xref: /freebsd/sys/kern/kern_exec.c (revision 718cf2ccb9956613756ab15d7a0e28f2c8e91cab)
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 EVENTHANDLER_LIST_DECLARE(process_exec);
148 
149 static int
150 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
151 {
152 	struct proc *p;
153 	int error;
154 
155 	p = curproc;
156 #ifdef SCTL_MASK32
157 	if (req->flags & SCTL_MASK32) {
158 		unsigned int val;
159 		val = (unsigned int)p->p_sysent->sv_psstrings;
160 		error = SYSCTL_OUT(req, &val, sizeof(val));
161 	} else
162 #endif
163 		error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
164 		   sizeof(p->p_sysent->sv_psstrings));
165 	return error;
166 }
167 
168 static int
169 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
170 {
171 	struct proc *p;
172 	int error;
173 
174 	p = curproc;
175 #ifdef SCTL_MASK32
176 	if (req->flags & SCTL_MASK32) {
177 		unsigned int val;
178 		val = (unsigned int)p->p_sysent->sv_usrstack;
179 		error = SYSCTL_OUT(req, &val, sizeof(val));
180 	} else
181 #endif
182 		error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
183 		    sizeof(p->p_sysent->sv_usrstack));
184 	return error;
185 }
186 
187 static int
188 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
189 {
190 	struct proc *p;
191 
192 	p = curproc;
193 	return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
194 	    sizeof(p->p_sysent->sv_stackprot)));
195 }
196 
197 /*
198  * Each of the items is a pointer to a `const struct execsw', hence the
199  * double pointer here.
200  */
201 static const struct execsw **execsw;
202 
203 #ifndef _SYS_SYSPROTO_H_
204 struct execve_args {
205 	char    *fname;
206 	char    **argv;
207 	char    **envv;
208 };
209 #endif
210 
211 int
212 sys_execve(struct thread *td, struct execve_args *uap)
213 {
214 	struct image_args args;
215 	struct vmspace *oldvmspace;
216 	int error;
217 
218 	error = pre_execve(td, &oldvmspace);
219 	if (error != 0)
220 		return (error);
221 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
222 	    uap->argv, uap->envv);
223 	if (error == 0)
224 		error = kern_execve(td, &args, NULL);
225 	post_execve(td, error, oldvmspace);
226 	return (error);
227 }
228 
229 #ifndef _SYS_SYSPROTO_H_
230 struct fexecve_args {
231 	int	fd;
232 	char	**argv;
233 	char	**envv;
234 }
235 #endif
236 int
237 sys_fexecve(struct thread *td, struct fexecve_args *uap)
238 {
239 	struct image_args args;
240 	struct vmspace *oldvmspace;
241 	int error;
242 
243 	error = pre_execve(td, &oldvmspace);
244 	if (error != 0)
245 		return (error);
246 	error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
247 	    uap->argv, uap->envv);
248 	if (error == 0) {
249 		args.fd = uap->fd;
250 		error = kern_execve(td, &args, NULL);
251 	}
252 	post_execve(td, error, oldvmspace);
253 	return (error);
254 }
255 
256 #ifndef _SYS_SYSPROTO_H_
257 struct __mac_execve_args {
258 	char	*fname;
259 	char	**argv;
260 	char	**envv;
261 	struct mac	*mac_p;
262 };
263 #endif
264 
265 int
266 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap)
267 {
268 #ifdef MAC
269 	struct image_args args;
270 	struct vmspace *oldvmspace;
271 	int error;
272 
273 	error = pre_execve(td, &oldvmspace);
274 	if (error != 0)
275 		return (error);
276 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
277 	    uap->argv, uap->envv);
278 	if (error == 0)
279 		error = kern_execve(td, &args, uap->mac_p);
280 	post_execve(td, error, oldvmspace);
281 	return (error);
282 #else
283 	return (ENOSYS);
284 #endif
285 }
286 
287 int
288 pre_execve(struct thread *td, struct vmspace **oldvmspace)
289 {
290 	struct proc *p;
291 	int error;
292 
293 	KASSERT(td == curthread, ("non-current thread %p", td));
294 	error = 0;
295 	p = td->td_proc;
296 	if ((p->p_flag & P_HADTHREADS) != 0) {
297 		PROC_LOCK(p);
298 		if (thread_single(p, SINGLE_BOUNDARY) != 0)
299 			error = ERESTART;
300 		PROC_UNLOCK(p);
301 	}
302 	KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0,
303 	    ("nested execve"));
304 	*oldvmspace = p->p_vmspace;
305 	return (error);
306 }
307 
308 void
309 post_execve(struct thread *td, int error, struct vmspace *oldvmspace)
310 {
311 	struct proc *p;
312 
313 	KASSERT(td == curthread, ("non-current thread %p", td));
314 	p = td->td_proc;
315 	if ((p->p_flag & P_HADTHREADS) != 0) {
316 		PROC_LOCK(p);
317 		/*
318 		 * If success, we upgrade to SINGLE_EXIT state to
319 		 * force other threads to suicide.
320 		 */
321 		if (error == EJUSTRETURN)
322 			thread_single(p, SINGLE_EXIT);
323 		else
324 			thread_single_end(p, SINGLE_BOUNDARY);
325 		PROC_UNLOCK(p);
326 	}
327 	if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
328 		KASSERT(p->p_vmspace != oldvmspace,
329 		    ("oldvmspace still used"));
330 		vmspace_free(oldvmspace);
331 		td->td_pflags &= ~TDP_EXECVMSPC;
332 	}
333 }
334 
335 /*
336  * XXX: kern_execve has the astonishing property of not always returning to
337  * the caller.  If sufficiently bad things happen during the call to
338  * do_execve(), it can end up calling exit1(); as a result, callers must
339  * avoid doing anything which they might need to undo (e.g., allocating
340  * memory).
341  */
342 int
343 kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p)
344 {
345 
346 	AUDIT_ARG_ARGV(args->begin_argv, args->argc,
347 	    args->begin_envv - args->begin_argv);
348 	AUDIT_ARG_ENVV(args->begin_envv, args->envc,
349 	    args->endp - args->begin_envv);
350 	return (do_execve(td, args, mac_p));
351 }
352 
353 /*
354  * In-kernel implementation of execve().  All arguments are assumed to be
355  * userspace pointers from the passed thread.
