xref: /freebsd/sys/kern/kern_exec.c (revision 86aa9539fef591a363b06a0ebd3aa7a07f4c1579)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 1993, David Greenman
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include "opt_capsicum.h"
33 #include "opt_hwpmc_hooks.h"
34 #include "opt_ktrace.h"
35 #include "opt_vm.h"
36 
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/acct.h>
40 #include <sys/capsicum.h>
41 #include <sys/eventhandler.h>
42 #include <sys/exec.h>
43 #include <sys/fcntl.h>
44 #include <sys/filedesc.h>
45 #include <sys/imgact.h>
46 #include <sys/imgact_elf.h>
47 #include <sys/kernel.h>
48 #include <sys/lock.h>
49 #include <sys/malloc.h>
50 #include <sys/mman.h>
51 #include <sys/mount.h>
52 #include <sys/mutex.h>
53 #include <sys/namei.h>
54 #include <sys/pioctl.h>
55 #include <sys/priv.h>
56 #include <sys/proc.h>
57 #include <sys/ptrace.h>
58 #include <sys/resourcevar.h>
59 #include <sys/rwlock.h>
60 #include <sys/sched.h>
61 #include <sys/sdt.h>
62 #include <sys/sf_buf.h>
63 #include <sys/shm.h>
64 #include <sys/signalvar.h>
65 #include <sys/smp.h>
66 #include <sys/stat.h>
67 #include <sys/syscallsubr.h>
68 #include <sys/sysctl.h>
69 #include <sys/sysent.h>
70 #include <sys/sysproto.h>
71 #include <sys/vnode.h>
72 #include <sys/wait.h>
73 #ifdef KTRACE
74 #include <sys/ktrace.h>
75 #endif
76 
77 #include <vm/vm.h>
78 #include <vm/vm_param.h>
79 #include <vm/pmap.h>
80 #include <vm/vm_page.h>
81 #include <vm/vm_map.h>
82 #include <vm/vm_kern.h>
83 #include <vm/vm_extern.h>
84 #include <vm/vm_object.h>
85 #include <vm/vm_pager.h>
86 
87 #ifdef	HWPMC_HOOKS
88 #include <sys/pmckern.h>
89 #endif
90 
91 #include <machine/reg.h>
92 
93 #include <security/audit/audit.h>
94 #include <security/mac/mac_framework.h>
95 
96 #ifdef KDTRACE_HOOKS
97 #include <sys/dtrace_bsd.h>
98 dtrace_execexit_func_t	dtrace_fasttrap_exec;
99 #endif
100 
101 SDT_PROVIDER_DECLARE(proc);
102 SDT_PROBE_DEFINE1(proc, , , exec, "char *");
103 SDT_PROBE_DEFINE1(proc, , , exec__failure, "int");
104 SDT_PROBE_DEFINE1(proc, , , exec__success, "char *");
105 
106 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
107 
108 int coredump_pack_fileinfo = 1;
109 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_fileinfo, CTLFLAG_RWTUN,
110     &coredump_pack_fileinfo, 0,
111     "Enable file path packing in 'procstat -f' coredump notes");
112 
113 int coredump_pack_vmmapinfo = 1;
114 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_vmmapinfo, CTLFLAG_RWTUN,
115     &coredump_pack_vmmapinfo, 0,
116     "Enable file path packing in 'procstat -v' coredump notes");
117 
118 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
119 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
120 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
121 static int do_execve(struct thread *td, struct image_args *args,
122     struct mac *mac_p);
123 
124 /* XXX This should be vm_size_t. */
125 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD|
126     CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_ps_strings, "LU", "");
127 
128 /* XXX This should be vm_size_t. */
129 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD|
130     CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_usrstack, "LU", "");
131 
132 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD|CTLFLAG_MPSAFE,
133     NULL, 0, sysctl_kern_stackprot, "I", "");
134 
135 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
136 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
137     &ps_arg_cache_limit, 0, "");
138 
139 static int disallow_high_osrel;
140 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW,
141     &disallow_high_osrel, 0,
142     "Disallow execution of binaries built for higher version of the world");
143 
144 static int map_at_zero = 0;
145 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RWTUN, &map_at_zero, 0,
146     "Permit processes to map an object at virtual address 0.");
147 
148 static int
149 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
150 {
151 	struct proc *p;
152 	int error;
153 
154 	p = curproc;
155 #ifdef SCTL_MASK32
156 	if (req->flags & SCTL_MASK32) {
157 		unsigned int val;
158 		val = (unsigned int)p->p_sysent->sv_psstrings;
159 		error = SYSCTL_OUT(req, &val, sizeof(val));
160 	} else
161 #endif
162 		error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
163 		   sizeof(p->p_sysent->sv_psstrings));
164 	return error;
165 }
166 
167 static int
168 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
169 {
170 	struct proc *p;
171 	int error;
172 
173 	p = curproc;
174 #ifdef SCTL_MASK32
175 	if (req->flags & SCTL_MASK32) {
176 		unsigned int val;
177 		val = (unsigned int)p->p_sysent->sv_usrstack;
178 		error = SYSCTL_OUT(req, &val, sizeof(val));
179 	} else
180 #endif
181 		error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
182 		    sizeof(p->p_sysent->sv_usrstack));
183 	return error;
184 }
185 
186 static int
187 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
188 {
189 	struct proc *p;
190 
191 	p = curproc;
192 	return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
193 	    sizeof(p->p_sysent->sv_stackprot)));
194 }
195 
196 /*
197  * Each of the items is a pointer to a `const struct execsw', hence the
198  * double pointer here.
199  */
200 static const struct execsw **execsw;
201 
202 #ifndef _SYS_SYSPROTO_H_
203 struct execve_args {
204 	char    *fname;
205 	char    **argv;
206 	char    **envv;
207 };
208 #endif
209 
210 int
211 sys_execve(struct thread *td, struct execve_args *uap)
212 {
213 	struct image_args args;
214 	struct vmspace *oldvmspace;
215 	int error;
216 
217 	error = pre_execve(td, &oldvmspace);
218 	if (error != 0)
219 		return (error);
220 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
221 	    uap->argv, uap->envv);
222 	if (error == 0)
223 		error = kern_execve(td, &args, NULL);
224 	post_execve(td, error, oldvmspace);
225 	return (error);
226 }
227 
228 #ifndef _SYS_SYSPROTO_H_
229 struct fexecve_args {
230 	int	fd;
231 	char	**argv;
232 	char	**envv;
233 }
234 #endif
235 int
236 sys_fexecve(struct thread *td, struct fexecve_args *uap)
237 {
238 	struct image_args args;
239 	struct vmspace *oldvmspace;
240 	int error;
241 
242 	error = pre_execve(td, &oldvmspace);
243 	if (error != 0)
244 		return (error);
245 	error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
246 	    uap->argv, uap->envv);
247 	if (error == 0) {
248 		args.fd = uap->fd;
249 		error = kern_execve(td, &args, NULL);
250 	}
251 	post_execve(td, error, oldvmspace);
252 	return (error);
253 }
254 
255 #ifndef _SYS_SYSPROTO_H_
256 struct __mac_execve_args {
257 	char	*fname;
258 	char	**argv;
259 	char	**envv;
260 	struct mac	*mac_p;
261 };
262 #endif
263 
264 int
265 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap)
266 {
267 #ifdef MAC
268 	struct image_args args;
269 	struct vmspace *oldvmspace;
270 	int error;
271 
272 	error = pre_execve(td, &oldvmspace);
273 	if (error != 0)
274 		return (error);
275 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
276 	    uap->argv, uap->envv);
277 	if (error == 0)
278 		error = kern_execve(td, &args, uap->mac_p);
279 	post_execve(td, error, oldvmspace);
280 	return (error);
281 #else
282 	return (ENOSYS);
283 #endif
284 }
285 
286 int
287 pre_execve(struct thread *td, struct vmspace **oldvmspace)
288 {
289 	struct proc *p;
290 	int error;
291 
292 	KASSERT(td == curthread, ("non-current thread %p", td));
293 	error = 0;
294 	p = td->td_proc;
295 	if ((p->p_flag & P_HADTHREADS) != 0) {
296 		PROC_LOCK(p);
297 		if (thread_single(p, SINGLE_BOUNDARY) != 0)
298 			error = ERESTART;
299 		PROC_UNLOCK(p);
300 	}
301 	KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0,
302 	    ("nested execve"));
303 	*oldvmspace = p->p_vmspace;
304 	return (error);
305 }
306 
307 void
308 post_execve(struct thread *td, int error, struct vmspace *oldvmspace)
309 {
310 	struct proc *p;
311 
312 	KASSERT(td == curthread, ("non-current thread %p", td));
313 	p = td->td_proc;
314 	if ((p->p_flag & P_HADTHREADS) != 0) {
315 		PROC_LOCK(p);
316 		/*
317 		 * If success, we upgrade to SINGLE_EXIT state to
318 		 * force other threads to suicide.
