xref: /freebsd/sys/kern/kern_exec.c (revision 72c59f1728c12fbdcd316c172708d8f3a60299be)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 1993, David Greenman
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include "opt_capsicum.h"
33 #include "opt_hwpmc_hooks.h"
34 #include "opt_ktrace.h"
35 #include "opt_vm.h"
36 
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/acct.h>
40 #include <sys/capsicum.h>
41 #include <sys/eventhandler.h>
42 #include <sys/exec.h>
43 #include <sys/fcntl.h>
44 #include <sys/filedesc.h>
45 #include <sys/imgact.h>
46 #include <sys/imgact_elf.h>
47 #include <sys/kernel.h>
48 #include <sys/lock.h>
49 #include <sys/malloc.h>
50 #include <sys/mman.h>
51 #include <sys/mount.h>
52 #include <sys/mutex.h>
53 #include <sys/namei.h>
54 #include <sys/priv.h>
55 #include <sys/proc.h>
56 #include <sys/ptrace.h>
57 #include <sys/resourcevar.h>
58 #include <sys/rwlock.h>
59 #include <sys/sched.h>
60 #include <sys/sdt.h>
61 #include <sys/sf_buf.h>
62 #include <sys/shm.h>
63 #include <sys/signalvar.h>
64 #include <sys/smp.h>
65 #include <sys/stat.h>
66 #include <sys/syscallsubr.h>
67 #include <sys/sysctl.h>
68 #include <sys/sysent.h>
69 #include <sys/sysproto.h>
70 #include <sys/timers.h>
71 #include <sys/umtx.h>
72 #include <sys/vnode.h>
73 #include <sys/wait.h>
74 #ifdef KTRACE
75 #include <sys/ktrace.h>
76 #endif
77 
78 #include <vm/vm.h>
79 #include <vm/vm_param.h>
80 #include <vm/pmap.h>
81 #include <vm/vm_page.h>
82 #include <vm/vm_map.h>
83 #include <vm/vm_kern.h>
84 #include <vm/vm_extern.h>
85 #include <vm/vm_object.h>
86 #include <vm/vm_pager.h>
87 
88 #ifdef	HWPMC_HOOKS
89 #include <sys/pmckern.h>
90 #endif
91 
92 #include <machine/reg.h>
93 
94 #include <security/audit/audit.h>
95 #include <security/mac/mac_framework.h>
96 
97 #ifdef KDTRACE_HOOKS
98 #include <sys/dtrace_bsd.h>
99 dtrace_execexit_func_t	dtrace_fasttrap_exec;
100 #endif
101 
102 SDT_PROVIDER_DECLARE(proc);
103 SDT_PROBE_DEFINE1(proc, , , exec, "char *");
104 SDT_PROBE_DEFINE1(proc, , , exec__failure, "int");
105 SDT_PROBE_DEFINE1(proc, , , exec__success, "char *");
106 
107 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
108 
109 int coredump_pack_fileinfo = 1;
110 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_fileinfo, CTLFLAG_RWTUN,
111     &coredump_pack_fileinfo, 0,
112     "Enable file path packing in 'procstat -f' coredump notes");
113 
114 int coredump_pack_vmmapinfo = 1;
115 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_vmmapinfo, CTLFLAG_RWTUN,
116     &coredump_pack_vmmapinfo, 0,
117     "Enable file path packing in 'procstat -v' coredump notes");
118 
119 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
120 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
121 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
122 static int do_execve(struct thread *td, struct image_args *args,
123     struct mac *mac_p, struct vmspace *oldvmspace);
124 
125 /* XXX This should be vm_size_t. */
126 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD|
127     CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_ps_strings, "LU",
128     "Location of process' ps_strings structure");
129 
130 /* XXX This should be vm_size_t. */
131 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD|
132     CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_usrstack, "LU",
133     "Top of process stack");
134 
135 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD|CTLFLAG_MPSAFE,
136     NULL, 0, sysctl_kern_stackprot, "I",
137     "Stack memory permissions");
138 
139 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
140 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
141     &ps_arg_cache_limit, 0,
142     "Process' command line characters cache limit");
143 
144 static int disallow_high_osrel;
145 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW,
146     &disallow_high_osrel, 0,
147     "Disallow execution of binaries built for higher version of the world");
148 
149 static int map_at_zero = 0;
150 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RWTUN, &map_at_zero, 0,
151     "Permit processes to map an object at virtual address 0.");
152 
153 static int
154 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
155 {
156 	struct proc *p;
157 	int error;
158 
159 	p = curproc;
160 #ifdef SCTL_MASK32
161 	if (req->flags & SCTL_MASK32) {
162 		unsigned int val;
163 		val = (unsigned int)p->p_sysent->sv_psstrings;
164 		error = SYSCTL_OUT(req, &val, sizeof(val));
165 	} else
166 #endif
167 		error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
168 		   sizeof(p->p_sysent->sv_psstrings));
169 	return error;
170 }
171 
172 static int
173 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
174 {
175 	struct proc *p;
176 	int error;
177 
178 	p = curproc;
179 #ifdef SCTL_MASK32
180 	if (req->flags & SCTL_MASK32) {
181 		unsigned int val;
182 		val = (unsigned int)p->p_sysent->sv_usrstack;
183 		error = SYSCTL_OUT(req, &val, sizeof(val));
184 	} else
185 #endif
186 		error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
187 		    sizeof(p->p_sysent->sv_usrstack));
188 	return error;
189 }
190 
191 static int
192 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
193 {
194 	struct proc *p;
195 
196 	p = curproc;
197 	return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
198 	    sizeof(p->p_sysent->sv_stackprot)));
199 }
200 
201 /*
202  * Each of the items is a pointer to a `const struct execsw', hence the
203  * double pointer here.
204  */
205 static const struct execsw **execsw;
206 
207 #ifndef _SYS_SYSPROTO_H_
208 struct execve_args {
209 	char    *fname;
210 	char    **argv;
211 	char    **envv;
212 };
213 #endif
214 
215 int
216 sys_execve(struct thread *td, struct execve_args *uap)
217 {
218 	struct image_args args;
219 	struct vmspace *oldvmspace;
220 	int error;
221 
222 	error = pre_execve(td, &oldvmspace);
223 	if (error != 0)
224 		return (error);
225 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
226 	    uap->argv, uap->envv);
227 	if (error == 0)
228 		error = kern_execve(td, &args, NULL, oldvmspace);
229 	post_execve(td, error, oldvmspace);
230 	AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
231 	return (error);
232 }
233 
234 #ifndef _SYS_SYSPROTO_H_
235 struct fexecve_args {
236 	int	fd;
237 	char	**argv;
238 	char	**envv;
239 };
240 #endif
241 int
242 sys_fexecve(struct thread *td, struct fexecve_args *uap)
243 {
244 	struct image_args args;
245 	struct vmspace *oldvmspace;
246 	int error;
247 
248 	error = pre_execve(td, &oldvmspace);
249 	if (error != 0)
250 		return (error);
251 	error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
252 	    uap->argv, uap->envv);
253 	if (error == 0) {
254 		args.fd = uap->fd;
255 		error = kern_execve(td, &args, NULL, oldvmspace);
256 	}
257 	post_execve(td, error, oldvmspace);
258 	AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
259 	return (error);
260 }
261 
262 #ifndef _SYS_SYSPROTO_H_
263 struct __mac_execve_args {
264 	char	*fname;
265 	char	**argv;
266 	char	**envv;
267 	struct mac	*mac_p;
268 };
269 #endif
270 
271 int
272 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap)
273 {
274 #ifdef MAC
275 	struct image_args args;
276 	struct vmspace *oldvmspace;
277 	int error;
278 
279 	error = pre_execve(td, &oldvmspace);
280 	if (error != 0)
281 		return (error);
282 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
283 	    uap->argv, uap->envv);
284 	if (error == 0)
285 		error = kern_execve(td, &args, uap->mac_p, oldvmspace);
286 	post_execve(td, error, oldvmspace);
287 	AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
288 	return (error);
289 #else
290 	return (ENOSYS);
291 #endif
292 }
293 
294 int
295 pre_execve(struct thread *td, struct vmspace **oldvmspace)
296 {
297 	struct proc *p;
298 	int error;
299 
300 	KASSERT(td == curthread, ("non-current thread %p", td));
301 	error = 0;
302 	p = td->td_proc;
303 	if ((p->p_flag & P_HADTHREADS) != 0) {
304 		PROC_LOCK(p);
305 		if (thread_single(p, SINGLE_BOUNDARY) != 0)
306 			error = ERESTART;
307 		PROC_UNLOCK(p);
308 	}
309 	KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0,
310 	    ("nested execve"));
311 	*oldvmspace = p->p_vmspace;
312 	return (error);
313 }
314 
315 void
316 post_execve(struct thread *td, int error, struct vmspace *oldvmspace)
317 {
318 	struct proc *p;
319 
320 	KASSERT(td == curthread, ("non-current thread %p", td));
321 	p = td->td_proc;
322 	if ((p->p_flag & P_HADTHREADS) != 0) {
323 		PROC_LOCK(p);
324 		/*
325 		 * If success, we upgrade to SINGLE_EXIT state to
326 		 * force other threads to suicide.