356  */
357 static int
358 do_execve(struct thread *td, struct image_args *args, struct mac *mac_p)
359 {
360 	struct proc *p = td->td_proc;
361 	struct nameidata nd;
362 	struct ucred *oldcred;
363 	struct uidinfo *euip = NULL;
364 	register_t *stack_base;
365 	int error, i;
366 	struct image_params image_params, *imgp;
367 	struct vattr attr;
368 	int (*img_first)(struct image_params *);
369 	struct pargs *oldargs = NULL, *newargs = NULL;
370 	struct sigacts *oldsigacts = NULL, *newsigacts = NULL;
371 #ifdef KTRACE
372 	struct vnode *tracevp = NULL;
373 	struct ucred *tracecred = NULL;
374 #endif
375 	struct vnode *oldtextvp = NULL, *newtextvp;
376 	cap_rights_t rights;
377 	int credential_changing;
378 	int textset;
379 #ifdef MAC
380 	struct label *interpvplabel = NULL;
381 	int will_transition;
382 #endif
383 #ifdef HWPMC_HOOKS
384 	struct pmckern_procexec pe;
385 #endif
386 	static const char fexecv_proc_title[] = "(fexecv)";
387 
388 	imgp = &image_params;
389 
390 	/*
391 	 * Lock the process and set the P_INEXEC flag to indicate that
392 	 * it should be left alone until we're done here.  This is
393 	 * necessary to avoid race conditions - e.g. in ptrace() -
394 	 * that might allow a local user to illicitly obtain elevated
395 	 * privileges.
396 	 */
397 	PROC_LOCK(p);
398 	KASSERT((p->p_flag & P_INEXEC) == 0,
399 	    ("%s(): process already has P_INEXEC flag", __func__));
400 	p->p_flag |= P_INEXEC;
401 	PROC_UNLOCK(p);
402 
403 	/*
404 	 * Initialize part of the common data
405 	 */
406 	bzero(imgp, sizeof(*imgp));
407 	imgp->proc = p;
408 	imgp->attr = &attr;
409 	imgp->args = args;
410 	oldcred = p->p_ucred;
411 
412 #ifdef MAC
413 	error = mac_execve_enter(imgp, mac_p);
414 	if (error)
415 		goto exec_fail;
416 #endif
417 
418 	/*
419 	 * Translate the file name. namei() returns a vnode pointer
420 	 *	in ni_vp among other things.
421 	 *
422 	 * XXXAUDIT: It would be desirable to also audit the name of the
423 	 * interpreter if this is an interpreted binary.
424 	 */
425 	if (args->fname != NULL) {
426 		NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME
427 		    | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
428 	}
429 
430 	SDT_PROBE1(proc, , , exec, args->fname);
431 
432 interpret:
433 	if (args->fname != NULL) {
434 #ifdef CAPABILITY_MODE
435 		/*
436 		 * While capability mode can't reach this point via direct
437 		 * path arguments to execve(), we also don't allow
438 		 * interpreters to be used in capability mode (for now).
439 		 * Catch indirect lookups and return a permissions error.
440 		 */
441 		if (IN_CAPABILITY_MODE(td)) {
442 			error = ECAPMODE;
443 			goto exec_fail;
444 		}
445 #endif
446 		error = namei(&nd);
447 		if (error)
448 			goto exec_fail;
449 
450 		newtextvp = nd.ni_vp;
451 		imgp->vp = newtextvp;
452 	} else {
453 		AUDIT_ARG_FD(args->fd);
454 		/*
455 		 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
456 		 */
457 		error = fgetvp_exec(td, args->fd,
458 		    cap_rights_init(&rights, CAP_FEXECVE), &newtextvp);
459 		if (error)
460 			goto exec_fail;
461 		vn_lock(newtextvp, LK_EXCLUSIVE | LK_RETRY);
462 		AUDIT_ARG_VNODE1(newtextvp);
463 		imgp->vp = newtextvp;
464 	}
465 
466 	/*
467 	 * Check file permissions (also 'opens' file)
468 	 */
469 	error = exec_check_permissions(imgp);
470 	if (error)
471 		goto exec_fail_dealloc;
472 
473 	imgp->object = imgp->vp->v_object;
474 	if (imgp->object != NULL)
475 		vm_object_reference(imgp->object);
476 
477 	/*
478 	 * Set VV_TEXT now so no one can write to the executable while we're
479 	 * activating it.
480 	 *
481 	 * Remember if this was set before and unset it in case this is not
482 	 * actually an executable image.
483 	 */
484 	textset = VOP_IS_TEXT(imgp->vp);
485 	VOP_SET_TEXT(imgp->vp);
486 
487 	error = exec_map_first_page(imgp);
488 	if (error)
489 		goto exec_fail_dealloc;
490 
491 	imgp->proc->p_osrel = 0;
492 
493 	/*
494 	 * Implement image setuid/setgid.
495 	 *
496 	 * Determine new credentials before attempting image activators
497 	 * so that it can be used by process_exec handlers to determine
498 	 * credential/setid changes.
499 	 *
500 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
501 	 * the process is being traced.
502 	 *
503 	 * We disable setuid/setgid/etc in capability mode on the basis
504 	 * that most setugid applications are not written with that
505 	 * environment in mind, and will therefore almost certainly operate
506 	 * incorrectly. In principle there's no reason that setugid
507 	 * applications might not be useful in capability mode, so we may want
508 	 * to reconsider this conservative design choice in the future.
509 	 *
510 	 * XXXMAC: For the time being, use NOSUID to also prohibit
511 	 * transitions on the file system.
512 	 */
513 	credential_changing = 0;
514 	credential_changing |= (attr.va_mode & S_ISUID) &&
515 	    oldcred->cr_uid != attr.va_uid;
516 	credential_changing |= (attr.va_mode & S_ISGID) &&
517 	    oldcred->cr_gid != attr.va_gid;
518 #ifdef MAC
519 	will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
520 	    interpvplabel, imgp);
521 	credential_changing |= will_transition;
522 #endif
523 
524 	if (credential_changing &&
525 #ifdef CAPABILITY_MODE
526 	    ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
527 #endif
528 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
529 	    (p->p_flag & P_TRACED) == 0) {
530 		imgp->credential_setid = true;
531 		VOP_UNLOCK(imgp->vp, 0);
532 		imgp->newcred = crdup(oldcred);
533 		if (attr.va_mode & S_ISUID) {
534 			euip = uifind(attr.va_uid);
535 			change_euid(imgp->newcred, euip);
536 		}
537 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
538 		if (attr.va_mode & S_ISGID)
539 			change_egid(imgp->newcred, attr.va_gid);
540 		/*
541 		 * Implement correct POSIX saved-id behavior.
542 		 *
543 		 * XXXMAC: Note that the current logic will save the
544 		 * uid and gid if a MAC domain transition occurs, even
545 		 * though maybe it shouldn't.
546 		 */
547 		change_svuid(imgp->newcred, imgp->newcred->cr_uid);
548 		change_svgid(imgp->newcred, imgp->newcred->cr_gid);
549 	} else {
550 		/*
551 		 * Implement correct POSIX saved-id behavior.
552 		 *
553 		 * XXX: It's not clear that the existing behavior is
554 		 * POSIX-compliant.  A number of sources indicate that the
555 		 * saved uid/gid should only be updated if the new ruid is
556 		 * not equal to the old ruid, or the new euid is not equal
557 		 * to the old euid and the new euid is not equal to the old
558 		 * ruid.  The FreeBSD code always updates the saved uid/gid.
559 		 * Also, this code uses the new (replaced) euid and egid as
560 		 * the source, which may or may not be the right ones to use.
