xref: /freebsd/sys/kern/kern_exec.c (revision 21b492ed51aa6ff8008a8aa83333b1de30288a15)
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/reg.h>
60 #include <sys/resourcevar.h>
61 #include <sys/rwlock.h>
62 #include <sys/sched.h>
63 #include <sys/sdt.h>
64 #include <sys/sf_buf.h>
65 #include <sys/shm.h>
66 #include <sys/signalvar.h>
67 #include <sys/smp.h>
68 #include <sys/stat.h>
69 #include <sys/syscallsubr.h>
70 #include <sys/sysctl.h>
71 #include <sys/sysent.h>
72 #include <sys/sysproto.h>
73 #include <sys/timers.h>
74 #include <sys/umtxvar.h>
75 #include <sys/vnode.h>
76 #include <sys/wait.h>
77 #ifdef KTRACE
78 #include <sys/ktrace.h>
79 #endif
80 
81 #include <vm/vm.h>
82 #include <vm/vm_param.h>
83 #include <vm/pmap.h>
84 #include <vm/vm_page.h>
85 #include <vm/vm_map.h>
86 #include <vm/vm_kern.h>
87 #include <vm/vm_extern.h>
88 #include <vm/vm_object.h>
89 #include <vm/vm_pager.h>
90 
91 #ifdef	HWPMC_HOOKS
92 #include <sys/pmckern.h>
93 #endif
94 
95 #include <security/audit/audit.h>
96 #include <security/mac/mac_framework.h>
97 
98 #ifdef KDTRACE_HOOKS
99 #include <sys/dtrace_bsd.h>
100 dtrace_execexit_func_t	dtrace_fasttrap_exec;
101 #endif
102 
103 SDT_PROVIDER_DECLARE(proc);
104 SDT_PROBE_DEFINE1(proc, , , exec, "char *");
105 SDT_PROBE_DEFINE1(proc, , , exec__failure, "int");
106 SDT_PROBE_DEFINE1(proc, , , exec__success, "char *");
107 
108 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
109 
110 int coredump_pack_fileinfo = 1;
111 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_fileinfo, CTLFLAG_RWTUN,
112     &coredump_pack_fileinfo, 0,
113     "Enable file path packing in 'procstat -f' coredump notes");
114 
115 int coredump_pack_vmmapinfo = 1;
116 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_vmmapinfo, CTLFLAG_RWTUN,
117     &coredump_pack_vmmapinfo, 0,
118     "Enable file path packing in 'procstat -v' coredump notes");
119 
120 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
121 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
122 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
123 static int do_execve(struct thread *td, struct image_args *args,
124     struct mac *mac_p, struct vmspace *oldvmspace);
125 
126 /* XXX This should be vm_size_t. */
127 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD|
128     CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_ps_strings, "LU",
129     "Location of process' ps_strings structure");
130 
131 /* XXX This should be vm_size_t. */
132 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD|
133     CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_usrstack, "LU",
134     "Top of process stack");
135 
136 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD|CTLFLAG_MPSAFE,
137     NULL, 0, sysctl_kern_stackprot, "I",
138     "Stack memory permissions");
139 
140 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
141 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
142     &ps_arg_cache_limit, 0,
143     "Process' command line characters cache limit");
144 
145 static int disallow_high_osrel;
146 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW,
147     &disallow_high_osrel, 0,
148     "Disallow execution of binaries built for higher version of the world");
149 
150 static int map_at_zero = 0;
151 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RWTUN, &map_at_zero, 0,
152     "Permit processes to map an object at virtual address 0.");
153 
154 static int
155 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
156 {
157 	struct proc *p;
158 	int error;
159 
160 	p = curproc;
161 #ifdef SCTL_MASK32
162 	if (req->flags & SCTL_MASK32) {
163 		unsigned int val;
164 		val = (unsigned int)p->p_sysent->sv_psstrings;
165 		error = SYSCTL_OUT(req, &val, sizeof(val));
166 	} else
167 #endif
168 		error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
169 		   sizeof(p->p_sysent->sv_psstrings));
170 	return error;
171 }
172 
173 static int
174 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
175 {
176 	struct proc *p;
177 	int error;
178 
179 	p = curproc;
180 #ifdef SCTL_MASK32
181 	if (req->flags & SCTL_MASK32) {
182 		unsigned int val;
183 		val = (unsigned int)p->p_sysent->sv_usrstack;
184 		error = SYSCTL_OUT(req, &val, sizeof(val));
185 	} else
186 #endif
187 		error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
188 		    sizeof(p->p_sysent->sv_usrstack));
189 	return error;
190 }
191 
192 static int
193 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
194 {
195 	struct proc *p;
196 
197 	p = curproc;
198 	return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
199 	    sizeof(p->p_sysent->sv_stackprot)));
200 }
201 
202 /*
203  * Each of the items is a pointer to a `const struct execsw', hence the
204  * double pointer here.
205  */
206 static const struct execsw **execsw;
207 
208 #ifndef _SYS_SYSPROTO_H_
209 struct execve_args {
210 	char    *fname;
211 	char    **argv;
212 	char    **envv;
213 };
214 #endif
215 
216 int
217 sys_execve(struct thread *td, struct execve_args *uap)
218 {
219 	struct image_args args;
220 	struct vmspace *oldvmspace;
221 	int error;
222 
223 	error = pre_execve(td, &oldvmspace);
224 	if (error != 0)
225 		return (error);
226 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
227 	    uap->argv, uap->envv);
228 	if (error == 0)
229 		error = kern_execve(td, &args, NULL, oldvmspace);
230 	post_execve(td, error, oldvmspace);
231 	AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
232 	return (error);
233 }
234 
235 #ifndef _SYS_SYSPROTO_H_
236 struct fexecve_args {
237 	int	fd;
238 	char	**argv;
239 	char	**envv;
240 };
241 #endif
242 int
243 sys_fexecve(struct thread *td, struct fexecve_args *uap)
244 {
245 	struct image_args args;
246 	struct vmspace *oldvmspace;
247 	int error;
248 
249 	error = pre_execve(td, &oldvmspace);
250 	if (error != 0)
251 		return (error);
252 	error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
253 	    uap->argv, uap->envv);
254 	if (error == 0) {
255 		args.fd = uap->fd;
256 		error = kern_execve(td, &args, NULL, oldvmspace);
257 	}
258 	post_execve(td, error, oldvmspace);
259 	AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
260 	return (error);
261 }
262 
263 #ifndef _SYS_SYSPROTO_H_
264 struct __mac_execve_args {
265 	char	*fname;
266 	char	**argv;
267 	char	**envv;
268 	struct mac	*mac_p;
269 };
270 #endif
271 
272 int
273 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap)
274 {
275 #ifdef MAC
276 	struct image_args args;
277 	struct vmspace *oldvmspace;
278 	int error;
279 
280 	error = pre_execve(td, &oldvmspace);
281 	if (error != 0)
282 		return (error);
283 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
284 	    uap->argv, uap->envv);
285 	if (error == 0)
286 		error = kern_execve(td, &args, uap->mac_p, oldvmspace);
287 	post_execve(td, error, oldvmspace);
288 	AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
289 	return (error);
290 #else
291 	return (ENOSYS);
292 #endif
293 }
294 
295 int
296 pre_execve(struct thread *td, struct vmspace **oldvmspace)
297 {
298 	struct proc *p;
299 	int error;
300 
301 	KASSERT(td == curthread, ("non-current thread %p", td));
302 	error = 0;
303 	p = td->td_proc;
304 	if ((p->p_flag & P_HADTHREADS) != 0) {
305 		PROC_LOCK(p);
306 		if (thread_single(p, SINGLE_BOUNDARY) != 0)
307 			error = ERESTART;
308 		PROC_UNLOCK(p);
309 	}
310 	KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0,
311 	    ("nested execve"));
312 	*oldvmspace = p->p_vmspace;
313 	return (error);
314 }
315 
316 void
317 post_execve(struct thread *td, int error, struct vmspace *oldvmspace)
318 {
319 	struct proc *p;
320 
321 	KASSERT(td == curthread, ("non-current thread %p", td));
322 	p = td->td_proc;
323 	if ((p->p_flag & P_HADTHREADS) != 0) {
324 		PROC_LOCK(p);
325 		/*
326 		 * If success, we upgrade to SINGLE_EXIT state to
327 		 * force other threads to suicide.