327 		 */
328 		if (error == EJUSTRETURN)
329 			thread_single(p, SINGLE_EXIT);
330 		else
331 			thread_single_end(p, SINGLE_BOUNDARY);
332 		PROC_UNLOCK(p);
333 	}
334 	exec_cleanup(td, oldvmspace);
335 }
336 
337 /*
338  * kern_execve() has the astonishing property of not always returning to
339  * the caller.  If sufficiently bad things happen during the call to
340  * do_execve(), it can end up calling exit1(); as a result, callers must
341  * avoid doing anything which they might need to undo (e.g., allocating
342  * memory).
343  */
344 int
345 kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p,
346     struct vmspace *oldvmspace)
347 {
348 
349 	AUDIT_ARG_ARGV(args->begin_argv, args->argc,
350 	    exec_args_get_begin_envv(args) - args->begin_argv);
351 	AUDIT_ARG_ENVV(exec_args_get_begin_envv(args), args->envc,
352 	    args->endp - exec_args_get_begin_envv(args));
353 	return (do_execve(td, args, mac_p, oldvmspace));
354 }
355 
356 /*
357  * In-kernel implementation of execve().  All arguments are assumed to be
358  * userspace pointers from the passed thread.
359  */
360 static int
361 do_execve(struct thread *td, struct image_args *args, struct mac *mac_p,
362     struct vmspace *oldvmspace)
363 {
364 	struct proc *p = td->td_proc;
365 	struct nameidata nd;
366 	struct ucred *oldcred;
367 	struct uidinfo *euip = NULL;
368 	uintptr_t stack_base;
369 	struct image_params image_params, *imgp;
370 	struct vattr attr;
371 	int (*img_first)(struct image_params *);
372 	struct pargs *oldargs = NULL, *newargs = NULL;
373 	struct sigacts *oldsigacts = NULL, *newsigacts = NULL;
374 #ifdef KTRACE
375 	struct vnode *tracevp = NULL;
376 	struct ucred *tracecred = NULL;
377 #endif
378 	struct vnode *oldtextvp = NULL, *newtextvp;
379 	int credential_changing;
380 #ifdef MAC
381 	struct label *interpvplabel = NULL;
382 	int will_transition;
383 #endif
384 #ifdef HWPMC_HOOKS
385 	struct pmckern_procexec pe;
386 #endif
387 	int error, i, orig_osrel;
388 	uint32_t orig_fctl0;
389 	static const char fexecv_proc_title[] = "(fexecv)";
390 
391 	imgp = &image_params;
392 
393 	/*
394 	 * Lock the process and set the P_INEXEC flag to indicate that
395 	 * it should be left alone until we're done here.  This is
396 	 * necessary to avoid race conditions - e.g. in ptrace() -
397 	 * that might allow a local user to illicitly obtain elevated
398 	 * privileges.
399 	 */
400 	PROC_LOCK(p);
401 	KASSERT((p->p_flag & P_INEXEC) == 0,
402 	    ("%s(): process already has P_INEXEC flag", __func__));
403 	p->p_flag |= P_INEXEC;
404 	PROC_UNLOCK(p);
405 
406 	/*
407 	 * Initialize part of the common data
408 	 */
409 	bzero(imgp, sizeof(*imgp));
410 	imgp->proc = p;
411 	imgp->attr = &attr;
412 	imgp->args = args;
413 	oldcred = p->p_ucred;
414 	orig_osrel = p->p_osrel;
415 	orig_fctl0 = p->p_fctl0;
416 
417 #ifdef MAC
418 	error = mac_execve_enter(imgp, mac_p);
419 	if (error)
420 		goto exec_fail;
421 #endif
422 
423 	/*
424 	 * Translate the file name. namei() returns a vnode pointer
425 	 *	in ni_vp among other things.
426 	 *
427 	 * XXXAUDIT: It would be desirable to also audit the name of the
428 	 * interpreter if this is an interpreted binary.
429 	 */
430 	if (args->fname != NULL) {
431 		NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW |
432 		    SAVENAME | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
433 	}
434 
435 	SDT_PROBE1(proc, , , exec, args->fname);
436 
437 interpret:
438 	if (args->fname != NULL) {
439 #ifdef CAPABILITY_MODE
440 		/*
441 		 * While capability mode can't reach this point via direct
442 		 * path arguments to execve(), we also don't allow
443 		 * interpreters to be used in capability mode (for now).
444 		 * Catch indirect lookups and return a permissions error.
445 		 */
446 		if (IN_CAPABILITY_MODE(td)) {
447 			error = ECAPMODE;
448 			goto exec_fail;
449 		}
450 #endif
451 		error = namei(&nd);
452 		if (error)
453 			goto exec_fail;
454 
455 		newtextvp = nd.ni_vp;
456 		imgp->vp = newtextvp;
457 	} else {
458 		AUDIT_ARG_FD(args->fd);
459 		/*
460 		 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
461 		 */
462 		error = fgetvp_exec(td, args->fd, &cap_fexecve_rights, &newtextvp);
463 		if (error)
464 			goto exec_fail;
465 		vn_lock(newtextvp, LK_SHARED | LK_RETRY);
466 		AUDIT_ARG_VNODE1(newtextvp);
467 		imgp->vp = newtextvp;
468 	}
469 
470 	/*
471 	 * Check file permissions.  Also 'opens' file and sets its vnode to
472 	 * text mode.
473 	 */
474 	error = exec_check_permissions(imgp);
475 	if (error)
476 		goto exec_fail_dealloc;
477 
478 	imgp->object = imgp->vp->v_object;
479 	if (imgp->object != NULL)
480 		vm_object_reference(imgp->object);
481 
482 	error = exec_map_first_page(imgp);
483 	if (error)
484 		goto exec_fail_dealloc;
485 
486 	imgp->proc->p_osrel = 0;
487 	imgp->proc->p_fctl0 = 0;
488 
489 	/*
490 	 * Implement image setuid/setgid.
