 	destp -= ARG_MAX - imgp->args->stringspace;
1528 	destp = rounddown2(destp, sizeof(void *));
1529 
1530 	/*
1531 	 * If we have a valid auxargs ptr, prepare some room
1532 	 * on the stack.
1533 	 */
1534 	if (imgp->auxargs) {
1535 		/*
1536 		 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1537 		 * lower compatibility.
1538 		 */
1539 		imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1540 		    (AT_COUNT * 2);
1541 		/*
1542 		 * The '+ 2' is for the null pointers at the end of each of
1543 		 * the arg and env vector sets,and imgp->auxarg_size is room
1544 		 * for argument of Runtime loader.
1545 		 */
1546 		vectp = (char **)(destp - (imgp->args->argc +
1547 		    imgp->args->envc + 2 + imgp->auxarg_size)
1548 		    * sizeof(char *));
1549 	} else {
1550 		/*
1551 		 * The '+ 2' is for the null pointers at the end of each of
1552 		 * the arg and env vector sets
1553 		 */
1554 		vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc
1555 		    + 2) * sizeof(char *));
1556 	}
1557 
1558 	/*
1559 	 * vectp also becomes our initial stack base
1560 	 */
1561 	stack_base = (register_t *)vectp;
1562 
1563 	stringp = imgp->args->begin_argv;
1564 	argc = imgp->args->argc;
1565 	envc = imgp->args->envc;
1566 
1567 	/*
1568 	 * Copy out strings - arguments and environment.
1569 	 */
1570 	copyout(stringp, (void *)destp, ARG_MAX - imgp->args->stringspace);
1571 
1572 	/*
1573 	 * Fill in "ps_strings" struct for ps, w, etc.
1574 	 */
1575 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1576 	suword32(&arginfo->ps_nargvstr, argc);
1577 
1578 	/*
1579 	 * Fill in argument portion of vector table.
1580 	 */
1581 	for (; argc > 0; --argc) {
1582 		suword(vectp++, (long)(intptr_t)destp);
1583 		while (*stringp++ != 0)
1584 			destp++;
1585 		destp++;
1586 	}
1587 
1588 	/* a null vector table pointer separates the argp's from the envp's */
1589 	suword(vectp++, 0);
1590 
1591 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1592 	suword32(&arginfo->ps_nenvstr, envc);
1593 
1594 	/*
1595 	 * Fill in environment portion of vector table.
1596 	 */
1597 	for (; envc > 0; --envc) {
1598 		suword(vectp++, (long)(intptr_t)destp);
1599 		while (*stringp++ != 0)
1600 			destp++;
1601 		destp++;
1602 	}
1603 
1604 	/* end of vector table is a null pointer */
1605 	suword(vectp, 0);
1606 
1607 	return (stack_base);
1608 }
1609 
1610 /*
1611  * Check permissions of file to execute.
1612  *	Called with imgp->vp locked.
1613  *	Return 0 for success or error code on failure.
1614  */
1615 int
1616 exec_check_permissions(struct image_params *imgp)
1617 {
1618 	struct vnode *vp = imgp->vp;
1619 	struct vattr *attr = imgp->attr;
1620 	struct thread *td;
1621 	int error, writecount;
1622 
1623 	td = curthread;
1624 
1625 	/* Get file attributes */
1626 	error = VOP_GETATTR(vp, attr, td->td_ucred);
1627 	if (error)
1628 		return (error);
1629 
1630 #ifdef MAC
1631 	error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1632 	if (error)
1633 		return (error);
1634 #endif
1635 
1636 	/*
1637 	 * 1) Check if file execution is disabled for the filesystem that
1638 	 *    this file resides on.
1639 	 * 2) Ensure that at least one execute bit is on. Otherwise, a
1640 	 *    privileged user will always succeed, and we don't want this
1641 	 *    to happen unless the file really is executable.
1642 	 * 3) Ensure that the file is a regular file.
1643 	 */
1644 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1645 	    (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1646 	    (attr->va_type != VREG))
1647 		return (EACCES);
1648 
1649 	/*
1650 	 * Zero length files can't be exec'd
1651 	 */
1652 	if (attr->va_size == 0)
1653 		return (ENOEXEC);
1654 
1655 	/*
1656 	 *  Check for execute permission to file based on current credentials.
1657 	 */
1658 	error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1659 	if (error)
1660 		return (error);
1661 
1662 	/*
1663 	 * Check number of open-for-writes on the file and deny execution
1664 	 * if there are any.
1665 	 */
1666 	error = VOP_GET_WRITECOUNT(vp, &writecount);
1667 	if (error != 0)
1668 		return (error);
1669 	if (writecount != 0)
1670 		return (ETXTBSY);
1671 
1672 	/*
1673 	 * Call filesystem specific open routine (which does nothing in the
1674 	 * general case).
1675 	 */
1676 	error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1677 	if (error == 0)
1678 		imgp->opened = 1;
1679 	return (error);
1680 }
1681 
1682 /*
1683  * Exec handler registration
1684  */
1685 int
1686 exec_register(const struct execsw *execsw_arg)
1687 {
1688 	const struct execsw **es, **xs, **newexecsw;
1689 	int count = 2;	/* New slot and trailing NULL */
1690 
1691 	if (execsw)
1692 		for (es = execsw; *es; es++)
1693 			count++;
1694 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1695 	xs = newexecsw;
1696 	if (execsw)
1697 		for (es = execsw; *es; es++)
1698 			*xs++ = *es;
1699 	*xs++ = execsw_arg;
1700 	*xs = NULL;
1701 	if (execsw)
1702 		free(execsw, M_TEMP);
1703 	execsw = newexecsw;
1704 	return (0);
1705 }
1706 
1707 int
1708 exec_unregister(const struct execsw *execsw_arg)
1709 {
1710 	const struct execsw **es, **xs, **newexecsw;
1711 	int count = 1;
1712 
1713 	if (execsw == NULL)
1714 		panic("unregister with no handlers left?\n");
1715 
1716 	for (es = execsw; *es; es++) {
1717 		if (*es == execsw_arg)
1718 			break;
1719 	}
1720 	if (*es == NULL)
1721 		return (ENOENT);
1722 	for (es = execsw; *es; es++)
1723 		if (*es != execsw_arg)
1724 			count++;
1725 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1726 	xs = newexecsw;
1727 	for (es = execsw; *es; es++)
1728 		if (*es != execsw_arg)
1729 			*xs++ = *es;
1730 	*xs = NULL;
1731 	if (execsw)
1732 		free(execsw, M_TEMP);
1733 	execsw = newexecsw;
1734 	return (0);
1735 }
1736