xref: /freebsd/sys/kern/kern_exec.c (revision 63d1fd5970ec814904aa0f4580b10a0d302d08b2)
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(td, args, mac_p)
357 	struct thread *td;
358 	struct image_args *args;
359 	struct mac *mac_p;
360 {
361 	struct proc *p = td->td_proc;
362 	struct nameidata nd;
363 	struct ucred *oldcred;
364 	struct uidinfo *euip = NULL;
365 	register_t *stack_base;
366 	int error, i;
367 	struct image_params image_params, *imgp;
368 	struct vattr attr;
369 	int (*img_first)(struct image_params *);
370 	struct pargs *oldargs = NULL, *newargs = NULL;
371 	struct sigacts *oldsigacts = NULL, *newsigacts = NULL;
372 #ifdef KTRACE
373 	struct vnode *tracevp = NULL;
374 	struct ucred *tracecred = NULL;
375 #endif
376 	struct vnode *oldtextvp = NULL, *newtextvp;
377 	cap_rights_t rights;
378 	int credential_changing;
379 	int textset;
380 #ifdef MAC
381 	struct label *interpvplabel = NULL;
382 	int will_transition;
383 #endif
384 #ifdef HWPMC_HOOKS
385 	struct pmckern_procexec pe;
386 #endif
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 
413 #ifdef MAC
414 	error = mac_execve_enter(imgp, mac_p);
415 	if (error)
416 		goto exec_fail;
417 #endif
418 
419 	/*
420 	 * Translate the file name. namei() returns a vnode pointer
421 	 *	in ni_vp among other things.
422 	 *
423 	 * XXXAUDIT: It would be desirable to also audit the name of the
424 	 * interpreter if this is an interpreted binary.
425 	 */
426 	if (args->fname != NULL) {
427 		NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME
428 		    | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
429 	}
430 
431 	SDT_PROBE1(proc, , , exec, args->fname);
432 
433 interpret:
434 	if (args->fname != NULL) {
435 #ifdef CAPABILITY_MODE
436 		/*
437 		 * While capability mode can't reach this point via direct
438 		 * path arguments to execve(), we also don't allow
439 		 * interpreters to be used in capability mode (for now).
440 		 * Catch indirect lookups and return a permissions error.
441 		 */
442 		if (IN_CAPABILITY_MODE(td)) {
443 			error = ECAPMODE;
444 			goto exec_fail;
445 		}
446 #endif
447 		error = namei(&nd);
448 		if (error)
449 			goto exec_fail;
450 
451 		newtextvp = nd.ni_vp;
452 		imgp->vp = newtextvp;
453 	} else {
454 		AUDIT_ARG_FD(args->fd);
455 		/*
456 		 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
457 		 */
458 		error = fgetvp_exec(td, args->fd,
459 		    cap_rights_init(&rights, CAP_FEXECVE), &newtextvp);
460 		if (error)
461 			goto exec_fail;
462 		vn_lock(newtextvp, LK_EXCLUSIVE | LK_RETRY);
463 		AUDIT_ARG_VNODE1(newtextvp);
464 		imgp->vp = newtextvp;
465 	}
466 
467 	/*
468 	 * Check file permissions (also 'opens' file)
469 	 */
470 	error = exec_check_permissions(imgp);
471 	if (error)
472 		goto exec_fail_dealloc;
473 
474 	imgp->object = imgp->vp->v_object;
475 	if (imgp->object != NULL)
476 		vm_object_reference(imgp->object);
477 
478 	/*
479 	 * Set VV_TEXT now so no one can write to the executable while we're
480 	 * activating it.
481 	 *
482 	 * Remember if this was set before and unset it in case this is not
483 	 * actually an executable image.
484 	 */
485 	textset = VOP_IS_TEXT(imgp->vp);
486 	VOP_SET_TEXT(imgp->vp);
487 
488 	error = exec_map_first_page(imgp);
489 	if (error)
490 		goto exec_fail_dealloc;
491 
492 	imgp->proc->p_osrel = 0;
493 
494 	/*
495 	 * Implement image setuid/setgid.
496 	 *
497 	 * Determine new credentials before attempting image activators
498 	 * so that it can be used by process_exec handlers to determine
499 	 * credential/setid changes.
500 	 *
501 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
502 	 * the process is being traced.
503 	 *
504 	 * We disable setuid/setgid/etc in capability mode on the basis
505 	 * that most setugid applications are not written with that
506 	 * environment in mind, and will therefore almost certainly operate
507 	 * incorrectly. In principle there's no reason that setugid
508 	 * applications might not be useful in capability mode, so we may want
509 	 * to reconsider this conservative design choice in the future.
510 	 *
511 	 * XXXMAC: For the time being, use NOSUID to also prohibit
512 	 * transitions on the file system.
513 	 */
514 	credential_changing = 0;
515 	credential_changing |= (attr.va_mode & S_ISUID) &&
516 	    oldcred->cr_uid != attr.va_uid;
517 	credential_changing |= (attr.va_mode & S_ISGID) &&
518 	    oldcred->cr_gid != attr.va_gid;
519 #ifdef MAC
520 	will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
521 	    interpvplabel, imgp);
522 	credential_changing |= will_transition;
523 #endif
524 
525 	if (credential_changing &&
526 #ifdef CAPABILITY_MODE
527 	    ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
528 #endif
529 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
530 	    (p->p_flag & P_TRACED) == 0) {
531 		imgp->credential_setid = true;
532 		VOP_UNLOCK(imgp->vp, 0);
533 		imgp->newcred = crdup(oldcred);
534 		if (attr.va_mode & S_ISUID) {
535 			euip = uifind(attr.va_uid);
536 			change_euid(imgp->newcred, euip);
537 		}
538 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
539 		if (attr.va_mode & S_ISGID)
540 			change_egid(imgp->newcred, attr.va_gid);
541 		/*
542 		 * Implement correct POSIX saved-id behavior.