491 	 *
492 	 * Determine new credentials before attempting image activators
493 	 * so that it can be used by process_exec handlers to determine
494 	 * credential/setid changes.
495 	 *
496 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
497 	 * the process is being traced.
498 	 *
499 	 * We disable setuid/setgid/etc in capability mode on the basis
500 	 * that most setugid applications are not written with that
501 	 * environment in mind, and will therefore almost certainly operate
502 	 * incorrectly. In principle there's no reason that setugid
503 	 * applications might not be useful in capability mode, so we may want
504 	 * to reconsider this conservative design choice in the future.
505 	 *
506 	 * XXXMAC: For the time being, use NOSUID to also prohibit
507 	 * transitions on the file system.
508 	 */
509 	credential_changing = 0;
510 	credential_changing |= (attr.va_mode & S_ISUID) &&
511 	    oldcred->cr_uid != attr.va_uid;
512 	credential_changing |= (attr.va_mode & S_ISGID) &&
513 	    oldcred->cr_gid != attr.va_gid;
514 #ifdef MAC
515 	will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
516 	    interpvplabel, imgp);
517 	credential_changing |= will_transition;
518 #endif
519 
520 	/* Don't inherit PROC_PDEATHSIG_CTL value if setuid/setgid. */
521 	if (credential_changing)
522 		imgp->proc->p_pdeathsig = 0;
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);
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_SHARED | 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);
565 			imgp->newcred = crdup(oldcred);
566 			vn_lock(imgp->vp, LK_SHARED | 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);
580 		if (vn_fullpath(imgp->vp, &imgp->execpath, &imgp->freepath) != 0)
581 			imgp->execpath = args->fname;
582 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
583 	}
584 
585 	/*
586 	 *	If the current process has a special image activator it
587 	 *	wants to try first, call it.   For example, emulating shell
588 	 *	scripts differently.
589 	 */
590 	error = -1;
591 	if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
592 		error = img_first(imgp);
593 
594 	/*
595 	 *	Loop through the list of image activators, calling each one.
596 	 *	An activator returns -1 if there is no match, 0 on success,
597 	 *	and an error otherwise.
598 	 */
599 	for (i = 0; error == -1 && execsw[i]; ++i) {
600 		if (execsw[i]->ex_imgact == NULL ||
601 		    execsw[i]->ex_imgact == img_first) {
602 			continue;
603 		}
604 		error = (*execsw[i]->ex_imgact)(imgp);
605 	}
606 
607 	if (error) {
608 		if (error == -1)
609 			error = ENOEXEC;
610 		goto exec_fail_dealloc;
611 	}
612 
613 	/*
614 	 * Special interpreter operation, cleanup and loop up to try to
615 	 * activate the interpreter.
616 	 */
617 	if (imgp->interpreted) {
618 		exec_unmap_first_page(imgp);
619 		/*
620 		 * The text reference needs to be removed for scripts.
621 		 * There is a short period before we determine that
622 		 * something is a script where text reference is active.
623 		 * The vnode lock is held over this entire period
624 		 * so nothing should illegitimately be blocked.
625 		 */
626 		MPASS(imgp->textset);
627 		VOP_UNSET_TEXT_CHECKED(newtextvp);
628 		imgp->textset = false;
629 		/* free name buffer and old vnode */
630 		if (args->fname != NULL)
631 			NDFREE(&nd, NDF_ONLY_PNBUF);
632 #ifdef MAC
633 		mac_execve_interpreter_enter(newtextvp, &interpvplabel);
634 #endif
635 		if (imgp->opened) {
636 			VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td);
637 			imgp->opened = 0;
638 		}
639 		vput(newtextvp);
640 		vm_object_deallocate(imgp->object);
641 		imgp->object = NULL;
642 		imgp->credential_setid = false;
643 		if (imgp->newcred != NULL) {
644 			crfree(imgp->newcred);
645 			imgp->newcred = NULL;
646 		}
647 		imgp->execpath = NULL;
648 		free(imgp->freepath, M_TEMP);
649 		imgp->freepath = NULL;
650 		/* set new name to that of the interpreter */
651 		NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW |
652 		    SAVENAME, UIO_SYSSPACE, imgp->interpreter_name, td);
653 		args->fname = imgp->interpreter_name;
654 		goto interpret;
655 	}
656 
657 	/*
658 	 * NB: We unlock the vnode here because it is believed that none
659 	 * of the sv_copyout_strings/sv_fixup operations require the vnode.
660 	 */
661 	VOP_UNLOCK(imgp->vp);
662 
663 	if (disallow_high_osrel &&
664 	    P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
665 		error = ENOEXEC;
666 		uprintf("Osrel %d for image %s too high\n", p->p_osrel,
667 		    imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
668 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
669 		goto exec_fail_dealloc;
670 	}
671 
672 	/* ABI enforces the use of Capsicum. Switch into capabilities mode. */
673 	if (SV_PROC_FLAG(p, SV_CAPSICUM))
674 		sys_cap_enter(td, NULL);
675 
676 	/*
677 	 * Copy out strings (args and env) and initialize stack base.
678 	 */
679 	error = (*p->p_sysent->sv_copyout_strings)(imgp, &stack_base);
680 	if (error != 0) {
681 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
682 		goto exec_fail_dealloc;
683 	}
684 
685 	/*
686 	 * Stack setup.
687 	 */
688 	error = (*p->p_sysent->sv_fixup)(&stack_base, imgp);
689 	if (error != 0) {
690 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
691 		goto exec_fail_dealloc;
692 	}
693 
694 	if (args->fdp != NULL) {
695 		/* Install a brand new file descriptor table. */
696 		fdinstall_remapped(td, args->fdp);
697 		args->fdp = NULL;
698 	} else {
699 		/*
700 		 * Keep on using the existing file descriptor table. For
701 		 * security and other reasons, the file descriptor table
702 		 * cannot be shared after an exec.
703 		 */
704 		fdunshare(td);
705 		pdunshare(td);
706 		/* close files on exec */
707 		fdcloseexec(td);
708 	}
709 
710 	/*
711 	 * Malloc things before we need locks.
712 	 */
713 	i = exec_args_get_begin_envv(imgp->args) - 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_flag2 & P2_STKGAP_DISABLE_EXEC) == 0)
763 		p->p_flag2 &= ~P2_STKGAP_DISABLE;
764 	if (p->p_flag & P_PPWAIT) {
765 		p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
766 		cv_broadcast(&p->p_pwait);
767 		/* STOPs are no longer ignored, arrange for AST */
768 		signotify(td);
769 	}
770 
771 	/*
772 	 * Implement image setuid/setgid installation.
773 	 */
774 	if (imgp->credential_setid) {
775 		/*
776 		 * Turn off syscall tracing for set-id programs, except for
777 		 * root.  Record any set-id flags first to make sure that
778 		 * we do not regain any tracing during a possible block.
779 		 */
780 		setsugid(p);
781 
782 #ifdef KTRACE
783 		if (p->p_tracecred != NULL &&
784 		    priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED))
785 			ktrprocexec(p, &tracecred, &tracevp);
786 #endif
787 		/*
788 		 * Close any file descriptors 0..2 that reference procfs,
789 		 * then make sure file descriptors 0..2 are in use.
790 		 *
791 		 * Both fdsetugidsafety() and fdcheckstd() may call functions
792 		 * taking sleepable locks, so temporarily drop our locks.