319 		 */
320 		if (error == EJUSTRETURN)
321 			thread_single(p, SINGLE_EXIT);
322 		else
323 			thread_single_end(p, SINGLE_BOUNDARY);
324 		PROC_UNLOCK(p);
325 	}
326 	if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
327 		KASSERT(p->p_vmspace != oldvmspace,
328 		    ("oldvmspace still used"));
329 		vmspace_free(oldvmspace);
330 		td->td_pflags &= ~TDP_EXECVMSPC;
331 	}
332 }
333 
334 /*
335  * XXX: kern_execve has the astonishing property of not always returning to
336  * the caller.  If sufficiently bad things happen during the call to
337  * do_execve(), it can end up calling exit1(); as a result, callers must
338  * avoid doing anything which they might need to undo (e.g., allocating
339  * memory).
340  */
341 int
342 kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p)
343 {
344 
345 	AUDIT_ARG_ARGV(args->begin_argv, args->argc,
346 	    exec_args_get_begin_envv(args) - args->begin_argv);
347 	AUDIT_ARG_ENVV(exec_args_get_begin_envv(args), args->envc,
348 	    args->endp - exec_args_get_begin_envv(args));
349 	return (do_execve(td, args, mac_p));
350 }
351 
352 /*
353  * In-kernel implementation of execve().  All arguments are assumed to be
354  * userspace pointers from the passed thread.
355  */
356 static int
357 do_execve(struct thread *td, struct image_args *args, struct mac *mac_p)
358 {
359 	struct proc *p = td->td_proc;
360 	struct nameidata nd;
361 	struct ucred *oldcred;
362 	struct uidinfo *euip = NULL;
363 	register_t *stack_base;
364 	int error, i;
365 	struct image_params image_params, *imgp;
366 	struct vattr attr;
367 	int (*img_first)(struct image_params *);
368 	struct pargs *oldargs = NULL, *newargs = NULL;
369 	struct sigacts *oldsigacts = NULL, *newsigacts = NULL;
370 #ifdef KTRACE
371 	struct vnode *tracevp = NULL;
372 	struct ucred *tracecred = NULL;
373 #endif
374 	struct vnode *oldtextvp = NULL, *newtextvp;
375 	int credential_changing;
376 #ifdef MAC
377 	struct label *interpvplabel = NULL;
378 	int will_transition;
379 #endif
380 #ifdef HWPMC_HOOKS
381 	struct pmckern_procexec pe;
382 #endif
383 	static const char fexecv_proc_title[] = "(fexecv)";
384 
385 	imgp = &image_params;
386 
387 	/*
388 	 * Lock the process and set the P_INEXEC flag to indicate that
389 	 * it should be left alone until we're done here.  This is
390 	 * necessary to avoid race conditions - e.g. in ptrace() -
391 	 * that might allow a local user to illicitly obtain elevated
392 	 * privileges.
393 	 */
394 	PROC_LOCK(p);
395 	KASSERT((p->p_flag & P_INEXEC) == 0,
396 	    ("%s(): process already has P_INEXEC flag", __func__));
397 	p->p_flag |= P_INEXEC;
398 	PROC_UNLOCK(p);
399 
400 	/*
401 	 * Initialize part of the common data
402 	 */
403 	bzero(imgp, sizeof(*imgp));
404 	imgp->proc = p;
405 	imgp->attr = &attr;
406 	imgp->args = args;
407 	oldcred = p->p_ucred;
408 
409 #ifdef MAC
410 	error = mac_execve_enter(imgp, mac_p);
411 	if (error)
412 		goto exec_fail;
413 #endif
414 
415 	/*
416 	 * Translate the file name. namei() returns a vnode pointer
417 	 *	in ni_vp among other things.
418 	 *
419 	 * XXXAUDIT: It would be desirable to also audit the name of the
420 	 * interpreter if this is an interpreted binary.
421 	 */
422 	if (args->fname != NULL) {
423 		NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW |
424 		    SAVENAME | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
425 	}
426 
427 	SDT_PROBE1(proc, , , exec, args->fname);
428 
429 interpret:
430 	if (args->fname != NULL) {
431 #ifdef CAPABILITY_MODE
432 		/*
433 		 * While capability mode can't reach this point via direct
434 		 * path arguments to execve(), we also don't allow
435 		 * interpreters to be used in capability mode (for now).
436 		 * Catch indirect lookups and return a permissions error.
437 		 */
438 		if (IN_CAPABILITY_MODE(td)) {
439 			error = ECAPMODE;
440 			goto exec_fail;
441 		}
442 #endif
443 		error = namei(&nd);
444 		if (error)
445 			goto exec_fail;
446 
447 		newtextvp = nd.ni_vp;
448 		imgp->vp = newtextvp;
449 	} else {
450 		AUDIT_ARG_FD(args->fd);
451 		/*
452 		 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
453 		 */
454 		error = fgetvp_exec(td, args->fd, &cap_fexecve_rights, &newtextvp);
455 		if (error)
456 			goto exec_fail;
457 		vn_lock(newtextvp, LK_SHARED | LK_RETRY);
458 		AUDIT_ARG_VNODE1(newtextvp);
459 		imgp->vp = newtextvp;
460 	}
461 
462 	/*
463 	 * Check file permissions.  Also 'opens' file and sets its vnode to
464 	 * text mode.
465 	 */
466 	error = exec_check_permissions(imgp);
467 	if (error)
468 		goto exec_fail_dealloc;
469 
470 	imgp->object = imgp->vp->v_object;
471 	if (imgp->object != NULL)
472 		vm_object_reference(imgp->object);
473 
474 	error = exec_map_first_page(imgp);
475 	if (error)
476 		goto exec_fail_dealloc;
477 
478 	imgp->proc->p_osrel = 0;
479 	imgp->proc->p_fctl0 = 0;
480 
481 	/*
482 	 * Implement image setuid/setgid.
483 	 *
484 	 * Determine new credentials before attempting image activators
485 	 * so that it can be used by process_exec handlers to determine
486 	 * credential/setid changes.
487 	 *
488 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
489 	 * the process is being traced.
490 	 *
491 	 * We disable setuid/setgid/etc in capability mode on the basis
492 	 * that most setugid applications are not written with that
493 	 * environment in mind, and will therefore almost certainly operate
494 	 * incorrectly. In principle there's no reason that setugid
495 	 * applications might not be useful in capability mode, so we may want
496 	 * to reconsider this conservative design choice in the future.
497 	 *
498 	 * XXXMAC: For the time being, use NOSUID to also prohibit
499 	 * transitions on the file system.