561 		 */
562 		if (oldcred->cr_svuid != oldcred->cr_uid ||
563 		    oldcred->cr_svgid != oldcred->cr_gid) {
564 			VOP_UNLOCK(imgp->vp, 0);
565 			imgp->newcred = crdup(oldcred);
566 			vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
567 			change_svuid(imgp->newcred, imgp->newcred->cr_uid);
568 			change_svgid(imgp->newcred, imgp->newcred->cr_gid);
569 		}
570 	}
571 	/* The new credentials are installed into the process later. */
572 
573 	/*
574 	 * Do the best to calculate the full path to the image file.
575 	 */
576 	if (args->fname != NULL && args->fname[0] == '/')
577 		imgp->execpath = args->fname;
578 	else {
579 		VOP_UNLOCK(imgp->vp, 0);
580 		if (vn_fullpath(td, imgp->vp, &imgp->execpath,
581 		    &imgp->freepath) != 0)
582 			imgp->execpath = args->fname;
583 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
584 	}
585 
586 	/*
587 	 *	If the current process has a special image activator it
588 	 *	wants to try first, call it.   For example, emulating shell
589 	 *	scripts differently.
590 	 */
591 	error = -1;
592 	if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
593 		error = img_first(imgp);
594 
595 	/*
596 	 *	Loop through the list of image activators, calling each one.
597 	 *	An activator returns -1 if there is no match, 0 on success,
598 	 *	and an error otherwise.
599 	 */
600 	for (i = 0; error == -1 && execsw[i]; ++i) {
601 		if (execsw[i]->ex_imgact == NULL ||
602 		    execsw[i]->ex_imgact == img_first) {
603 			continue;
604 		}
605 		error = (*execsw[i]->ex_imgact)(imgp);
606 	}
607 
608 	if (error) {
609 		if (error == -1) {
610 			if (textset == 0)
611 				VOP_UNSET_TEXT(imgp->vp);
612 			error = ENOEXEC;
613 		}
614 		goto exec_fail_dealloc;
615 	}
616 
617 	/*
618 	 * Special interpreter operation, cleanup and loop up to try to
619 	 * activate the interpreter.
620 	 */
621 	if (imgp->interpreted) {
622 		exec_unmap_first_page(imgp);
623 		/*
624 		 * VV_TEXT needs to be unset for scripts.  There is a short
625 		 * period before we determine that something is a script where
626 		 * VV_TEXT will be set. The vnode lock is held over this
627 		 * entire period so nothing should illegitimately be blocked.
628 		 */
629 		VOP_UNSET_TEXT(imgp->vp);
630 		/* free name buffer and old vnode */
631 		if (args->fname != NULL)
632 			NDFREE(&nd, NDF_ONLY_PNBUF);
633 #ifdef MAC
634 		mac_execve_interpreter_enter(newtextvp, &interpvplabel);
635 #endif
636 		if (imgp->opened) {
637 			VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td);
638 			imgp->opened = 0;
639 		}
640 		vput(newtextvp);
641 		vm_object_deallocate(imgp->object);
642 		imgp->object = NULL;
643 		imgp->credential_setid = false;
644 		if (imgp->newcred != NULL) {
645 			crfree(imgp->newcred);
646 			imgp->newcred = NULL;
647 		}
648 		imgp->execpath = NULL;
649 		free(imgp->freepath, M_TEMP);
650 		imgp->freepath = NULL;
651 		/* set new name to that of the interpreter */
652 		NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME,
653 		    UIO_SYSSPACE, imgp->interpreter_name, td);
654 		args->fname = imgp->interpreter_name;
655 		goto interpret;
656 	}
657 
658 	/*
659 	 * NB: We unlock the vnode here because it is believed that none
660 	 * of the sv_copyout_strings/sv_fixup operations require the vnode.
661 	 */
662 	VOP_UNLOCK(imgp->vp, 0);
663 
664 	if (disallow_high_osrel &&
665 	    P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
666 		error = ENOEXEC;
667 		uprintf("Osrel %d for image %s too high\n", p->p_osrel,
668 		    imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
669 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
670 		goto exec_fail_dealloc;
671 	}
672 
673 	/* ABI enforces the use of Capsicum. Switch into capabilities mode. */
674 	if (SV_PROC_FLAG(p, SV_CAPSICUM))
675 		sys_cap_enter(td, NULL);
676 
677 	/*
678 	 * Copy out strings (args and env) and initialize stack base
679 	 */
680 	if (p->p_sysent->sv_copyout_strings)
681 		stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
682 	else
683 		stack_base = exec_copyout_strings(imgp);
684 
685 	/*
686 	 * If custom stack fixup routine present for this process
687 	 * let it do the stack setup.
688 	 * Else stuff argument count as first item on stack
689 	 */
690 	if (p->p_sysent->sv_fixup != NULL)
691 		(*p->p_sysent->sv_fixup)(&stack_base, imgp);
692 	else
693 		suword(--stack_base, imgp->args->argc);
694 
695 	if (args->fdp != NULL) {
696 		/* Install a brand new file descriptor table. */
697 		fdinstall_remapped(td, args->fdp);
698 		args->fdp = NULL;
699 	} else {
700 		/*
701 		 * Keep on using the existing file descriptor table. For
702 		 * security and other reasons, the file descriptor table
703 		 * cannot be shared after an exec.
704 		 */
705 		fdunshare(td);
706 		/* close files on exec */
707 		fdcloseexec(td);
708 	}
709 
710 	/*
711 	 * Malloc things before we need locks.
712 	 */
713 	i = imgp->args->begin_envv - imgp->args->begin_argv;
714 	/* Cache arguments if they fit inside our allowance */
715 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
716 		newargs = pargs_alloc(i);
717 		bcopy(imgp->args->begin_argv, newargs->ar_args, i);
718 	}
719 
720 	/*
721 	 * For security and other reasons, signal handlers cannot
722 	 * be shared after an exec. The new process gets a copy of the old
723 	 * handlers. In execsigs(), the new process will have its signals
724 	 * reset.
725 	 */
726 	if (sigacts_shared(p->p_sigacts)) {
727 		oldsigacts = p->p_sigacts;
728 		newsigacts = sigacts_alloc();
729 		sigacts_copy(newsigacts, oldsigacts);
730 	}
731 
732 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
733 
734 	PROC_LOCK(p);
735 	if (oldsigacts)
736 		p->p_sigacts = newsigacts;
737 	/* Stop profiling */
738 	stopprofclock(p);
739 
740 	/* reset caught signals */
741 	execsigs(p);
742 
743 	/* name this process - nameiexec(p, ndp) */
744 	bzero(p->p_comm, sizeof(p->p_comm));
745 	if (args->fname)
746 		bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
747 		    min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
748 	else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0)
749 		bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
750 	bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
751 #ifdef KTR
752 	sched_clear_tdname(td);
753 #endif
754 
755 	/*
756 	 * mark as execed, wakeup the process that vforked (if any) and tell
757 	 * it that it now has its own resources back
758 	 */
759 	p->p_flag |= P_EXEC;
760 	if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0)
761 		p->p_flag2 &= ~P2_NOTRACE;
762 	if (p->p_flag & P_PPWAIT) {
763 		p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
764 		cv_broadcast(&p->p_pwait);
765 		/* STOPs are no longer ignored, arrange for AST */
766 		signotify(td);
767 	}
768 
769 	/*
770 	 * Implement image setuid/setgid installation.