793 		 */
794 		PROC_UNLOCK(p);
795 		VOP_UNLOCK(imgp->vp);
796 		fdsetugidsafety(td);
797 		error = fdcheckstd(td);
798 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
799 		if (error != 0)
800 			goto exec_fail_dealloc;
801 		PROC_LOCK(p);
802 #ifdef MAC
803 		if (will_transition) {
804 			mac_vnode_execve_transition(oldcred, imgp->newcred,
805 			    imgp->vp, interpvplabel, imgp);
806 		}
807 #endif
808 	} else {
809 		if (oldcred->cr_uid == oldcred->cr_ruid &&
810 		    oldcred->cr_gid == oldcred->cr_rgid)
811 			p->p_flag &= ~P_SUGID;
812 	}
813 	/*
814 	 * Set the new credentials.
815 	 */
816 	if (imgp->newcred != NULL) {
817 		proc_set_cred(p, imgp->newcred);
818 		crfree(oldcred);
819 		oldcred = NULL;
820 	}
821 
822 	/*
823 	 * Store the vp for use in procfs.  This vnode was referenced by namei
824 	 * or fgetvp_exec.
825 	 */
826 	oldtextvp = p->p_textvp;
827 	p->p_textvp = newtextvp;
828 
829 #ifdef KDTRACE_HOOKS
830 	/*
831 	 * Tell the DTrace fasttrap provider about the exec if it
832 	 * has declared an interest.
833 	 */
834 	if (dtrace_fasttrap_exec)
835 		dtrace_fasttrap_exec(p);
836 #endif
837 
838 	/*
839 	 * Notify others that we exec'd, and clear the P_INEXEC flag
840 	 * as we're now a bona fide freshly-execed process.
841 	 */
842 	KNOTE_LOCKED(p->p_klist, NOTE_EXEC);
843 	p->p_flag &= ~P_INEXEC;
844 
845 	/* clear "fork but no exec" flag, as we _are_ execing */
846 	p->p_acflag &= ~AFORK;
847 
848 	/*
849 	 * Free any previous argument cache and replace it with
850 	 * the new argument cache, if any.
851 	 */
852 	oldargs = p->p_args;
853 	p->p_args = newargs;
854 	newargs = NULL;
855 
856 	PROC_UNLOCK(p);
857 
858 #ifdef	HWPMC_HOOKS
859 	/*
860 	 * Check if system-wide sampling is in effect or if the
861 	 * current process is using PMCs.  If so, do exec() time
862 	 * processing.  This processing needs to happen AFTER the
863 	 * P_INEXEC flag is cleared.
864 	 */
865 	if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
866 		VOP_UNLOCK(imgp->vp);
867 		pe.pm_credentialschanged = credential_changing;
868 		pe.pm_entryaddr = imgp->entry_addr;
869 
870 		PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
871 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
872 	}
873 #endif
874 
875 	/* Set values passed into the program in registers. */
876 	(*p->p_sysent->sv_setregs)(td, imgp, stack_base);
877 
878 	VOP_MMAPPED(imgp->vp);
879 
880 	SDT_PROBE1(proc, , , exec__success, args->fname);
881 
882 exec_fail_dealloc:
883 	if (error != 0) {
884 		p->p_osrel = orig_osrel;
885 		p->p_fctl0 = orig_fctl0;
886 	}
887 
888 	if (imgp->firstpage != NULL)
889 		exec_unmap_first_page(imgp);
890 
891 	if (imgp->vp != NULL) {
892 		if (args->fname)
893 			NDFREE(&nd, NDF_ONLY_PNBUF);
894 		if (imgp->opened)
895 			VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
896 		if (imgp->textset)
897 			VOP_UNSET_TEXT_CHECKED(imgp->vp);
898 		if (error != 0)
899 			vput(imgp->vp);
900 		else
901 			VOP_UNLOCK(imgp->vp);
902 	}
903 
904 	if (imgp->object != NULL)
905 		vm_object_deallocate(imgp->object);
906 
907 	free(imgp->freepath, M_TEMP);
908 
909 	if (error == 0) {
910 		if (p->p_ptevents & PTRACE_EXEC) {
911 			PROC_LOCK(p);
912 			if (p->p_ptevents & PTRACE_EXEC)
913 				td->td_dbgflags |= TDB_EXEC;
914 			PROC_UNLOCK(p);
915 		}
916 	} else {
917 exec_fail:
918 		/* we're done here, clear P_INEXEC */
919 		PROC_LOCK(p);
920 		p->p_flag &= ~P_INEXEC;
921 		PROC_UNLOCK(p);
922 
923 		SDT_PROBE1(proc, , , exec__failure, error);
924 	}
925 
926 	if (imgp->newcred != NULL && oldcred != NULL)
927 		crfree(imgp->newcred);
928 
929 #ifdef MAC
930 	mac_execve_exit(imgp);
931 	mac_execve_interpreter_exit(interpvplabel);
932 #endif
933 	exec_free_args(args);
934 
935 	/*
936 	 * Handle deferred decrement of ref counts.
937 	 */
938 	if (oldtextvp != NULL)
939 		vrele(oldtextvp);
940 #ifdef KTRACE
941 	if (tracevp != NULL)
942 		vrele(tracevp);
943 	if (tracecred != NULL)
944 		crfree(tracecred);
945 #endif
946 	pargs_drop(oldargs);
947 	pargs_drop(newargs);
948 	if (oldsigacts != NULL)
949 		sigacts_free(oldsigacts);
950 	if (euip != NULL)
951 		uifree(euip);
952 
953 	if (error && imgp->vmspace_destroyed) {
954 		/* sorry, no more process anymore. exit gracefully */
955 		exec_cleanup(td, oldvmspace);
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 void
975 exec_cleanup(struct thread *td, struct vmspace *oldvmspace)
976 {
977 	if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
978 		KASSERT(td->td_proc->p_vmspace != oldvmspace,
979 		    ("oldvmspace still used"));
980 		vmspace_free(oldvmspace);
981 		td->td_pflags &= ~TDP_EXECVMSPC;
982 	}
983 }
984 
985 int
986 exec_map_first_page(struct image_params *imgp)
987 {
988 	vm_object_t object;
989 	vm_page_t m;
990 	int error;
991 
992 	if (imgp->firstpage != NULL)
993 		exec_unmap_first_page(imgp);
994 
995 	object = imgp->vp->v_object;
996 	if (object == NULL)
997 		return (EACCES);
998 #if VM_NRESERVLEVEL > 0
999 	if ((object->flags & OBJ_COLORED) == 0) {
1000 		VM_OBJECT_WLOCK(object);
1001 		vm_object_color(object, 0);
1002 		VM_OBJECT_WUNLOCK(object);
1003 	}
1004 #endif
1005 	error = vm_page_grab_valid_unlocked(&m, object, 0,
1006 	    VM_ALLOC_COUNT(VM_INITIAL_PAGEIN) |
1007             VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED);
1008 
1009 	if (error != VM_PAGER_OK)
1010 		return (EIO);
1011 	imgp->firstpage = sf_buf_alloc(m, 0);
1012 	imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
1013 
1014 	return (0);
1015 }
1016 
1017 void
1018 exec_unmap_first_page(struct image_params *imgp)
1019 {
1020 	vm_page_t m;
1021 
1022 	if (imgp->firstpage != NULL) {
1023 		m = sf_buf_page(imgp->firstpage);
1024 		sf_buf_free(imgp->firstpage);
1025 		imgp->firstpage = NULL;
1026 		vm_page_unwire(m, PQ_ACTIVE);
1027 	}
1028 }
1029 
1030 /*
1031  * Destroy old address space, and allocate a new stack.