543 		 *
544 		 * XXXMAC: Note that the current logic will save the
545 		 * uid and gid if a MAC domain transition occurs, even
546 		 * though maybe it shouldn't.
547 		 */
548 		change_svuid(imgp->newcred, imgp->newcred->cr_uid);
549 		change_svgid(imgp->newcred, imgp->newcred->cr_gid);
550 	} else {
551 		/*
552 		 * Implement correct POSIX saved-id behavior.
553 		 *
554 		 * XXX: It's not clear that the existing behavior is
555 		 * POSIX-compliant.  A number of sources indicate that the
556 		 * saved uid/gid should only be updated if the new ruid is
557 		 * not equal to the old ruid, or the new euid is not equal
558 		 * to the old euid and the new euid is not equal to the old
559 		 * ruid.  The FreeBSD code always updates the saved uid/gid.
560 		 * Also, this code uses the new (replaced) euid and egid as
561 		 * the source, which may or may not be the right ones to use.
562 		 */
563 		if (oldcred->cr_svuid != oldcred->cr_uid ||
564 		    oldcred->cr_svgid != oldcred->cr_gid) {
565 			VOP_UNLOCK(imgp->vp, 0);
566 			imgp->newcred = crdup(oldcred);
567 			vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
568 			change_svuid(imgp->newcred, imgp->newcred->cr_uid);
569 			change_svgid(imgp->newcred, imgp->newcred->cr_gid);
570 		}
571 	}
572 	/* The new credentials are installed into the process later. */
573 
574 	/*
575 	 * Do the best to calculate the full path to the image file.
576 	 */
577 	if (args->fname != NULL && args->fname[0] == '/')
578 		imgp->execpath = args->fname;
579 	else {
580 		VOP_UNLOCK(imgp->vp, 0);
581 		if (vn_fullpath(td, imgp->vp, &imgp->execpath,
582 		    &imgp->freepath) != 0)
583 			imgp->execpath = args->fname;
584 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
585 	}
586 
587 	/*
588 	 *	If the current process has a special image activator it
589 	 *	wants to try first, call it.   For example, emulating shell
590 	 *	scripts differently.
591 	 */
592 	error = -1;
593 	if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
594 		error = img_first(imgp);
595 
596 	/*
597 	 *	Loop through the list of image activators, calling each one.
598 	 *	An activator returns -1 if there is no match, 0 on success,
599 	 *	and an error otherwise.
600 	 */
601 	for (i = 0; error == -1 && execsw[i]; ++i) {
602 		if (execsw[i]->ex_imgact == NULL ||
603 		    execsw[i]->ex_imgact == img_first) {
604 			continue;
605 		}
606 		error = (*execsw[i]->ex_imgact)(imgp);
607 	}
608 
609 	if (error) {
610 		if (error == -1) {
611 			if (textset == 0)
612 				VOP_UNSET_TEXT(imgp->vp);
613 			error = ENOEXEC;
614 		}
615 		goto exec_fail_dealloc;
616 	}
617 
618 	/*
619 	 * Special interpreter operation, cleanup and loop up to try to
620 	 * activate the interpreter.
621 	 */
622 	if (imgp->interpreted) {
623 		exec_unmap_first_page(imgp);
624 		/*
625 		 * VV_TEXT needs to be unset for scripts.  There is a short
626 		 * period before we determine that something is a script where
627 		 * VV_TEXT will be set. The vnode lock is held over this
628 		 * entire period so nothing should illegitimately be blocked.
629 		 */
630 		VOP_UNSET_TEXT(imgp->vp);
631 		/* free name buffer and old vnode */
632 		if (args->fname != NULL)
633 			NDFREE(&nd, NDF_ONLY_PNBUF);
634 #ifdef MAC
635 		mac_execve_interpreter_enter(newtextvp, &interpvplabel);
636 #endif
637 		if (imgp->opened) {
638 			VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td);
639 			imgp->opened = 0;
640 		}
641 		vput(newtextvp);
642 		vm_object_deallocate(imgp->object);
643 		imgp->object = NULL;
644 		imgp->credential_setid = false;
645 		if (imgp->newcred != NULL) {
646 			crfree(imgp->newcred);
647 			imgp->newcred = NULL;
648 		}
649 		imgp->execpath = NULL;
650 		free(imgp->freepath, M_TEMP);
651 		imgp->freepath = NULL;
652 		/* set new name to that of the interpreter */
653 		NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
654 		    UIO_SYSSPACE, imgp->interpreter_name, td);
655 		args->fname = imgp->interpreter_name;
656 		goto interpret;
657 	}
658 
659 	/*
660 	 * NB: We unlock the vnode here because it is believed that none
661 	 * of the sv_copyout_strings/sv_fixup operations require the vnode.
662 	 */
663 	VOP_UNLOCK(imgp->vp, 0);
664 
665 	if (disallow_high_osrel &&
666 	    P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
667 		error = ENOEXEC;
668 		uprintf("Osrel %d for image %s too high\n", p->p_osrel,
669 		    imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
670 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
671 		goto exec_fail_dealloc;
672 	}
673 
674 	/* ABI enforces the use of Capsicum. Switch into capabilities mode. */
675 	if (SV_PROC_FLAG(p, SV_CAPSICUM))
676 		sys_cap_enter(td, NULL);
677 
678 	/*
679 	 * Copy out strings (args and env) and initialize stack base
680 	 */
681 	if (p->p_sysent->sv_copyout_strings)
682 		stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
683 	else
684 		stack_base = exec_copyout_strings(imgp);
685 
686 	/*
687 	 * If custom stack fixup routine present for this process
688 	 * let it do the stack setup.