328 		 */
329 		if (error == EJUSTRETURN)
330 			thread_single(p, SINGLE_EXIT);
331 		else
332 			thread_single_end(p, SINGLE_BOUNDARY);
333 		PROC_UNLOCK(p);
334 	}
335 	exec_cleanup(td, oldvmspace);
336 }
337 
338 /*
339  * kern_execve() has the astonishing property of not always returning to
340  * the caller.  If sufficiently bad things happen during the call to
341  * do_execve(), it can end up calling exit1(); as a result, callers must
342  * avoid doing anything which they might need to undo (e.g., allocating
343  * memory).
344  */
345 int
346 kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p,
347     struct vmspace *oldvmspace)
348 {
349 
350 	AUDIT_ARG_ARGV(args->begin_argv, args->argc,
351 	    exec_args_get_begin_envv(args) - args->begin_argv);
352 	AUDIT_ARG_ENVV(exec_args_get_begin_envv(args), args->envc,
353 	    args->endp - exec_args_get_begin_envv(args));
354 	return (do_execve(td, args, mac_p, oldvmspace));
355 }
356 
357 static void
358 execve_nosetid(struct image_params *imgp)
359 {
360 	imgp->credential_setid = false;
361 	if (imgp->newcred != NULL) {
362 		crfree(imgp->newcred);
363 		imgp->newcred = NULL;
364 	}
365 }
366 
367 /*
368  * In-kernel implementation of execve().  All arguments are assumed to be
369  * userspace pointers from the passed thread.
370  */
371 static int
372 do_execve(struct thread *td, struct image_args *args, struct mac *mac_p,
373     struct vmspace *oldvmspace)
374 {
375 	struct proc *p = td->td_proc;
376 	struct nameidata nd;
377 	struct ucred *oldcred;
378 	struct uidinfo *euip = NULL;
379 	uintptr_t stack_base;
380 	struct image_params image_params, *imgp;
381 	struct vattr attr;
382 	int (*img_first)(struct image_params *);
383 	struct pargs *oldargs = NULL, *newargs = NULL;
384 	struct sigacts *oldsigacts = NULL, *newsigacts = NULL;
385 #ifdef KTRACE
386 	struct ktr_io_params *kiop;
387 #endif
388 	struct vnode *oldtextvp = NULL, *newtextvp;
389 	int credential_changing;
390 #ifdef MAC
391 	struct label *interpvplabel = NULL;
392 	int will_transition;
393 #endif
394 #ifdef HWPMC_HOOKS
395 	struct pmckern_procexec pe;
396 #endif
397 	int error, i, orig_osrel;
398 	uint32_t orig_fctl0;
399 	Elf_Brandinfo *orig_brandinfo;
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 	orig_brandinfo = p->p_elf_brandinfo;
431 
432 #ifdef MAC
433 	error = mac_execve_enter(imgp, mac_p);
434 	if (error)
435 		goto exec_fail;
436 #endif
437 
438 	/*
439 	 * Translate the file name. namei() returns a vnode pointer
440 	 *	in ni_vp among other things.
441 	 *
442 	 * XXXAUDIT: It would be desirable to also audit the name of the
443 	 * interpreter if this is an interpreted binary.
444 	 */
445 	if (args->fname != NULL) {
446 		NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW |
447 		    SAVENAME | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
448 	}
449 
450 	SDT_PROBE1(proc, , , exec, args->fname);
451 
452 interpret:
453 	if (args->fname != NULL) {
454 #ifdef CAPABILITY_MODE
455 		/*
456 		 * While capability mode can't reach this point via direct
457 		 * path arguments to execve(), we also don't allow
458 		 * interpreters to be used in capability mode (for now).
459 		 * Catch indirect lookups and return a permissions error.
460 		 */
461 		if (IN_CAPABILITY_MODE(td)) {
462 			error = ECAPMODE;
463 			goto exec_fail;
464 		}
465 #endif
466 		error = namei(&nd);
467 		if (error)
468 			goto exec_fail;
469 
470 		newtextvp = nd.ni_vp;
471 		imgp->vp = newtextvp;
472 	} else {
473 		AUDIT_ARG_FD(args->fd);
474 		/*
475 		 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
476 		 */
477 		error = fgetvp_exec(td, args->fd, &cap_fexecve_rights, &newtextvp);
478 		if (error)
479 			goto exec_fail;
480 		vn_lock(newtextvp, LK_SHARED | LK_RETRY);
481 		AUDIT_ARG_VNODE1(newtextvp);
482 		imgp->vp = newtextvp;
483 	}
484 
485 	/*
486 	 * Check file permissions.  Also 'opens' file and sets its vnode to
487 	 * text mode.
488 	 */
489 	error = exec_check_permissions(imgp);
490 	if (error)
491 		goto exec_fail_dealloc;
492 
493 	imgp->object = imgp->vp->v_object;
494 	if (imgp->object != NULL)
495 		vm_object_reference(imgp->object);
496 
497 	error = exec_map_first_page(imgp);
498 	if (error)
499 		goto exec_fail_dealloc;
500 
501 	imgp->proc->p_osrel = 0;
502 	imgp->proc->p_fctl0 = 0;
503 	imgp->proc->p_elf_brandinfo = NULL;
504 
505 	/*
506 	 * Implement image setuid/setgid.
507 	 *
508 	 * Determine new credentials before attempting image activators
509 	 * so that it can be used by process_exec handlers to determine
510 	 * credential/setid changes.
511 	 *
512 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
513 	 * the process is being traced.
514 	 *
515 	 * We disable setuid/setgid/etc in capability mode on the basis
516 	 * that most setugid applications are not written with that
517 	 * environment in mind, and will therefore almost certainly operate
518 	 * incorrectly. In principle there's no reason that setugid
519 	 * applications might not be useful in capability mode, so we may want
520 	 * to reconsider this conservative design choice in the future.
521 	 *
522 	 * XXXMAC: For the time being, use NOSUID to also prohibit
523 	 * transitions on the file system.
524 	 */
525 	credential_changing = 0;
526 	credential_changing |= (attr.va_mode & S_ISUID) &&
527 	    oldcred->cr_uid != attr.va_uid;
528 	credential_changing |= (attr.va_mode & S_ISGID) &&
529 	    oldcred->cr_gid != attr.va_gid;
530 #ifdef MAC
531 	will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
532 	    interpvplabel, imgp);
533 	credential_changing |= will_transition;
534 #endif
535 
536 	/* Don't inherit PROC_PDEATHSIG_CTL value if setuid/setgid. */
537 	if (credential_changing)
538 		imgp->proc->p_pdeathsig = 0;
539 
540 	if (credential_changing &&
541 #ifdef CAPABILITY_MODE
542 	    ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
543 #endif
544 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
545 	    (p->p_flag & P_TRACED) == 0) {
546 		imgp->credential_setid = true;
547 		VOP_UNLOCK(imgp->vp);
548 		imgp->newcred = crdup(oldcred);
549 		if (attr.va_mode & S_ISUID) {
550 			euip = uifind(attr.va_uid);
551 			change_euid(imgp->newcred, euip);
552 		}
553 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
554 		if (attr.va_mode & S_ISGID)
555 			change_egid(imgp->newcred, attr.va_gid);
556 		/*
557 		 * Implement correct POSIX saved-id behavior.