1032  *	The new stack is only sgrowsiz large because it is grown
1033  *	automatically on a page fault.
1034  */
1035 int
1036 exec_new_vmspace(struct image_params *imgp, struct sysentvec *sv)
1037 {
1038 	int error;
1039 	struct proc *p = imgp->proc;
1040 	struct vmspace *vmspace = p->p_vmspace;
1041 	struct thread *td = curthread;
1042 	vm_object_t obj;
1043 	struct rlimit rlim_stack;
1044 	vm_offset_t sv_minuser, stack_addr;
1045 	vm_map_t map;
1046 	u_long ssiz;
1047 
1048 	imgp->vmspace_destroyed = 1;
1049 	imgp->sysent = sv;
1050 
1051 	sigfastblock_clear(td);
1052 	umtx_exec(p);
1053 	itimers_exec(p);
1054 	if (sv->sv_onexec != NULL)
1055 		sv->sv_onexec(p, imgp);
1056 
1057 	EVENTHANDLER_DIRECT_INVOKE(process_exec, p, imgp);
1058 
1059 	/*
1060 	 * Blow away entire process VM, if address space not shared,
1061 	 * otherwise, create a new VM space so that other threads are
1062 	 * not disrupted
1063 	 */
1064 	map = &vmspace->vm_map;
1065 	if (map_at_zero)
1066 		sv_minuser = sv->sv_minuser;
1067 	else
1068 		sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1069 	if (refcount_load(&vmspace->vm_refcnt) == 1 &&
1070 	    vm_map_min(map) == sv_minuser &&
1071 	    vm_map_max(map) == sv->sv_maxuser &&
1072 	    cpu_exec_vmspace_reuse(p, map)) {
1073 		shmexit(vmspace);
1074 		pmap_remove_pages(vmspace_pmap(vmspace));
1075 		vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1076 		/*
1077 		 * An exec terminates mlockall(MCL_FUTURE), ASLR state
1078 		 * must be re-evaluated.
1079 		 */
1080 		vm_map_lock(map);
1081 		vm_map_modflags(map, 0, MAP_WIREFUTURE | MAP_ASLR |
1082 		    MAP_ASLR_IGNSTART);
1083 		vm_map_unlock(map);
1084 	} else {
1085 		error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1086 		if (error)
1087 			return (error);
1088 		vmspace = p->p_vmspace;
1089 		map = &vmspace->vm_map;
1090 	}
1091 	map->flags |= imgp->map_flags;
1092 
1093 	/* Map a shared page */
1094 	obj = sv->sv_shared_page_obj;
1095 	if (obj != NULL) {
1096 		vm_object_reference(obj);
1097 		error = vm_map_fixed(map, obj, 0,
1098 		    sv->sv_shared_page_base, sv->sv_shared_page_len,
1099 		    VM_PROT_READ | VM_PROT_EXECUTE,
1100 		    VM_PROT_READ | VM_PROT_EXECUTE,
1101 		    MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1102 		if (error != KERN_SUCCESS) {
1103 			vm_object_deallocate(obj);
1104 			return (vm_mmap_to_errno(error));
1105 		}
1106 	}
1107 
1108 	/* Allocate a new stack */
1109 	if (imgp->stack_sz != 0) {
1110 		ssiz = trunc_page(imgp->stack_sz);
1111 		PROC_LOCK(p);
1112 		lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack);
1113 		PROC_UNLOCK(p);
1114 		if (ssiz > rlim_stack.rlim_max)
1115 			ssiz = rlim_stack.rlim_max;
1116 		if (ssiz > rlim_stack.rlim_cur) {
1117 			rlim_stack.rlim_cur = ssiz;
1118 			kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack);
1119 		}
1120 	} else if (sv->sv_maxssiz != NULL) {
1121 		ssiz = *sv->sv_maxssiz;
1122 	} else {
1123 		ssiz = maxssiz;
1124 	}
1125 	imgp->eff_stack_sz = lim_cur(curthread, RLIMIT_STACK);
1126 	if (ssiz < imgp->eff_stack_sz)
1127 		imgp->eff_stack_sz = ssiz;
1128 	stack_addr = sv->sv_usrstack - ssiz;
1129 	error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1130 	    obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1131 	    sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1132 	if (error != KERN_SUCCESS)
1133 		return (vm_mmap_to_errno(error));
1134 
1135 	/*
1136 	 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they
1137 	 * are still used to enforce the stack rlimit on the process stack.
1138 	 */
1139 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1140 	vmspace->vm_maxsaddr = (char *)stack_addr;
1141 
1142 	return (0);
1143 }
1144 
1145 /*
1146  * Copy out argument and environment strings from the old process address
1147  * space into the temporary string buffer.
1148  */
1149 int
1150 exec_copyin_args(struct image_args *args, const char *fname,
1151     enum uio_seg segflg, char **argv, char **envv)
1152 {
1153 	u_long arg, env;
1154 	int error;
1155 
1156 	bzero(args, sizeof(*args));
1157 	if (argv == NULL)
1158 		return (EFAULT);
1159 
1160 	/*
1161 	 * Allocate demand-paged memory for the file name, argument, and
1162 	 * environment strings.
1163 	 */
1164 	error = exec_alloc_args(args);
1165 	if (error != 0)
1166 		return (error);
1167 
1168 	/*
1169 	 * Copy the file name.
1170 	 */
1171 	error = exec_args_add_fname(args, fname, segflg);
1172 	if (error != 0)
1173 		goto err_exit;
1174 
1175 	/*
1176 	 * extract arguments first
1177 	 */
1178 	for (;;) {
1179 		error = fueword(argv++, &arg);
1180 		if (error == -1) {
1181 			error = EFAULT;
1182 			goto err_exit;
1183 		}
1184 		if (arg == 0)
1185 			break;
1186 		error = exec_args_add_arg(args, (char *)(uintptr_t)arg,
1187 		    UIO_USERSPACE);
1188 		if (error != 0)
1189 			goto err_exit;
1190 	}
1191 
1192 	/*
1193 	 * extract environment strings
1194 	 */
1195 	if (envv) {
1196 		for (;;) {
1197 			error = fueword(envv++, &env);
1198 			if (error == -1) {
1199 				error = EFAULT;
1200 				goto err_exit;
1201 			}
1202 			if (env == 0)
1203 				break;
1204 			error = exec_args_add_env(args,
1205 			    (char *)(uintptr_t)env, UIO_USERSPACE);
1206 			if (error != 0)
1207 				goto err_exit;
1208 		}
1209 	}
1210 
1211 	return (0);
1212 
1213 err_exit:
1214 	exec_free_args(args);
1215 	return (error);
1216 }
1217 
1218 int
1219 exec_copyin_data_fds(struct thread *td, struct image_args *args,
1220     const void *data, size_t datalen, const int *fds, size_t fdslen)
1221 {
1222 	struct filedesc *ofdp;
1223 	const char *p;
1224 	int *kfds;
1225 	int error;
1226 
1227 	memset(args, '\0', sizeof(*args));
1228 	ofdp = td->td_proc->p_fd;
1229 	if (datalen >= ARG_MAX || fdslen >= ofdp->fd_nfiles)
1230 		return (E2BIG);
1231 	error = exec_alloc_args(args);
1232 	if (error != 0)
1233 		return (error);
1234 
1235 	args->begin_argv = args->buf;
1236 	args->stringspace = ARG_MAX;
1237 
1238 	if (datalen > 0) {
1239 		/*
1240 		 * Argument buffer has been provided. Copy it into the
1241 		 * kernel as a single string and add a terminating null
1242 		 * byte.
