xref: /freebsd/sys/kern/kern_exec.c (revision 79ac3c12a714bcd3f2354c52d948aed9575c46d6)
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/asan.h>
41 #include <sys/capsicum.h>
42 #include <sys/compressor.h>
43 #include <sys/eventhandler.h>
44 #include <sys/exec.h>
45 #include <sys/fcntl.h>
46 #include <sys/filedesc.h>
47 #include <sys/imgact.h>
48 #include <sys/imgact_elf.h>
49 #include <sys/kernel.h>
50 #include <sys/lock.h>
51 #include <sys/malloc.h>
52 #include <sys/mman.h>
53 #include <sys/mount.h>
54 #include <sys/mutex.h>
55 #include <sys/namei.h>
56 #include <sys/priv.h>
57 #include <sys/proc.h>
58 #include <sys/ptrace.h>
59 #include <sys/resourcevar.h>
60 #include <sys/rwlock.h>
61 #include <sys/sched.h>
62 #include <sys/sdt.h>
63 #include <sys/sf_buf.h>
64 #include <sys/shm.h>
65 #include <sys/signalvar.h>
66 #include <sys/smp.h>
67 #include <sys/stat.h>
68 #include <sys/syscallsubr.h>
69 #include <sys/sysctl.h>
70 #include <sys/sysent.h>
71 #include <sys/sysproto.h>
72 #include <sys/timers.h>
73 #include <sys/umtx.h>
74 #include <sys/vnode.h>
75 #include <sys/wait.h>
76 #ifdef KTRACE
77 #include <sys/ktrace.h>
78 #endif
79 
80 #include <vm/vm.h>
81 #include <vm/vm_param.h>
82 #include <vm/pmap.h>
83 #include <vm/vm_page.h>
84 #include <vm/vm_map.h>
85 #include <vm/vm_kern.h>
86 #include <vm/vm_extern.h>
87 #include <vm/vm_object.h>
88 #include <vm/vm_pager.h>
89 
90 #ifdef	HWPMC_HOOKS
91 #include <sys/pmckern.h>
92 #endif
93 
94 #include <machine/reg.h>
95 
96 #include <security/audit/audit.h>
97 #include <security/mac/mac_framework.h>
98 
99 #ifdef KDTRACE_HOOKS
100 #include <sys/dtrace_bsd.h>
101 dtrace_execexit_func_t	dtrace_fasttrap_exec;
102 #endif
103 
104 SDT_PROVIDER_DECLARE(proc);
105 SDT_PROBE_DEFINE1(proc, , , exec, "char *");
106 SDT_PROBE_DEFINE1(proc, , , exec__failure, "int");
107 SDT_PROBE_DEFINE1(proc, , , exec__success, "char *");
108 
109 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
110 
111 int coredump_pack_fileinfo = 1;
112 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_fileinfo, CTLFLAG_RWTUN,
113     &coredump_pack_fileinfo, 0,
114     "Enable file path packing in 'procstat -f' coredump notes");
115 
116 int coredump_pack_vmmapinfo = 1;
117 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_vmmapinfo, CTLFLAG_RWTUN,
118     &coredump_pack_vmmapinfo, 0,
119     "Enable file path packing in 'procstat -v' coredump notes");
120 
121 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
122 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
123 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
124 static int do_execve(struct thread *td, struct image_args *args,
125     struct mac *mac_p, struct vmspace *oldvmspace);
126 
127 /* XXX This should be vm_size_t. */
128 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD|
129     CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_ps_strings, "LU",
130     "Location of process' ps_strings structure");
131 
132 /* XXX This should be vm_size_t. */
133 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD|
134     CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_usrstack, "LU",
135     "Top of process stack");
136 
137 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD|CTLFLAG_MPSAFE,
138     NULL, 0, sysctl_kern_stackprot, "I",
139     "Stack memory permissions");
140 
141 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
142 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
143     &ps_arg_cache_limit, 0,
144     "Process' command line characters cache limit");
145 
146 static int disallow_high_osrel;
147 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW,
148     &disallow_high_osrel, 0,
149     "Disallow execution of binaries built for higher version of the world");
150 
151 static int map_at_zero = 0;
152 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RWTUN, &map_at_zero, 0,
153     "Permit processes to map an object at virtual address 0.");
154 
155 static int
156 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
157 {
158 	struct proc *p;
159 	int error;
160 
161 	p = curproc;
162 #ifdef SCTL_MASK32
163 	if (req->flags & SCTL_MASK32) {
164 		unsigned int val;
165 		val = (unsigned int)p->p_sysent->sv_psstrings;
166 		error = SYSCTL_OUT(req, &val, sizeof(val));
167 	} else
168 #endif
169 		error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
170 		   sizeof(p->p_sysent->sv_psstrings));
171 	return error;
172 }
173 
174 static int
175 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
176 {
177 	struct proc *p;
178 	int error;
179 
180 	p = curproc;
181 #ifdef SCTL_MASK32
182 	if (req->flags & SCTL_MASK32) {
183 		unsigned int val;
184 		val = (unsigned int)p->p_sysent->sv_usrstack;
185 		error = SYSCTL_OUT(req, &val, sizeof(val));
186 	} else
187 #endif
188 		error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
189 		    sizeof(p->p_sysent->sv_usrstack));
190 	return error;
191 }
192 
193 static int
194 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
195 {
196 	struct proc *p;
197 
198 	p = curproc;
199 	return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
200 	    sizeof(p->p_sysent->sv_stackprot)));
201 }
202 
203 /*
204  * Each of the items is a pointer to a `const struct execsw', hence the
205  * double pointer here.
206  */
207 static const struct execsw **execsw;
208 
209 #ifndef _SYS_SYSPROTO_H_
210 struct execve_args {
211 	char    *fname;
212 	char    **argv;
213 	char    **envv;
214 };
215 #endif
216 
217 int
218 sys_execve(struct thread *td, struct execve_args *uap)
219 {
220 	struct image_args args;
221 	struct vmspace *oldvmspace;
222 	int error;
223 
224 	error = pre_execve(td, &oldvmspace);
225 	if (error != 0)
226 		return (error);
227 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
228 	    uap->argv, uap->envv);
229 	if (error == 0)
230 		error = kern_execve(td, &args, NULL, oldvmspace);
231 	post_execve(td, error, oldvmspace);
232 	AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
233 	return (error);
234 }
235 
236 #ifndef _SYS_SYSPROTO_H_
237 struct fexecve_args {
238 	int	fd;
239 	char	**argv;
240 	char	**envv;
241 };
242 #endif
243 int
244 sys_fexecve(struct thread *td, struct fexecve_args *uap)
245 {
246 	struct image_args args;
247 	struct vmspace *oldvmspace;
248 	int error;
249 
250 	error = pre_execve(td, &oldvmspace);
251 	if (error != 0)
252 		return (error);
253 	error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
254 	    uap->argv, uap->envv);
255 	if (error == 0) {
256 		args.fd = uap->fd;
257 		error = kern_execve(td, &args, NULL, oldvmspace);
258 	}
259 	post_execve(td, error, oldvmspace);
260 	AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
261 	return (error);
262 }
263 
264 #ifndef _SYS_SYSPROTO_H_
265 struct __mac_execve_args {
266 	char	*fname;
267 	char	**argv;
268 	char	**envv;
269 	struct mac	*mac_p;
270 };
271 #endif
272 
273 int
274 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap)
275 {
276 #ifdef MAC
277 	struct image_args args;
278 	struct vmspace *oldvmspace;
279 	int error;
280 
281 	error = pre_execve(td, &oldvmspace);
282 	if (error != 0)
283 		return (error);
284 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
285 	    uap->argv, uap->envv);
286 	if (error == 0)
287 		error = kern_execve(td, &args, uap->mac_p, oldvmspace);
288 	post_execve(td, error, oldvmspace);
289 	AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
290 	return (error);
291 #else
292 	return (ENOSYS);
293 #endif
294 }
295 
296 int
297 pre_execve(struct thread *td, struct vmspace **oldvmspace)
298 {
299 	struct proc *p;
300 	int error;
301 
302 	KASSERT(td == curthread, ("non-current thread %p", td));
303 	error = 0;
304 	p = td->td_proc;
305 	if ((p->p_flag & P_HADTHREADS) != 0) {
306 		PROC_LOCK(p);
307 		if (thread_single(p, SINGLE_BOUNDARY) != 0)
308 			error = ERESTART;
309 		PROC_UNLOCK(p);
310 	}
311 	KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0,
312 	    ("nested execve"));
313 	*oldvmspace = p->p_vmspace;
314 	return (error);
315 }
316 
317 void
318 post_execve(struct thread *td, int error, struct vmspace *oldvmspace)
319 {
320 	struct proc *p;
321 
322 	KASSERT(td == curthread, ("non-current thread %p", td));
323 	p = td->td_proc;
324 	if ((p->p_flag & P_HADTHREADS) != 0) {
325 		PROC_LOCK(p);
326 		/*
327 		 * If success, we upgrade to SINGLE_EXIT state to
328 		 * force other threads to suicide.