771 	 */
772 	if (imgp->credential_setid) {
773 		/*
774 		 * Turn off syscall tracing for set-id programs, except for
775 		 * root.  Record any set-id flags first to make sure that
776 		 * we do not regain any tracing during a possible block.
777 		 */
778 		setsugid(p);
779 
780 #ifdef KTRACE
781 		if (p->p_tracecred != NULL &&
782 		    priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED, 0))
783 			ktrprocexec(p, &tracecred, &tracevp);
784 #endif
785 		/*
786 		 * Close any file descriptors 0..2 that reference procfs,
787 		 * then make sure file descriptors 0..2 are in use.
788 		 *
789 		 * Both fdsetugidsafety() and fdcheckstd() may call functions
790 		 * taking sleepable locks, so temporarily drop our locks.
791 		 */
792 		PROC_UNLOCK(p);
793 		VOP_UNLOCK(imgp->vp, 0);
794 		fdsetugidsafety(td);
795 		error = fdcheckstd(td);
796 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
797 		if (error != 0)
798 			goto exec_fail_dealloc;
799 		PROC_LOCK(p);
800 #ifdef MAC
801 		if (will_transition) {
802 			mac_vnode_execve_transition(oldcred, imgp->newcred,
803 			    imgp->vp, interpvplabel, imgp);
804 		}
805 #endif
806 	} else {
807 		if (oldcred->cr_uid == oldcred->cr_ruid &&
808 		    oldcred->cr_gid == oldcred->cr_rgid)
809 			p->p_flag &= ~P_SUGID;
810 	}
811 	/*
812 	 * Set the new credentials.
813 	 */
814 	if (imgp->newcred != NULL) {
815 		proc_set_cred(p, imgp->newcred);
816 		crfree(oldcred);
817 		oldcred = NULL;
818 	}
819 
820 	/*
821 	 * Store the vp for use in procfs.  This vnode was referenced by namei
822 	 * or fgetvp_exec.
823 	 */
824 	oldtextvp = p->p_textvp;
825 	p->p_textvp = newtextvp;
826 
827 #ifdef KDTRACE_HOOKS
828 	/*
829 	 * Tell the DTrace fasttrap provider about the exec if it
830 	 * has declared an interest.
831 	 */
832 	if (dtrace_fasttrap_exec)
833 		dtrace_fasttrap_exec(p);
834 #endif
835 
836 	/*
837 	 * Notify others that we exec'd, and clear the P_INEXEC flag
838 	 * as we're now a bona fide freshly-execed process.
839 	 */
840 	KNOTE_LOCKED(p->p_klist, NOTE_EXEC);
841 	p->p_flag &= ~P_INEXEC;
842 
843 	/* clear "fork but no exec" flag, as we _are_ execing */
844 	p->p_acflag &= ~AFORK;
845 
846 	/*
847 	 * Free any previous argument cache and replace it with
848 	 * the new argument cache, if any.
849 	 */
850 	oldargs = p->p_args;
851 	p->p_args = newargs;
852 	newargs = NULL;
853 
854 	PROC_UNLOCK(p);
855 
856 #ifdef	HWPMC_HOOKS
857 	/*
858 	 * Check if system-wide sampling is in effect or if the
859 	 * current process is using PMCs.  If so, do exec() time
860 	 * processing.  This processing needs to happen AFTER the
861 	 * P_INEXEC flag is cleared.
862 	 */
863 	if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(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 	}
871 #endif
872 
873 	/* Set values passed into the program in registers. */
874 	if (p->p_sysent->sv_setregs)
875 		(*p->p_sysent->sv_setregs)(td, imgp,
876 		    (u_long)(uintptr_t)stack_base);
877 	else
878 		exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base);
879 
880 	vfs_mark_atime(imgp->vp, td->td_ucred);
881 
882 	SDT_PROBE1(proc, , , exec__success, args->fname);
883 
884 exec_fail_dealloc:
885 	if (imgp->firstpage != NULL)
886 		exec_unmap_first_page(imgp);
887 
888 	if (imgp->vp != NULL) {
889 		if (args->fname)
890 			NDFREE(&nd, NDF_ONLY_PNBUF);
891 		if (imgp->opened)
892 			VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
893 		if (error != 0)
894 			vput(imgp->vp);
895 		else
896 			VOP_UNLOCK(imgp->vp, 0);
897 	}
898 
899 	if (imgp->object != NULL)
900 		vm_object_deallocate(imgp->object);
901 
902 	free(imgp->freepath, M_TEMP);
903 
904 	if (error == 0) {
905 		if (p->p_ptevents & PTRACE_EXEC) {
906 			PROC_LOCK(p);
907 			if (p->p_ptevents & PTRACE_EXEC)
908 				td->td_dbgflags |= TDB_EXEC;
909 			PROC_UNLOCK(p);
910 		}
911 
912 		/*
913 		 * Stop the process here if its stop event mask has
914 		 * the S_EXEC bit set.
915 		 */
916 		STOPEVENT(p, S_EXEC, 0);
917 	} else {
918 exec_fail:
919 		/* we're done here, clear P_INEXEC */
920 		PROC_LOCK(p);
921 		p->p_flag &= ~P_INEXEC;
922 		PROC_UNLOCK(p);
923 
924 		SDT_PROBE1(proc, , , exec__failure, error);
925 	}
926 
927 	if (imgp->newcred != NULL && oldcred != NULL)
928 		crfree(imgp->newcred);
929 
930 #ifdef MAC
931 	mac_execve_exit(imgp);
932 	mac_execve_interpreter_exit(interpvplabel);
933 #endif
934 	exec_free_args(args);
935 
936 	/*
937 	 * Handle deferred decrement of ref counts.
938 	 */
939 	if (oldtextvp != NULL)
940 		vrele(oldtextvp);
941 #ifdef KTRACE
942 	if (tracevp != NULL)
943 		vrele(tracevp);
944 	if (tracecred != NULL)
945 		crfree(tracecred);
946 #endif
947 	pargs_drop(oldargs);
948 	pargs_drop(newargs);
949 	if (oldsigacts != NULL)
950 		sigacts_free(oldsigacts);
951 	if (euip != NULL)
952 		uifree(euip);
953 
954 	if (error && imgp->vmspace_destroyed) {
955 		/* sorry, no more process anymore. exit gracefully */
956 		exit1(td, 0, SIGABRT);
957 		/* NOT REACHED */
958 	}
959 
960 #ifdef KTRACE
961 	if (error == 0)
962 		ktrprocctor(p);
963 #endif
964 
965 	/*
966 	 * We don't want cpu_set_syscall_retval() to overwrite any of
967 	 * the register values put in place by exec_setregs().