689 	 * Else stuff argument count as first item on stack
690 	 */
691 	if (p->p_sysent->sv_fixup != NULL)
692 		(*p->p_sysent->sv_fixup)(&stack_base, imgp);
693 	else
694 		suword(--stack_base, imgp->args->argc);
695 
696 	if (args->fdp != NULL) {
697 		/* Install a brand new file descriptor table. */
698 		fdinstall_remapped(td, args->fdp);
699 		args->fdp = NULL;
700 	} else {
701 		/*
702 		 * Keep on using the existing file descriptor table. For
703 		 * security and other reasons, the file descriptor table
704 		 * cannot be shared after an exec.
705 		 */
706 		fdunshare(td);
707 		/* close files on exec */
708 		fdcloseexec(td);
709 	}
710 
711 	/*
712 	 * Malloc things before we need locks.
713 	 */
714 	i = imgp->args->begin_envv - imgp->args->begin_argv;
715 	/* Cache arguments if they fit inside our allowance */
716 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
717 		newargs = pargs_alloc(i);
718 		bcopy(imgp->args->begin_argv, newargs->ar_args, i);
719 	}
720 
721 	/*
722 	 * For security and other reasons, signal handlers cannot
723 	 * be shared after an exec. The new process gets a copy of the old
724 	 * handlers. In execsigs(), the new process will have its signals
725 	 * reset.
726 	 */
727 	if (sigacts_shared(p->p_sigacts)) {
728 		oldsigacts = p->p_sigacts;
729 		newsigacts = sigacts_alloc();
730 		sigacts_copy(newsigacts, oldsigacts);
731 	}
732 
733 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
734 
735 	PROC_LOCK(p);
736 	if (oldsigacts)
737 		p->p_sigacts = newsigacts;
738 	/* Stop profiling */
739 	stopprofclock(p);
740 
741 	/* reset caught signals */
742 	execsigs(p);
743 
744 	/* name this process - nameiexec(p, ndp) */
745 	bzero(p->p_comm, sizeof(p->p_comm));
746 	if (args->fname)
747 		bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
748 		    min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
749 	else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0)
750 		bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
751 	bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
752 #ifdef KTR
753 	sched_clear_tdname(td);
754 #endif
755 
756 	/*
757 	 * mark as execed, wakeup the process that vforked (if any) and tell
758 	 * it that it now has its own resources back
759 	 */
760 	p->p_flag |= P_EXEC;
761 	if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0)
762 		p->p_flag2 &= ~P2_NOTRACE;
763 	if (p->p_flag & P_PPWAIT) {
764 		p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
765 		cv_broadcast(&p->p_pwait);
766 		/* STOPs are no longer ignored, arrange for AST */
767 		signotify(td);
768 	}
769 
770 	/*
771 	 * Implement image setuid/setgid installation.
772 	 */
773 	if (imgp->credential_setid) {
774 		/*
775 		 * Turn off syscall tracing for set-id programs, except for
776 		 * root.  Record any set-id flags first to make sure that
777 		 * we do not regain any tracing during a possible block.
778 		 */
779 		setsugid(p);
780 
781 #ifdef KTRACE
782 		if (p->p_tracecred != NULL &&
783 		    priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED, 0))
784 			ktrprocexec(p, &tracecred, &tracevp);
785 #endif
786 		/*
787 		 * Close any file descriptors 0..2 that reference procfs,
788 		 * then make sure file descriptors 0..2 are in use.
789 		 *
790 		 * Both fdsetugidsafety() and fdcheckstd() may call functions
791 		 * taking sleepable locks, so temporarily drop our locks.
792 		 */
793 		PROC_UNLOCK(p);
794 		VOP_UNLOCK(imgp->vp, 0);
795 		fdsetugidsafety(td);
796 		error = fdcheckstd(td);
797 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
798 		if (error != 0)
799 			goto exec_fail_dealloc;
800 		PROC_LOCK(p);
801 #ifdef MAC
802 		if (will_transition) {
803 			mac_vnode_execve_transition(oldcred, imgp->newcred,
804 			    imgp->vp, interpvplabel, imgp);
805 		}
806 #endif
807 	} else {
808 		if (oldcred->cr_uid == oldcred->cr_ruid &&
809 		    oldcred->cr_gid == oldcred->cr_rgid)
810 			p->p_flag &= ~P_SUGID;
811 	}
812 	/*
813 	 * Set the new credentials.
814 	 */
815 	if (imgp->newcred != NULL) {
816 		proc_set_cred(p, imgp->newcred);
817 		crfree(oldcred);
818 		oldcred = NULL;
819 	}
820 
821 	/*
822 	 * Store the vp for use in procfs.  This vnode was referenced by namei
823 	 * or fgetvp_exec.
824 	 */
825 	oldtextvp = p->p_textvp;
826 	p->p_textvp = newtextvp;
827 
828 #ifdef KDTRACE_HOOKS
829 	/*
830 	 * Tell the DTrace fasttrap provider about the exec if it
831 	 * has declared an interest.
832 	 */
833 	if (dtrace_fasttrap_exec)
834 		dtrace_fasttrap_exec(p);
835 #endif
836 
837 	/*
838 	 * Notify others that we exec'd, and clear the P_INEXEC flag
839 	 * as we're now a bona fide freshly-execed process.