558 		 *
559 		 * XXXMAC: Note that the current logic will save the
560 		 * uid and gid if a MAC domain transition occurs, even
561 		 * though maybe it shouldn't.
562 		 */
563 		change_svuid(imgp->newcred, imgp->newcred->cr_uid);
564 		change_svgid(imgp->newcred, imgp->newcred->cr_gid);
565 	} else {
566 		/*
567 		 * Implement correct POSIX saved-id behavior.
568 		 *
569 		 * XXX: It's not clear that the existing behavior is
570 		 * POSIX-compliant.  A number of sources indicate that the
571 		 * saved uid/gid should only be updated if the new ruid is
572 		 * not equal to the old ruid, or the new euid is not equal
573 		 * to the old euid and the new euid is not equal to the old
574 		 * ruid.  The FreeBSD code always updates the saved uid/gid.
575 		 * Also, this code uses the new (replaced) euid and egid as
576 		 * the source, which may or may not be the right ones to use.
577 		 */
578 		if (oldcred->cr_svuid != oldcred->cr_uid ||
579 		    oldcred->cr_svgid != oldcred->cr_gid) {
580 			VOP_UNLOCK(imgp->vp);
581 			imgp->newcred = crdup(oldcred);
582 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
583 			change_svuid(imgp->newcred, imgp->newcred->cr_uid);
584 			change_svgid(imgp->newcred, imgp->newcred->cr_gid);
585 		}
586 	}
587 	/* The new credentials are installed into the process later. */
588 
589 	/*
590 	 * Do the best to calculate the full path to the image file.
591 	 */
592 	if (args->fname != NULL && args->fname[0] == '/')
593 		imgp->execpath = args->fname;
594 	else {
595 		VOP_UNLOCK(imgp->vp);
596 		if (vn_fullpath(imgp->vp, &imgp->execpath, &imgp->freepath) != 0)
597 			imgp->execpath = args->fname;
598 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
599 	}
600 
601 	/*
602 	 *	If the current process has a special image activator it
603 	 *	wants to try first, call it.   For example, emulating shell
604 	 *	scripts differently.
605 	 */
606 	error = -1;
607 	if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
608 		error = img_first(imgp);
609 
610 	/*
611 	 *	Loop through the list of image activators, calling each one.
612 	 *	An activator returns -1 if there is no match, 0 on success,
613 	 *	and an error otherwise.
614 	 */
615 	for (i = 0; error == -1 && execsw[i]; ++i) {
616 		if (execsw[i]->ex_imgact == NULL ||
617 		    execsw[i]->ex_imgact == img_first) {
618 			continue;
619 		}
620 		error = (*execsw[i]->ex_imgact)(imgp);
621 	}
622 
623 	if (error) {
624 		if (error == -1)
625 			error = ENOEXEC;
626 		goto exec_fail_dealloc;
627 	}
628 
629 	/*
630 	 * Special interpreter operation, cleanup and loop up to try to
631 	 * activate the interpreter.
632 	 */
633 	if (imgp->interpreted) {
634 		exec_unmap_first_page(imgp);
635 		/*
636 		 * The text reference needs to be removed for scripts.
637 		 * There is a short period before we determine that
638 		 * something is a script where text reference is active.
639 		 * The vnode lock is held over this entire period
640 		 * so nothing should illegitimately be blocked.
641 		 */
642 		MPASS(imgp->textset);
643 		VOP_UNSET_TEXT_CHECKED(newtextvp);
644 		imgp->textset = false;
645 		/* free name buffer and old vnode */
646 		if (args->fname != NULL)
647 			NDFREE(&nd, NDF_ONLY_PNBUF);
648 #ifdef MAC
649 		mac_execve_interpreter_enter(newtextvp, &interpvplabel);
650 #endif
651 		if (imgp->opened) {
652 			VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td);
653 			imgp->opened = 0;
654 		}
655 		vput(newtextvp);
656 		vm_object_deallocate(imgp->object);
657 		imgp->object = NULL;
658 		execve_nosetid(imgp);
659 		imgp->execpath = NULL;
660 		free(imgp->freepath, M_TEMP);
661 		imgp->freepath = NULL;
662 		/* set new name to that of the interpreter */
663 		NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW |
664 		    SAVENAME, UIO_SYSSPACE, imgp->interpreter_name, td);
665 		args->fname = imgp->interpreter_name;
666 		goto interpret;
667 	}
668 
669 	/*
670 	 * NB: We unlock the vnode here because it is believed that none
671 	 * of the sv_copyout_strings/sv_fixup operations require the vnode.
672 	 */
673 	VOP_UNLOCK(imgp->vp);
674 
675 	if (disallow_high_osrel &&
676 	    P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
677 		error = ENOEXEC;
678 		uprintf("Osrel %d for image %s too high\n", p->p_osrel,
679 		    imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
680 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
681 		goto exec_fail_dealloc;
682 	}
683 
684 	/* ABI enforces the use of Capsicum. Switch into capabilities mode. */
685 	if (SV_PROC_FLAG(p, SV_CAPSICUM))
686 		sys_cap_enter(td, NULL);
687 
688 	/*
689 	 * Copy out strings (args and env) and initialize stack base.
690 	 */
691 	error = (*p->p_sysent->sv_copyout_strings)(imgp, &stack_base);
692 	if (error != 0) {
693 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
694 		goto exec_fail_dealloc;
695 	}
696 
697 	/*
698 	 * Stack setup.
699 	 */
700 	error = (*p->p_sysent->sv_fixup)(&stack_base, imgp);
701 	if (error != 0) {
702 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
703 		goto exec_fail_dealloc;
704 	}
705 
706 	if (args->fdp != NULL) {
707 		/* Install a brand new file descriptor table. */
708 		fdinstall_remapped(td, args->fdp);
709 		args->fdp = NULL;
710 	} else {
711 		/*
712 		 * Keep on using the existing file descriptor table. For
713 		 * security and other reasons, the file descriptor table
714 		 * cannot be shared after an exec.
715 		 */
716 		fdunshare(td);
717 		pdunshare(td);
718 		/* close files on exec */
719 		fdcloseexec(td);
720 	}
721 
722 	/*
723 	 * Malloc things before we need locks.
724 	 */
725 	i = exec_args_get_begin_envv(imgp->args) - imgp->args->begin_argv;
726 	/* Cache arguments if they fit inside our allowance */
727 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
728 		newargs = pargs_alloc(i);
729 		bcopy(imgp->args->begin_argv, newargs->ar_args, i);
730 	}
731 
732 	/*
733 	 * For security and other reasons, signal handlers cannot
734 	 * be shared after an exec. The new process gets a copy of the old
735 	 * handlers. In execsigs(), the new process will have its signals
736 	 * reset.