968 	 * Implementations of cpu_set_syscall_retval() will leave
969 	 * registers unmodified when returning EJUSTRETURN.
970 	 */
971 	return (error == 0 ? EJUSTRETURN : error);
972 }
973 
974 int
975 exec_map_first_page(imgp)
976 	struct image_params *imgp;
977 {
978 	int rv, i, after, initial_pagein;
979 	vm_page_t ma[VM_INITIAL_PAGEIN];
980 	vm_object_t object;
981 
982 	if (imgp->firstpage != NULL)
983 		exec_unmap_first_page(imgp);
984 
985 	object = imgp->vp->v_object;
986 	if (object == NULL)
987 		return (EACCES);
988 	VM_OBJECT_WLOCK(object);
989 #if VM_NRESERVLEVEL > 0
990 	vm_object_color(object, 0);
991 #endif
992 	ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY);
993 	if (ma[0]->valid != VM_PAGE_BITS_ALL) {
994 		vm_page_xbusy(ma[0]);
995 		if (!vm_pager_has_page(object, 0, NULL, &after)) {
996 			vm_page_lock(ma[0]);
997 			vm_page_free(ma[0]);
998 			vm_page_unlock(ma[0]);
999 			VM_OBJECT_WUNLOCK(object);
1000 			return (EIO);
1001 		}
1002 		initial_pagein = min(after, VM_INITIAL_PAGEIN);
1003 		KASSERT(initial_pagein <= object->size,
1004 		    ("%s: initial_pagein %d object->size %ju",
1005 		    __func__, initial_pagein, (uintmax_t )object->size));
1006 		for (i = 1; i < initial_pagein; i++) {
1007 			if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) {
1008 				if (ma[i]->valid)
1009 					break;
1010 				if (vm_page_tryxbusy(ma[i]))
1011 					break;
1012 			} else {
1013 				ma[i] = vm_page_alloc(object, i,
1014 				    VM_ALLOC_NORMAL);
1015 				if (ma[i] == NULL)
1016 					break;
1017 			}
1018 		}
1019 		initial_pagein = i;
1020 		rv = vm_pager_get_pages(object, ma, initial_pagein, NULL, NULL);
1021 		if (rv != VM_PAGER_OK) {
1022 			for (i = 0; i < initial_pagein; i++) {
1023 				vm_page_lock(ma[i]);
1024 				vm_page_free(ma[i]);
1025 				vm_page_unlock(ma[i]);
1026 			}
1027 			VM_OBJECT_WUNLOCK(object);
1028 			return (EIO);
1029 		}
1030 		vm_page_xunbusy(ma[0]);
1031 		for (i = 1; i < initial_pagein; i++)
1032 			vm_page_readahead_finish(ma[i]);
1033 	}
1034 	vm_page_lock(ma[0]);
1035 	vm_page_hold(ma[0]);
1036 	vm_page_activate(ma[0]);
1037 	vm_page_unlock(ma[0]);
1038 	VM_OBJECT_WUNLOCK(object);
1039 
1040 	imgp->firstpage = sf_buf_alloc(ma[0], 0);
1041 	imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
1042 
1043 	return (0);
1044 }
1045 
1046 void
1047 exec_unmap_first_page(struct image_params *imgp)
1048 {
1049 	vm_page_t m;
1050 
1051 	if (imgp->firstpage != NULL) {
1052 		m = sf_buf_page(imgp->firstpage);
1053 		sf_buf_free(imgp->firstpage);
1054 		imgp->firstpage = NULL;
1055 		vm_page_lock(m);
1056 		vm_page_unhold(m);
1057 		vm_page_unlock(m);
1058 	}
1059 }
1060 
1061 /*
1062  * Destroy old address space, and allocate a new stack.
1063  *	The new stack is only sgrowsiz large because it is grown
1064  *	automatically on a page fault.
1065  */
1066 int
1067 exec_new_vmspace(struct image_params *imgp, struct sysentvec *sv)
1068 {
1069 	int error;
1070 	struct proc *p = imgp->proc;
1071 	struct vmspace *vmspace = p->p_vmspace;
1072 	vm_object_t obj;
1073 	struct rlimit rlim_stack;
1074 	vm_offset_t sv_minuser, stack_addr;
1075 	vm_map_t map;
1076 	u_long ssiz;
1077 
1078 	imgp->vmspace_destroyed = 1;
1079 	imgp->sysent = sv;
1080 
1081 	/* May be called with Giant held */
1082 	EVENTHANDLER_DIRECT_INVOKE(process_exec, p, imgp);
1083 
1084 	/*
1085 	 * Blow away entire process VM, if address space not shared,
1086 	 * otherwise, create a new VM space so that other threads are
1087 	 * not disrupted
1088 	 */
1089 	map = &vmspace->vm_map;
1090 	if (map_at_zero)
1091 		sv_minuser = sv->sv_minuser;
1092 	else
1093 		sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1094 	if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser &&
1095 	    vm_map_max(map) == sv->sv_maxuser) {
1096 		shmexit(vmspace);
1097 		pmap_remove_pages(vmspace_pmap(vmspace));
1098 		vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1099 		/* An exec terminates mlockall(MCL_FUTURE). */
1100 		vm_map_lock(map);
1101 		vm_map_modflags(map, 0, MAP_WIREFUTURE);
1102 		vm_map_unlock(map);
1103 	} else {
1104 		error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1105 		if (error)
1106 			return (error);
1107 		vmspace = p->p_vmspace;
1108 		map = &vmspace->vm_map;
1109 	}
1110 
1111 	/* Map a shared page */
1112 	obj = sv->sv_shared_page_obj;
1113 	if (obj != NULL) {
1114 		vm_object_reference(obj);
1115 		error = vm_map_fixed(map, obj, 0,
1116 		    sv->sv_shared_page_base, sv->sv_shared_page_len,
1117 		    VM_PROT_READ | VM_PROT_EXECUTE,
1118 		    VM_PROT_READ | VM_PROT_EXECUTE,
1119 		    MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1120 		if (error != KERN_SUCCESS) {
1121 			vm_object_deallocate(obj);
1122 			return (vm_mmap_to_errno(error));
1123 		}
1124 	}
1125 
1126 	/* Allocate a new stack */
1127 	if (imgp->stack_sz != 0) {
1128 		ssiz = trunc_page(imgp->stack_sz);
1129 		PROC_LOCK(p);
1130 		lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack);
1131 		PROC_UNLOCK(p);
1132 		if (ssiz > rlim_stack.rlim_max)
1133 			ssiz = rlim_stack.rlim_max;
1134 		if (ssiz > rlim_stack.rlim_cur) {
1135 			rlim_stack.rlim_cur = ssiz;
1136 			kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack);
1137 		}
1138 	} else if (sv->sv_maxssiz != NULL) {
1139 		ssiz = *sv->sv_maxssiz;
1140 	} else {
1141 		ssiz = maxssiz;
1142 	}
1143 	stack_addr = sv->sv_usrstack - ssiz;
1144 	error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1145 	    obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1146 	    sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1147 	if (error != KERN_SUCCESS)
1148 		return (vm_mmap_to_errno(error));
1149 
1150 	/*
1151 	 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they
1152 	 * are still used to enforce the stack rlimit on the process stack.