500 	 */
501 	credential_changing = 0;
502 	credential_changing |= (attr.va_mode & S_ISUID) &&
503 	    oldcred->cr_uid != attr.va_uid;
504 	credential_changing |= (attr.va_mode & S_ISGID) &&
505 	    oldcred->cr_gid != attr.va_gid;
506 #ifdef MAC
507 	will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
508 	    interpvplabel, imgp);
509 	credential_changing |= will_transition;
510 #endif
511 
512 	/* Don't inherit PROC_PDEATHSIG_CTL value if setuid/setgid. */
513 	if (credential_changing)
514 		imgp->proc->p_pdeathsig = 0;
515 
516 	if (credential_changing &&
517 #ifdef CAPABILITY_MODE
518 	    ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
519 #endif
520 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
521 	    (p->p_flag & P_TRACED) == 0) {
522 		imgp->credential_setid = true;
523 		VOP_UNLOCK(imgp->vp, 0);
524 		imgp->newcred = crdup(oldcred);
525 		if (attr.va_mode & S_ISUID) {
526 			euip = uifind(attr.va_uid);
527 			change_euid(imgp->newcred, euip);
528 		}
529 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
530 		if (attr.va_mode & S_ISGID)
531 			change_egid(imgp->newcred, attr.va_gid);
532 		/*
533 		 * Implement correct POSIX saved-id behavior.
534 		 *
535 		 * XXXMAC: Note that the current logic will save the
536 		 * uid and gid if a MAC domain transition occurs, even
537 		 * though maybe it shouldn't.
538 		 */
539 		change_svuid(imgp->newcred, imgp->newcred->cr_uid);
540 		change_svgid(imgp->newcred, imgp->newcred->cr_gid);
541 	} else {
542 		/*
543 		 * Implement correct POSIX saved-id behavior.
544 		 *
545 		 * XXX: It's not clear that the existing behavior is
546 		 * POSIX-compliant.  A number of sources indicate that the
547 		 * saved uid/gid should only be updated if the new ruid is
548 		 * not equal to the old ruid, or the new euid is not equal
549 		 * to the old euid and the new euid is not equal to the old
550 		 * ruid.  The FreeBSD code always updates the saved uid/gid.
551 		 * Also, this code uses the new (replaced) euid and egid as
552 		 * the source, which may or may not be the right ones to use.
553 		 */
554 		if (oldcred->cr_svuid != oldcred->cr_uid ||
555 		    oldcred->cr_svgid != oldcred->cr_gid) {
556 			VOP_UNLOCK(imgp->vp, 0);
557 			imgp->newcred = crdup(oldcred);
558 			vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
559 			change_svuid(imgp->newcred, imgp->newcred->cr_uid);
560 			change_svgid(imgp->newcred, imgp->newcred->cr_gid);
561 		}
562 	}
563 	/* The new credentials are installed into the process later. */
564 
565 	/*
566 	 * Do the best to calculate the full path to the image file.
567 	 */
568 	if (args->fname != NULL && args->fname[0] == '/')
569 		imgp->execpath = args->fname;
570 	else {
571 		VOP_UNLOCK(imgp->vp, 0);
572 		if (vn_fullpath(td, imgp->vp, &imgp->execpath,
573 		    &imgp->freepath) != 0)
574 			imgp->execpath = args->fname;
575 		vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY);
576 	}
577 
578 	/*
579 	 *	If the current process has a special image activator it
580 	 *	wants to try first, call it.   For example, emulating shell
581 	 *	scripts differently.
582 	 */
583 	error = -1;
584 	if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
585 		error = img_first(imgp);
586 
587 	/*
588 	 *	Loop through the list of image activators, calling each one.
589 	 *	An activator returns -1 if there is no match, 0 on success,
590 	 *	and an error otherwise.
591 	 */
592 	for (i = 0; error == -1 && execsw[i]; ++i) {
593 		if (execsw[i]->ex_imgact == NULL ||
594 		    execsw[i]->ex_imgact == img_first) {
595 			continue;
596 		}
597 		error = (*execsw[i]->ex_imgact)(imgp);
598 	}
599 
600 	if (error) {
601 		if (error == -1)
602 			error = ENOEXEC;
603 		goto exec_fail_dealloc;
604 	}
605 
606 	/*
607 	 * Special interpreter operation, cleanup and loop up to try to
608 	 * activate the interpreter.
609 	 */
610 	if (imgp->interpreted) {
611 		exec_unmap_first_page(imgp);
612 		/*
613 		 * The text reference needs to be removed for scripts.
614 		 * There is a short period before we determine that
615 		 * something is a script where text reference is active.
616 		 * The vnode lock is held over this entire period
617 		 * so nothing should illegitimately be blocked.
618 		 */
619 		VOP_UNSET_TEXT_CHECKED(imgp->vp);
620 		/* free name buffer and old vnode */
621 		if (args->fname != NULL)
622 			NDFREE(&nd, NDF_ONLY_PNBUF);
623 #ifdef MAC
624 		mac_execve_interpreter_enter(newtextvp, &interpvplabel);
625 #endif
626 		if (imgp->opened) {
627 			VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td);
628 			imgp->opened = 0;
629 		}
630 		vput(newtextvp);
631 		vm_object_deallocate(imgp->object);
632 		imgp->object = NULL;
633 		imgp->credential_setid = false;
634 		if (imgp->newcred != NULL) {
635 			crfree(imgp->newcred);
636 			imgp->newcred = NULL;
637 		}
638 		imgp->execpath = NULL;
639 		free(imgp->freepath, M_TEMP);
640 		imgp->freepath = NULL;
641 		/* set new name to that of the interpreter */
642 		NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME,
643 		    UIO_SYSSPACE, imgp->interpreter_name, td);
644 		args->fname = imgp->interpreter_name;
645 		goto interpret;
646 	}
647 
648 	/*
649 	 * NB: We unlock the vnode here because it is believed that none
650 	 * of the sv_copyout_strings/sv_fixup operations require the vnode.
651 	 */
652 	VOP_UNLOCK(imgp->vp, 0);
653 
654 	if (disallow_high_osrel &&
655 	    P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
656 		error = ENOEXEC;
657 		uprintf("Osrel %d for image %s too high\n", p->p_osrel,
658 		    imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
659 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
660 		goto exec_fail_dealloc;
661 	}
662 
663 	/* ABI enforces the use of Capsicum. Switch into capabilities mode. */
664 	if (SV_PROC_FLAG(p, SV_CAPSICUM))
665 		sys_cap_enter(td, NULL);
666 
667 	/*
668 	 * Copy out strings (args and env) and initialize stack base.
669 	 */
670 	stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
671 
672 	/*
673 	 * Stack setup.
674 	 */
675 	error = (*p->p_sysent->sv_fixup)(&stack_base, imgp);
676 	if (error != 0) {
677 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
678 		goto exec_fail_dealloc;
679 	}
680 
681 	if (args->fdp != NULL) {
682 		/* Install a brand new file descriptor table. */
683 		fdinstall_remapped(td, args->fdp);
684 		args->fdp = NULL;
685 	} else {
686 		/*
687 		 * Keep on using the existing file descriptor table. For
688 		 * security and other reasons, the file descriptor table
689 		 * cannot be shared after an exec.
690 		 */
691 		fdunshare(td);
692 		/* close files on exec */
693 		fdcloseexec(td);
694 	}
695 
696 	/*
697 	 * Malloc things before we need locks.
698 	 */
699 	i = exec_args_get_begin_envv(imgp->args) - imgp->args->begin_argv;
700 	/* Cache arguments if they fit inside our allowance */
701 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
702 		newargs = pargs_alloc(i);
703 		bcopy(imgp->args->begin_argv, newargs->ar_args, i);
704 	}
705 
706 	/*
707 	 * For security and other reasons, signal handlers cannot
708 	 * be shared after an exec. The new process gets a copy of the old
709 	 * handlers. In execsigs(), the new process will have its signals
710 	 * reset.