737 	 */
738 	if (sigacts_shared(p->p_sigacts)) {
739 		oldsigacts = p->p_sigacts;
740 		newsigacts = sigacts_alloc();
741 		sigacts_copy(newsigacts, oldsigacts);
742 	}
743 
744 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
745 
746 	PROC_LOCK(p);
747 	if (oldsigacts)
748 		p->p_sigacts = newsigacts;
749 	/* Stop profiling */
750 	stopprofclock(p);
751 
752 	/* reset caught signals */
753 	execsigs(p);
754 
755 	/* name this process - nameiexec(p, ndp) */
756 	bzero(p->p_comm, sizeof(p->p_comm));
757 	if (args->fname)
758 		bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
759 		    min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
760 	else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0)
761 		bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
762 	bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
763 #ifdef KTR
764 	sched_clear_tdname(td);
765 #endif
766 
767 	/*
768 	 * mark as execed, wakeup the process that vforked (if any) and tell
769 	 * it that it now has its own resources back
770 	 */
771 	p->p_flag |= P_EXEC;
772 	if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0)
773 		p->p_flag2 &= ~P2_NOTRACE;
774 	if ((p->p_flag2 & P2_STKGAP_DISABLE_EXEC) == 0)
775 		p->p_flag2 &= ~P2_STKGAP_DISABLE;
776 	if (p->p_flag & P_PPWAIT) {
777 		p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
778 		cv_broadcast(&p->p_pwait);
779 		/* STOPs are no longer ignored, arrange for AST */
780 		signotify(td);
781 	}
782 
783 	if ((imgp->sysent->sv_setid_allowed != NULL &&
784 	    !(*imgp->sysent->sv_setid_allowed)(td, imgp)) ||
785 	    (p->p_flag2 & P2_NO_NEW_PRIVS) != 0)
786 		execve_nosetid(imgp);
787 
788 	/*
789 	 * Implement image setuid/setgid installation.
790 	 */
791 	if (imgp->credential_setid) {
792 		/*
793 		 * Turn off syscall tracing for set-id programs, except for
794 		 * root.  Record any set-id flags first to make sure that
795 		 * we do not regain any tracing during a possible block.
796 		 */
797 		setsugid(p);
798 #ifdef KTRACE
799 		kiop = ktrprocexec(p);
800 #endif
801 		/*
802 		 * Close any file descriptors 0..2 that reference procfs,
803 		 * then make sure file descriptors 0..2 are in use.
804 		 *
805 		 * Both fdsetugidsafety() and fdcheckstd() may call functions
806 		 * taking sleepable locks, so temporarily drop our locks.
807 		 */
808 		PROC_UNLOCK(p);
809 		VOP_UNLOCK(imgp->vp);
810 		fdsetugidsafety(td);
811 		error = fdcheckstd(td);
812 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
813 		if (error != 0)
814 			goto exec_fail_dealloc;
815 		PROC_LOCK(p);
816 #ifdef MAC
817 		if (will_transition) {
818 			mac_vnode_execve_transition(oldcred, imgp->newcred,
819 			    imgp->vp, interpvplabel, imgp);
820 		}
821 #endif
822 	} else {
823 		if (oldcred->cr_uid == oldcred->cr_ruid &&
824 		    oldcred->cr_gid == oldcred->cr_rgid)
825 			p->p_flag &= ~P_SUGID;
826 	}
827 	/*
828 	 * Set the new credentials.
829 	 */
830 	if (imgp->newcred != NULL) {
831 		proc_set_cred(p, imgp->newcred);
832 		crfree(oldcred);
833 		oldcred = NULL;
834 	}
835 
836 	/*
837 	 * Store the vp for use in procfs.  This vnode was referenced by namei
838 	 * or fgetvp_exec.
839 	 */
840 	oldtextvp = p->p_textvp;
841 	p->p_textvp = newtextvp;
842 
843 #ifdef KDTRACE_HOOKS
844 	/*
845 	 * Tell the DTrace fasttrap provider about the exec if it
846 	 * has declared an interest.
847 	 */
848 	if (dtrace_fasttrap_exec)
849 		dtrace_fasttrap_exec(p);
850 #endif
851 
852 	/*
853 	 * Notify others that we exec'd, and clear the P_INEXEC flag
854 	 * as we're now a bona fide freshly-execed process.
855 	 */
856 	KNOTE_LOCKED(p->p_klist, NOTE_EXEC);
857 	p->p_flag &= ~P_INEXEC;
858 
859 	/* clear "fork but no exec" flag, as we _are_ execing */
860 	p->p_acflag &= ~AFORK;
861 
862 	/*
863 	 * Free any previous argument cache and replace it with
864 	 * the new argument cache, if any.
865 	 */
866 	oldargs = p->p_args;
867 	p->p_args = newargs;
868 	newargs = NULL;
869 
870 	PROC_UNLOCK(p);
871 
872 #ifdef	HWPMC_HOOKS
873 	/*
874 	 * Check if system-wide sampling is in effect or if the
875 	 * current process is using PMCs.  If so, do exec() time
876 	 * processing.  This processing needs to happen AFTER the
877 	 * P_INEXEC flag is cleared.
878 	 */
879 	if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
880 		VOP_UNLOCK(imgp->vp);
881 		pe.pm_credentialschanged = credential_changing;
882 		pe.pm_entryaddr = imgp->entry_addr;
883 
884 		PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
885 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
886 	}
887 #endif
888 
889 	/* Set values passed into the program in registers. */
890 	(*p->p_sysent->sv_setregs)(td, imgp, stack_base);
891 
892 	VOP_MMAPPED(imgp->vp);
893 
894 	SDT_PROBE1(proc, , , exec__success, args->fname);
895 
896 exec_fail_dealloc:
897 	if (error != 0) {
898 		p->p_osrel = orig_osrel;
899 		p->p_fctl0 = orig_fctl0;
900 		p->p_elf_brandinfo = orig_brandinfo;
901 	}
902 
903 	if (imgp->firstpage != NULL)
904 		exec_unmap_first_page(imgp);
905 
906 	if (imgp->vp != NULL) {
907 		if (args->fname)
908 			NDFREE(&nd, NDF_ONLY_PNBUF);
909 		if (imgp->opened)
910 			VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
911 		if (imgp->textset)
912 			VOP_UNSET_TEXT_CHECKED(imgp->vp);
913 		if (error != 0)
914 			vput(imgp->vp);
915 		else
916 			VOP_UNLOCK(imgp->vp);
917 	}
918 
919 	if (imgp->object != NULL)
920 		vm_object_deallocate(imgp->object);
921 
922 	free(imgp->freepath, M_TEMP);
923 
924 	if (error == 0) {
925 		if (p->p_ptevents & PTRACE_EXEC) {
926 			PROC_LOCK(p);
927 			if (p->p_ptevents & PTRACE_EXEC)
928 				td->td_dbgflags |= TDB_EXEC;
929 			PROC_UNLOCK(p);
930 		}
931 	} else {
932 exec_fail:
933 		/* we're done here, clear P_INEXEC */
934 		PROC_LOCK(p);
935 		p->p_flag &= ~P_INEXEC;
936 		PROC_UNLOCK(p);
937 
938 		SDT_PROBE1(proc, , , exec__failure, error);
939 	}
940 
941 	if (imgp->newcred != NULL && oldcred != NULL)
942 		crfree(imgp->newcred);
943 
944 #ifdef MAC
945 	mac_execve_exit(imgp);
946 	mac_execve_interpreter_exit(interpvplabel);
947 #endif
948 	exec_free_args(args);
949 
950 	/*
951 	 * Handle deferred decrement of ref counts.
952 	 */
953 	if (oldtextvp != NULL)
954 		vrele(oldtextvp);
955 #ifdef KTRACE
956 	ktr_io_params_free(kiop);
957 #endif
958 	pargs_drop(oldargs);
959 	pargs_drop(newargs);
960 	if (oldsigacts != NULL)
961 		sigacts_free(oldsigacts);
962 	if (euip != NULL)
963 		uifree(euip);
964 
965 	if (error && imgp->vmspace_destroyed) {
966 		/* sorry, no more process anymore. exit gracefully */
967 		exec_cleanup(td, oldvmspace);
968 		exit1(td, 0, SIGABRT);
969 		/* NOT REACHED */
970 	}
971 
972 #ifdef KTRACE
973 	if (error == 0)
974 		ktrprocctor(p);
975 #endif
976 
977 	/*
978 	 * We don't want cpu_set_syscall_retval() to overwrite any of
979 	 * the register values put in place by exec_setregs().