329 		 */
330 		if (error == EJUSTRETURN)
331 			thread_single(p, SINGLE_EXIT);
332 		else
333 			thread_single_end(p, SINGLE_BOUNDARY);
334 		PROC_UNLOCK(p);
335 	}
336 	exec_cleanup(td, oldvmspace);
337 }
338 
339 /*
340  * kern_execve() has the astonishing property of not always returning to
341  * the caller.  If sufficiently bad things happen during the call to
342  * do_execve(), it can end up calling exit1(); as a result, callers must
343  * avoid doing anything which they might need to undo (e.g., allocating
344  * memory).
345  */
346 int
347 kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p,
348     struct vmspace *oldvmspace)
349 {
350 
351 	AUDIT_ARG_ARGV(args->begin_argv, args->argc,
352 	    exec_args_get_begin_envv(args) - args->begin_argv);
353 	AUDIT_ARG_ENVV(exec_args_get_begin_envv(args), args->envc,
354 	    args->endp - exec_args_get_begin_envv(args));
355 	return (do_execve(td, args, mac_p, oldvmspace));
356 }
357 
358 static void
359 execve_nosetid(struct image_params *imgp)
360 {
361 	imgp->credential_setid = false;
362 	if (imgp->newcred != NULL) {
363 		crfree(imgp->newcred);
364 		imgp->newcred = NULL;
365 	}
366 }
367 
368 /*
369  * In-kernel implementation of execve().  All arguments are assumed to be
370  * userspace pointers from the passed thread.
371  */
372 static int
373 do_execve(struct thread *td, struct image_args *args, struct mac *mac_p,
374     struct vmspace *oldvmspace)
375 {
376 	struct proc *p = td->td_proc;
377 	struct nameidata nd;
378 	struct ucred *oldcred;
379 	struct uidinfo *euip = NULL;
380 	uintptr_t stack_base;
381 	struct image_params image_params, *imgp;
382 	struct vattr attr;
383 	int (*img_first)(struct image_params *);
384 	struct pargs *oldargs = NULL, *newargs = NULL;
385 	struct sigacts *oldsigacts = NULL, *newsigacts = NULL;
386 #ifdef KTRACE
387 	struct ktr_io_params *kiop;
388 #endif
389 	struct vnode *oldtextvp = NULL, *newtextvp;
390 	int credential_changing;
391 #ifdef MAC
392 	struct label *interpvplabel = NULL;
393 	int will_transition;
394 #endif
395 #ifdef HWPMC_HOOKS
396 	struct pmckern_procexec pe;
397 #endif
398 	int error, i, orig_osrel;
399 	uint32_t orig_fctl0;
400 	static const char fexecv_proc_title[] = "(fexecv)";
401 
402 	imgp = &image_params;
403 #ifdef KTRACE
404 	kiop = NULL;
405 #endif
406 
407 	/*
408 	 * Lock the process and set the P_INEXEC flag to indicate that
409 	 * it should be left alone until we're done here.  This is
410 	 * necessary to avoid race conditions - e.g. in ptrace() -
411 	 * that might allow a local user to illicitly obtain elevated
412 	 * privileges.
413 	 */
414 	PROC_LOCK(p);
415 	KASSERT((p->p_flag & P_INEXEC) == 0,
416 	    ("%s(): process already has P_INEXEC flag", __func__));
417 	p->p_flag |= P_INEXEC;
418 	PROC_UNLOCK(p);
419 
420 	/*
421 	 * Initialize part of the common data
422 	 */
423 	bzero(imgp, sizeof(*imgp));
424 	imgp->proc = p;
425 	imgp->attr = &attr;
426 	imgp->args = args;
427 	oldcred = p->p_ucred;
428 	orig_osrel = p->p_osrel;
429 	orig_fctl0 = p->p_fctl0;
430 
431 #ifdef MAC
432 	error = mac_execve_enter(imgp, mac_p);
433 	if (error)
434 		goto exec_fail;
435 #endif
436 
437 	/*
438 	 * Translate the file name. namei() returns a vnode pointer
439 	 *	in ni_vp among other things.
440 	 *
441 	 * XXXAUDIT: It would be desirable to also audit the name of the
442 	 * interpreter if this is an interpreted binary.
443 	 */
444 	if (args->fname != NULL) {
445 		NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW |
446 		    SAVENAME | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
447 	}
448 
449 	SDT_PROBE1(proc, , , exec, args->fname);
450 
451 interpret:
452 	if (args->fname != NULL) {
453 #ifdef CAPABILITY_MODE
454 		/*
455 		 * While capability mode can't reach this point via direct
456 		 * path arguments to execve(), we also don't allow
457 		 * interpreters to be used in capability mode (for now).
458 		 * Catch indirect lookups and return a permissions error.
459 		 */
460 		if (IN_CAPABILITY_MODE(td)) {
461 			error = ECAPMODE;
462 			goto exec_fail;
463 		}
464 #endif
465 		error = namei(&nd);
466 		if (error)
467 			goto exec_fail;
468 
469 		newtextvp = nd.ni_vp;
470 		imgp->vp = newtextvp;
471 	} else {
472 		AUDIT_ARG_FD(args->fd);
473 		/*
474 		 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
475 		 */
476 		error = fgetvp_exec(td, args->fd, &cap_fexecve_rights, &newtextvp);
477 		if (error)
478 			goto exec_fail;
479 		vn_lock(newtextvp, LK_SHARED | LK_RETRY);
480 		AUDIT_ARG_VNODE1(newtextvp);
481 		imgp->vp = newtextvp;
482 	}
483 
484 	/*
485 	 * Check file permissions.  Also 'opens' file and sets its vnode to
486 	 * text mode.
487 	 */
488 	error = exec_check_permissions(imgp);
489 	if (error)
490 		goto exec_fail_dealloc;
491 
492 	imgp->object = imgp->vp->v_object;
493 	if (imgp->object != NULL)
494 		vm_object_reference(imgp->object);
495 
496 	error = exec_map_first_page(imgp);
497 	if (error)
498 		goto exec_fail_dealloc;
499 
500 	imgp->proc->p_osrel = 0;
501 	imgp->proc->p_fctl0 = 0;
502 
503 	/*
504 	 * Implement image setuid/setgid.
505 	 *
506 	 * Determine new credentials before attempting image activators
507 	 * so that it can be used by process_exec handlers to determine
508 	 * credential/setid changes.
509 	 *
510 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
511 	 * the process is being traced.
512 	 *
513 	 * We disable setuid/setgid/etc in capability mode on the basis
514 	 * that most setugid applications are not written with that
515 	 * environment in mind, and will therefore almost certainly operate
516 	 * incorrectly. In principle there's no reason that setugid
517 	 * applications might not be useful in capability mode, so we may want
518 	 * to reconsider this conservative design choice in the future.
519 	 *
520 	 * XXXMAC: For the time being, use NOSUID to also prohibit
521 	 * transitions on the file system.
522 	 */
523 	credential_changing = 0;
524 	credential_changing |= (attr.va_mode & S_ISUID) &&
525 	    oldcred->cr_uid != attr.va_uid;
526 	credential_changing |= (attr.va_mode & S_ISGID) &&
527 	    oldcred->cr_gid != attr.va_gid;
528 #ifdef MAC
529 	will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
530 	    interpvplabel, imgp);
531 	credential_changing |= will_transition;
532 #endif
533 
534 	/* Don't inherit PROC_PDEATHSIG_CTL value if setuid/setgid. */
535 	if (credential_changing)
536 		imgp->proc->p_pdeathsig = 0;
537 
538 	if (credential_changing &&
539 #ifdef CAPABILITY_MODE
540 	    ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
541 #endif
542 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
543 	    (p->p_flag & P_TRACED) == 0) {
544 		imgp->credential_setid = true;
545 		VOP_UNLOCK(imgp->vp);
546 		imgp->newcred = crdup(oldcred);
547 		if (attr.va_mode & S_ISUID) {
548 			euip = uifind(attr.va_uid);
549 			change_euid(imgp->newcred, euip);
550 		}
551 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
552 		if (attr.va_mode & S_ISGID)
553 			change_egid(imgp->newcred, attr.va_gid);
554 		/*
555 		 * Implement correct POSIX saved-id behavior.
556 		 *
557 		 * XXXMAC: Note that the current logic will save the
558 		 * uid and gid if a MAC domain transition occurs, even
559 		 * though maybe it shouldn't.
560 		 */
561 		change_svuid(imgp->newcred, imgp->newcred->cr_uid);
562 		change_svgid(imgp->newcred, imgp->newcred->cr_gid);
563 	} else {
564 		/*
565 		 * Implement correct POSIX saved-id behavior.