1243 		 */
1244 		error = copyin(data, args->begin_argv, datalen);
1245 		if (error != 0)
1246 			goto err_exit;
1247 		args->begin_argv[datalen] = '\0';
1248 		args->endp = args->begin_argv + datalen + 1;
1249 		args->stringspace -= datalen + 1;
1250 
1251 		/*
1252 		 * Traditional argument counting. Count the number of
1253 		 * null bytes.
1254 		 */
1255 		for (p = args->begin_argv; p < args->endp; ++p)
1256 			if (*p == '\0')
1257 				++args->argc;
1258 	} else {
1259 		/* No argument buffer provided. */
1260 		args->endp = args->begin_argv;
1261 	}
1262 
1263 	/* Create new file descriptor table. */
1264 	kfds = malloc(fdslen * sizeof(int), M_TEMP, M_WAITOK);
1265 	error = copyin(fds, kfds, fdslen * sizeof(int));
1266 	if (error != 0) {
1267 		free(kfds, M_TEMP);
1268 		goto err_exit;
1269 	}
1270 	error = fdcopy_remapped(ofdp, kfds, fdslen, &args->fdp);
1271 	free(kfds, M_TEMP);
1272 	if (error != 0)
1273 		goto err_exit;
1274 
1275 	return (0);
1276 err_exit:
1277 	exec_free_args(args);
1278 	return (error);
1279 }
1280 
1281 struct exec_args_kva {
1282 	vm_offset_t addr;
1283 	u_int gen;
1284 	SLIST_ENTRY(exec_args_kva) next;
1285 };
1286 
1287 DPCPU_DEFINE_STATIC(struct exec_args_kva *, exec_args_kva);
1288 
1289 static SLIST_HEAD(, exec_args_kva) exec_args_kva_freelist;
1290 static struct mtx exec_args_kva_mtx;
1291 static u_int exec_args_gen;
1292 
1293 static void
1294 exec_prealloc_args_kva(void *arg __unused)
1295 {
1296 	struct exec_args_kva *argkva;
1297 	u_int i;
1298 
1299 	SLIST_INIT(&exec_args_kva_freelist);
1300 	mtx_init(&exec_args_kva_mtx, "exec args kva", NULL, MTX_DEF);
1301 	for (i = 0; i < exec_map_entries; i++) {
1302 		argkva = malloc(sizeof(*argkva), M_PARGS, M_WAITOK);
1303 		argkva->addr = kmap_alloc_wait(exec_map, exec_map_entry_size);
1304 		argkva->gen = exec_args_gen;
1305 		SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1306 	}
1307 }
1308 SYSINIT(exec_args_kva, SI_SUB_EXEC, SI_ORDER_ANY, exec_prealloc_args_kva, NULL);
1309 
1310 static vm_offset_t
1311 exec_alloc_args_kva(void **cookie)
1312 {
1313 	struct exec_args_kva *argkva;
1314 
1315 	argkva = (void *)atomic_readandclear_ptr(
1316 	    (uintptr_t *)DPCPU_PTR(exec_args_kva));
1317 	if (argkva == NULL) {
1318 		mtx_lock(&exec_args_kva_mtx);
1319 		while ((argkva = SLIST_FIRST(&exec_args_kva_freelist)) == NULL)
1320 			(void)mtx_sleep(&exec_args_kva_freelist,
1321 			    &exec_args_kva_mtx, 0, "execkva", 0);
1322 		SLIST_REMOVE_HEAD(&exec_args_kva_freelist, next);
1323 		mtx_unlock(&exec_args_kva_mtx);
1324 	}
1325 	*(struct exec_args_kva **)cookie = argkva;
1326 	return (argkva->addr);
1327 }
1328 
1329 static void
1330 exec_release_args_kva(struct exec_args_kva *argkva, u_int gen)
1331 {
1332 	vm_offset_t base;
1333 
1334 	base = argkva->addr;
1335 	if (argkva->gen != gen) {
1336 		(void)vm_map_madvise(exec_map, base, base + exec_map_entry_size,
1337 		    MADV_FREE);
1338 		argkva->gen = gen;
1339 	}
1340 	if (!atomic_cmpset_ptr((uintptr_t *)DPCPU_PTR(exec_args_kva),
1341 	    (uintptr_t)NULL, (uintptr_t)argkva)) {
1342 		mtx_lock(&exec_args_kva_mtx);
1343 		SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1344 		wakeup_one(&exec_args_kva_freelist);
1345 		mtx_unlock(&exec_args_kva_mtx);
1346 	}
1347 }
1348 
1349 static void
1350 exec_free_args_kva(void *cookie)
1351 {
1352 
1353 	exec_release_args_kva(cookie, exec_args_gen);
1354 }
1355 
1356 static void
1357 exec_args_kva_lowmem(void *arg __unused)
1358 {
1359 	SLIST_HEAD(, exec_args_kva) head;
1360 	struct exec_args_kva *argkva;
1361 	u_int gen;
1362 	int i;
1363 
1364 	gen = atomic_fetchadd_int(&exec_args_gen, 1) + 1;
1365 
1366 	/*
1367 	 * Force an madvise of each KVA range. Any currently allocated ranges
1368 	 * will have MADV_FREE applied once they are freed.
1369 	 */
1370 	SLIST_INIT(&head);
1371 	mtx_lock(&exec_args_kva_mtx);
1372 	SLIST_SWAP(&head, &exec_args_kva_freelist, exec_args_kva);
1373 	mtx_unlock(&exec_args_kva_mtx);
1374 	while ((argkva = SLIST_FIRST(&head)) != NULL) {
1375 		SLIST_REMOVE_HEAD(&head, next);
1376 		exec_release_args_kva(argkva, gen);
1377 	}
1378 
1379 	CPU_FOREACH(i) {
1380 		argkva = (void *)atomic_readandclear_ptr(
1381 		    (uintptr_t *)DPCPU_ID_PTR(i, exec_args_kva));
1382 		if (argkva != NULL)
1383 			exec_release_args_kva(argkva, gen);
1384 	}
1385 }
1386 EVENTHANDLER_DEFINE(vm_lowmem, exec_args_kva_lowmem, NULL,
1387     EVENTHANDLER_PRI_ANY);
1388 
1389 /*
1390  * Allocate temporary demand-paged, zero-filled memory for the file name,
1391  * argument, and environment strings.
1392  */
1393 int
1394 exec_alloc_args(struct image_args *args)
1395 {
1396 
1397 	args->buf = (char *)exec_alloc_args_kva(&args->bufkva);
1398 	return (0);
1399 }
1400 
1401 void
1402 exec_free_args(struct image_args *args)
1403 {
1404 
1405 	if (args->buf != NULL) {
1406 		exec_free_args_kva(args->bufkva);
1407 		args->buf = NULL;
1408 	}
1409 	if (args->fname_buf != NULL) {
1410 		free(args->fname_buf, M_TEMP);
1411 		args->fname_buf = NULL;
1412 	}
1413 	if (args->fdp != NULL)
1414 		fdescfree_remapped(args->fdp);
1415 }
1416 
1417 /*
1418  * A set to functions to fill struct image args.
1419  *
1420  * NOTE: exec_args_add_fname() must be called (possibly with a NULL
1421  * fname) before the other functions.  All exec_args_add_arg() calls must
1422  * be made before any exec_args_add_env() calls.  exec_args_adjust_args()
1423  * may be called any time after exec_args_add_fname().