711 	 */
712 	if (sigacts_shared(p->p_sigacts)) {
713 		oldsigacts = p->p_sigacts;
714 		newsigacts = sigacts_alloc();
715 		sigacts_copy(newsigacts, oldsigacts);
716 	}
717 
718 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
719 
720 	PROC_LOCK(p);
721 	if (oldsigacts)
722 		p->p_sigacts = newsigacts;
723 	/* Stop profiling */
724 	stopprofclock(p);
725 
726 	/* reset caught signals */
727 	execsigs(p);
728 
729 	/* name this process - nameiexec(p, ndp) */
730 	bzero(p->p_comm, sizeof(p->p_comm));
731 	if (args->fname)
732 		bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
733 		    min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
734 	else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0)
735 		bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
736 	bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
737 #ifdef KTR
738 	sched_clear_tdname(td);
739 #endif
740 
741 	/*
742 	 * mark as execed, wakeup the process that vforked (if any) and tell
743 	 * it that it now has its own resources back
744 	 */
745 	p->p_flag |= P_EXEC;
746 	if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0)
747 		p->p_flag2 &= ~P2_NOTRACE;
748 	if (p->p_flag & P_PPWAIT) {
749 		p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
750 		cv_broadcast(&p->p_pwait);
751 		/* STOPs are no longer ignored, arrange for AST */
752 		signotify(td);
753 	}
754 
755 	/*
756 	 * Implement image setuid/setgid installation.
757 	 */
758 	if (imgp->credential_setid) {
759 		/*
760 		 * Turn off syscall tracing for set-id programs, except for
761 		 * root.  Record any set-id flags first to make sure that
762 		 * we do not regain any tracing during a possible block.
763 		 */
764 		setsugid(p);
765 
766 #ifdef KTRACE
767 		if (p->p_tracecred != NULL &&
768 		    priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED))
769 			ktrprocexec(p, &tracecred, &tracevp);
770 #endif
771 		/*
772 		 * Close any file descriptors 0..2 that reference procfs,
773 		 * then make sure file descriptors 0..2 are in use.
774 		 *
775 		 * Both fdsetugidsafety() and fdcheckstd() may call functions
776 		 * taking sleepable locks, so temporarily drop our locks.
777 		 */
778 		PROC_UNLOCK(p);
779 		VOP_UNLOCK(imgp->vp, 0);
780 		fdsetugidsafety(td);
781 		error = fdcheckstd(td);
782 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
783 		if (error != 0)
784 			goto exec_fail_dealloc;
785 		PROC_LOCK(p);
786 #ifdef MAC
787 		if (will_transition) {
788 			mac_vnode_execve_transition(oldcred, imgp->newcred,
789 			    imgp->vp, interpvplabel, imgp);
790 		}
791 #endif
792 	} else {
793 		if (oldcred->cr_uid == oldcred->cr_ruid &&
794 		    oldcred->cr_gid == oldcred->cr_rgid)
795 			p->p_flag &= ~P_SUGID;
796 	}
797 	/*
798 	 * Set the new credentials.
799 	 */
800 	if (imgp->newcred != NULL) {
801 		proc_set_cred(p, imgp->newcred);
802 		crfree(oldcred);
803 		oldcred = NULL;
804 	}
805 
806 	/*
807 	 * Store the vp for use in procfs.  This vnode was referenced by namei
808 	 * or fgetvp_exec.
809 	 */
810 	oldtextvp = p->p_textvp;
811 	p->p_textvp = newtextvp;
812 
813 #ifdef KDTRACE_HOOKS
814 	/*
815 	 * Tell the DTrace fasttrap provider about the exec if it
816 	 * has declared an interest.
817 	 */
818 	if (dtrace_fasttrap_exec)
819 		dtrace_fasttrap_exec(p);
820 #endif
821 
822 	/*
823 	 * Notify others that we exec'd, and clear the P_INEXEC flag
824 	 * as we're now a bona fide freshly-execed process.
825 	 */
826 	KNOTE_LOCKED(p->p_klist, NOTE_EXEC);
827 	p->p_flag &= ~P_INEXEC;
828 
829 	/* clear "fork but no exec" flag, as we _are_ execing */
830 	p->p_acflag &= ~AFORK;
831 
832 	/*
833 	 * Free any previous argument cache and replace it with
834 	 * the new argument cache, if any.
835 	 */
836 	oldargs = p->p_args;
837 	p->p_args = newargs;
838 	newargs = NULL;
839 
840 	PROC_UNLOCK(p);
841 
842 #ifdef	HWPMC_HOOKS
843 	/*
844 	 * Check if system-wide sampling is in effect or if the
845 	 * current process is using PMCs.  If so, do exec() time
846 	 * processing.  This processing needs to happen AFTER the
847 	 * P_INEXEC flag is cleared.
848 	 */
849 	if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
850 		VOP_UNLOCK(imgp->vp, 0);
851 		pe.pm_credentialschanged = credential_changing;
852 		pe.pm_entryaddr = imgp->entry_addr;
853 
854 		PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
855 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
856 	}
857 #endif
858 
859 	/* Set values passed into the program in registers. */
860 	(*p->p_sysent->sv_setregs)(td, imgp, (u_long)(uintptr_t)stack_base);
861 
862 	vfs_mark_atime(imgp->vp, td->td_ucred);
863 
864 	SDT_PROBE1(proc, , , exec__success, args->fname);
865 
866 exec_fail_dealloc:
867 	if (imgp->firstpage != NULL)
868 		exec_unmap_first_page(imgp);
869 
870 	if (imgp->vp != NULL) {
871 		if (args->fname)
872 			NDFREE(&nd, NDF_ONLY_PNBUF);
873 		if (imgp->opened)
874 			VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
875 		if (imgp->textset)
876 			VOP_UNSET_TEXT_CHECKED(imgp->vp);
877 		if (error != 0)
878 			vput(imgp->vp);
879 		else
880 			VOP_UNLOCK(imgp->vp, 0);
881 	}
882 
883 	if (imgp->object != NULL)
884 		vm_object_deallocate(imgp->object);
885 
886 	free(imgp->freepath, M_TEMP);
887 
888 	if (error == 0) {
889 		if (p->p_ptevents & PTRACE_EXEC) {
890 			PROC_LOCK(p);
891 			if (p->p_ptevents & PTRACE_EXEC)
892 				td->td_dbgflags |= TDB_EXEC;
893 			PROC_UNLOCK(p);
894 		}
895 
896 		/*
897 		 * Stop the process here if its stop event mask has
898 		 * the S_EXEC bit set.
899 		 */
900 		STOPEVENT(p, S_EXEC, 0);
901 	} else {
902 exec_fail:
903 		/* we're done here, clear P_INEXEC */
904 		PROC_LOCK(p);
905 		p->p_flag &= ~P_INEXEC;
906 		PROC_UNLOCK(p);
907 
908 		SDT_PROBE1(proc, , , exec__failure, error);
909 	}
910 
911 	if (imgp->newcred != NULL && oldcred != NULL)
912 		crfree(imgp->newcred);
913 
914 #ifdef MAC
915 	mac_execve_exit(imgp);
916 	mac_execve_interpreter_exit(interpvplabel);
917 #endif
918 	exec_free_args(args);
919 
920 	/*
921 	 * Handle deferred decrement of ref counts.
922 	 */
923 	if (oldtextvp != NULL)
924 		vrele(oldtextvp);
925 #ifdef KTRACE
926 	if (tracevp != NULL)
927 		vrele(tracevp);
928 	if (tracecred != NULL)
929 		crfree(tracecred);
930 #endif
931 	pargs_drop(oldargs);
932 	pargs_drop(newargs);
933 	if (oldsigacts != NULL)
934 		sigacts_free(oldsigacts);
935 	if (euip != NULL)
936 		uifree(euip);
937 
938 	if (error && imgp->vmspace_destroyed) {
939 		/* sorry, no more process anymore. exit gracefully */
940 		exit1(td, 0, SIGABRT);
941 		/* NOT REACHED */
942 	}
943 
944 #ifdef KTRACE
945 	if (error == 0)
946 		ktrprocctor(p);
947 #endif
948 
949 	/*
950 	 * We don't want cpu_set_syscall_retval() to overwrite any of
951 	 * the register values put in place by exec_setregs().