566 		 *
567 		 * XXX: It's not clear that the existing behavior is
568 		 * POSIX-compliant.  A number of sources indicate that the
569 		 * saved uid/gid should only be updated if the new ruid is
570 		 * not equal to the old ruid, or the new euid is not equal
571 		 * to the old euid and the new euid is not equal to the old
572 		 * ruid.  The FreeBSD code always updates the saved uid/gid.
573 		 * Also, this code uses the new (replaced) euid and egid as
574 		 * the source, which may or may not be the right ones to use.
575 		 */
576 		if (oldcred->cr_svuid != oldcred->cr_uid ||
577 		    oldcred->cr_svgid != oldcred->cr_gid) {
578 			VOP_UNLOCK(imgp->vp);
579 			imgp->newcred = crdup(oldcred);
580 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
581 			change_svuid(imgp->newcred, imgp->newcred->cr_uid);
582 			change_svgid(imgp->newcred, imgp->newcred->cr_gid);
583 		}
584 	}
585 	/* The new credentials are installed into the process later. */
586 
587 	/*
588 	 * Do the best to calculate the full path to the image file.
589 	 */
590 	if (args->fname != NULL && args->fname[0] == '/')
591 		imgp->execpath = args->fname;
592 	else {
593 		VOP_UNLOCK(imgp->vp);
594 		if (vn_fullpath(imgp->vp, &imgp->execpath, &imgp->freepath) != 0)
595 			imgp->execpath = args->fname;
596 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
597 	}
598 
599 	/*
600 	 *	If the current process has a special image activator it
601 	 *	wants to try first, call it.   For example, emulating shell
602 	 *	scripts differently.
603 	 */
604 	error = -1;
605 	if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
606 		error = img_first(imgp);
607 
608 	/*
609 	 *	Loop through the list of image activators, calling each one.
610 	 *	An activator returns -1 if there is no match, 0 on success,
611 	 *	and an error otherwise.
612 	 */
613 	for (i = 0; error == -1 && execsw[i]; ++i) {
614 		if (execsw[i]->ex_imgact == NULL ||
615 		    execsw[i]->ex_imgact == img_first) {
616 			continue;
617 		}
618 		error = (*execsw[i]->ex_imgact)(imgp);
619 	}
620 
621 	if (error) {
622 		if (error == -1)
623 			error = ENOEXEC;
624 		goto exec_fail_dealloc;
625 	}
626 
627 	/*
628 	 * Special interpreter operation, cleanup and loop up to try to
629 	 * activate the interpreter.
630 	 */
631 	if (imgp->interpreted) {
632 		exec_unmap_first_page(imgp);
633 		/*
634 		 * The text reference needs to be removed for scripts.
635 		 * There is a short period before we determine that
636 		 * something is a script where text reference is active.
637 		 * The vnode lock is held over this entire period
638 		 * so nothing should illegitimately be blocked.
639 		 */
640 		MPASS(imgp->textset);
641 		VOP_UNSET_TEXT_CHECKED(newtextvp);
642 		imgp->textset = false;
643 		/* free name buffer and old vnode */
644 		if (args->fname != NULL)
645 			NDFREE(&nd, NDF_ONLY_PNBUF);
646 #ifdef MAC
647 		mac_execve_interpreter_enter(newtextvp, &interpvplabel);
648 #endif
649 		if (imgp->opened) {
650 			VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td);
651 			imgp->opened = 0;
652 		}
653 		vput(newtextvp);
654 		vm_object_deallocate(imgp->object);
655 		imgp->object = NULL;
656 		execve_nosetid(imgp);
657 		imgp->execpath = NULL;
658 		free(imgp->freepath, M_TEMP);
659 		imgp->freepath = NULL;
660 		/* set new name to that of the interpreter */
661 		NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW |
662 		    SAVENAME, UIO_SYSSPACE, imgp->interpreter_name, td);
663 		args->fname = imgp->interpreter_name;
664 		goto interpret;
665 	}
666 
667 	/*
668 	 * NB: We unlock the vnode here because it is believed that none
669 	 * of the sv_copyout_strings/sv_fixup operations require the vnode.
670 	 */
671 	VOP_UNLOCK(imgp->vp);
672 
673 	if (disallow_high_osrel &&
674 	    P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
675 		error = ENOEXEC;
676 		uprintf("Osrel %d for image %s too high\n", p->p_osrel,
677 		    imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
678 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
679 		goto exec_fail_dealloc;
680 	}
681 
682 	/* ABI enforces the use of Capsicum. Switch into capabilities mode. */
683 	if (SV_PROC_FLAG(p, SV_CAPSICUM))
684 		sys_cap_enter(td, NULL);
685 
686 	/*
687 	 * Copy out strings (args and env) and initialize stack base.
688 	 */
689 	error = (*p->p_sysent->sv_copyout_strings)(imgp, &stack_base);
690 	if (error != 0) {
691 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
692 		goto exec_fail_dealloc;
693 	}
694 
695 	/*
696 	 * Stack setup.
697 	 */
698 	error = (*p->p_sysent->sv_fixup)(&stack_base, imgp);
699 	if (error != 0) {
700 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
701 		goto exec_fail_dealloc;
702 	}
703 
704 	if (args->fdp != NULL) {
705 		/* Install a brand new file descriptor table. */
706 		fdinstall_remapped(td, args->fdp);
707 		args->fdp = NULL;
708 	} else {
709 		/*
710 		 * Keep on using the existing file descriptor table. For
711 		 * security and other reasons, the file descriptor table
712 		 * cannot be shared after an exec.
713 		 */
714 		fdunshare(td);
715 		pdunshare(td);
716 		/* close files on exec */
717 		fdcloseexec(td);
718 	}
719 
720 	/*
721 	 * Malloc things before we need locks.
722 	 */
723 	i = exec_args_get_begin_envv(imgp->args) - imgp->args->begin_argv;
724 	/* Cache arguments if they fit inside our allowance */
725 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
726 		newargs = pargs_alloc(i);
727 		bcopy(imgp->args->begin_argv, newargs->ar_args, i);
728 	}
729 
730 	/*
731 	 * For security and other reasons, signal handlers cannot
732 	 * be shared after an exec. The new process gets a copy of the old
733 	 * handlers. In execsigs(), the new process will have its signals
734 	 * reset.
735 	 */
736 	if (sigacts_shared(p->p_sigacts)) {
737 		oldsigacts = p->p_sigacts;
738 		newsigacts = sigacts_alloc();
739 		sigacts_copy(newsigacts, oldsigacts);
740 	}
741 
742 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
743 
744 	PROC_LOCK(p);
745 	if (oldsigacts)
746 		p->p_sigacts = newsigacts;
747 	/* Stop profiling */
748 	stopprofclock(p);
749 
750 	/* reset caught signals */
751 	execsigs(p);
752 
753 	/* name this process - nameiexec(p, ndp) */
754 	bzero(p->p_comm, sizeof(p->p_comm));
755 	if (args->fname)
756 		bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
757 		    min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
758 	else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0)
759 		bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
760 	bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
761 #ifdef KTR
762 	sched_clear_tdname(td);
763 #endif
764 
765 	/*
766 	 * mark as execed, wakeup the process that vforked (if any) and tell
767 	 * it that it now has its own resources back
768 	 */
769 	p->p_flag |= P_EXEC;
770 	if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0)
771 		p->p_flag2 &= ~P2_NOTRACE;
772 	if ((p->p_flag2 & P2_STKGAP_DISABLE_EXEC) == 0)
773 		p->p_flag2 &= ~P2_STKGAP_DISABLE;
774 	if (p->p_flag & P_PPWAIT) {
775 		p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
776 		cv_broadcast(&p->p_pwait);
777 		/* STOPs are no longer ignored, arrange for AST */
778 		signotify(td);
779 	}
780 
781 	if (imgp->sysent->sv_setid_allowed != NULL &&
782 	    !(*imgp->sysent->sv_setid_allowed)(td, imgp))
783 		execve_nosetid(imgp);
784 
785 	/*
786 	 * Implement image setuid/setgid installation.
787 	 */
788 	if (imgp->credential_setid) {
789 		/*
790 		 * Turn off syscall tracing for set-id programs, except for
791 		 * root.  Record any set-id flags first to make sure that
792 		 * we do not regain any tracing during a possible block.
793 		 */
794 		setsugid(p);
795 #ifdef KTRACE
796 		kiop = ktrprocexec(p);
797 #endif
798 		/*
799 		 * Close any file descriptors 0..2 that reference procfs,
800 		 * then make sure file descriptors 0..2 are in use.
801 		 *
802 		 * Both fdsetugidsafety() and fdcheckstd() may call functions
803 		 * taking sleepable locks, so temporarily drop our locks.