840 	 */
841 	KNOTE_LOCKED(p->p_klist, NOTE_EXEC);
842 	p->p_flag &= ~P_INEXEC;
843 
844 	/* clear "fork but no exec" flag, as we _are_ execing */
845 	p->p_acflag &= ~AFORK;
846 
847 	/*
848 	 * Free any previous argument cache and replace it with
849 	 * the new argument cache, if any.
850 	 */
851 	oldargs = p->p_args;
852 	p->p_args = newargs;
853 	newargs = NULL;
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 	 * The proc lock needs to be released before taking the PMC
863 	 * SX.
864 	 */
865 	if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
866 		PROC_UNLOCK(p);
867 		VOP_UNLOCK(imgp->vp, 0);
868 		pe.pm_credentialschanged = credential_changing;
869 		pe.pm_entryaddr = imgp->entry_addr;
870 
871 		PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
872 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
873 	} else
874 		PROC_UNLOCK(p);
875 #else  /* !HWPMC_HOOKS */
876 	PROC_UNLOCK(p);
877 #endif
878 
879 	/* Set values passed into the program in registers. */
880 	if (p->p_sysent->sv_setregs)
881 		(*p->p_sysent->sv_setregs)(td, imgp,
882 		    (u_long)(uintptr_t)stack_base);
883 	else
884 		exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base);
885 
886 	vfs_mark_atime(imgp->vp, td->td_ucred);
887 
888 	SDT_PROBE1(proc, , , exec__success, args->fname);
889 
890 exec_fail_dealloc:
891 	if (imgp->firstpage != NULL)
892 		exec_unmap_first_page(imgp);
893 
894 	if (imgp->vp != NULL) {
895 		if (args->fname)
896 			NDFREE(&nd, NDF_ONLY_PNBUF);
897 		if (imgp->opened)
898 			VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
899 		if (error != 0)
900 			vput(imgp->vp);
901 		else
902 			VOP_UNLOCK(imgp->vp, 0);
903 	}
904 
905 	if (imgp->object != NULL)
906 		vm_object_deallocate(imgp->object);
907 
908 	free(imgp->freepath, M_TEMP);
909 
910 	if (error == 0) {
911 		PROC_LOCK(p);
912 		if (p->p_ptevents & PTRACE_EXEC)
913 			td->td_dbgflags |= TDB_EXEC;
914 		PROC_UNLOCK(p);
915 
916 		/*
917 		 * Stop the process here if its stop event mask has
918 		 * the S_EXEC bit set.
919 		 */
920 		STOPEVENT(p, S_EXEC, 0);
921 	} else {
922 exec_fail:
923 		/* we're done here, clear P_INEXEC */
924 		PROC_LOCK(p);
925 		p->p_flag &= ~P_INEXEC;
926 		PROC_UNLOCK(p);
927 
928 		SDT_PROBE1(proc, , , exec__failure, error);
929 	}
930 
931 	if (imgp->newcred != NULL && oldcred != NULL)
932 		crfree(imgp->newcred);
933 
934 #ifdef MAC
935 	mac_execve_exit(imgp);
936 	mac_execve_interpreter_exit(interpvplabel);
937 #endif
938 	exec_free_args(args);
939 
940 	/*
941 	 * Handle deferred decrement of ref counts.
942 	 */
943 	if (oldtextvp != NULL)
944 		vrele(oldtextvp);
945 #ifdef KTRACE
946 	if (tracevp != NULL)
947 		vrele(tracevp);
948 	if (tracecred != NULL)
949 		crfree(tracecred);
950 #endif
951 	pargs_drop(oldargs);
952 	pargs_drop(newargs);
953 	if (oldsigacts != NULL)
954 		sigacts_free(oldsigacts);
955 	if (euip != NULL)
956 		uifree(euip);
957 
958 	if (error && imgp->vmspace_destroyed) {
959 		/* sorry, no more process anymore. exit gracefully */
960 		exit1(td, 0, SIGABRT);
961 		/* NOT REACHED */
962 	}
963 
964 #ifdef KTRACE
965 	if (error == 0)
966 		ktrprocctor(p);
967 #endif
968 
969 	return (error);
970 }
971 
972 int
973 exec_map_first_page(imgp)
974 	struct image_params *imgp;
975 {
976 	int rv, i, after, initial_pagein;
977 	vm_page_t ma[VM_INITIAL_PAGEIN];
978 	vm_object_t object;
979 
980 	if (imgp->firstpage != NULL)
981 		exec_unmap_first_page(imgp);
982 
983 	object = imgp->vp->v_object;
984 	if (object == NULL)
985 		return (EACCES);
986 	VM_OBJECT_WLOCK(object);
987 #if VM_NRESERVLEVEL > 0
988 	vm_object_color(object, 0);
989 #endif
990 	ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY);
991 	if (ma[0]->valid != VM_PAGE_BITS_ALL) {
992 		vm_page_xbusy(ma[0]);
993 		if (!vm_pager_has_page(object, 0, NULL, &after)) {
994 			vm_page_lock(ma[0]);
995 			vm_page_free(ma[0]);
996 			vm_page_unlock(ma[0]);
997 			VM_OBJECT_WUNLOCK(object);
998 			return (EIO);
999 		}
1000 		initial_pagein = min(after, VM_INITIAL_PAGEIN);
1001 		KASSERT(initial_pagein <= object->size,
1002 		    ("%s: initial_pagein %d object->size %ju",
1003 		    __func__, initial_pagein, (uintmax_t )object->size));
1004 		for (i = 1; i < initial_pagein; i++) {
1005 			if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) {
1006 				if (ma[i]->valid)
1007 					break;
1008 				if (vm_page_tryxbusy(ma[i]))
1009 					break;
1010 			} else {
1011 				ma[i] = vm_page_alloc(object, i,
1012 				    VM_ALLOC_NORMAL);
1013 				if (ma[i] == NULL)
1014 					break;
1015 			}
1016 		}
1017 		initial_pagein = i;
1018 		rv = vm_pager_get_pages(object, ma, initial_pagein, NULL, NULL);