1424  *
1425  * exec_args_add_fname() - install path to be executed
1426  * exec_args_add_arg() - append an argument string
1427  * exec_args_add_env() - append an env string
1428  * exec_args_adjust_args() - adjust location of the argument list to
1429  *                           allow new arguments to be prepended
1430  */
1431 int
1432 exec_args_add_fname(struct image_args *args, const char *fname,
1433     enum uio_seg segflg)
1434 {
1435 	int error;
1436 	size_t length;
1437 
1438 	KASSERT(args->fname == NULL, ("fname already appended"));
1439 	KASSERT(args->endp == NULL, ("already appending to args"));
1440 
1441 	if (fname != NULL) {
1442 		args->fname = args->buf;
1443 		error = segflg == UIO_SYSSPACE ?
1444 		    copystr(fname, args->fname, PATH_MAX, &length) :
1445 		    copyinstr(fname, args->fname, PATH_MAX, &length);
1446 		if (error != 0)
1447 			return (error == ENAMETOOLONG ? E2BIG : error);
1448 	} else
1449 		length = 0;
1450 
1451 	/* Set up for _arg_*()/_env_*() */
1452 	args->endp = args->buf + length;
1453 	/* begin_argv must be set and kept updated */
1454 	args->begin_argv = args->endp;
1455 	KASSERT(exec_map_entry_size - length >= ARG_MAX,
1456 	    ("too little space remaining for arguments %zu < %zu",
1457 	    exec_map_entry_size - length, (size_t)ARG_MAX));
1458 	args->stringspace = ARG_MAX;
1459 
1460 	return (0);
1461 }
1462 
1463 static int
1464 exec_args_add_str(struct image_args *args, const char *str,
1465     enum uio_seg segflg, int *countp)
1466 {
1467 	int error;
1468 	size_t length;
1469 
1470 	KASSERT(args->endp != NULL, ("endp not initialized"));
1471 	KASSERT(args->begin_argv != NULL, ("begin_argp not initialized"));
1472 
1473 	error = (segflg == UIO_SYSSPACE) ?
1474 	    copystr(str, args->endp, args->stringspace, &length) :
1475 	    copyinstr(str, args->endp, args->stringspace, &length);
1476 	if (error != 0)
1477 		return (error == ENAMETOOLONG ? E2BIG : error);
1478 	args->stringspace -= length;
1479 	args->endp += length;
1480 	(*countp)++;
1481 
1482 	return (0);
1483 }
1484 
1485 int
1486 exec_args_add_arg(struct image_args *args, const char *argp,
1487     enum uio_seg segflg)
1488 {
1489 
1490 	KASSERT(args->envc == 0, ("appending args after env"));
1491 
1492 	return (exec_args_add_str(args, argp, segflg, &args->argc));
1493 }
1494 
1495 int
1496 exec_args_add_env(struct image_args *args, const char *envp,
1497     enum uio_seg segflg)
1498 {
1499 
1500 	if (args->envc == 0)
1501 		args->begin_envv = args->endp;
1502 
1503 	return (exec_args_add_str(args, envp, segflg, &args->envc));
1504 }
1505 
1506 int
1507 exec_args_adjust_args(struct image_args *args, size_t consume, ssize_t extend)
1508 {
1509 	ssize_t offset;
1510 
1511 	KASSERT(args->endp != NULL, ("endp not initialized"));
1512 	KASSERT(args->begin_argv != NULL, ("begin_argp not initialized"));
1513 
1514 	offset = extend - consume;
1515 	if (args->stringspace < offset)
1516 		return (E2BIG);
1517 	memmove(args->begin_argv + extend, args->begin_argv + consume,
1518 	    args->endp - args->begin_argv + consume);
1519 	if (args->envc > 0)
1520 		args->begin_envv += offset;
1521 	args->endp += offset;
1522 	args->stringspace -= offset;
1523 	return (0);
1524 }
1525 
1526 char *
1527 exec_args_get_begin_envv(struct image_args *args)
1528 {
1529 
1530 	KASSERT(args->endp != NULL, ("endp not initialized"));
1531 
1532 	if (args->envc > 0)
1533 		return (args->begin_envv);
1534 	return (args->endp);
1535 }
1536 
1537 void
1538 exec_stackgap(struct image_params *imgp, uintptr_t *dp)
1539 {
1540 	if (imgp->sysent->sv_stackgap == NULL ||
1541 	    (imgp->proc->p_fctl0 & (NT_FREEBSD_FCTL_ASLR_DISABLE |
1542 	    NT_FREEBSD_FCTL_ASG_DISABLE)) != 0 ||
1543 	    (imgp->map_flags & MAP_ASLR) == 0)
1544 		return;
1545 	imgp->sysent->sv_stackgap(imgp, dp);
1546 }
1547 
1548 /*
1549  * Copy strings out to the new process address space, constructing new arg
1550  * and env vector tables. Return a pointer to the base so that it can be used
1551  * as the initial stack pointer.
1552  */
1553 int
1554 exec_copyout_strings(struct image_params *imgp, uintptr_t *stack_base)
1555 {
1556 	int argc, envc;
1557 	char **vectp;
1558 	char *stringp;
1559 	uintptr_t destp, ustringp;
1560 	struct ps_strings *arginfo;
1561 	struct proc *p;
1562 	size_t execpath_len;
1563 	int error, szsigcode, szps;
1564 	char canary[sizeof(long) * 8];
1565 
1566 	szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1567 	/*
1568 	 * Calculate string base and vector table pointers.
1569 	 * Also deal with signal trampoline code for this exec type.
1570 	 */
1571 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
1572 		execpath_len = strlen(imgp->execpath) + 1;
1573 	else
1574 		execpath_len = 0;
1575 	p = imgp->proc;
1576 	szsigcode = 0;
1577 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1578 	imgp->ps_strings = arginfo;
1579 	if (p->p_sysent->sv_sigcode_base == 0) {
1580 		if (p->p_sysent->sv_szsigcode != NULL)
1581 			szsigcode = *(p->p_sysent->sv_szsigcode);
1582 	}
1583 	destp =	(uintptr_t)arginfo;
1584 
1585 	/*
1586 	 * install sigcode
1587 	 */
1588 	if (szsigcode != 0) {
1589 		destp -= szsigcode;
1590 		destp = rounddown2(destp, sizeof(void *));
1591 		error = copyout(p->p_sysent->sv_sigcode, (void *)destp,
1592 		    szsigcode);
1593 		if (error != 0)
1594 			return (error);
1595 	}
1596 
1597 	/*
1598 	 * Copy the image path for the rtld.
1599 	 */
1600 	if (execpath_len != 0) {
1601 		destp -= execpath_len;
1602 		destp = rounddown2(destp, sizeof(void *));
1603 		imgp->execpathp = (void *)destp;
1604 		error = copyout(imgp->execpath, imgp->execpathp, execpath_len);
1605 		if (error != 0)
1606 			return (error);
1607 	}
1608 
1609 	/*
1610 	 * Prepare the canary for SSP.
1611 	 */
1612 	arc4rand(canary, sizeof(canary), 0);
1613 	destp -= sizeof(canary);
1614 	imgp->canary = (void *)destp;
1615 	error = copyout(canary, imgp->canary, sizeof(canary));
1616 	if (error != 0)
1617 		return (error);
1618 	imgp->canarylen = sizeof(canary);
1619 
1620 	/*
1621 	 * Prepare the pagesizes array.