804 		 */
805 		PROC_UNLOCK(p);
806 		VOP_UNLOCK(imgp->vp);
807 		fdsetugidsafety(td);
808 		error = fdcheckstd(td);
809 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
810 		if (error != 0)
811 			goto exec_fail_dealloc;
812 		PROC_LOCK(p);
813 #ifdef MAC
814 		if (will_transition) {
815 			mac_vnode_execve_transition(oldcred, imgp->newcred,
816 			    imgp->vp, interpvplabel, imgp);
817 		}
818 #endif
819 	} else {
820 		if (oldcred->cr_uid == oldcred->cr_ruid &&
821 		    oldcred->cr_gid == oldcred->cr_rgid)
822 			p->p_flag &= ~P_SUGID;
823 	}
824 	/*
825 	 * Set the new credentials.
826 	 */
827 	if (imgp->newcred != NULL) {
828 		proc_set_cred(p, imgp->newcred);
829 		crfree(oldcred);
830 		oldcred = NULL;
831 	}
832 
833 	/*
834 	 * Store the vp for use in procfs.  This vnode was referenced by namei
835 	 * or fgetvp_exec.
836 	 */
837 	oldtextvp = p->p_textvp;
838 	p->p_textvp = newtextvp;
839 
840 #ifdef KDTRACE_HOOKS
841 	/*
842 	 * Tell the DTrace fasttrap provider about the exec if it
843 	 * has declared an interest.
844 	 */
845 	if (dtrace_fasttrap_exec)
846 		dtrace_fasttrap_exec(p);
847 #endif
848 
849 	/*
850 	 * Notify others that we exec'd, and clear the P_INEXEC flag
851 	 * as we're now a bona fide freshly-execed process.
852 	 */
853 	KNOTE_LOCKED(p->p_klist, NOTE_EXEC);
854 	p->p_flag &= ~P_INEXEC;
855 
856 	/* clear "fork but no exec" flag, as we _are_ execing */
857 	p->p_acflag &= ~AFORK;
858 
859 	/*
860 	 * Free any previous argument cache and replace it with
861 	 * the new argument cache, if any.
862 	 */
863 	oldargs = p->p_args;
864 	p->p_args = newargs;
865 	newargs = NULL;
866 
867 	PROC_UNLOCK(p);
868 
869 #ifdef	HWPMC_HOOKS
870 	/*
871 	 * Check if system-wide sampling is in effect or if the
872 	 * current process is using PMCs.  If so, do exec() time
873 	 * processing.  This processing needs to happen AFTER the
874 	 * P_INEXEC flag is cleared.
875 	 */
876 	if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
877 		VOP_UNLOCK(imgp->vp);
878 		pe.pm_credentialschanged = credential_changing;
879 		pe.pm_entryaddr = imgp->entry_addr;
880 
881 		PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
882 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
883 	}
884 #endif
885 
886 	/* Set values passed into the program in registers. */
887 	(*p->p_sysent->sv_setregs)(td, imgp, stack_base);
888 
889 	VOP_MMAPPED(imgp->vp);
890 
891 	SDT_PROBE1(proc, , , exec__success, args->fname);
892 
893 exec_fail_dealloc:
894 	if (error != 0) {
895 		p->p_osrel = orig_osrel;
896 		p->p_fctl0 = orig_fctl0;
897 	}
898 
899 	if (imgp->firstpage != NULL)
900 		exec_unmap_first_page(imgp);
901 
902 	if (imgp->vp != NULL) {
903 		if (args->fname)
904 			NDFREE(&nd, NDF_ONLY_PNBUF);
905 		if (imgp->opened)
906 			VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
907 		if (imgp->textset)
908 			VOP_UNSET_TEXT_CHECKED(imgp->vp);
909 		if (error != 0)
910 			vput(imgp->vp);
911 		else
912 			VOP_UNLOCK(imgp->vp);
913 	}
914 
915 	if (imgp->object != NULL)
916 		vm_object_deallocate(imgp->object);
917 
918 	free(imgp->freepath, M_TEMP);
919 
920 	if (error == 0) {
921 		if (p->p_ptevents & PTRACE_EXEC) {
922 			PROC_LOCK(p);
923 			if (p->p_ptevents & PTRACE_EXEC)
924 				td->td_dbgflags |= TDB_EXEC;
925 			PROC_UNLOCK(p);
926 		}
927 	} else {
928 exec_fail:
929 		/* we're done here, clear P_INEXEC */
930 		PROC_LOCK(p);
931 		p->p_flag &= ~P_INEXEC;
932 		PROC_UNLOCK(p);
933 
934 		SDT_PROBE1(proc, , , exec__failure, error);
935 	}
936 
937 	if (imgp->newcred != NULL && oldcred != NULL)
938 		crfree(imgp->newcred);
939 
940 #ifdef MAC
941 	mac_execve_exit(imgp);
942 	mac_execve_interpreter_exit(interpvplabel);
943 #endif
944 	exec_free_args(args);
945 
946 	/*
947 	 * Handle deferred decrement of ref counts.
948 	 */
949 	if (oldtextvp != NULL)
950 		vrele(oldtextvp);
951 #ifdef KTRACE
952 	ktr_io_params_free(kiop);
953 #endif
954 	pargs_drop(oldargs);
955 	pargs_drop(newargs);
956 	if (oldsigacts != NULL)
957 		sigacts_free(oldsigacts);
958 	if (euip != NULL)
959 		uifree(euip);
960 
961 	if (error && imgp->vmspace_destroyed) {
962 		/* sorry, no more process anymore. exit gracefully */
963 		exec_cleanup(td, oldvmspace);
964 		exit1(td, 0, SIGABRT);
965 		/* NOT REACHED */
966 	}
967 
968 #ifdef KTRACE
969 	if (error == 0)
970 		ktrprocctor(p);
971 #endif
972 
973 	/*
974 	 * We don't want cpu_set_syscall_retval() to overwrite any of
975 	 * the register values put in place by exec_setregs().
976 	 * Implementations of cpu_set_syscall_retval() will leave
977 	 * registers unmodified when returning EJUSTRETURN.
978 	 */
979 	return (error == 0 ? EJUSTRETURN : error);
980 }
981 
982 void
983 exec_cleanup(struct thread *td, struct vmspace *oldvmspace)
984 {
985 	if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
986 		KASSERT(td->td_proc->p_vmspace != oldvmspace,
987 		    ("oldvmspace still used"));
988 		vmspace_free(oldvmspace);
989 		td->td_pflags &= ~TDP_EXECVMSPC;
990 	}
991 }
992 
993 int
994 exec_map_first_page(struct image_params *imgp)
995 {
996 	vm_object_t object;
997 	vm_page_t m;
998 	int error;
999 
1000 	if (imgp->firstpage != NULL)
1001 		exec_unmap_first_page(imgp);
1002 
1003 	object = imgp->vp->v_object;
1004 	if (object == NULL)
1005 		return (EACCES);
1006 #if VM_NRESERVLEVEL > 0
1007 	if ((object->flags & OBJ_COLORED) == 0) {
1008 		VM_OBJECT_WLOCK(object);
1009 		vm_object_color(object, 0);
1010 		VM_OBJECT_WUNLOCK(object);
1011 	}
1012 #endif
1013 	error = vm_page_grab_valid_unlocked(&m, object, 0,
1014 	    VM_ALLOC_COUNT(VM_INITIAL_PAGEIN) |
1015             VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED);
1016 
1017 	if (error != VM_PAGER_OK)
1018 		return (EIO);
1019 	imgp->firstpage = sf_buf_alloc(m, 0);
1020 	imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
1021 
1022 	return (0);
1023 }
1024 
1025 void
1026 exec_unmap_first_page(struct image_params *imgp)
1027 {
1028 	vm_page_t m;
1029 
1030 	if (imgp->firstpage != NULL) {
1031 		m = sf_buf_page(imgp->firstpage);
1032 		sf_buf_free(imgp->firstpage);
1033 		imgp->firstpage = NULL;
1034 		vm_page_unwire(m, PQ_ACTIVE);
1035 	}
1036 }
1037 
1038 /*
1039  * Destroy old address space, and allocate a new stack.
1040  *	The new stack is only sgrowsiz large because it is grown
1041  *	automatically on a page fault.