1019 		if (rv != VM_PAGER_OK) {
1020 			for (i = 0; i < initial_pagein; i++) {
1021 				vm_page_lock(ma[i]);
1022 				vm_page_free(ma[i]);
1023 				vm_page_unlock(ma[i]);
1024 			}
1025 			VM_OBJECT_WUNLOCK(object);
1026 			return (EIO);
1027 		}
1028 		vm_page_xunbusy(ma[0]);
1029 		for (i = 1; i < initial_pagein; i++)
1030 			vm_page_readahead_finish(ma[i]);
1031 	}
1032 	vm_page_lock(ma[0]);
1033 	vm_page_hold(ma[0]);
1034 	vm_page_activate(ma[0]);
1035 	vm_page_unlock(ma[0]);
1036 	VM_OBJECT_WUNLOCK(object);
1037 
1038 	imgp->firstpage = sf_buf_alloc(ma[0], 0);
1039 	imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
1040 
1041 	return (0);
1042 }
1043 
1044 void
1045 exec_unmap_first_page(imgp)
1046 	struct image_params *imgp;
1047 {
1048 	vm_page_t m;
1049 
1050 	if (imgp->firstpage != NULL) {
1051 		m = sf_buf_page(imgp->firstpage);
1052 		sf_buf_free(imgp->firstpage);
1053 		imgp->firstpage = NULL;
1054 		vm_page_lock(m);
1055 		vm_page_unhold(m);
1056 		vm_page_unlock(m);
1057 	}
1058 }
1059 
1060 /*
1061  * Destroy old address space, and allocate a new stack
1062  *	The new stack is only SGROWSIZ large because it is grown
1063  *	automatically in trap.c.
1064  */
1065 int
1066 exec_new_vmspace(imgp, sv)
1067 	struct image_params *imgp;
1068 	struct sysentvec *sv;
1069 {
1070 	int error;
1071 	struct proc *p = imgp->proc;
1072 	struct vmspace *vmspace = p->p_vmspace;
1073 	vm_object_t obj;
1074 	struct rlimit rlim_stack;
1075 	vm_offset_t sv_minuser, stack_addr;
1076 	vm_map_t map;
1077 	u_long ssiz;
1078 
1079 	imgp->vmspace_destroyed = 1;
1080 	imgp->sysent = sv;
1081 
1082 	/* May be called with Giant held */
1083 	EVENTHANDLER_INVOKE(process_exec, p, imgp);
1084 
1085 	/*
1086 	 * Blow away entire process VM, if address space not shared,
1087 	 * otherwise, create a new VM space so that other threads are
1088 	 * not disrupted
1089 	 */
1090 	map = &vmspace->vm_map;
1091 	if (map_at_zero)
1092 		sv_minuser = sv->sv_minuser;
1093 	else
1094 		sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1095 	if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser &&
1096 	    vm_map_max(map) == sv->sv_maxuser) {
1097 		shmexit(vmspace);
1098 		pmap_remove_pages(vmspace_pmap(vmspace));
1099 		vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1100 	} else {
1101 		error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1102 		if (error)
1103 			return (error);
1104 		vmspace = p->p_vmspace;
1105 		map = &vmspace->vm_map;
1106 	}
1107 
1108 	/* Map a shared page */
1109 	obj = sv->sv_shared_page_obj;
1110 	if (obj != NULL) {
1111 		vm_object_reference(obj);
1112 		error = vm_map_fixed(map, obj, 0,
1113 		    sv->sv_shared_page_base, sv->sv_shared_page_len,
1114 		    VM_PROT_READ | VM_PROT_EXECUTE,
1115 		    VM_PROT_READ | VM_PROT_EXECUTE,
1116 		    MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1117 		if (error) {
1118 			vm_object_deallocate(obj);
1119 			return (error);
1120 		}
1121 	}
1122 
1123 	/* Allocate a new stack */
1124 	if (imgp->stack_sz != 0) {
1125 		ssiz = trunc_page(imgp->stack_sz);
1126 		PROC_LOCK(p);
1127 		lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack);
1128 		PROC_UNLOCK(p);
1129 		if (ssiz > rlim_stack.rlim_max)
1130 			ssiz = rlim_stack.rlim_max;
1131 		if (ssiz > rlim_stack.rlim_cur) {
1132 			rlim_stack.rlim_cur = ssiz;
1133 			kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack);
1134 		}
1135 	} else if (sv->sv_maxssiz != NULL) {
1136 		ssiz = *sv->sv_maxssiz;
1137 	} else {
1138 		ssiz = maxssiz;
1139 	}
1140 	stack_addr = sv->sv_usrstack - ssiz;
1141 	error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1142 	    obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1143 		sv->sv_stackprot,
1144 	    VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1145 	if (error)
1146 		return (error);
1147 
1148 	/*
1149 	 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they
1150 	 * are still used to enforce the stack rlimit on the process stack.
1151 	 */
1152 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1153 	vmspace->vm_maxsaddr = (char *)stack_addr;
1154 
1155 	return (0);
1156 }
1157 
1158 /*
1159  * Copy out argument and environment strings from the old process address
1160  * space into the temporary string buffer.
1161  */
1162 int
1163 exec_copyin_args(struct image_args *args, char *fname,
1164     enum uio_seg segflg, char **argv, char **envv)
1165 {
1166 	u_long argp, envp;
1167 	int error;
1168 	size_t length;
1169 
1170 	bzero(args, sizeof(*args));
1171 	if (argv == NULL)
1172 		return (EFAULT);
1173 
1174 	/*
1175 	 * Allocate demand-paged memory for the file name, argument, and
1176 	 * environment strings.