980 	 * Implementations of cpu_set_syscall_retval() will leave
981 	 * registers unmodified when returning EJUSTRETURN.
982 	 */
983 	return (error == 0 ? EJUSTRETURN : error);
984 }
985 
986 void
987 exec_cleanup(struct thread *td, struct vmspace *oldvmspace)
988 {
989 	if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
990 		KASSERT(td->td_proc->p_vmspace != oldvmspace,
991 		    ("oldvmspace still used"));
992 		vmspace_free(oldvmspace);
993 		td->td_pflags &= ~TDP_EXECVMSPC;
994 	}
995 }
996 
997 int
998 exec_map_first_page(struct image_params *imgp)
999 {
1000 	vm_object_t object;
1001 	vm_page_t m;
1002 	int error;
1003 
1004 	if (imgp->firstpage != NULL)
1005 		exec_unmap_first_page(imgp);
1006 
1007 	object = imgp->vp->v_object;
1008 	if (object == NULL)
1009 		return (EACCES);
1010 #if VM_NRESERVLEVEL > 0
1011 	if ((object->flags & OBJ_COLORED) == 0) {
1012 		VM_OBJECT_WLOCK(object);
1013 		vm_object_color(object, 0);
1014 		VM_OBJECT_WUNLOCK(object);
1015 	}
1016 #endif
1017 	error = vm_page_grab_valid_unlocked(&m, object, 0,
1018 	    VM_ALLOC_COUNT(VM_INITIAL_PAGEIN) |
1019             VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED);
1020 
1021 	if (error != VM_PAGER_OK)
1022 		return (EIO);
1023 	imgp->firstpage = sf_buf_alloc(m, 0);
1024 	imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
1025 
1026 	return (0);
1027 }
1028 
1029 void
1030 exec_unmap_first_page(struct image_params *imgp)
1031 {
1032 	vm_page_t m;
1033 
1034 	if (imgp->firstpage != NULL) {
1035 		m = sf_buf_page(imgp->firstpage);
1036 		sf_buf_free(imgp->firstpage);
1037 		imgp->firstpage = NULL;
1038 		vm_page_unwire(m, PQ_ACTIVE);
1039 	}
1040 }
1041 
1042 void
1043 exec_onexec_old(struct thread *td)
1044 {
1045 	sigfastblock_clear(td);
1046 	umtx_exec(td->td_proc);
1047 }
1048 
1049 /*
1050  * Destroy old address space, and allocate a new stack.
1051  *	The new stack is only sgrowsiz large because it is grown
1052  *	automatically on a page fault.
1053  */
1054 int
1055 exec_new_vmspace(struct image_params *imgp, struct sysentvec *sv)
1056 {
1057 	int error;
1058 	struct proc *p = imgp->proc;
1059 	struct vmspace *vmspace = p->p_vmspace;
1060 	struct thread *td = curthread;
1061 	vm_object_t obj;
1062 	struct rlimit rlim_stack;
1063 	vm_offset_t sv_minuser, stack_addr;
1064 	vm_map_t map;
1065 	vm_prot_t stack_prot;
1066 	u_long ssiz;
1067 
1068 	imgp->vmspace_destroyed = 1;
1069 	imgp->sysent = sv;
1070 
1071 	if (p->p_sysent->sv_onexec_old != NULL)
1072 		p->p_sysent->sv_onexec_old(td);
1073 	itimers_exec(p);
1074 
1075 	EVENTHANDLER_DIRECT_INVOKE(process_exec, p, imgp);
1076 
1077 	/*
1078 	 * Blow away entire process VM, if address space not shared,
1079 	 * otherwise, create a new VM space so that other threads are
1080 	 * not disrupted
1081 	 */
1082 	map = &vmspace->vm_map;
1083 	if (map_at_zero)
1084 		sv_minuser = sv->sv_minuser;
1085 	else
1086 		sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1087 	if (refcount_load(&vmspace->vm_refcnt) == 1 &&
1088 	    vm_map_min(map) == sv_minuser &&
1089 	    vm_map_max(map) == sv->sv_maxuser &&
1090 	    cpu_exec_vmspace_reuse(p, map)) {
1091 		shmexit(vmspace);
1092 		pmap_remove_pages(vmspace_pmap(vmspace));
1093 		vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1094 		/*
1095 		 * An exec terminates mlockall(MCL_FUTURE).
1096 		 * ASLR and W^X states must be re-evaluated.
1097 		 */
1098 		vm_map_lock(map);
1099 		vm_map_modflags(map, 0, MAP_WIREFUTURE | MAP_ASLR |
1100 		    MAP_ASLR_IGNSTART | MAP_WXORX);
1101 		vm_map_unlock(map);
1102 	} else {
1103 		error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1104 		if (error)
1105 			return (error);
1106 		vmspace = p->p_vmspace;
1107 		map = &vmspace->vm_map;
1108 	}
1109 	map->flags |= imgp->map_flags;
1110 
1111 	/* Map a shared page */
1112 	obj = sv->sv_shared_page_obj;
1113 	if (obj != NULL) {
1114 		vm_object_reference(obj);
1115 		error = vm_map_fixed(map, obj, 0,
1116 		    sv->sv_shared_page_base, sv->sv_shared_page_len,
1117 		    VM_PROT_READ | VM_PROT_EXECUTE,
1118 		    VM_PROT_READ | VM_PROT_EXECUTE,
1119 		    MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1120 		if (error != KERN_SUCCESS) {
1121 			vm_object_deallocate(obj);
1122 			return (vm_mmap_to_errno(error));
1123 		}
1124 	}
1125 
1126 	/* Allocate a new stack */
1127 	if (imgp->stack_sz != 0) {
1128 		ssiz = trunc_page(imgp->stack_sz);
1129 		PROC_LOCK(p);
1130 		lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack);
1131 		PROC_UNLOCK(p);
1132 		if (ssiz > rlim_stack.rlim_max)
1133 			ssiz = rlim_stack.rlim_max;
1134 		if (ssiz > rlim_stack.rlim_cur) {
1135 			rlim_stack.rlim_cur = ssiz;
1136 			kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack);
1137 		}
1138 	} else if (sv->sv_maxssiz != NULL) {
1139 		ssiz = *sv->sv_maxssiz;
1140 	} else {
1141 		ssiz = maxssiz;
1142 	}
1143 	imgp->eff_stack_sz = lim_cur(curthread, RLIMIT_STACK);
1144 	if (ssiz < imgp->eff_stack_sz)
1145 		imgp->eff_stack_sz = ssiz;
1146 	stack_addr = sv->sv_usrstack - ssiz;
1147 	stack_prot = obj != NULL && imgp->stack_prot != 0 ?
1148 	    imgp->stack_prot : sv->sv_stackprot;
1149 	error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz, stack_prot,
1150 	    VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1151 	if (error != KERN_SUCCESS) {
1152 		uprintf("exec_new_vmspace: mapping stack size %#jx prot %#x "
1153 		    "failed mach error %d errno %d\n", (uintmax_t)ssiz,
1154 		    stack_prot, error, vm_mmap_to_errno(error));
1155 		return (vm_mmap_to_errno(error));
1156 	}
1157 
1158 	/*
1159 	 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they
1160 	 * are still used to enforce the stack rlimit on the process stack.