1042  */
1043 int
1044 exec_new_vmspace(struct image_params *imgp, struct sysentvec *sv)
1045 {
1046 	int error;
1047 	struct proc *p = imgp->proc;
1048 	struct vmspace *vmspace = p->p_vmspace;
1049 	struct thread *td = curthread;
1050 	vm_object_t obj;
1051 	struct rlimit rlim_stack;
1052 	vm_offset_t sv_minuser, stack_addr;
1053 	vm_map_t map;
1054 	vm_prot_t stack_prot;
1055 	u_long ssiz;
1056 
1057 	imgp->vmspace_destroyed = 1;
1058 	imgp->sysent = sv;
1059 
1060 	sigfastblock_clear(td);
1061 	umtx_exec(p);
1062 	itimers_exec(p);
1063 	if (sv->sv_onexec != NULL)
1064 		sv->sv_onexec(p, imgp);
1065 
1066 	EVENTHANDLER_DIRECT_INVOKE(process_exec, p, imgp);
1067 
1068 	/*
1069 	 * Blow away entire process VM, if address space not shared,
1070 	 * otherwise, create a new VM space so that other threads are
1071 	 * not disrupted
1072 	 */
1073 	map = &vmspace->vm_map;
1074 	if (map_at_zero)
1075 		sv_minuser = sv->sv_minuser;
1076 	else
1077 		sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1078 	if (refcount_load(&vmspace->vm_refcnt) == 1 &&
1079 	    vm_map_min(map) == sv_minuser &&
1080 	    vm_map_max(map) == sv->sv_maxuser &&
1081 	    cpu_exec_vmspace_reuse(p, map)) {
1082 		shmexit(vmspace);
1083 		pmap_remove_pages(vmspace_pmap(vmspace));
1084 		vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1085 		/*
1086 		 * An exec terminates mlockall(MCL_FUTURE).
1087 		 * ASLR and W^X states must be re-evaluated.
1088 		 */
1089 		vm_map_lock(map);
1090 		vm_map_modflags(map, 0, MAP_WIREFUTURE | MAP_ASLR |
1091 		    MAP_ASLR_IGNSTART | MAP_WXORX);
1092 		vm_map_unlock(map);
1093 	} else {
1094 		error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1095 		if (error)
1096 			return (error);
1097 		vmspace = p->p_vmspace;
1098 		map = &vmspace->vm_map;
1099 	}
1100 	map->flags |= imgp->map_flags;
1101 
1102 	/* Map a shared page */
1103 	obj = sv->sv_shared_page_obj;
1104 	if (obj != NULL) {
1105 		vm_object_reference(obj);
1106 		error = vm_map_fixed(map, obj, 0,
1107 		    sv->sv_shared_page_base, sv->sv_shared_page_len,
1108 		    VM_PROT_READ | VM_PROT_EXECUTE,
1109 		    VM_PROT_READ | VM_PROT_EXECUTE,
1110 		    MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1111 		if (error != KERN_SUCCESS) {
1112 			vm_object_deallocate(obj);
1113 			return (vm_mmap_to_errno(error));
1114 		}
1115 	}
1116 
1117 	/* Allocate a new stack */
1118 	if (imgp->stack_sz != 0) {
1119 		ssiz = trunc_page(imgp->stack_sz);
1120 		PROC_LOCK(p);
1121 		lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack);
1122 		PROC_UNLOCK(p);
1123 		if (ssiz > rlim_stack.rlim_max)
1124 			ssiz = rlim_stack.rlim_max;
1125 		if (ssiz > rlim_stack.rlim_cur) {
1126 			rlim_stack.rlim_cur = ssiz;
1127 			kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack);
1128 		}
1129 	} else if (sv->sv_maxssiz != NULL) {
1130 		ssiz = *sv->sv_maxssiz;
1131 	} else {
1132 		ssiz = maxssiz;
1133 	}
1134 	imgp->eff_stack_sz = lim_cur(curthread, RLIMIT_STACK);
1135 	if (ssiz < imgp->eff_stack_sz)
1136 		imgp->eff_stack_sz = ssiz;
1137 	stack_addr = sv->sv_usrstack - ssiz;
1138 	stack_prot = obj != NULL && imgp->stack_prot != 0 ?
1139 	    imgp->stack_prot : sv->sv_stackprot;
1140 	error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz, stack_prot,
1141 	    VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1142 	if (error != KERN_SUCCESS) {
1143 		uprintf("exec_new_vmspace: mapping stack size %#jx prot %#x "
1144 		    "failed mach error %d errno %d\n", (uintmax_t)ssiz,
1145 		    stack_prot, error, vm_mmap_to_errno(error));
1146 		return (vm_mmap_to_errno(error));
1147 	}
1148 
1149 	/*
1150 	 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they
1151 	 * are still used to enforce the stack rlimit on the process stack.
1152 	 */
1153 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1154 	vmspace->vm_maxsaddr = (char *)stack_addr;
1155 
1156 	return (0);
1157 }
1158 
1159 /*
1160  * Copy out argument and environment strings from the old process address
1161  * space into the temporary string buffer.
1162  */
1163 int
1164 exec_copyin_args(struct image_args *args, const char *fname,
1165     enum uio_seg segflg, char **argv, char **envv)
1166 {
1167 	u_long arg, env;
1168 	int error;
1169 
1170 	bzero(args, sizeof(*args));
1171 	if (argv == NULL)
1172 		return (EFAULT);
1173 
1174 	/*
1175 	 * Allocate demand-paged memory for the file name, argument, and
1176 	 * environment strings.
1177 	 */
1178 	error = exec_alloc_args(args);
1179 	if (error != 0)
1180 		return (error);
1181 
1182 	/*
1183 	 * Copy the file name.
1184 	 */
1185 	error = exec_args_add_fname(args, fname, segflg);
1186 	if (error != 0)
1187 		goto err_exit;
1188 
1189 	/*
1190 	 * extract arguments first
1191 	 */
1192 	for (;;) {
1193 		error = fueword(argv++, &arg);
1194 		if (error == -1) {
1195 			error = EFAULT;
1196 			goto err_exit;
1197 		}
1198 		if (arg == 0)
1199 			break;
1200 		error = exec_args_add_arg(args, (char *)(uintptr_t)arg,
1201 		    UIO_USERSPACE);
1202 		if (error != 0)
1203 			goto err_exit;
1204 	}
1205 
1206 	/*
1207 	 * extract environment strings
1208 	 */
1209 	if (envv) {
1210 		for (;;) {
1211 			error = fueword(envv++, &env);
1212 			if (error == -1) {
1213 				error = EFAULT;
1214 				goto err_exit;
1215 			}
1216 			if (env == 0)
1217 				break;
1218 			error = exec_args_add_env(args,
1219 			    (char *)(uintptr_t)env, UIO_USERSPACE);
1220 			if (error != 0)
1221 				goto err_exit;
1222 		}
1223 	}
1224 
1225 	return (0);
1226 
1227 err_exit:
1228 	exec_free_args(args);
1229 	return (error);
1230 }
1231 
1232 int
1233 exec_copyin_data_fds(struct thread *td, struct image_args *args,
1234     const void *data, size_t datalen, const int *fds, size_t fdslen)
1235 {
1236 	struct filedesc *ofdp;
1237 	const char *p;
1238 	int *kfds;
1239 	int error;
1240 
1241 	memset(args, '\0', sizeof(*args));
1242 	ofdp = td->td_proc->p_fd;
1243 	if (datalen >= ARG_MAX || fdslen >= ofdp->fd_nfiles)
1244 		return (E2BIG);
1245 	error = exec_alloc_args(args);
1246 	if (error != 0)
1247 		return (error);
1248 
1249 	args->begin_argv = args->buf;
1250 	args->stringspace = ARG_MAX;
1251 
1252 	if (datalen > 0) {
1253 		/*
1254 		 * Argument buffer has been provided. Copy it into the
1255 		 * kernel as a single string and add a terminating null
1256 		 * byte.
1257 		 */
1258 		error = copyin(data, args->begin_argv, datalen);
1259 		if (error != 0)
1260 			goto err_exit;
1261 		args->begin_argv[datalen] = '\0';
1262 		args->endp = args->begin_argv + datalen + 1;
1263 		args->stringspace -= datalen + 1;
1264 
1265 		/*
1266 		 * Traditional argument counting. Count the number of
1267 		 * null bytes.