1177 	 */
1178 	error = exec_alloc_args(args);
1179 	if (error != 0)
1180 		return (error);
1181 
1182 	/*
1183 	 * Copy the file name.
1184 	 */
1185 	if (fname != NULL) {
1186 		args->fname = args->buf;
1187 		error = (segflg == UIO_SYSSPACE) ?
1188 		    copystr(fname, args->fname, PATH_MAX, &length) :
1189 		    copyinstr(fname, args->fname, PATH_MAX, &length);
1190 		if (error != 0)
1191 			goto err_exit;
1192 	} else
1193 		length = 0;
1194 
1195 	args->begin_argv = args->buf + length;
1196 	args->endp = args->begin_argv;
1197 	args->stringspace = ARG_MAX;
1198 
1199 	/*
1200 	 * extract arguments first
1201 	 */
1202 	for (;;) {
1203 		error = fueword(argv++, &argp);
1204 		if (error == -1) {
1205 			error = EFAULT;
1206 			goto err_exit;
1207 		}
1208 		if (argp == 0)
1209 			break;
1210 		error = copyinstr((void *)(uintptr_t)argp, args->endp,
1211 		    args->stringspace, &length);
1212 		if (error != 0) {
1213 			if (error == ENAMETOOLONG)
1214 				error = E2BIG;
1215 			goto err_exit;
1216 		}
1217 		args->stringspace -= length;
1218 		args->endp += length;
1219 		args->argc++;
1220 	}
1221 
1222 	args->begin_envv = args->endp;
1223 
1224 	/*
1225 	 * extract environment strings
1226 	 */
1227 	if (envv) {
1228 		for (;;) {
1229 			error = fueword(envv++, &envp);
1230 			if (error == -1) {
1231 				error = EFAULT;
1232 				goto err_exit;
1233 			}
1234 			if (envp == 0)
1235 				break;
1236 			error = copyinstr((void *)(uintptr_t)envp,
1237 			    args->endp, args->stringspace, &length);
1238 			if (error != 0) {
1239 				if (error == ENAMETOOLONG)
1240 					error = E2BIG;
1241 				goto err_exit;
1242 			}
1243 			args->stringspace -= length;
1244 			args->endp += length;
1245 			args->envc++;
1246 		}
1247 	}
1248 
1249 	return (0);
1250 
1251 err_exit:
1252 	exec_free_args(args);
1253 	return (error);
1254 }
1255 
1256 int
1257 exec_copyin_data_fds(struct thread *td, struct image_args *args,
1258     const void *data, size_t datalen, const int *fds, size_t fdslen)
1259 {
1260 	struct filedesc *ofdp;
1261 	const char *p;
1262 	int *kfds;
1263 	int error;
1264 
1265 	memset(args, '\0', sizeof(*args));
1266 	ofdp = td->td_proc->p_fd;
1267 	if (datalen >= ARG_MAX || fdslen > ofdp->fd_lastfile + 1)
1268 		return (E2BIG);
1269 	error = exec_alloc_args(args);
1270 	if (error != 0)
1271 		return (error);
1272 
1273 	args->begin_argv = args->buf;
1274 	args->stringspace = ARG_MAX;
1275 
1276 	if (datalen > 0) {
1277 		/*
1278 		 * Argument buffer has been provided. Copy it into the
1279 		 * kernel as a single string and add a terminating null
1280 		 * byte.
1281 		 */
1282 		error = copyin(data, args->begin_argv, datalen);
1283 		if (error != 0)
1284 			goto err_exit;
1285 		args->begin_argv[datalen] = '\0';
1286 		args->endp = args->begin_argv + datalen + 1;
1287 		args->stringspace -= datalen + 1;
1288 
1289 		/*
1290 		 * Traditional argument counting. Count the number of
1291 		 * null bytes.
1292 		 */
1293 		for (p = args->begin_argv; p < args->endp; ++p)
1294 			if (*p == '\0')
1295 				++args->argc;
1296 	} else {
1297 		/* No argument buffer provided. */
1298 		args->endp = args->begin_argv;
1299 	}
1300 	/* There are no environment variables. */
1301 	args->begin_envv = args->endp;
1302 
1303 	/* Create new file descriptor table. */
1304 	kfds = malloc(fdslen * sizeof(int), M_TEMP, M_WAITOK);
1305 	error = copyin(fds, kfds, fdslen * sizeof(int));
1306 	if (error != 0) {
1307 		free(kfds, M_TEMP);
1308 		goto err_exit;
1309 	}
1310 	error = fdcopy_remapped(ofdp, kfds, fdslen, &args->fdp);
1311 	free(kfds, M_TEMP);
1312 	if (error != 0)
1313 		goto err_exit;
1314 
1315 	return (0);
1316 err_exit:
1317 	exec_free_args(args);
1318 	return (error);
1319 }
1320 
1321 struct exec_args_kva {
1322 	vm_offset_t addr;
1323 	SLIST_ENTRY(exec_args_kva) next;
1324 };
1325 
1326 static DPCPU_DEFINE(struct exec_args_kva *, exec_args_kva);
1327 
1328 static SLIST_HEAD(, exec_args_kva) exec_args_kva_freelist;
1329 static struct mtx exec_args_kva_mtx;
1330 
1331 static void
1332 exec_prealloc_args_kva(void *arg __unused)
1333 {
1334 	struct exec_args_kva *argkva;
1335 	u_int i;
1336 
1337 	SLIST_INIT(&exec_args_kva_freelist);
1338 	mtx_init(&exec_args_kva_mtx, "exec args kva", NULL, MTX_DEF);
1339 	for (i = 0; i < exec_map_entries; i++) {
1340 		argkva = malloc(sizeof(*argkva), M_PARGS, M_WAITOK);
1341 		argkva->addr = kmap_alloc_wait(exec_map, exec_map_entry_size);
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_free_args_kva(void *cookie)
1368 {
1369 	struct exec_args_kva *argkva;
1370 	vm_offset_t base;
1371 
1372 	argkva = cookie;
1373 	base = argkva->addr;
1374 
1375 	vm_map_madvise(exec_map, base, base + exec_map_entry_size, MADV_FREE);
1376 	if (!atomic_cmpset_ptr((uintptr_t *)DPCPU_PTR(exec_args_kva),
1377 	    (uintptr_t)NULL, (uintptr_t)argkva)) {
1378 		mtx_lock(&exec_args_kva_mtx);
1379 		SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1380 		wakeup_one(&exec_args_kva_freelist);
1381 		mtx_unlock(&exec_args_kva_mtx);
1382 	}
1383 }
1384 
1385 /*
1386  * Allocate temporary demand-paged, zero-filled memory for the file name,
1387  * argument, and environment strings.