1622 	 */
1623 	destp -= szps;
1624 	destp = rounddown2(destp, sizeof(void *));
1625 	imgp->pagesizes = (void *)destp;
1626 	error = copyout(pagesizes, imgp->pagesizes, szps);
1627 	if (error != 0)
1628 		return (error);
1629 	imgp->pagesizeslen = szps;
1630 
1631 	/*
1632 	 * Allocate room for the argument and environment strings.
1633 	 */
1634 	destp -= ARG_MAX - imgp->args->stringspace;
1635 	destp = rounddown2(destp, sizeof(void *));
1636 	ustringp = destp;
1637 
1638 	exec_stackgap(imgp, &destp);
1639 
1640 	if (imgp->auxargs) {
1641 		/*
1642 		 * Allocate room on the stack for the ELF auxargs
1643 		 * array.  It has up to AT_COUNT entries.
1644 		 */
1645 		destp -= AT_COUNT * sizeof(Elf_Auxinfo);
1646 		destp = rounddown2(destp, sizeof(void *));
1647 	}
1648 
1649 	vectp = (char **)destp;
1650 
1651 	/*
1652 	 * Allocate room for the argv[] and env vectors including the
1653 	 * terminating NULL pointers.
1654 	 */
1655 	vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
1656 
1657 	/*
1658 	 * vectp also becomes our initial stack base
1659 	 */
1660 	*stack_base = (uintptr_t)vectp;
1661 
1662 	stringp = imgp->args->begin_argv;
1663 	argc = imgp->args->argc;
1664 	envc = imgp->args->envc;
1665 
1666 	/*
1667 	 * Copy out strings - arguments and environment.
1668 	 */
1669 	error = copyout(stringp, (void *)ustringp,
1670 	    ARG_MAX - imgp->args->stringspace);
1671 	if (error != 0)
1672 		return (error);
1673 
1674 	/*
1675 	 * Fill in "ps_strings" struct for ps, w, etc.
1676 	 */
1677 	imgp->argv = vectp;
1678 	if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 ||
1679 	    suword32(&arginfo->ps_nargvstr, argc) != 0)
1680 		return (EFAULT);
1681 
1682 	/*
1683 	 * Fill in argument portion of vector table.
1684 	 */
1685 	for (; argc > 0; --argc) {
1686 		if (suword(vectp++, ustringp) != 0)
1687 			return (EFAULT);
1688 		while (*stringp++ != 0)
1689 			ustringp++;
1690 		ustringp++;
1691 	}
1692 
1693 	/* a null vector table pointer separates the argp's from the envp's */
1694 	if (suword(vectp++, 0) != 0)
1695 		return (EFAULT);
1696 
1697 	imgp->envv = vectp;
1698 	if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 ||
1699 	    suword32(&arginfo->ps_nenvstr, envc) != 0)
1700 		return (EFAULT);
1701 
1702 	/*
1703 	 * Fill in environment portion of vector table.
1704 	 */
1705 	for (; envc > 0; --envc) {
1706 		if (suword(vectp++, ustringp) != 0)
1707 			return (EFAULT);
1708 		while (*stringp++ != 0)
1709 			ustringp++;
1710 		ustringp++;
1711 	}
1712 
1713 	/* end of vector table is a null pointer */
1714 	if (suword(vectp, 0) != 0)
1715 		return (EFAULT);
1716 
1717 	if (imgp->auxargs) {
1718 		vectp++;
1719 		error = imgp->sysent->sv_copyout_auxargs(imgp,
1720 		    (uintptr_t)vectp);
1721 		if (error != 0)
1722 			return (error);
1723 	}
1724 
1725 	return (0);
1726 }
1727 
1728 /*
1729  * Check permissions of file to execute.
1730  *	Called with imgp->vp locked.
1731  *	Return 0 for success or error code on failure.
1732  */
1733 int
1734 exec_check_permissions(struct image_params *imgp)
1735 {
1736 	struct vnode *vp = imgp->vp;
1737 	struct vattr *attr = imgp->attr;
1738 	struct thread *td;
1739 	int error;
1740 
1741 	td = curthread;
1742 
1743 	/* Get file attributes */
1744 	error = VOP_GETATTR(vp, attr, td->td_ucred);
1745 	if (error)
1746 		return (error);
1747 
1748 #ifdef MAC
1749 	error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1750 	if (error)
1751 		return (error);
1752 #endif
1753 
1754 	/*
1755 	 * 1) Check if file execution is disabled for the filesystem that
1756 	 *    this file resides on.
1757 	 * 2) Ensure that at least one execute bit is on. Otherwise, a
1758 	 *    privileged user will always succeed, and we don't want this
1759 	 *    to happen unless the file really is executable.
1760 	 * 3) Ensure that the file is a regular file.
1761 	 */
1762 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1763 	    (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1764 	    (attr->va_type != VREG))
1765 		return (EACCES);
1766 
1767 	/*
1768 	 * Zero length files can't be exec'd
1769 	 */
1770 	if (attr->va_size == 0)
1771 		return (ENOEXEC);
1772 
1773 	/*
1774 	 *  Check for execute permission to file based on current credentials.
1775 	 */
1776 	error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1777 	if (error)
1778 		return (error);
1779 
1780 	/*
1781 	 * Check number of open-for-writes on the file and deny execution
1782 	 * if there are any.
1783 	 *
1784 	 * Add a text reference now so no one can write to the
1785 	 * executable while we're activating it.
1786 	 *
1787 	 * Remember if this was set before and unset it in case this is not
1788 	 * actually an executable image.
1789 	 */
1790 	error = VOP_SET_TEXT(vp);
1791 	if (error != 0)
1792 		return (error);
1793 	imgp->textset = true;
1794 
1795 	/*
1796 	 * Call filesystem specific open routine (which does nothing in the
1797 	 * general case).
1798 	 */
1799 	error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1800 	if (error == 0)
1801 		imgp->opened = 1;
1802 	return (error);
1803 }
1804 
1805 /*
1806  * Exec handler registration
1807  */
1808 int
1809 exec_register(const struct execsw *execsw_arg)
1810 {
1811 	const struct execsw **es, **xs, **newexecsw;
1812 	u_int count = 2;	/* New slot and trailing NULL */
1813 
1814 	if (execsw)
1815 		for (es = execsw; *es; es++)
1816 			count++;
1817 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1818 	xs = newexecsw;
1819 	if (execsw)
1820 		for (es = execsw; *es; es++)
1821 			*xs++ = *es;
1822 	*xs++ = execsw_arg;
1823 	*xs = NULL;
1824 	if (execsw)
1825 		free(execsw, M_TEMP);
1826 	execsw = newexecsw;
1827 	return (0);
1828 }
1829 
1830 int
1831 exec_unregister(const struct execsw *execsw_arg)
1832 {
1833 	const struct execsw **es, **xs, **newexecsw;
1834 	int count = 1;
1835 
1836 	if (execsw == NULL)
1837 		panic("unregister with no handlers left?\n");
1838 
1839 	for (es = execsw; *es; es++) {
1840 		if (*es == execsw_arg)
1841 			break;
1842 	}
1843 	if (*es == NULL)
1844 		return (ENOENT);
1845 	for (es = execsw; *es; es++)
1846 		if (*es != execsw_arg)
1847 			count++;
1848 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1849 	xs = newexecsw;
1850 	for (es = execsw; *es; es++)
1851 		if (*es != execsw_arg)
1852 			*xs++ = *es;
1853 	*xs = NULL;
1854 	if (execsw)
1855 		free(execsw, M_TEMP);
1856 	execsw = newexecsw;
1857 	return (0);
1858 }
1859