1161 	 */
1162 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1163 	vmspace->vm_maxsaddr = (char *)stack_addr;
1164 
1165 	return (sv->sv_onexec != NULL ? sv->sv_onexec(p, imgp) : 0);
1166 }
1167 
1168 /*
1169  * Copy out argument and environment strings from the old process address
1170  * space into the temporary string buffer.
1171  */
1172 int
1173 exec_copyin_args(struct image_args *args, const char *fname,
1174     enum uio_seg segflg, char **argv, char **envv)
1175 {
1176 	u_long arg, env;
1177 	int error;
1178 
1179 	bzero(args, sizeof(*args));
1180 	if (argv == NULL)
1181 		return (EFAULT);
1182 
1183 	/*
1184 	 * Allocate demand-paged memory for the file name, argument, and
1185 	 * environment strings.
1186 	 */
1187 	error = exec_alloc_args(args);
1188 	if (error != 0)
1189 		return (error);
1190 
1191 	/*
1192 	 * Copy the file name.
1193 	 */
1194 	error = exec_args_add_fname(args, fname, segflg);
1195 	if (error != 0)
1196 		goto err_exit;
1197 
1198 	/*
1199 	 * extract arguments first
1200 	 */
1201 	for (;;) {
1202 		error = fueword(argv++, &arg);
1203 		if (error == -1) {
1204 			error = EFAULT;
1205 			goto err_exit;
1206 		}
1207 		if (arg == 0)
1208 			break;
1209 		error = exec_args_add_arg(args, (char *)(uintptr_t)arg,
1210 		    UIO_USERSPACE);
1211 		if (error != 0)
1212 			goto err_exit;
1213 	}
1214 
1215 	/*
1216 	 * extract environment strings
1217 	 */
1218 	if (envv) {
1219 		for (;;) {
1220 			error = fueword(envv++, &env);
1221 			if (error == -1) {
1222 				error = EFAULT;
1223 				goto err_exit;
1224 			}
1225 			if (env == 0)
1226 				break;
1227 			error = exec_args_add_env(args,
1228 			    (char *)(uintptr_t)env, UIO_USERSPACE);
1229 			if (error != 0)
1230 				goto err_exit;
1231 		}
1232 	}
1233 
1234 	return (0);
1235 
1236 err_exit:
1237 	exec_free_args(args);
1238 	return (error);
1239 }
1240 
1241 int
1242 exec_copyin_data_fds(struct thread *td, struct image_args *args,
1243     const void *data, size_t datalen, const int *fds, size_t fdslen)
1244 {
1245 	struct filedesc *ofdp;
1246 	const char *p;
1247 	int *kfds;
1248 	int error;
1249 
1250 	memset(args, '\0', sizeof(*args));
1251 	ofdp = td->td_proc->p_fd;
1252 	if (datalen >= ARG_MAX || fdslen >= ofdp->fd_nfiles)
1253 		return (E2BIG);
1254 	error = exec_alloc_args(args);
1255 	if (error != 0)
1256 		return (error);
1257 
1258 	args->begin_argv = args->buf;
1259 	args->stringspace = ARG_MAX;
1260 
1261 	if (datalen > 0) {
1262 		/*
1263 		 * Argument buffer has been provided. Copy it into the
1264 		 * kernel as a single string and add a terminating null
1265 		 * byte.
1266 		 */
1267 		error = copyin(data, args->begin_argv, datalen);
1268 		if (error != 0)
1269 			goto err_exit;
1270 		args->begin_argv[datalen] = '\0';
1271 		args->endp = args->begin_argv + datalen + 1;
1272 		args->stringspace -= datalen + 1;
1273 
1274 		/*
1275 		 * Traditional argument counting. Count the number of
1276 		 * null bytes.
1277 		 */
1278 		for (p = args->begin_argv; p < args->endp; ++p)
1279 			if (*p == '\0')
1280 				++args->argc;
1281 	} else {
1282 		/* No argument buffer provided. */
1283 		args->endp = args->begin_argv;
1284 	}
1285 
1286 	/* Create new file descriptor table. */
1287 	kfds = malloc(fdslen * sizeof(int), M_TEMP, M_WAITOK);
1288 	error = copyin(fds, kfds, fdslen * sizeof(int));
1289 	if (error != 0) {
1290 		free(kfds, M_TEMP);
1291 		goto err_exit;
1292 	}
1293 	error = fdcopy_remapped(ofdp, kfds, fdslen, &args->fdp);
1294 	free(kfds, M_TEMP);
1295 	if (error != 0)
1296 		goto err_exit;
1297 
1298 	return (0);
1299 err_exit:
1300 	exec_free_args(args);
1301 	return (error);
1302 }
1303 
1304 struct exec_args_kva {
1305 	vm_offset_t addr;
1306 	u_int gen;
1307 	SLIST_ENTRY(exec_args_kva) next;
1308 };
1309 
1310 DPCPU_DEFINE_STATIC(struct exec_args_kva *, exec_args_kva);
1311 
1312 static SLIST_HEAD(, exec_args_kva) exec_args_kva_freelist;
1313 static struct mtx exec_args_kva_mtx;
1314 static u_int exec_args_gen;
1315 
1316 static void
1317 exec_prealloc_args_kva(void *arg __unused)
1318 {
1319 	struct exec_args_kva *argkva;
1320 	u_int i;
1321 
1322 	SLIST_INIT(&exec_args_kva_freelist);
1323 	mtx_init(&exec_args_kva_mtx, "exec args kva", NULL, MTX_DEF);
1324 	for (i = 0; i < exec_map_entries; i++) {
1325 		argkva = malloc(sizeof(*argkva), M_PARGS, M_WAITOK);
1326 		argkva->addr = kmap_alloc_wait(exec_map, exec_map_entry_size);
1327 		argkva->gen = exec_args_gen;
1328 		SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1329 	}
1330 }
1331 SYSINIT(exec_args_kva, SI_SUB_EXEC, SI_ORDER_ANY, exec_prealloc_args_kva, NULL);
1332 
1333 static vm_offset_t
1334 exec_alloc_args_kva(void **cookie)
1335 {
1336 	struct exec_args_kva *argkva;
1337 
1338 	argkva = (void *)atomic_readandclear_ptr(
1339 	    (uintptr_t *)DPCPU_PTR(exec_args_kva));
1340 	if (argkva == NULL) {
1341 		mtx_lock(&exec_args_kva_mtx);
1342 		while ((argkva = SLIST_FIRST(&exec_args_kva_freelist)) == NULL)
1343 			(void)mtx_sleep(&exec_args_kva_freelist,
1344 			    &exec_args_kva_mtx, 0, "execkva", 0);
1345 		SLIST_REMOVE_HEAD(&exec_args_kva_freelist, next);
1346 		mtx_unlock(&exec_args_kva_mtx);
1347 	}
1348 	kasan_mark((void *)argkva->addr, exec_map_entry_size,
1349 	    exec_map_entry_size, 0);
1350 	*(struct exec_args_kva **)cookie = argkva;
1351 	return (argkva->addr);
1352 }
1353 
1354 static void
1355 exec_release_args_kva(struct exec_args_kva *argkva, u_int gen)
1356 {
1357 	vm_offset_t base;
1358 
1359 	base = argkva->addr;
1360 	kasan_mark((void *)argkva->addr, 0, exec_map_entry_size,
1361 	    KASAN_EXEC_ARGS_FREED);
1362 	if (argkva->gen != gen) {
1363 		(void)vm_map_madvise(exec_map, base, base + exec_map_entry_size,
1364 		    MADV_FREE);
1365 		argkva->gen = gen;
1366 	}
1367 	if (!atomic_cmpset_ptr((uintptr_t *)DPCPU_PTR(exec_args_kva),
1368 	    (uintptr_t)NULL, (uintptr_t)argkva)) {
1369 		mtx_lock(&exec_args_kva_mtx);
1370 		SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1371 		wakeup_one(&exec_args_kva_freelist);
1372 		mtx_unlock(&exec_args_kva_mtx);
1373 	}
1374 }
1375 
1376 static void
1377 exec_free_args_kva(void *cookie)
1378 {
1379 
1380 	exec_release_args_kva(cookie, exec_args_gen);
1381 }
1382 
1383 static void
1384 exec_args_kva_lowmem(void *arg __unused)
1385 {
1386 	SLIST_HEAD(, exec_args_kva) head;
1387 	struct exec_args_kva *argkva;
1388 	u_int gen;
1389 	int i;
1390 
1391 	gen = atomic_fetchadd_int(&exec_args_gen, 1) + 1;
1392 
1393 	/*
1394 	 * Force an madvise of each KVA range. Any currently allocated ranges
1395 	 * will have MADV_FREE applied once they are freed.