1388  */
1389 int
1390 exec_alloc_args(struct image_args *args)
1391 {
1392 
1393 	args->buf = (char *)exec_alloc_args_kva(&args->bufkva);
1394 	return (0);
1395 }
1396 
1397 void
1398 exec_free_args(struct image_args *args)
1399 {
1400 
1401 	if (args->buf != NULL) {
1402 		exec_free_args_kva(args->bufkva);
1403 		args->buf = NULL;
1404 	}
1405 	if (args->fname_buf != NULL) {
1406 		free(args->fname_buf, M_TEMP);
1407 		args->fname_buf = NULL;
1408 	}
1409 	if (args->fdp != NULL)
1410 		fdescfree_remapped(args->fdp);
1411 }
1412 
1413 /*
1414  * Copy strings out to the new process address space, constructing new arg
1415  * and env vector tables. Return a pointer to the base so that it can be used
1416  * as the initial stack pointer.
1417  */
1418 register_t *
1419 exec_copyout_strings(imgp)
1420 	struct image_params *imgp;
1421 {
1422 	int argc, envc;
1423 	char **vectp;
1424 	char *stringp;
1425 	uintptr_t destp;
1426 	register_t *stack_base;
1427 	struct ps_strings *arginfo;
1428 	struct proc *p;
1429 	size_t execpath_len;
1430 	int szsigcode, szps;
1431 	char canary[sizeof(long) * 8];
1432 
1433 	szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1434 	/*
1435 	 * Calculate string base and vector table pointers.
1436 	 * Also deal with signal trampoline code for this exec type.
1437 	 */
1438 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
1439 		execpath_len = strlen(imgp->execpath) + 1;
1440 	else
1441 		execpath_len = 0;
1442 	p = imgp->proc;
1443 	szsigcode = 0;
1444 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1445 	if (p->p_sysent->sv_sigcode_base == 0) {
1446 		if (p->p_sysent->sv_szsigcode != NULL)
1447 			szsigcode = *(p->p_sysent->sv_szsigcode);
1448 	}
1449 	destp =	(uintptr_t)arginfo;
1450 
1451 	/*
1452 	 * install sigcode
1453 	 */
1454 	if (szsigcode != 0) {
1455 		destp -= szsigcode;
1456 		destp = rounddown2(destp, sizeof(void *));
1457 		copyout(p->p_sysent->sv_sigcode, (void *)destp, szsigcode);
1458 	}
1459 
1460 	/*
1461 	 * Copy the image path for the rtld.
1462 	 */
1463 	if (execpath_len != 0) {
1464 		destp -= execpath_len;
1465 		imgp->execpathp = destp;
1466 		copyout(imgp->execpath, (void *)destp, execpath_len);
1467 	}
1468 
1469 	/*
1470 	 * Prepare the canary for SSP.
1471 	 */
1472 	arc4rand(canary, sizeof(canary), 0);
1473 	destp -= sizeof(canary);
1474 	imgp->canary = destp;
1475 	copyout(canary, (void *)destp, sizeof(canary));
1476 	imgp->canarylen = sizeof(canary);
1477 
1478 	/*
1479 	 * Prepare the pagesizes array.
1480 	 */
1481 	destp -= szps;
1482 	destp = rounddown2(destp, sizeof(void *));
1483 	imgp->pagesizes = destp;
1484 	copyout(pagesizes, (void *)destp, szps);
1485 	imgp->pagesizeslen = szps;
1486 
1487 	destp -= ARG_MAX - imgp->args->stringspace;
1488 	destp = rounddown2(destp, sizeof(void *));
1489 
1490 	/*
1491 	 * If we have a valid auxargs ptr, prepare some room
1492 	 * on the stack.
1493 	 */
1494 	if (imgp->auxargs) {
1495 		/*
1496 		 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1497 		 * lower compatibility.
1498 		 */
1499 		imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1500 		    (AT_COUNT * 2);
1501 		/*
1502 		 * The '+ 2' is for the null pointers at the end of each of
1503 		 * the arg and env vector sets,and imgp->auxarg_size is room
1504 		 * for argument of Runtime loader.