952 	 * Implementations of cpu_set_syscall_retval() will leave
953 	 * registers unmodified when returning EJUSTRETURN.
954 	 */
955 	return (error == 0 ? EJUSTRETURN : error);
956 }
957 
958 int
959 exec_map_first_page(struct image_params *imgp)
960 {
961 	int rv, i, after, initial_pagein;
962 	vm_page_t ma[VM_INITIAL_PAGEIN];
963 	vm_object_t object;
964 
965 	if (imgp->firstpage != NULL)
966 		exec_unmap_first_page(imgp);
967 
968 	object = imgp->vp->v_object;
969 	if (object == NULL)
970 		return (EACCES);
971 	VM_OBJECT_WLOCK(object);
972 #if VM_NRESERVLEVEL > 0
973 	vm_object_color(object, 0);
974 #endif
975 	ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY);
976 	if (ma[0]->valid != VM_PAGE_BITS_ALL) {
977 		vm_page_xbusy(ma[0]);
978 		if (!vm_pager_has_page(object, 0, NULL, &after)) {
979 			vm_page_lock(ma[0]);
980 			vm_page_free(ma[0]);
981 			vm_page_unlock(ma[0]);
982 			VM_OBJECT_WUNLOCK(object);
983 			return (EIO);
984 		}
985 		initial_pagein = min(after, VM_INITIAL_PAGEIN);
986 		KASSERT(initial_pagein <= object->size,
987 		    ("%s: initial_pagein %d object->size %ju",
988 		    __func__, initial_pagein, (uintmax_t )object->size));
989 		for (i = 1; i < initial_pagein; i++) {
990 			if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) {
991 				if (ma[i]->valid)
992 					break;
993 				if (!vm_page_tryxbusy(ma[i]))
994 					break;
995 			} else {
996 				ma[i] = vm_page_alloc(object, i,
997 				    VM_ALLOC_NORMAL);
998 				if (ma[i] == NULL)
999 					break;
1000 			}
1001 		}
1002 		initial_pagein = i;
1003 		rv = vm_pager_get_pages(object, ma, initial_pagein, NULL, NULL);
1004 		if (rv != VM_PAGER_OK) {
1005 			for (i = 0; i < initial_pagein; i++) {
1006 				vm_page_lock(ma[i]);
1007 				vm_page_free(ma[i]);
1008 				vm_page_unlock(ma[i]);
1009 			}
1010 			VM_OBJECT_WUNLOCK(object);
1011 			return (EIO);
1012 		}
1013 		vm_page_xunbusy(ma[0]);
1014 		for (i = 1; i < initial_pagein; i++)
1015 			vm_page_readahead_finish(ma[i]);
1016 	}
1017 	vm_page_lock(ma[0]);
1018 	vm_page_wire(ma[0]);
1019 	vm_page_unlock(ma[0]);
1020 	VM_OBJECT_WUNLOCK(object);
1021 
1022 	imgp->firstpage = sf_buf_alloc(ma[0], 0);
1023 	imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
1024 
1025 	return (0);
1026 }
1027 
1028 void
1029 exec_unmap_first_page(struct image_params *imgp)
1030 {
1031 	vm_page_t m;
1032 
1033 	if (imgp->firstpage != NULL) {
1034 		m = sf_buf_page(imgp->firstpage);
1035 		sf_buf_free(imgp->firstpage);
1036 		imgp->firstpage = NULL;
1037 		vm_page_lock(m);
1038 		vm_page_unwire(m, PQ_ACTIVE);
1039 		vm_page_unlock(m);
1040 	}
1041 }
1042 
1043 /*
1044  * Destroy old address space, and allocate a new stack.
1045  *	The new stack is only sgrowsiz large because it is grown
1046  *	automatically on a page fault.
1047  */
1048 int
1049 exec_new_vmspace(struct image_params *imgp, struct sysentvec *sv)
1050 {
1051 	int error;
1052 	struct proc *p = imgp->proc;
1053 	struct vmspace *vmspace = p->p_vmspace;
1054 	vm_object_t obj;
1055 	struct rlimit rlim_stack;
1056 	vm_offset_t sv_minuser, stack_addr;
1057 	vm_map_t map;
1058 	u_long ssiz;
1059 
1060 	imgp->vmspace_destroyed = 1;
1061 	imgp->sysent = sv;
1062 
1063 	/* May be called with Giant held */
1064 	EVENTHANDLER_DIRECT_INVOKE(process_exec, p, imgp);
1065 
1066 	/*
1067 	 * Blow away entire process VM, if address space not shared,
1068 	 * otherwise, create a new VM space so that other threads are
1069 	 * not disrupted
1070 	 */
1071 	map = &vmspace->vm_map;
1072 	if (map_at_zero)
1073 		sv_minuser = sv->sv_minuser;
1074 	else
1075 		sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1076 	if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser &&
1077 	    vm_map_max(map) == sv->sv_maxuser &&
1078 	    cpu_exec_vmspace_reuse(p, map)) {
1079 		shmexit(vmspace);
1080 		pmap_remove_pages(vmspace_pmap(vmspace));
1081 		vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1082 		/*
1083 		 * An exec terminates mlockall(MCL_FUTURE), ASLR state
1084 		 * must be re-evaluated.
1085 		 */
1086 		vm_map_lock(map);
1087 		vm_map_modflags(map, 0, MAP_WIREFUTURE | MAP_ASLR |
1088 		    MAP_ASLR_IGNSTART);
1089 		vm_map_unlock(map);
1090 	} else {
1091 		error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1092 		if (error)
1093 			return (error);
1094 		vmspace = p->p_vmspace;
1095 		map = &vmspace->vm_map;
1096 	}
1097 	map->flags |= imgp->map_flags;
1098 
1099 	/* Map a shared page */
1100 	obj = sv->sv_shared_page_obj;
1101 	if (obj != NULL) {
1102 		vm_object_reference(obj);
1103 		error = vm_map_fixed(map, obj, 0,
1104 		    sv->sv_shared_page_base, sv->sv_shared_page_len,
1105 		    VM_PROT_READ | VM_PROT_EXECUTE,
1106 		    VM_PROT_READ | VM_PROT_EXECUTE,
1107 		    MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1108 		if (error != KERN_SUCCESS) {
1109 			vm_object_deallocate(obj);
1110 			return (vm_mmap_to_errno(error));
1111 		}
1112 	}
1113 
1114 	/* Allocate a new stack */
1115 	if (imgp->stack_sz != 0) {
1116 		ssiz = trunc_page(imgp->stack_sz);
1117 		PROC_LOCK(p);
1118 		lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack);
1119 		PROC_UNLOCK(p);
1120 		if (ssiz > rlim_stack.rlim_max)
1121 			ssiz = rlim_stack.rlim_max;
1122 		if (ssiz > rlim_stack.rlim_cur) {
1123 			rlim_stack.rlim_cur = ssiz;
1124 			kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack);
1125 		}
1126 	} else if (sv->sv_maxssiz != NULL) {
1127 		ssiz = *sv->sv_maxssiz;
1128 	} else {
1129 		ssiz = maxssiz;
1130 	}
1131 	imgp->eff_stack_sz = lim_cur(curthread, RLIMIT_STACK);
1132 	if (ssiz < imgp->eff_stack_sz)
1133 		imgp->eff_stack_sz = ssiz;
1134 	stack_addr = sv->sv_usrstack - ssiz;
1135 	error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1136 	    obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1137 	    sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1138 	if (error != KERN_SUCCESS)
1139 		return (vm_mmap_to_errno(error));
1140 
1141 	/*
1142 	 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they
1143 	 * are still used to enforce the stack rlimit on the process stack.
1144 	 */
1145 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1146 	vmspace->vm_maxsaddr = (char *)stack_addr;
1147 
1148 	return (0);
1149 }
1150 
1151 /*
1152  * Copy out argument and environment strings from the old process address
1153  * space into the temporary string buffer.