1268 		 */
1269 		for (p = args->begin_argv; p < args->endp; ++p)
1270 			if (*p == '\0')
1271 				++args->argc;
1272 	} else {
1273 		/* No argument buffer provided. */
1274 		args->endp = args->begin_argv;
1275 	}
1276 
1277 	/* Create new file descriptor table. */
1278 	kfds = malloc(fdslen * sizeof(int), M_TEMP, M_WAITOK);
1279 	error = copyin(fds, kfds, fdslen * sizeof(int));
1280 	if (error != 0) {
1281 		free(kfds, M_TEMP);
1282 		goto err_exit;
1283 	}
1284 	error = fdcopy_remapped(ofdp, kfds, fdslen, &args->fdp);
1285 	free(kfds, M_TEMP);
1286 	if (error != 0)
1287 		goto err_exit;
1288 
1289 	return (0);
1290 err_exit:
1291 	exec_free_args(args);
1292 	return (error);
1293 }
1294 
1295 struct exec_args_kva {
1296 	vm_offset_t addr;
1297 	u_int gen;
1298 	SLIST_ENTRY(exec_args_kva) next;
1299 };
1300 
1301 DPCPU_DEFINE_STATIC(struct exec_args_kva *, exec_args_kva);
1302 
1303 static SLIST_HEAD(, exec_args_kva) exec_args_kva_freelist;
1304 static struct mtx exec_args_kva_mtx;
1305 static u_int exec_args_gen;
1306 
1307 static void
1308 exec_prealloc_args_kva(void *arg __unused)
1309 {
1310 	struct exec_args_kva *argkva;
1311 	u_int i;
1312 
1313 	SLIST_INIT(&exec_args_kva_freelist);
1314 	mtx_init(&exec_args_kva_mtx, "exec args kva", NULL, MTX_DEF);
1315 	for (i = 0; i < exec_map_entries; i++) {
1316 		argkva = malloc(sizeof(*argkva), M_PARGS, M_WAITOK);
1317 		argkva->addr = kmap_alloc_wait(exec_map, exec_map_entry_size);
1318 		argkva->gen = exec_args_gen;
1319 		SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1320 	}
1321 }
1322 SYSINIT(exec_args_kva, SI_SUB_EXEC, SI_ORDER_ANY, exec_prealloc_args_kva, NULL);
1323 
1324 static vm_offset_t
1325 exec_alloc_args_kva(void **cookie)
1326 {
1327 	struct exec_args_kva *argkva;
1328 
1329 	argkva = (void *)atomic_readandclear_ptr(
1330 	    (uintptr_t *)DPCPU_PTR(exec_args_kva));
1331 	if (argkva == NULL) {
1332 		mtx_lock(&exec_args_kva_mtx);
1333 		while ((argkva = SLIST_FIRST(&exec_args_kva_freelist)) == NULL)
1334 			(void)mtx_sleep(&exec_args_kva_freelist,
1335 			    &exec_args_kva_mtx, 0, "execkva", 0);
1336 		SLIST_REMOVE_HEAD(&exec_args_kva_freelist, next);
1337 		mtx_unlock(&exec_args_kva_mtx);
1338 	}
1339 	kasan_mark((void *)argkva->addr, exec_map_entry_size,
1340 	    exec_map_entry_size, 0);
1341 	*(struct exec_args_kva **)cookie = argkva;
1342 	return (argkva->addr);
1343 }
1344 
1345 static void
1346 exec_release_args_kva(struct exec_args_kva *argkva, u_int gen)
1347 {
1348 	vm_offset_t base;
1349 
1350 	base = argkva->addr;
1351 	kasan_mark((void *)argkva->addr, 0, exec_map_entry_size,
1352 	    KASAN_EXEC_ARGS_FREED);
1353 	if (argkva->gen != gen) {
1354 		(void)vm_map_madvise(exec_map, base, base + exec_map_entry_size,
1355 		    MADV_FREE);
1356 		argkva->gen = gen;
1357 	}
1358 	if (!atomic_cmpset_ptr((uintptr_t *)DPCPU_PTR(exec_args_kva),
1359 	    (uintptr_t)NULL, (uintptr_t)argkva)) {
1360 		mtx_lock(&exec_args_kva_mtx);
1361 		SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1362 		wakeup_one(&exec_args_kva_freelist);
1363 		mtx_unlock(&exec_args_kva_mtx);
1364 	}
1365 }
1366 
1367 static void
1368 exec_free_args_kva(void *cookie)
1369 {
1370 
1371 	exec_release_args_kva(cookie, exec_args_gen);
1372 }
1373 
1374 static void
1375 exec_args_kva_lowmem(void *arg __unused)
1376 {
1377 	SLIST_HEAD(, exec_args_kva) head;
1378 	struct exec_args_kva *argkva;
1379 	u_int gen;
1380 	int i;
1381 
1382 	gen = atomic_fetchadd_int(&exec_args_gen, 1) + 1;
1383 
1384 	/*
1385 	 * Force an madvise of each KVA range. Any currently allocated ranges
1386 	 * will have MADV_FREE applied once they are freed.
1387 	 */
1388 	SLIST_INIT(&head);
1389 	mtx_lock(&exec_args_kva_mtx);
1390 	SLIST_SWAP(&head, &exec_args_kva_freelist, exec_args_kva);
1391 	mtx_unlock(&exec_args_kva_mtx);
1392 	while ((argkva = SLIST_FIRST(&head)) != NULL) {
1393 		SLIST_REMOVE_HEAD(&head, next);
1394 		exec_release_args_kva(argkva, gen);
1395 	}
1396 
1397 	CPU_FOREACH(i) {
1398 		argkva = (void *)atomic_readandclear_ptr(
1399 		    (uintptr_t *)DPCPU_ID_PTR(i, exec_args_kva));
1400 		if (argkva != NULL)
1401 			exec_release_args_kva(argkva, gen);
1402 	}
1403 }
1404 EVENTHANDLER_DEFINE(vm_lowmem, exec_args_kva_lowmem, NULL,
1405     EVENTHANDLER_PRI_ANY);
1406 
1407 /*
1408  * Allocate temporary demand-paged, zero-filled memory for the file name,
1409  * argument, and environment strings.
1410  */
1411 int
1412 exec_alloc_args(struct image_args *args)
1413 {
1414 
1415 	args->buf = (char *)exec_alloc_args_kva(&args->bufkva);
1416 	return (0);
1417 }
1418 
1419 void
1420 exec_free_args(struct image_args *args)
1421 {
1422 
1423 	if (args->buf != NULL) {
1424 		exec_free_args_kva(args->bufkva);
1425 		args->buf = NULL;
1426 	}
1427 	if (args->fname_buf != NULL) {
1428 		free(args->fname_buf, M_TEMP);
1429 		args->fname_buf = NULL;
1430 	}
1431 	if (args->fdp != NULL)
1432 		fdescfree_remapped(args->fdp);
1433 }
1434 
1435 /*
1436  * A set to functions to fill struct image args.
1437  *
1438  * NOTE: exec_args_add_fname() must be called (possibly with a NULL
1439  * fname) before the other functions.  All exec_args_add_arg() calls must
1440  * be made before any exec_args_add_env() calls.  exec_args_adjust_args()
1441  * may be called any time after exec_args_add_fname().
1442  *
1443  * exec_args_add_fname() - install path to be executed
1444  * exec_args_add_arg() - append an argument string
1445  * exec_args_add_env() - append an env string
1446  * exec_args_adjust_args() - adjust location of the argument list to
1447  *                           allow new arguments to be prepended
1448  */
1449 int
1450 exec_args_add_fname(struct image_args *args, const char *fname,
1451     enum uio_seg segflg)
1452 {
1453 	int error;
1454 	size_t length;
1455 
1456 	KASSERT(args->fname == NULL, ("fname already appended"));
1457 	KASSERT(args->endp == NULL, ("already appending to args"));
1458 
1459 	if (fname != NULL) {
1460 		args->fname = args->buf;
1461 		error = segflg == UIO_SYSSPACE ?
1462 		    copystr(fname, args->fname, PATH_MAX, &length) :
1463 		    copyinstr(fname, args->fname, PATH_MAX, &length);
1464 		if (error != 0)
1465 			return (error == ENAMETOOLONG ? E2BIG : error);
1466 	} else
1467 		length = 0;
1468 
1469 	/* Set up for _arg_*()/_env_*() */
1470 	args->endp = args->buf + length;
1471 	/* begin_argv must be set and kept updated */
1472 	args->begin_argv = args->endp;
1473 	KASSERT(exec_map_entry_size - length >= ARG_MAX,
1474 	    ("too little space remaining for arguments %zu < %zu",
1475 	    exec_map_entry_size - length, (size_t)ARG_MAX));
1476 	args->stringspace = ARG_MAX;
1477 
1478 	return (0);
1479 }
1480 
1481 static int
1482 exec_args_add_str(struct image_args *args, const char *str,
1483     enum uio_seg segflg, int *countp)
1484 {
1485 	int error;
1486 	size_t length;
1487 
1488 	KASSERT(args->endp != NULL, ("endp not initialized"));
1489 	KASSERT(args->begin_argv != NULL, ("begin_argp not initialized"));
1490 
1491 	error = (segflg == UIO_SYSSPACE) ?