1505 		 */
1506 		vectp = (char **)(destp - (imgp->args->argc +
1507 		    imgp->args->envc + 2 + imgp->auxarg_size)
1508 		    * sizeof(char *));
1509 	} else {
1510 		/*
1511 		 * The '+ 2' is for the null pointers at the end of each of
1512 		 * the arg and env vector sets
1513 		 */
1514 		vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc
1515 		    + 2) * sizeof(char *));
1516 	}
1517 
1518 	/*
1519 	 * vectp also becomes our initial stack base
1520 	 */
1521 	stack_base = (register_t *)vectp;
1522 
1523 	stringp = imgp->args->begin_argv;
1524 	argc = imgp->args->argc;
1525 	envc = imgp->args->envc;
1526 
1527 	/*
1528 	 * Copy out strings - arguments and environment.
1529 	 */
1530 	copyout(stringp, (void *)destp, ARG_MAX - imgp->args->stringspace);
1531 
1532 	/*
1533 	 * Fill in "ps_strings" struct for ps, w, etc.
1534 	 */
1535 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1536 	suword32(&arginfo->ps_nargvstr, argc);
1537 
1538 	/*
1539 	 * Fill in argument portion of vector table.
1540 	 */
1541 	for (; argc > 0; --argc) {
1542 		suword(vectp++, (long)(intptr_t)destp);
1543 		while (*stringp++ != 0)
1544 			destp++;
1545 		destp++;
1546 	}
1547 
1548 	/* a null vector table pointer separates the argp's from the envp's */
1549 	suword(vectp++, 0);
1550 
1551 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1552 	suword32(&arginfo->ps_nenvstr, envc);
1553 
1554 	/*
1555 	 * Fill in environment portion of vector table.
1556 	 */
1557 	for (; envc > 0; --envc) {
1558 		suword(vectp++, (long)(intptr_t)destp);
1559 		while (*stringp++ != 0)
1560 			destp++;
1561 		destp++;
1562 	}
1563 
1564 	/* end of vector table is a null pointer */
1565 	suword(vectp, 0);
1566 
1567 	return (stack_base);
1568 }
1569 
1570 /*
1571  * Check permissions of file to execute.
1572  *	Called with imgp->vp locked.
1573  *	Return 0 for success or error code on failure.
1574  */
1575 int
1576 exec_check_permissions(imgp)
1577 	struct image_params *imgp;
1578 {
1579 	struct vnode *vp = imgp->vp;
1580 	struct vattr *attr = imgp->attr;
1581 	struct thread *td;
1582 	int error, writecount;
1583 
1584 	td = curthread;
1585 
1586 	/* Get file attributes */
1587 	error = VOP_GETATTR(vp, attr, td->td_ucred);
1588 	if (error)
1589 		return (error);
1590 
1591 #ifdef MAC
1592 	error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1593 	if (error)
1594 		return (error);
1595 #endif
1596 
1597 	/*
1598 	 * 1) Check if file execution is disabled for the filesystem that
1599 	 *    this file resides on.
1600 	 * 2) Ensure that at least one execute bit is on. Otherwise, a
1601 	 *    privileged user will always succeed, and we don't want this
1602 	 *    to happen unless the file really is executable.
1603 	 * 3) Ensure that the file is a regular file.
1604 	 */
1605 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1606 	    (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1607 	    (attr->va_type != VREG))
1608 		return (EACCES);
1609 
1610 	/*
1611 	 * Zero length files can't be exec'd
1612 	 */
1613 	if (attr->va_size == 0)
1614 		return (ENOEXEC);
1615 
1616 	/*
1617 	 *  Check for execute permission to file based on current credentials.
1618 	 */
1619 	error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1620 	if (error)
1621 		return (error);
1622 
1623 	/*
1624 	 * Check number of open-for-writes on the file and deny execution
1625 	 * if there are any.
1626 	 */
1627 	error = VOP_GET_WRITECOUNT(vp, &writecount);
1628 	if (error != 0)
1629 		return (error);
1630 	if (writecount != 0)
1631 		return (ETXTBSY);
1632 
1633 	/*
1634 	 * Call filesystem specific open routine (which does nothing in the
1635 	 * general case).
1636 	 */
1637 	error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1638 	if (error == 0)
1639 		imgp->opened = 1;
1640 	return (error);
1641 }
1642 
1643 /*
1644  * Exec handler registration
1645  */
1646 int
1647 exec_register(execsw_arg)
1648 	const struct execsw *execsw_arg;
1649 {
1650 	const struct execsw **es, **xs, **newexecsw;
1651 	int count = 2;	/* New slot and trailing NULL */
1652 
1653 	if (execsw)
1654 		for (es = execsw; *es; es++)
1655 			count++;
1656 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1657 	xs = newexecsw;
1658 	if (execsw)
1659 		for (es = execsw; *es; es++)
1660 			*xs++ = *es;
1661 	*xs++ = execsw_arg;
1662 	*xs = NULL;
1663 	if (execsw)
1664 		free(execsw, M_TEMP);
1665 	execsw = newexecsw;
1666 	return (0);
1667 }
1668 
1669 int
1670 exec_unregister(execsw_arg)
1671 	const struct execsw *execsw_arg;
1672 {
1673 	const struct execsw **es, **xs, **newexecsw;
1674 	int count = 1;
1675 
1676 	if (execsw == NULL)
1677 		panic("unregister with no handlers left?\n");
1678 
1679 	for (es = execsw; *es; es++) {
1680 		if (*es == execsw_arg)
1681 			break;
1682 	}
1683 	if (*es == NULL)
1684 		return (ENOENT);
1685 	for (es = execsw; *es; es++)
1686 		if (*es != execsw_arg)
1687 			count++;
1688 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1689 	xs = newexecsw;
1690 	for (es = execsw; *es; es++)
1691 		if (*es != execsw_arg)
1692 			*xs++ = *es;
1693 	*xs = NULL;
1694 	if (execsw)
1695 		free(execsw, M_TEMP);
1696 	execsw = newexecsw;
1697 	return (0);
1698 }
1699