1153 	 */
1154 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1155 	vmspace->vm_maxsaddr = (char *)stack_addr;
1156 
1157 	return (0);
1158 }
1159 
1160 /*
1161  * Copy out argument and environment strings from the old process address
1162  * space into the temporary string buffer.
1163  */
1164 int
1165 exec_copyin_args(struct image_args *args, char *fname,
1166     enum uio_seg segflg, char **argv, char **envv)
1167 {
1168 	u_long argp, envp;
1169 	int error;
1170 	size_t length;
1171 
1172 	bzero(args, sizeof(*args));
1173 	if (argv == NULL)
1174 		return (EFAULT);
1175 
1176 	/*
1177 	 * Allocate demand-paged memory for the file name, argument, and
1178 	 * environment strings.
1179 	 */
1180 	error = exec_alloc_args(args);
1181 	if (error != 0)
1182 		return (error);
1183 
1184 	/*
1185 	 * Copy the file name.
1186 	 */
1187 	if (fname != NULL) {
1188 		args->fname = args->buf;
1189 		error = (segflg == UIO_SYSSPACE) ?
1190 		    copystr(fname, args->fname, PATH_MAX, &length) :
1191 		    copyinstr(fname, args->fname, PATH_MAX, &length);
1192 		if (error != 0)
1193 			goto err_exit;
1194 	} else
1195 		length = 0;
1196 
1197 	args->begin_argv = args->buf + length;
1198 	args->endp = args->begin_argv;
1199 	args->stringspace = ARG_MAX;
1200 
1201 	/*
1202 	 * extract arguments first
1203 	 */
1204 	for (;;) {
1205 		error = fueword(argv++, &argp);
1206 		if (error == -1) {
1207 			error = EFAULT;
1208 			goto err_exit;
1209 		}
1210 		if (argp == 0)
1211 			break;
1212 		error = copyinstr((void *)(uintptr_t)argp, args->endp,
1213 		    args->stringspace, &length);
1214 		if (error != 0) {
1215 			if (error == ENAMETOOLONG)
1216 				error = E2BIG;
1217 			goto err_exit;
1218 		}
1219 		args->stringspace -= length;
1220 		args->endp += length;
1221 		args->argc++;
1222 	}
1223 
1224 	args->begin_envv = args->endp;
1225 
1226 	/*
1227 	 * extract environment strings
1228 	 */
1229 	if (envv) {
1230 		for (;;) {
1231 			error = fueword(envv++, &envp);
1232 			if (error == -1) {
1233 				error = EFAULT;
1234 				goto err_exit;
1235 			}
1236 			if (envp == 0)
1237 				break;
1238 			error = copyinstr((void *)(uintptr_t)envp,
1239 			    args->endp, args->stringspace, &length);
1240 			if (error != 0) {
1241 				if (error == ENAMETOOLONG)
1242 					error = E2BIG;
1243 				goto err_exit;
1244 			}
1245 			args->stringspace -= length;
1246 			args->endp += length;
1247 			args->envc++;
1248 		}
1249 	}
1250 
1251 	return (0);
1252 
1253 err_exit:
1254 	exec_free_args(args);
1255 	return (error);
1256 }
1257 
1258 int
1259 exec_copyin_data_fds(struct thread *td, struct image_args *args,
1260     const void *data, size_t datalen, const int *fds, size_t fdslen)
1261 {
1262 	struct filedesc *ofdp;
1263 	const char *p;
1264 	int *kfds;
1265 	int error;
1266 
1267 	memset(args, '\0', sizeof(*args));
1268 	ofdp = td->td_proc->p_fd;
1269 	if (datalen >= ARG_MAX || fdslen > ofdp->fd_lastfile + 1)
1270 		return (E2BIG);
1271 	error = exec_alloc_args(args);
1272 	if (error != 0)
1273 		return (error);
1274 
1275 	args->begin_argv = args->buf;
1276 	args->stringspace = ARG_MAX;
1277 
1278 	if (datalen > 0) {
1279 		/*
1280 		 * Argument buffer has been provided. Copy it into the
1281 		 * kernel as a single string and add a terminating null
1282 		 * byte.
1283 		 */
1284 		error = copyin(data, args->begin_argv, datalen);
1285 		if (error != 0)
1286 			goto err_exit;
1287 		args->begin_argv[datalen] = '\0';
1288 		args->endp = args->begin_argv + datalen + 1;
1289 		args->stringspace -= datalen + 1;
1290 
1291 		/*
1292 		 * Traditional argument counting. Count the number of
1293 		 * null bytes.