1154  */
1155 int
1156 exec_copyin_args(struct image_args *args, const char *fname,
1157     enum uio_seg segflg, char **argv, char **envv)
1158 {
1159 	u_long arg, env;
1160 	int error;
1161 
1162 	bzero(args, sizeof(*args));
1163 	if (argv == NULL)
1164 		return (EFAULT);
1165 
1166 	/*
1167 	 * Allocate demand-paged memory for the file name, argument, and
1168 	 * environment strings.
1169 	 */
1170 	error = exec_alloc_args(args);
1171 	if (error != 0)
1172 		return (error);
1173 
1174 	/*
1175 	 * Copy the file name.
1176 	 */
1177 	error = exec_args_add_fname(args, fname, segflg);
1178 	if (error != 0)
1179 		goto err_exit;
1180 
1181 	/*
1182 	 * extract arguments first
1183 	 */
1184 	for (;;) {
1185 		error = fueword(argv++, &arg);
1186 		if (error == -1) {
1187 			error = EFAULT;
1188 			goto err_exit;
1189 		}
1190 		if (arg == 0)
1191 			break;
1192 		error = exec_args_add_arg(args, (char *)(uintptr_t)arg,
1193 		    UIO_USERSPACE);
1194 		if (error != 0)
1195 			goto err_exit;
1196 	}
1197 
1198 	/*
1199 	 * extract environment strings
1200 	 */
1201 	if (envv) {
1202 		for (;;) {
1203 			error = fueword(envv++, &env);
1204 			if (error == -1) {
1205 				error = EFAULT;
1206 				goto err_exit;
1207 			}
1208 			if (env == 0)
1209 				break;
1210 			error = exec_args_add_env(args,
1211 			    (char *)(uintptr_t)env, UIO_USERSPACE);
1212 			if (error != 0)
1213 				goto err_exit;
1214 		}
1215 	}
1216 
1217 	return (0);
1218 
1219 err_exit:
1220 	exec_free_args(args);
1221 	return (error);
1222 }
1223 
1224 int
1225 exec_copyin_data_fds(struct thread *td, struct image_args *args,
1226     const void *data, size_t datalen, const int *fds, size_t fdslen)
1227 {
1228 	struct filedesc *ofdp;
1229 	const char *p;
1230 	int *kfds;
1231 	int error;
1232 
1233 	memset(args, '\0', sizeof(*args));
1234 	ofdp = td->td_proc->p_fd;
1235 	if (datalen >= ARG_MAX || fdslen > ofdp->fd_lastfile + 1)
1236 		return (E2BIG);
1237 	error = exec_alloc_args(args);
1238 	if (error != 0)
1239 		return (error);
1240 
1241 	args->begin_argv = args->buf;
1242 	args->stringspace = ARG_MAX;
1243 
1244 	if (datalen > 0) {
1245 		/*
1246 		 * Argument buffer has been provided. Copy it into the
1247 		 * kernel as a single string and add a terminating null
1248 		 * byte.
1249 		 */
1250 		error = copyin(data, args->begin_argv, datalen);
1251 		if (error != 0)
1252 			goto err_exit;
1253 		args->begin_argv[datalen] = '\0';
1254 		args->endp = args->begin_argv + datalen + 1;
1255 		args->stringspace -= datalen + 1;
1256 
1257 		/*
1258 		 * Traditional argument counting. Count the number of
1259 		 * null bytes.
1260 		 */
1261 		for (p = args->begin_argv; p < args->endp; ++p)
1262 			if (*p == '\0')
1263 				++args->argc;
1264 	} else {
1265 		/* No argument buffer provided. */
1266 		args->endp = args->begin_argv;
1267 	}
1268 
1269 	/* Create new file descriptor table. */
1270 	kfds = malloc(fdslen * sizeof(int), M_TEMP, M_WAITOK);
1271 	error = copyin(fds, kfds, fdslen * sizeof(int));
1272 	if (error != 0) {
1273 		free(kfds, M_TEMP);
1274 		goto err_exit;
1275 	}
1276 	error = fdcopy_remapped(ofdp, kfds, fdslen, &args->fdp);
1277 	free(kfds, M_TEMP);
1278 	if (error != 0)
1279 		goto err_exit;
1280 
1281 	return (0);
1282 err_exit:
1283 	exec_free_args(args);
1284 	return (error);
1285 }
1286 
1287 struct exec_args_kva {
1288 	vm_offset_t addr;
1289 	u_int gen;
1290 	SLIST_ENTRY(exec_args_kva) next;
1291 };
1292 
1293 DPCPU_DEFINE_STATIC(struct exec_args_kva *, exec_args_kva);
1294 
1295 static SLIST_HEAD(, exec_args_kva) exec_args_kva_freelist;
1296 static struct mtx exec_args_kva_mtx;
1297 static u_int exec_args_gen;
1298 
1299 static void
1300 exec_prealloc_args_kva(void *arg __unused)
1301 {
1302 	struct exec_args_kva *argkva;
1303 	u_int i;
1304 
1305 	SLIST_INIT(&exec_args_kva_freelist);
1306 	mtx_init(&exec_args_kva_mtx, "exec args kva", NULL, MTX_DEF);
1307 	for (i = 0; i < exec_map_entries; i++) {
1308 		argkva = malloc(sizeof(*argkva), M_PARGS, M_WAITOK);
1309 		argkva->addr = kmap_alloc_wait(exec_map, exec_map_entry_size);
1310 		argkva->gen = exec_args_gen;
1311 		SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1312 	}
1313 }
1314 SYSINIT(exec_args_kva, SI_SUB_EXEC, SI_ORDER_ANY, exec_prealloc_args_kva, NULL);
1315 
1316 static vm_offset_t
1317 exec_alloc_args_kva(void **cookie)
1318 {
1319 	struct exec_args_kva *argkva;
1320 
1321 	argkva = (void *)atomic_readandclear_ptr(
1322 	    (uintptr_t *)DPCPU_PTR(exec_args_kva));
1323 	if (argkva == NULL) {
1324 		mtx_lock(&exec_args_kva_mtx);
1325 		while ((argkva = SLIST_FIRST(&exec_args_kva_freelist)) == NULL)
1326 			(void)mtx_sleep(&exec_args_kva_freelist,
1327 			    &exec_args_kva_mtx, 0, "execkva", 0);
1328 		SLIST_REMOVE_HEAD(&exec_args_kva_freelist, next);
1329 		mtx_unlock(&exec_args_kva_mtx);
1330 	}
1331 	*(struct exec_args_kva **)cookie = argkva;
1332 	return (argkva->addr);
1333 }
1334 
1335 static void
1336 exec_release_args_kva(struct exec_args_kva *argkva, u_int gen)
1337 {
1338 	vm_offset_t base;
1339 
1340 	base = argkva->addr;
1341 	if (argkva->gen != gen) {
1342 		(void)vm_map_madvise(exec_map, base, base + exec_map_entry_size,
1343 		    MADV_FREE);
1344 		argkva->gen = gen;
1345 	}
1346 	if (!atomic_cmpset_ptr((uintptr_t *)DPCPU_PTR(exec_args_kva),
1347 	    (uintptr_t)NULL, (uintptr_t)argkva)) {
1348 		mtx_lock(&exec_args_kva_mtx);
1349 		SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1350 		wakeup_one(&exec_args_kva_freelist);
1351 		mtx_unlock(&exec_args_kva_mtx);
1352 	}
1353 }
1354 
1355 static void
1356 exec_free_args_kva(void *cookie)
1357 {
1358 
1359 	exec_release_args_kva(cookie, exec_args_gen);
1360 }
1361 
1362 static void
1363 exec_args_kva_lowmem(void *arg __unused)
1364 {
1365 	SLIST_HEAD(, exec_args_kva) head;
1366 	struct exec_args_kva *argkva;
1367 	u_int gen;
1368 	int i;
1369 
1370 	gen = atomic_fetchadd_int(&exec_args_gen, 1) + 1;
1371 
1372 	/*
1373 	 * Force an madvise of each KVA range. Any currently allocated ranges
1374 	 * will have MADV_FREE applied once they are freed.