1492 	    copystr(str, args->endp, args->stringspace, &length) :
1493 	    copyinstr(str, args->endp, args->stringspace, &length);
1494 	if (error != 0)
1495 		return (error == ENAMETOOLONG ? E2BIG : error);
1496 	args->stringspace -= length;
1497 	args->endp += length;
1498 	(*countp)++;
1499 
1500 	return (0);
1501 }
1502 
1503 int
1504 exec_args_add_arg(struct image_args *args, const char *argp,
1505     enum uio_seg segflg)
1506 {
1507 
1508 	KASSERT(args->envc == 0, ("appending args after env"));
1509 
1510 	return (exec_args_add_str(args, argp, segflg, &args->argc));
1511 }
1512 
1513 int
1514 exec_args_add_env(struct image_args *args, const char *envp,
1515     enum uio_seg segflg)
1516 {
1517 
1518 	if (args->envc == 0)
1519 		args->begin_envv = args->endp;
1520 
1521 	return (exec_args_add_str(args, envp, segflg, &args->envc));
1522 }
1523 
1524 int
1525 exec_args_adjust_args(struct image_args *args, size_t consume, ssize_t extend)
1526 {
1527 	ssize_t offset;
1528 
1529 	KASSERT(args->endp != NULL, ("endp not initialized"));
1530 	KASSERT(args->begin_argv != NULL, ("begin_argp not initialized"));
1531 
1532 	offset = extend - consume;
1533 	if (args->stringspace < offset)
1534 		return (E2BIG);
1535 	memmove(args->begin_argv + extend, args->begin_argv + consume,
1536 	    args->endp - args->begin_argv + consume);
1537 	if (args->envc > 0)
1538 		args->begin_envv += offset;
1539 	args->endp += offset;
1540 	args->stringspace -= offset;
1541 	return (0);
1542 }
1543 
1544 char *
1545 exec_args_get_begin_envv(struct image_args *args)
1546 {
1547 
1548 	KASSERT(args->endp != NULL, ("endp not initialized"));
1549 
1550 	if (args->envc > 0)
1551 		return (args->begin_envv);
1552 	return (args->endp);
1553 }
1554 
1555 void
1556 exec_stackgap(struct image_params *imgp, uintptr_t *dp)
1557 {
1558 	if (imgp->sysent->sv_stackgap == NULL ||
1559 	    (imgp->proc->p_fctl0 & (NT_FREEBSD_FCTL_ASLR_DISABLE |
1560 	    NT_FREEBSD_FCTL_ASG_DISABLE)) != 0 ||
1561 	    (imgp->map_flags & MAP_ASLR) == 0)
1562 		return;
1563 	imgp->sysent->sv_stackgap(imgp, dp);
1564 }
1565 
1566 /*
1567  * Copy strings out to the new process address space, constructing new arg
1568  * and env vector tables. Return a pointer to the base so that it can be used
1569  * as the initial stack pointer.
1570  */
1571 int
1572 exec_copyout_strings(struct image_params *imgp, uintptr_t *stack_base)
1573 {
1574 	int argc, envc;
1575 	char **vectp;
1576 	char *stringp;
1577 	uintptr_t destp, ustringp;
1578 	struct ps_strings *arginfo;
1579 	struct proc *p;
1580 	size_t execpath_len;
1581 	int error, szsigcode, szps;
1582 	char canary[sizeof(long) * 8];
1583 
1584 	szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1585 	/*
1586 	 * Calculate string base and vector table pointers.
1587 	 * Also deal with signal trampoline code for this exec type.
1588 	 */
1589 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
1590 		execpath_len = strlen(imgp->execpath) + 1;
1591 	else
1592 		execpath_len = 0;
1593 	p = imgp->proc;
1594 	szsigcode = 0;
1595 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1596 	imgp->ps_strings = arginfo;
1597 	if (p->p_sysent->sv_sigcode_base == 0) {
1598 		if (p->p_sysent->sv_szsigcode != NULL)
1599 			szsigcode = *(p->p_sysent->sv_szsigcode);
1600 	}
1601 	destp =	(uintptr_t)arginfo;
1602 
1603 	/*
1604 	 * install sigcode
1605 	 */
1606 	if (szsigcode != 0) {
1607 		destp -= szsigcode;
1608 		destp = rounddown2(destp, sizeof(void *));
1609 		error = copyout(p->p_sysent->sv_sigcode, (void *)destp,
1610 		    szsigcode);
1611 		if (error != 0)
1612 			return (error);
1613 	}
1614 
1615 	/*
1616 	 * Copy the image path for the rtld.
1617 	 */
1618 	if (execpath_len != 0) {
1619 		destp -= execpath_len;
1620 		destp = rounddown2(destp, sizeof(void *));
1621 		imgp->execpathp = (void *)destp;
1622 		error = copyout(imgp->execpath, imgp->execpathp, execpath_len);
1623 		if (error != 0)
1624 			return (error);
1625 	}
1626 
1627 	/*
1628 	 * Prepare the canary for SSP.
1629 	 */
1630 	arc4rand(canary, sizeof(canary), 0);
1631 	destp -= sizeof(canary);
1632 	imgp->canary = (void *)destp;
1633 	error = copyout(canary, imgp->canary, sizeof(canary));
1634 	if (error != 0)
1635 		return (error);
1636 	imgp->canarylen = sizeof(canary);
1637 
1638 	/*
1639 	 * Prepare the pagesizes array.
1640 	 */
1641 	destp -= szps;
1642 	destp = rounddown2(destp, sizeof(void *));
1643 	imgp->pagesizes = (void *)destp;
1644 	error = copyout(pagesizes, imgp->pagesizes, szps);
1645 	if (error != 0)
1646 		return (error);
1647 	imgp->pagesizeslen = szps;
1648 
1649 	/*
1650 	 * Allocate room for the argument and environment strings.
1651 	 */
1652 	destp -= ARG_MAX - imgp->args->stringspace;
1653 	destp = rounddown2(destp, sizeof(void *));
1654 	ustringp = destp;
1655 
1656 	exec_stackgap(imgp, &destp);
1657 
1658 	if (imgp->auxargs) {
1659 		/*
1660 		 * Allocate room on the stack for the ELF auxargs
1661 		 * array.  It has up to AT_COUNT entries.
1662 		 */
1663 		destp -= AT_COUNT * sizeof(Elf_Auxinfo);
1664 		destp = rounddown2(destp, sizeof(void *));
1665 	}
1666 
1667 	vectp = (char **)destp;
1668 
1669 	/*
1670 	 * Allocate room for the argv[] and env vectors including the
1671 	 * terminating NULL pointers.
1672 	 */
1673 	vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
1674 
1675 	/*
1676 	 * vectp also becomes our initial stack base
1677 	 */
1678 	*stack_base = (uintptr_t)vectp;
1679 
1680 	stringp = imgp->args->begin_argv;
1681 	argc = imgp->args->argc;
1682 	envc = imgp->args->envc;
1683 
1684 	/*
1685 	 * Copy out strings - arguments and environment.
1686 	 */
1687 	error = copyout(stringp, (void *)ustringp,
1688 	    ARG_MAX - imgp->args->stringspace);
1689 	if (error != 0)
1690 		return (error);
1691 
1692 	/*
1693 	 * Fill in "ps_strings" struct for ps, w, etc.
1694 	 */
1695 	imgp->argv = vectp;
1696 	if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 ||
1697 	    suword32(&arginfo->ps_nargvstr, argc) != 0)
1698 		return (EFAULT);
1699 
1700 	/*
1701 	 * Fill in argument portion of vector table.
1702 	 */
1703 	for (; argc > 0; --argc) {
1704 		if (suword(vectp++, ustringp) != 0)
1705 			return (EFAULT);
1706 		while (*stringp++ != 0)
1707 			ustringp++;
1708 		ustringp++;
1709 	}
1710 
1711 	/* a null vector table pointer separates the argp's from the envp's */
1712 	if (suword(vectp++, 0) != 0)
1713 		return (EFAULT);
1714 
1715 	imgp->envv = vectp;
1716 	if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 ||
1717 	    suword32(&arginfo->ps_nenvstr, envc) != 0)
1718 		return (EFAULT);
1719 
1720 	/*
1721 	 * Fill in environment portion of vector table.
1722 	 */
1723 	for (; envc > 0; --envc) {
1724 		if (suword(vectp++, ustringp) != 0)
1725 			return (EFAULT);
1726 		while (*stringp++ != 0)
1727 			ustringp++;
1728 		ustringp++;
1729 	}
1730 
1731 	/* end of vector table is a null pointer */
1732 	if (suword(vectp, 0) != 0)
1733 		return (EFAULT);
1734 
1735 	if (imgp->auxargs) {
1736 		vectp++;
1737 		error = imgp->sysent->sv_copyout_auxargs(imgp,
1738 		    (uintptr_t)vectp);
1739 		if (error != 0)
1740 			return (error);
1741 	}
1742 
1743 	return (0);
1744 }
1745 
1746 /*
1747  * Check permissions of file to execute.
1748  *	Called with imgp->vp locked.
1749  *	Return 0 for success or error code on failure.
1750  */
1751 int
1752 exec_check_permissions(struct image_params *imgp)
1753 {
1754 	struct vnode *vp = imgp->vp;
1755 	struct vattr *attr = imgp->attr;
1756 	struct thread *td;
1757 	int error;
1758 
1759 	td = curthread;
1760 
1761 	/* Get file attributes */
1762 	error = VOP_GETATTR(vp, attr, td->td_ucred);
1763 	if (error)
1764 		return (error);
1765 
1766 #ifdef MAC
1767 	error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1768 	if (error)
1769 		return (error);
1770 #endif
1771 
1772 	/*
1773 	 * 1) Check if file execution is disabled for the filesystem that
1774 	 *    this file resides on.