1294 		 */
1295 		for (p = args->begin_argv; p < args->endp; ++p)
1296 			if (*p == '\0')
1297 				++args->argc;
1298 	} else {
1299 		/* No argument buffer provided. */
1300 		args->endp = args->begin_argv;
1301 	}
1302 	/* There are no environment variables. */
1303 	args->begin_envv = args->endp;
1304 
1305 	/* Create new file descriptor table. */
1306 	kfds = malloc(fdslen * sizeof(int), M_TEMP, M_WAITOK);
1307 	error = copyin(fds, kfds, fdslen * sizeof(int));
1308 	if (error != 0) {
1309 		free(kfds, M_TEMP);
1310 		goto err_exit;
1311 	}
1312 	error = fdcopy_remapped(ofdp, kfds, fdslen, &args->fdp);
1313 	free(kfds, M_TEMP);
1314 	if (error != 0)
1315 		goto err_exit;
1316 
1317 	return (0);
1318 err_exit:
1319 	exec_free_args(args);
1320 	return (error);
1321 }
1322 
1323 struct exec_args_kva {
1324 	vm_offset_t addr;
1325 	u_int gen;
1326 	SLIST_ENTRY(exec_args_kva) next;
1327 };
1328 
1329 static DPCPU_DEFINE(struct exec_args_kva *, exec_args_kva);
1330 
1331 static SLIST_HEAD(, exec_args_kva) exec_args_kva_freelist;
1332 static struct mtx exec_args_kva_mtx;
1333 static u_int exec_args_gen;
1334 
1335 static void
1336 exec_prealloc_args_kva(void *arg __unused)
1337 {
1338 	struct exec_args_kva *argkva;
1339 	u_int i;
1340 
1341 	SLIST_INIT(&exec_args_kva_freelist);
1342 	mtx_init(&exec_args_kva_mtx, "exec args kva", NULL, MTX_DEF);
1343 	for (i = 0; i < exec_map_entries; i++) {
1344 		argkva = malloc(sizeof(*argkva), M_PARGS, M_WAITOK);
1345 		argkva->addr = kmap_alloc_wait(exec_map, exec_map_entry_size);
1346 		argkva->gen = exec_args_gen;
1347 		SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1348 	}
1349 }
1350 SYSINIT(exec_args_kva, SI_SUB_EXEC, SI_ORDER_ANY, exec_prealloc_args_kva, NULL);
1351 
1352 static vm_offset_t
1353 exec_alloc_args_kva(void **cookie)
1354 {
1355 	struct exec_args_kva *argkva;
1356 
1357 	argkva = (void *)atomic_readandclear_ptr(
1358 	    (uintptr_t *)DPCPU_PTR(exec_args_kva));
1359 	if (argkva == NULL) {
1360 		mtx_lock(&exec_args_kva_mtx);
1361 		while ((argkva = SLIST_FIRST(&exec_args_kva_freelist)) == NULL)
1362 			(void)mtx_sleep(&exec_args_kva_freelist,
1363 			    &exec_args_kva_mtx, 0, "execkva", 0);
1364 		SLIST_REMOVE_HEAD(&exec_args_kva_freelist, next);
1365 		mtx_unlock(&exec_args_kva_mtx);
1366 	}
1367 	*(struct exec_args_kva **)cookie = argkva;
1368 	return (argkva->addr);
1369 }
1370 
1371 static void
1372 exec_release_args_kva(struct exec_args_kva *argkva, u_int gen)
1373 {
1374 	vm_offset_t base;
1375 
1376 	base = argkva->addr;
1377 	if (argkva->gen != gen) {
1378 		vm_map_madvise(exec_map, base, base + exec_map_entry_size,
1379 		    MADV_FREE);
1380 		argkva->gen = gen;
1381 	}
1382 	if (!atomic_cmpset_ptr((uintptr_t *)DPCPU_PTR(exec_args_kva),
1383 	    (uintptr_t)NULL, (uintptr_t)argkva)) {
1384 		mtx_lock(&exec_args_kva_mtx);
1385 		SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1386 		wakeup_one(&exec_args_kva_freelist);
1387 		mtx_unlock(&exec_args_kva_mtx);
1388 	}
1389 }
1390 
1391 static void
1392 exec_free_args_kva(void *cookie)
1393 {
1394 
1395 	exec_release_args_kva(cookie, exec_args_gen);
1396 }
1397 
1398 static void
1399 exec_args_kva_lowmem(void *arg __unused)
1400 {
1401 	SLIST_HEAD(, exec_args_kva) head;
1402 	struct exec_args_kva *argkva;
1403 	u_int gen;
1404 	int i;
1405 
1406 	gen = atomic_fetchadd_int(&exec_args_gen, 1) + 1;
1407 
1408 	/*
1409 	 * Force an madvise of each KVA range. Any currently allocated ranges
1410 	 * will have MADV_FREE applied once they are freed.
1411 	 */
1412 	SLIST_INIT(&head);
1413 	mtx_lock(&exec_args_kva_mtx);
1414 	SLIST_SWAP(&head, &exec_args_kva_freelist, exec_args_kva);
1415 	mtx_unlock(&exec_args_kva_mtx);
1416 	while ((argkva = SLIST_FIRST(&head)) != NULL) {
1417 		SLIST_REMOVE_HEAD(&head, next);
1418 		exec_release_args_kva(argkva, gen);
1419 	}
1420 
1421 	CPU_FOREACH(i) {
1422 		argkva = (void *)atomic_readandclear_ptr(
1423 		    (uintptr_t *)DPCPU_ID_PTR(i, exec_args_kva));
1424 		if (argkva != NULL)
1425 			exec_release_args_kva(argkva, gen);
1426 	}
1427 }
1428 EVENTHANDLER_DEFINE(vm_lowmem, exec_args_kva_lowmem, NULL,
1429     EVENTHANDLER_PRI_ANY);
1430 
1431 /*
1432  * Allocate temporary demand-paged, zero-filled memory for the file name,
1433  * argument, and environment strings.
1434  */
1435 int
1436 exec_alloc_args(struct image_args *args)
1437 {
1438 
1439 	args->buf = (char *)exec_alloc_args_kva(&args->bufkva);
1440 	return (0);
1441 }
1442 
1443 void
1444 exec_free_args(struct image_args *args)
1445 {
1446 
1447 	if (args->buf != NULL) {
1448 		exec_free_args_kva(args->bufkva);
1449 		args->buf = NULL;
1450 	}
1451 	if (args->fname_buf != NULL) {
1452 		free(args->fname_buf, M_TEMP);
1453 		args->fname_buf = NULL;
1454 	}
1455 	if (args->fdp != NULL)
1456 		fdescfree_remapped(args->fdp);
1457 }
1458 
1459 /*
1460  * Copy strings out to the new process address space, constructing new arg
1461  * and env vector tables. Return a pointer to the base so that it can be used
1462  * as the initial stack pointer.
1463  */
1464 register_t *
1465 exec_copyout_strings(struct image_params *imgp)
1466 {
1467 	int argc, envc;
1468 	char **vectp;
1469 	char *stringp;
1470 	uintptr_t destp;
1471 	register_t *stack_base;
1472 	struct ps_strings *arginfo;
1473 	struct proc *p;
1474 	size_t execpath_len;
1475 	int szsigcode, szps;
1476 	char canary[sizeof(long) * 8];
1477 
1478 	szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1479 	/*
1480 	 * Calculate string base and vector table pointers.
1481 	 * Also deal with signal trampoline code for this exec type.
1482 	 */
1483 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
1484 		execpath_len = strlen(imgp->execpath) + 1;
1485 	else
1486 		execpath_len = 0;
1487 	p = imgp->proc;
1488 	szsigcode = 0;
1489 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1490 	if (p->p_sysent->sv_sigcode_base == 0) {
1491 		if (p->p_sysent->sv_szsigcode != NULL)
1492 			szsigcode = *(p->p_sysent->sv_szsigcode);
1493 	}
1494 	destp =	(uintptr_t)arginfo;
1495 
1496 	/*
1497 	 * install sigcode
1498 	 */
1499 	if (szsigcode != 0) {
1500 		destp -= szsigcode;
1501 		destp = rounddown2(destp, sizeof(void *));
1502 		copyout(p->p_sysent->sv_sigcode, (void *)destp, szsigcode);
1503 	}
1504 
1505 	/*
1506 	 * Copy the image path for the rtld.
1507 	 */
1508 	if (execpath_len != 0) {
1509 		destp -= execpath_len;
1510 		imgp->execpathp = destp;
1511 		copyout(imgp->execpath, (void *)destp, execpath_len);
1512 	}
1513 
1514 	/*
1515 	 * Prepare the canary for SSP.
1516 	 */
1517 	arc4rand(canary, sizeof(canary), 0);
1518 	destp -= sizeof(canary);
1519 	imgp->canary = destp;
1520 	copyout(canary, (void *)destp, sizeof(canary));
1521 	imgp->canarylen = sizeof(canary);
1522 
1523 	/*
1524 	 * Prepare the pagesizes array.