1375 	 */
1376 	SLIST_INIT(&head);
1377 	mtx_lock(&exec_args_kva_mtx);
1378 	SLIST_SWAP(&head, &exec_args_kva_freelist, exec_args_kva);
1379 	mtx_unlock(&exec_args_kva_mtx);
1380 	while ((argkva = SLIST_FIRST(&head)) != NULL) {
1381 		SLIST_REMOVE_HEAD(&head, next);
1382 		exec_release_args_kva(argkva, gen);
1383 	}
1384 
1385 	CPU_FOREACH(i) {
1386 		argkva = (void *)atomic_readandclear_ptr(
1387 		    (uintptr_t *)DPCPU_ID_PTR(i, exec_args_kva));
1388 		if (argkva != NULL)
1389 			exec_release_args_kva(argkva, gen);
1390 	}
1391 }
1392 EVENTHANDLER_DEFINE(vm_lowmem, exec_args_kva_lowmem, NULL,
1393     EVENTHANDLER_PRI_ANY);
1394 
1395 /*
1396  * Allocate temporary demand-paged, zero-filled memory for the file name,
1397  * argument, and environment strings.
1398  */
1399 int
1400 exec_alloc_args(struct image_args *args)
1401 {
1402 
1403 	args->buf = (char *)exec_alloc_args_kva(&args->bufkva);
1404 	return (0);
1405 }
1406 
1407 void
1408 exec_free_args(struct image_args *args)
1409 {
1410 
1411 	if (args->buf != NULL) {
1412 		exec_free_args_kva(args->bufkva);
1413 		args->buf = NULL;
1414 	}
1415 	if (args->fname_buf != NULL) {
1416 		free(args->fname_buf, M_TEMP);
1417 		args->fname_buf = NULL;
1418 	}
1419 	if (args->fdp != NULL)
1420 		fdescfree_remapped(args->fdp);
1421 }
1422 
1423 /*
1424  * A set to functions to fill struct image args.
1425  *
1426  * NOTE: exec_args_add_fname() must be called (possibly with a NULL
1427  * fname) before the other functions.  All exec_args_add_arg() calls must
1428  * be made before any exec_args_add_env() calls.  exec_args_adjust_args()
1429  * may be called any time after exec_args_add_fname().
1430  *
1431  * exec_args_add_fname() - install path to be executed
1432  * exec_args_add_arg() - append an argument string
1433  * exec_args_add_env() - append an env string
1434  * exec_args_adjust_args() - adjust location of the argument list to
1435  *                           allow new arguments to be prepended
1436  */
1437 int
1438 exec_args_add_fname(struct image_args *args, const char *fname,
1439     enum uio_seg segflg)
1440 {
1441 	int error;
1442 	size_t length;
1443 
1444 	KASSERT(args->fname == NULL, ("fname already appended"));
1445 	KASSERT(args->endp == NULL, ("already appending to args"));
1446 
1447 	if (fname != NULL) {
1448 		args->fname = args->buf;
1449 		error = segflg == UIO_SYSSPACE ?
1450 		    copystr(fname, args->fname, PATH_MAX, &length) :
1451 		    copyinstr(fname, args->fname, PATH_MAX, &length);
1452 		if (error != 0)
1453 			return (error == ENAMETOOLONG ? E2BIG : error);
1454 	} else
1455 		length = 0;
1456 
1457 	/* Set up for _arg_*()/_env_*() */
1458 	args->endp = args->buf + length;
1459 	/* begin_argv must be set and kept updated */
1460 	args->begin_argv = args->endp;
1461 	KASSERT(exec_map_entry_size - length >= ARG_MAX,
1462 	    ("too little space remaining for arguments %zu < %zu",
1463 	    exec_map_entry_size - length, (size_t)ARG_MAX));
1464 	args->stringspace = ARG_MAX;
1465 
1466 	return (0);
1467 }
1468 
1469 static int
1470 exec_args_add_str(struct image_args *args, const char *str,
1471     enum uio_seg segflg, int *countp)
1472 {
1473 	int error;
1474 	size_t length;
1475 
1476 	KASSERT(args->endp != NULL, ("endp not initialized"));
1477 	KASSERT(args->begin_argv != NULL, ("begin_argp not initialized"));
1478 
1479 	error = (segflg == UIO_SYSSPACE) ?
1480 	    copystr(str, args->endp, args->stringspace, &length) :
1481 	    copyinstr(str, args->endp, args->stringspace, &length);
1482 	if (error != 0)
1483 		return (error == ENAMETOOLONG ? E2BIG : error);
1484 	args->stringspace -= length;
1485 	args->endp += length;
1486 	(*countp)++;
1487 
1488 	return (0);
1489 }
1490 
1491 int
1492 exec_args_add_arg(struct image_args *args, const char *argp,
1493     enum uio_seg segflg)
1494 {
1495 
1496 	KASSERT(args->envc == 0, ("appending args after env"));
1497 
1498 	return (exec_args_add_str(args, argp, segflg, &args->argc));
1499 }
1500 
1501 int
1502 exec_args_add_env(struct image_args *args, const char *envp,
1503     enum uio_seg segflg)
1504 {
1505 
1506 	if (args->envc == 0)
1507 		args->begin_envv = args->endp;
1508 
1509 	return (exec_args_add_str(args, envp, segflg, &args->envc));
1510 }
1511 
1512 int
1513 exec_args_adjust_args(struct image_args *args, size_t consume, ssize_t extend)
1514 {
1515 	ssize_t offset;
1516 
1517 	KASSERT(args->endp != NULL, ("endp not initialized"));
1518 	KASSERT(args->begin_argv != NULL, ("begin_argp not initialized"));
1519 
1520 	offset = extend - consume;
1521 	if (args->stringspace < offset)
1522 		return (E2BIG);
1523 	memmove(args->begin_argv + extend, args->begin_argv + consume,
1524 	    args->endp - args->begin_argv + consume);
1525 	if (args->envc > 0)
1526 		args->begin_envv += offset;
1527 	args->endp += offset;
1528 	args->stringspace -= offset;
1529 	return (0);
1530 }
1531 
1532 char *
1533 exec_args_get_begin_envv(struct image_args *args)
1534 {
1535 
1536 	KASSERT(args->endp != NULL, ("endp not initialized"));
1537 
1538 	if (args->envc > 0)
1539 		return (args->begin_envv);
1540 	return (args->endp);
1541 }
1542 
1543 /*
1544  * Copy strings out to the new process address space, constructing new arg
1545  * and env vector tables. Return a pointer to the base so that it can be used
1546  * as the initial stack pointer.
1547  */
1548 register_t *
1549 exec_copyout_strings(struct image_params *imgp)
1550 {
1551 	int argc, envc;
1552 	char **vectp;
1553 	char *stringp;
1554 	uintptr_t destp;
1555 	register_t *stack_base;
1556 	struct ps_strings *arginfo;
1557 	struct proc *p;
1558 	size_t execpath_len;
1559 	int szsigcode, szps;
1560 	char canary[sizeof(long) * 8];
1561 
1562 	szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1563 	/*
1564 	 * Calculate string base and vector table pointers.
1565 	 * Also deal with signal trampoline code for this exec type.
1566 	 */
1567 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
1568 		execpath_len = strlen(imgp->execpath) + 1;
1569 	else
1570 		execpath_len = 0;
1571 	p = imgp->proc;
1572 	szsigcode = 0;
1573 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1574 	if (p->p_sysent->sv_sigcode_base == 0) {
1575 		if (p->p_sysent->sv_szsigcode != NULL)
1576 			szsigcode = *(p->p_sysent->sv_szsigcode);
1577 	}
1578 	destp =	(uintptr_t)arginfo;
1579 
1580 	/*
1581 	 * install sigcode
1582 	 */
1583 	if (szsigcode != 0) {
1584 		destp -= szsigcode;
1585 		destp = rounddown2(destp, sizeof(void *));
1586 		copyout(p->p_sysent->sv_sigcode, (void *)destp, szsigcode);
1587 	}
1588 
1589 	/*
1590 	 * Copy the image path for the rtld.