1396 	 */
1397 	SLIST_INIT(&head);
1398 	mtx_lock(&exec_args_kva_mtx);
1399 	SLIST_SWAP(&head, &exec_args_kva_freelist, exec_args_kva);
1400 	mtx_unlock(&exec_args_kva_mtx);
1401 	while ((argkva = SLIST_FIRST(&head)) != NULL) {
1402 		SLIST_REMOVE_HEAD(&head, next);
1403 		exec_release_args_kva(argkva, gen);
1404 	}
1405 
1406 	CPU_FOREACH(i) {
1407 		argkva = (void *)atomic_readandclear_ptr(
1408 		    (uintptr_t *)DPCPU_ID_PTR(i, exec_args_kva));
1409 		if (argkva != NULL)
1410 			exec_release_args_kva(argkva, gen);
1411 	}
1412 }
1413 EVENTHANDLER_DEFINE(vm_lowmem, exec_args_kva_lowmem, NULL,
1414     EVENTHANDLER_PRI_ANY);
1415 
1416 /*
1417  * Allocate temporary demand-paged, zero-filled memory for the file name,
1418  * argument, and environment strings.
1419  */
1420 int
1421 exec_alloc_args(struct image_args *args)
1422 {
1423 
1424 	args->buf = (char *)exec_alloc_args_kva(&args->bufkva);
1425 	return (0);
1426 }
1427 
1428 void
1429 exec_free_args(struct image_args *args)
1430 {
1431 
1432 	if (args->buf != NULL) {
1433 		exec_free_args_kva(args->bufkva);
1434 		args->buf = NULL;
1435 	}
1436 	if (args->fname_buf != NULL) {
1437 		free(args->fname_buf, M_TEMP);
1438 		args->fname_buf = NULL;
1439 	}
1440 	if (args->fdp != NULL)
1441 		fdescfree_remapped(args->fdp);
1442 }
1443 
1444 /*
1445  * A set to functions to fill struct image args.
1446  *
1447  * NOTE: exec_args_add_fname() must be called (possibly with a NULL
1448  * fname) before the other functions.  All exec_args_add_arg() calls must
1449  * be made before any exec_args_add_env() calls.  exec_args_adjust_args()
1450  * may be called any time after exec_args_add_fname().
1451  *
1452  * exec_args_add_fname() - install path to be executed
1453  * exec_args_add_arg() - append an argument string
1454  * exec_args_add_env() - append an env string
1455  * exec_args_adjust_args() - adjust location of the argument list to
1456  *                           allow new arguments to be prepended
1457  */
1458 int
1459 exec_args_add_fname(struct image_args *args, const char *fname,
1460     enum uio_seg segflg)
1461 {
1462 	int error;
1463 	size_t length;
1464 
1465 	KASSERT(args->fname == NULL, ("fname already appended"));
1466 	KASSERT(args->endp == NULL, ("already appending to args"));
1467 
1468 	if (fname != NULL) {
1469 		args->fname = args->buf;
1470 		error = segflg == UIO_SYSSPACE ?
1471 		    copystr(fname, args->fname, PATH_MAX, &length) :
1472 		    copyinstr(fname, args->fname, PATH_MAX, &length);
1473 		if (error != 0)
1474 			return (error == ENAMETOOLONG ? E2BIG : error);
1475 	} else
1476 		length = 0;
1477 
1478 	/* Set up for _arg_*()/_env_*() */
1479 	args->endp = args->buf + length;
1480 	/* begin_argv must be set and kept updated */
1481 	args->begin_argv = args->endp;
1482 	KASSERT(exec_map_entry_size - length >= ARG_MAX,
1483 	    ("too little space remaining for arguments %zu < %zu",
1484 	    exec_map_entry_size - length, (size_t)ARG_MAX));
1485 	args->stringspace = ARG_MAX;
1486 
1487 	return (0);
1488 }
1489 
1490 static int
1491 exec_args_add_str(struct image_args *args, const char *str,
1492     enum uio_seg segflg, int *countp)
1493 {
1494 	int error;
1495 	size_t length;
1496 
1497 	KASSERT(args->endp != NULL, ("endp not initialized"));
1498 	KASSERT(args->begin_argv != NULL, ("begin_argp not initialized"));
1499 
1500 	error = (segflg == UIO_SYSSPACE) ?
1501 	    copystr(str, args->endp, args->stringspace, &length) :
1502 	    copyinstr(str, args->endp, args->stringspace, &length);
1503 	if (error != 0)
1504 		return (error == ENAMETOOLONG ? E2BIG : error);
1505 	args->stringspace -= length;
1506 	args->endp += length;
1507 	(*countp)++;
1508 
1509 	return (0);
1510 }
1511 
1512 int
1513 exec_args_add_arg(struct image_args *args, const char *argp,
1514     enum uio_seg segflg)
1515 {
1516 
1517 	KASSERT(args->envc == 0, ("appending args after env"));
1518 
1519 	return (exec_args_add_str(args, argp, segflg, &args->argc));
1520 }
1521 
1522 int
1523 exec_args_add_env(struct image_args *args, const char *envp,
1524     enum uio_seg segflg)
1525 {
1526 
1527 	if (args->envc == 0)
1528 		args->begin_envv = args->endp;
1529 
1530 	return (exec_args_add_str(args, envp, segflg, &args->envc));
1531 }
1532 
1533 int
1534 exec_args_adjust_args(struct image_args *args, size_t consume, ssize_t extend)
1535 {
1536 	ssize_t offset;
1537 
1538 	KASSERT(args->endp != NULL, ("endp not initialized"));
1539 	KASSERT(args->begin_argv != NULL, ("begin_argp not initialized"));
1540 
1541 	offset = extend - consume;
1542 	if (args->stringspace < offset)
1543 		return (E2BIG);
1544 	memmove(args->begin_argv + extend, args->begin_argv + consume,
1545 	    args->endp - args->begin_argv + consume);
1546 	if (args->envc > 0)
1547 		args->begin_envv += offset;
1548 	args->endp += offset;
1549 	args->stringspace -= offset;
1550 	return (0);
1551 }
1552 
1553 char *
1554 exec_args_get_begin_envv(struct image_args *args)
1555 {
1556 
1557 	KASSERT(args->endp != NULL, ("endp not initialized"));
1558 
1559 	if (args->envc > 0)
1560 		return (args->begin_envv);
1561 	return (args->endp);
1562 }
1563 
1564 void
1565 exec_stackgap(struct image_params *imgp, uintptr_t *dp)
1566 {
1567 	if (imgp->sysent->sv_stackgap == NULL ||
1568 	    (imgp->proc->p_fctl0 & (NT_FREEBSD_FCTL_ASLR_DISABLE |
1569 	    NT_FREEBSD_FCTL_ASG_DISABLE)) != 0 ||
1570 	    (imgp->map_flags & MAP_ASLR) == 0)
1571 		return;
1572 	imgp->sysent->sv_stackgap(imgp, dp);
1573 }
1574 
1575 /*
1576  * Copy strings out to the new process address space, constructing new arg
1577  * and env vector tables. Return a pointer to the base so that it can be used
1578  * as the initial stack pointer.