1775 	 * 2) Ensure that at least one execute bit is on. Otherwise, a
1776 	 *    privileged user will always succeed, and we don't want this
1777 	 *    to happen unless the file really is executable.
1778 	 * 3) Ensure that the file is a regular file.
1779 	 */
1780 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1781 	    (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1782 	    (attr->va_type != VREG))
1783 		return (EACCES);
1784 
1785 	/*
1786 	 * Zero length files can't be exec'd
1787 	 */
1788 	if (attr->va_size == 0)
1789 		return (ENOEXEC);
1790 
1791 	/*
1792 	 *  Check for execute permission to file based on current credentials.
1793 	 */
1794 	error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1795 	if (error)
1796 		return (error);
1797 
1798 	/*
1799 	 * Check number of open-for-writes on the file and deny execution
1800 	 * if there are any.
1801 	 *
1802 	 * Add a text reference now so no one can write to the
1803 	 * executable while we're activating it.
1804 	 *
1805 	 * Remember if this was set before and unset it in case this is not
1806 	 * actually an executable image.
1807 	 */
1808 	error = VOP_SET_TEXT(vp);
1809 	if (error != 0)
1810 		return (error);
1811 	imgp->textset = true;
1812 
1813 	/*
1814 	 * Call filesystem specific open routine (which does nothing in the
1815 	 * general case).
1816 	 */
1817 	error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1818 	if (error == 0)
1819 		imgp->opened = 1;
1820 	return (error);
1821 }
1822 
1823 /*
1824  * Exec handler registration
1825  */
1826 int
1827 exec_register(const struct execsw *execsw_arg)
1828 {
1829 	const struct execsw **es, **xs, **newexecsw;
1830 	u_int count = 2;	/* New slot and trailing NULL */
1831 
1832 	if (execsw)
1833 		for (es = execsw; *es; es++)
1834 			count++;
1835 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1836 	xs = newexecsw;
1837 	if (execsw)
1838 		for (es = execsw; *es; es++)
1839 			*xs++ = *es;
1840 	*xs++ = execsw_arg;
1841 	*xs = NULL;
1842 	if (execsw)
1843 		free(execsw, M_TEMP);
1844 	execsw = newexecsw;
1845 	return (0);
1846 }
1847 
1848 int
1849 exec_unregister(const struct execsw *execsw_arg)
1850 {
1851 	const struct execsw **es, **xs, **newexecsw;
1852 	int count = 1;
1853 
1854 	if (execsw == NULL)
1855 		panic("unregister with no handlers left?\n");
1856 
1857 	for (es = execsw; *es; es++) {
1858 		if (*es == execsw_arg)
1859 			break;
1860 	}
1861 	if (*es == NULL)
1862 		return (ENOENT);
1863 	for (es = execsw; *es; es++)
1864 		if (*es != execsw_arg)
1865 			count++;
1866 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1867 	xs = newexecsw;
1868 	for (es = execsw; *es; es++)
1869 		if (*es != execsw_arg)
1870 			*xs++ = *es;
1871 	*xs = NULL;
1872 	if (execsw)
1873 		free(execsw, M_TEMP);
1874 	execsw = newexecsw;
1875 	return (0);
1876 }
1877 
1878 /*
1879  * Write out a core segment to the compression stream.
1880  */
1881 static int
1882 compress_chunk(struct coredump_params *cp, char *base, char *buf, u_int len)
1883 {
1884 	u_int chunk_len;
1885 	int error;
1886 
1887 	while (len > 0) {
1888 		chunk_len = MIN(len, CORE_BUF_SIZE);
1889 
1890 		/*
1891 		 * We can get EFAULT error here.
1892 		 * In that case zero out the current chunk of the segment.
1893 		 */
1894 		error = copyin(base, buf, chunk_len);
1895 		if (error != 0)
1896 			bzero(buf, chunk_len);
1897 		error = compressor_write(cp->comp, buf, chunk_len);
1898 		if (error != 0)
1899 			break;
1900 		base += chunk_len;
1901 		len -= chunk_len;
1902 	}
1903 	return (error);
1904 }
1905 
1906 int
1907 core_write(struct coredump_params *cp, const void *base, size_t len,
1908     off_t offset, enum uio_seg seg, size_t *resid)
1909 {
1910 
1911 	return (vn_rdwr_inchunks(UIO_WRITE, cp->vp, __DECONST(void *, base),
1912 	    len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1913 	    cp->active_cred, cp->file_cred, resid, cp->td));
1914 }
1915 
1916 int
1917 core_output(char *base, size_t len, off_t offset, struct coredump_params *cp,
1918     void *tmpbuf)
1919 {
1920 	vm_map_t map;
1921 	struct mount *mp;
1922 	size_t resid, runlen;
1923 	int error;
1924 	bool success;
1925 
1926 	KASSERT((uintptr_t)base % PAGE_SIZE == 0,
1927 	    ("%s: user address %p is not page-aligned", __func__, base));
1928 
1929 	if (cp->comp != NULL)
1930 		return (compress_chunk(cp, base, tmpbuf, len));
1931 
1932 	map = &cp->td->td_proc->p_vmspace->vm_map;
1933 	for (; len > 0; base += runlen, offset += runlen, len -= runlen) {
1934 		/*
1935 		 * Attempt to page in all virtual pages in the range.  If a
1936 		 * virtual page is not backed by the pager, it is represented as
1937 		 * a hole in the file.  This can occur with zero-filled
1938 		 * anonymous memory or truncated files, for example.
1939 		 */
1940 		for (runlen = 0; runlen < len; runlen += PAGE_SIZE) {
1941 			error = vm_fault(map, (uintptr_t)base + runlen,
1942 			    VM_PROT_READ, VM_FAULT_NOFILL, NULL);
1943 			if (runlen == 0)
1944 				success = error == KERN_SUCCESS;
1945 			else if ((error == KERN_SUCCESS) != success)
1946 				break;
1947 		}
1948 
1949 		if (success) {
1950 			error = core_write(cp, base, runlen, offset,
1951 			    UIO_USERSPACE, &resid);
1952 			if (error != 0) {
1953 				if (error != EFAULT)
1954 					break;
1955 
1956 				/*
1957 				 * EFAULT may be returned if the user mapping
1958 				 * could not be accessed, e.g., because a mapped
1959 				 * file has been truncated.  Skip the page if no
1960 				 * progress was made, to protect against a
1961 				 * hypothetical scenario where vm_fault() was
1962 				 * successful but core_write() returns EFAULT
1963 				 * anyway.
1964 				 */
1965 				runlen -= resid;
1966 				if (runlen == 0) {
1967 					success = false;
1968 					runlen = PAGE_SIZE;
1969 				}
1970 			}
1971 		}
1972 		if (!success) {
1973 			error = vn_start_write(cp->vp, &mp, V_WAIT);
1974 			if (error != 0)
1975 				break;
1976 			vn_lock(cp->vp, LK_EXCLUSIVE | LK_RETRY);
1977 			error = vn_truncate_locked(cp->vp, offset + runlen,
1978 			    false, cp->td->td_ucred);
1979 			VOP_UNLOCK(cp->vp);
1980 			vn_finished_write(mp);
1981 			if (error != 0)
1982 				break;
1983 		}
1984 	}
1985 	return (error);
1986 }
1987 
1988 /*
1989  * Drain into a core file.
1990  */
1991 int
1992 sbuf_drain_core_output(void *arg, const char *data, int len)
1993 {
1994 	struct coredump_params *cp;
1995 	struct proc *p;
1996 	int error, locked;
1997 
1998 	cp = arg;
1999 	p = cp->td->td_proc;
2000 
2001 	/*
2002 	 * Some kern_proc out routines that print to this sbuf may
2003 	 * call us with the process lock held. Draining with the
2004 	 * non-sleepable lock held is unsafe. The lock is needed for
2005 	 * those routines when dumping a live process. In our case we
2006 	 * can safely release the lock before draining and acquire
2007 	 * again after.
2008 	 */
2009 	locked = PROC_LOCKED(p);
2010 	if (locked)
2011 		PROC_UNLOCK(p);
2012 	if (cp->comp != NULL)
2013 		error = compressor_write(cp->comp, __DECONST(char *, data), len);
2014 	else
2015 		error = core_write(cp, __DECONST(void *, data), len, cp->offset,
2016 		    UIO_SYSSPACE, NULL);
2017 	if (locked)
2018 		PROC_LOCK(p);
2019 	if (error != 0)
2020 		return (-error);
2021 	cp->offset += len;
2022 	return (len);
2023 }
2024