1591 	 */
1592 	if (execpath_len != 0) {
1593 		destp -= execpath_len;
1594 		destp = rounddown2(destp, sizeof(void *));
1595 		imgp->execpathp = destp;
1596 		copyout(imgp->execpath, (void *)destp, execpath_len);
1597 	}
1598 
1599 	/*
1600 	 * Prepare the canary for SSP.
1601 	 */
1602 	arc4rand(canary, sizeof(canary), 0);
1603 	destp -= sizeof(canary);
1604 	imgp->canary = destp;
1605 	copyout(canary, (void *)destp, sizeof(canary));
1606 	imgp->canarylen = sizeof(canary);
1607 
1608 	/*
1609 	 * Prepare the pagesizes array.
1610 	 */
1611 	destp -= szps;
1612 	destp = rounddown2(destp, sizeof(void *));
1613 	imgp->pagesizes = destp;
1614 	copyout(pagesizes, (void *)destp, szps);
1615 	imgp->pagesizeslen = szps;
1616 
1617 	destp -= ARG_MAX - imgp->args->stringspace;
1618 	destp = rounddown2(destp, sizeof(void *));
1619 
1620 	vectp = (char **)destp;
1621 	if (imgp->sysent->sv_stackgap != NULL)
1622 		imgp->sysent->sv_stackgap(imgp, (u_long *)&vectp);
1623 
1624 	if (imgp->auxargs) {
1625 		/*
1626 		 * Allocate room on the stack for the ELF auxargs
1627 		 * array.  It has up to AT_COUNT entries.
1628 		 */
1629 		vectp -= howmany(AT_COUNT * sizeof(Elf_Auxinfo),
1630 		    sizeof(*vectp));
1631 	}
1632 
1633 	/*
1634 	 * Allocate room for the argv[] and env vectors including the
1635 	 * terminating NULL pointers.
1636 	 */
1637 	vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
1638 
1639 	/*
1640 	 * vectp also becomes our initial stack base
1641 	 */
1642 	stack_base = (register_t *)vectp;
1643 
1644 	stringp = imgp->args->begin_argv;
1645 	argc = imgp->args->argc;
1646 	envc = imgp->args->envc;
1647 
1648 	/*
1649 	 * Copy out strings - arguments and environment.
1650 	 */
1651 	copyout(stringp, (void *)destp, ARG_MAX - imgp->args->stringspace);
1652 
1653 	/*
1654 	 * Fill in "ps_strings" struct for ps, w, etc.
1655 	 */
1656 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1657 	suword32(&arginfo->ps_nargvstr, argc);
1658 
1659 	/*
1660 	 * Fill in argument portion of vector table.
1661 	 */
1662 	for (; argc > 0; --argc) {
1663 		suword(vectp++, (long)(intptr_t)destp);
1664 		while (*stringp++ != 0)
1665 			destp++;
1666 		destp++;
1667 	}
1668 
1669 	/* a null vector table pointer separates the argp's from the envp's */
1670 	suword(vectp++, 0);
1671 
1672 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1673 	suword32(&arginfo->ps_nenvstr, envc);
1674 
1675 	/*
1676 	 * Fill in environment portion of vector table.
1677 	 */
1678 	for (; envc > 0; --envc) {
1679 		suword(vectp++, (long)(intptr_t)destp);
1680 		while (*stringp++ != 0)
1681 			destp++;
1682 		destp++;
1683 	}
1684 
1685 	/* end of vector table is a null pointer */
1686 	suword(vectp, 0);
1687 
1688 	return (stack_base);
1689 }
1690 
1691 /*
1692  * Check permissions of file to execute.
1693  *	Called with imgp->vp locked.
1694  *	Return 0 for success or error code on failure.
1695  */
1696 int
1697 exec_check_permissions(struct image_params *imgp)
1698 {
1699 	struct vnode *vp = imgp->vp;
1700 	struct vattr *attr = imgp->attr;
1701 	struct thread *td;
1702 	int error;
1703 
1704 	td = curthread;
1705 
1706 	/* Get file attributes */
1707 	error = VOP_GETATTR(vp, attr, td->td_ucred);
1708 	if (error)
1709 		return (error);
1710 
1711 #ifdef MAC
1712 	error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1713 	if (error)
1714 		return (error);
1715 #endif
1716 
1717 	/*
1718 	 * 1) Check if file execution is disabled for the filesystem that
1719 	 *    this file resides on.
1720 	 * 2) Ensure that at least one execute bit is on. Otherwise, a
1721 	 *    privileged user will always succeed, and we don't want this
1722 	 *    to happen unless the file really is executable.
1723 	 * 3) Ensure that the file is a regular file.
1724 	 */
1725 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1726 	    (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1727 	    (attr->va_type != VREG))
1728 		return (EACCES);
1729 
1730 	/*
1731 	 * Zero length files can't be exec'd
1732 	 */
1733 	if (attr->va_size == 0)
1734 		return (ENOEXEC);
1735 
1736 	/*
1737 	 *  Check for execute permission to file based on current credentials.
1738 	 */
1739 	error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1740 	if (error)
1741 		return (error);
1742 
1743 	/*
1744 	 * Check number of open-for-writes on the file and deny execution
1745 	 * if there are any.
1746 	 *
1747 	 * Add a text reference now so no one can write to the
1748 	 * executable while we're activating it.
1749 	 *
1750 	 * Remember if this was set before and unset it in case this is not
1751 	 * actually an executable image.
1752 	 */
1753 	error = VOP_SET_TEXT(vp);
1754 	if (error != 0)
1755 		return (error);
1756 	imgp->textset = true;
1757 
1758 	/*
1759 	 * Call filesystem specific open routine (which does nothing in the
1760 	 * general case).
1761 	 */
1762 	error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1763 	if (error == 0)
1764 		imgp->opened = 1;
1765 	return (error);
1766 }
1767 
1768 /*
1769  * Exec handler registration
1770  */
1771 int
1772 exec_register(const struct execsw *execsw_arg)
1773 {
1774 	const struct execsw **es, **xs, **newexecsw;
1775 	u_int count = 2;	/* New slot and trailing NULL */
1776 
1777 	if (execsw)
1778 		for (es = execsw; *es; es++)
1779 			count++;
1780 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1781 	xs = newexecsw;
1782 	if (execsw)
1783 		for (es = execsw; *es; es++)
1784 			*xs++ = *es;
1785 	*xs++ = execsw_arg;
1786 	*xs = NULL;
1787 	if (execsw)
1788 		free(execsw, M_TEMP);
1789 	execsw = newexecsw;
1790 	return (0);
1791 }
1792 
1793 int
1794 exec_unregister(const struct execsw *execsw_arg)
1795 {
1796 	const struct execsw **es, **xs, **newexecsw;
1797 	int count = 1;
1798 
1799 	if (execsw == NULL)
1800 		panic("unregister with no handlers left?\n");
1801 
1802 	for (es = execsw; *es; es++) {
1803 		if (*es == execsw_arg)
1804 			break;
1805 	}
1806 	if (*es == NULL)
1807 		return (ENOENT);
1808 	for (es = execsw; *es; es++)
1809 		if (*es != execsw_arg)
1810 			count++;
1811 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1812 	xs = newexecsw;
1813 	for (es = execsw; *es; es++)
1814 		if (*es != execsw_arg)
1815 			*xs++ = *es;
1816 	*xs = NULL;
1817 	if (execsw)
1818 		free(execsw, M_TEMP);
1819 	execsw = newexecsw;
1820 	return (0);
1821 }
1822