1525 	 */
1526 	destp -= szps;
1527 	destp = rounddown2(destp, sizeof(void *));
1528 	imgp->pagesizes = destp;
1529 	copyout(pagesizes, (void *)destp, szps);
1530 	imgp->pagesizeslen = szps;
1531 
1532 	destp -= ARG_MAX - imgp->args->stringspace;
1533 	destp = rounddown2(destp, sizeof(void *));
1534 
1535 	/*
1536 	 * If we have a valid auxargs ptr, prepare some room
1537 	 * on the stack.
1538 	 */
1539 	if (imgp->auxargs) {
1540 		/*
1541 		 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for
1542 		 * lower compatibility.
1543 		 */
1544 		imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size :
1545 		    (AT_COUNT * 2);
1546 		/*
1547 		 * The '+ 2' is for the null pointers at the end of each of
1548 		 * the arg and env vector sets,and imgp->auxarg_size is room
1549 		 * for argument of Runtime loader.
1550 		 */
1551 		vectp = (char **)(destp - (imgp->args->argc +
1552 		    imgp->args->envc + 2 + imgp->auxarg_size)
1553 		    * sizeof(char *));
1554 	} else {
1555 		/*
1556 		 * The '+ 2' is for the null pointers at the end of each of
1557 		 * the arg and env vector sets
1558 		 */
1559 		vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc
1560 		    + 2) * sizeof(char *));
1561 	}
1562 
1563 	/*
1564 	 * vectp also becomes our initial stack base
1565 	 */
1566 	stack_base = (register_t *)vectp;
1567 
1568 	stringp = imgp->args->begin_argv;
1569 	argc = imgp->args->argc;
1570 	envc = imgp->args->envc;
1571 
1572 	/*
1573 	 * Copy out strings - arguments and environment.
1574 	 */
1575 	copyout(stringp, (void *)destp, ARG_MAX - imgp->args->stringspace);
1576 
1577 	/*
1578 	 * Fill in "ps_strings" struct for ps, w, etc.
1579 	 */
1580 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1581 	suword32(&arginfo->ps_nargvstr, argc);
1582 
1583 	/*
1584 	 * Fill in argument portion of vector table.
1585 	 */
1586 	for (; argc > 0; --argc) {
1587 		suword(vectp++, (long)(intptr_t)destp);
1588 		while (*stringp++ != 0)
1589 			destp++;
1590 		destp++;
1591 	}
1592 
1593 	/* a null vector table pointer separates the argp's from the envp's */
1594 	suword(vectp++, 0);
1595 
1596 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1597 	suword32(&arginfo->ps_nenvstr, envc);
1598 
1599 	/*
1600 	 * Fill in environment portion of vector table.
1601 	 */
1602 	for (; envc > 0; --envc) {
1603 		suword(vectp++, (long)(intptr_t)destp);
1604 		while (*stringp++ != 0)
1605 			destp++;
1606 		destp++;
1607 	}
1608 
1609 	/* end of vector table is a null pointer */
1610 	suword(vectp, 0);
1611 
1612 	return (stack_base);
1613 }
1614 
1615 /*
1616  * Check permissions of file to execute.
1617  *	Called with imgp->vp locked.
1618  *	Return 0 for success or error code on failure.
1619  */
1620 int
1621 exec_check_permissions(struct image_params *imgp)
1622 {
1623 	struct vnode *vp = imgp->vp;
1624 	struct vattr *attr = imgp->attr;
1625 	struct thread *td;
1626 	int error, writecount;
1627 
1628 	td = curthread;
1629 
1630 	/* Get file attributes */
1631 	error = VOP_GETATTR(vp, attr, td->td_ucred);
1632 	if (error)
1633 		return (error);
1634 
1635 #ifdef MAC
1636 	error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1637 	if (error)
1638 		return (error);
1639 #endif
1640 
1641 	/*
1642 	 * 1) Check if file execution is disabled for the filesystem that
1643 	 *    this file resides on.
1644 	 * 2) Ensure that at least one execute bit is on. Otherwise, a
1645 	 *    privileged user will always succeed, and we don't want this
1646 	 *    to happen unless the file really is executable.
1647 	 * 3) Ensure that the file is a regular file.
1648 	 */
1649 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1650 	    (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1651 	    (attr->va_type != VREG))
1652 		return (EACCES);
1653 
1654 	/*
1655 	 * Zero length files can't be exec'd
1656 	 */
1657 	if (attr->va_size == 0)
1658 		return (ENOEXEC);
1659 
1660 	/*
1661 	 *  Check for execute permission to file based on current credentials.
1662 	 */
1663 	error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1664 	if (error)
1665 		return (error);
1666 
1667 	/*
1668 	 * Check number of open-for-writes on the file and deny execution
1669 	 * if there are any.
1670 	 */
1671 	error = VOP_GET_WRITECOUNT(vp, &writecount);
1672 	if (error != 0)
1673 		return (error);
1674 	if (writecount != 0)
1675 		return (ETXTBSY);
1676 
1677 	/*
1678 	 * Call filesystem specific open routine (which does nothing in the
1679 	 * general case).
1680 	 */
1681 	error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1682 	if (error == 0)
1683 		imgp->opened = 1;
1684 	return (error);
1685 }
1686 
1687 /*
1688  * Exec handler registration
1689  */
1690 int
1691 exec_register(const struct execsw *execsw_arg)
1692 {
1693 	const struct execsw **es, **xs, **newexecsw;
1694 	int count = 2;	/* New slot and trailing NULL */
1695 
1696 	if (execsw)
1697 		for (es = execsw; *es; es++)
1698 			count++;
1699 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1700 	xs = newexecsw;
1701 	if (execsw)
1702 		for (es = execsw; *es; es++)
1703 			*xs++ = *es;
1704 	*xs++ = execsw_arg;
1705 	*xs = NULL;
1706 	if (execsw)
1707 		free(execsw, M_TEMP);
1708 	execsw = newexecsw;
1709 	return (0);
1710 }
1711 
1712 int
1713 exec_unregister(const struct execsw *execsw_arg)
1714 {
1715 	const struct execsw **es, **xs, **newexecsw;
1716 	int count = 1;
1717 
1718 	if (execsw == NULL)
1719 		panic("unregister with no handlers left?\n");
1720 
1721 	for (es = execsw; *es; es++) {
1722 		if (*es == execsw_arg)
1723 			break;
1724 	}
1725 	if (*es == NULL)
1726 		return (ENOENT);
1727 	for (es = execsw; *es; es++)
1728 		if (*es != execsw_arg)
1729 			count++;
1730 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1731 	xs = newexecsw;
1732 	for (es = execsw; *es; es++)
1733 		if (*es != execsw_arg)
1734 			*xs++ = *es;
1735 	*xs = NULL;
1736 	if (execsw)
1737 		free(execsw, M_TEMP);
1738 	execsw = newexecsw;
1739 	return (0);
1740 }
1741