1579  */
1580 int
1581 exec_copyout_strings(struct image_params *imgp, uintptr_t *stack_base)
1582 {
1583 	int argc, envc;
1584 	char **vectp;
1585 	char *stringp;
1586 	uintptr_t destp, ustringp;
1587 	struct ps_strings *arginfo;
1588 	struct proc *p;
1589 	size_t execpath_len;
1590 	int error, szsigcode, szps;
1591 	char canary[sizeof(long) * 8];
1592 
1593 	szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1594 	/*
1595 	 * Calculate string base and vector table pointers.
1596 	 * Also deal with signal trampoline code for this exec type.
1597 	 */
1598 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
1599 		execpath_len = strlen(imgp->execpath) + 1;
1600 	else
1601 		execpath_len = 0;
1602 	p = imgp->proc;
1603 	szsigcode = 0;
1604 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1605 	imgp->ps_strings = arginfo;
1606 	if (p->p_sysent->sv_sigcode_base == 0) {
1607 		if (p->p_sysent->sv_szsigcode != NULL)
1608 			szsigcode = *(p->p_sysent->sv_szsigcode);
1609 	}
1610 	destp =	(uintptr_t)arginfo;
1611 
1612 	/*
1613 	 * install sigcode
1614 	 */
1615 	if (szsigcode != 0) {
1616 		destp -= szsigcode;
1617 		destp = rounddown2(destp, sizeof(void *));
1618 		error = copyout(p->p_sysent->sv_sigcode, (void *)destp,
1619 		    szsigcode);
1620 		if (error != 0)
1621 			return (error);
1622 	}
1623 
1624 	/*
1625 	 * Copy the image path for the rtld.
1626 	 */
1627 	if (execpath_len != 0) {
1628 		destp -= execpath_len;
1629 		destp = rounddown2(destp, sizeof(void *));
1630 		imgp->execpathp = (void *)destp;
1631 		error = copyout(imgp->execpath, imgp->execpathp, execpath_len);
1632 		if (error != 0)
1633 			return (error);
1634 	}
1635 
1636 	/*
1637 	 * Prepare the canary for SSP.
1638 	 */
1639 	arc4rand(canary, sizeof(canary), 0);
1640 	destp -= sizeof(canary);
1641 	imgp->canary = (void *)destp;
1642 	error = copyout(canary, imgp->canary, sizeof(canary));
1643 	if (error != 0)
1644 		return (error);
1645 	imgp->canarylen = sizeof(canary);
1646 
1647 	/*
1648 	 * Prepare the pagesizes array.
1649 	 */
1650 	destp -= szps;
1651 	destp = rounddown2(destp, sizeof(void *));
1652 	imgp->pagesizes = (void *)destp;
1653 	error = copyout(pagesizes, imgp->pagesizes, szps);
1654 	if (error != 0)
1655 		return (error);
1656 	imgp->pagesizeslen = szps;
1657 
1658 	/*
1659 	 * Allocate room for the argument and environment strings.
1660 	 */
1661 	destp -= ARG_MAX - imgp->args->stringspace;
1662 	destp = rounddown2(destp, sizeof(void *));
1663 	ustringp = destp;
1664 
1665 	exec_stackgap(imgp, &destp);
1666 
1667 	if (imgp->auxargs) {
1668 		/*
1669 		 * Allocate room on the stack for the ELF auxargs
1670 		 * array.  It has up to AT_COUNT entries.
1671 		 */
1672 		destp -= AT_COUNT * sizeof(Elf_Auxinfo);
1673 		destp = rounddown2(destp, sizeof(void *));
1674 	}
1675 
1676 	vectp = (char **)destp;
1677 
1678 	/*
1679 	 * Allocate room for the argv[] and env vectors including the
1680 	 * terminating NULL pointers.
1681 	 */
1682 	vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
1683 
1684 	/*
1685 	 * vectp also becomes our initial stack base
1686 	 */
1687 	*stack_base = (uintptr_t)vectp;
1688 
1689 	stringp = imgp->args->begin_argv;
1690 	argc = imgp->args->argc;
1691 	envc = imgp->args->envc;
1692 
1693 	/*
1694 	 * Copy out strings - arguments and environment.
1695 	 */
1696 	error = copyout(stringp, (void *)ustringp,
1697 	    ARG_MAX - imgp->args->stringspace);
1698 	if (error != 0)
1699 		return (error);
1700 
1701 	/*
1702 	 * Fill in "ps_strings" struct for ps, w, etc.
1703 	 */
1704 	imgp->argv = vectp;
1705 	if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 ||
1706 	    suword32(&arginfo->ps_nargvstr, argc) != 0)
1707 		return (EFAULT);
1708 
1709 	/*
1710 	 * Fill in argument portion of vector table.
1711 	 */
1712 	for (; argc > 0; --argc) {
1713 		if (suword(vectp++, ustringp) != 0)
1714 			return (EFAULT);
1715 		while (*stringp++ != 0)
1716 			ustringp++;
1717 		ustringp++;
1718 	}
1719 
1720 	/* a null vector table pointer separates the argp's from the envp's */
1721 	if (suword(vectp++, 0) != 0)
1722 		return (EFAULT);
1723 
1724 	imgp->envv = vectp;
1725 	if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 ||
1726 	    suword32(&arginfo->ps_nenvstr, envc) != 0)
1727 		return (EFAULT);
1728 
1729 	/*
1730 	 * Fill in environment portion of vector table.
1731 	 */
1732 	for (; envc > 0; --envc) {
1733 		if (suword(vectp++, ustringp) != 0)
1734 			return (EFAULT);
1735 		while (*stringp++ != 0)
1736 			ustringp++;
1737 		ustringp++;
1738 	}
1739 
1740 	/* end of vector table is a null pointer */
1741 	if (suword(vectp, 0) != 0)
1742 		return (EFAULT);
1743 
1744 	if (imgp->auxargs) {
1745 		vectp++;
1746 		error = imgp->sysent->sv_copyout_auxargs(imgp,
1747 		    (uintptr_t)vectp);
1748 		if (error != 0)
1749 			return (error);
1750 	}
1751 
1752 	return (0);
1753 }
1754 
1755 /*
1756  * Check permissions of file to execute.
1757  *	Called with imgp->vp locked.
1758  *	Return 0 for success or error code on failure.
1759  */
1760 int
1761 exec_check_permissions(struct image_params *imgp)
1762 {
1763 	struct vnode *vp = imgp->vp;
1764 	struct vattr *attr = imgp->attr;
1765 	struct thread *td;
1766 	int error;
1767 
1768 	td = curthread;
1769 
1770 	/* Get file attributes */
1771 	error = VOP_GETATTR(vp, attr, td->td_ucred);
1772 	if (error)
1773 		return (error);
1774 
1775 #ifdef MAC
1776 	error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1777 	if (error)
1778 		return (error);
1779 #endif
1780 
1781 	/*
1782 	 * 1) Check if file execution is disabled for the filesystem that
1783 	 *    this file resides on.
1784 	 * 2) Ensure that at least one execute bit is on. Otherwise, a
1785 	 *    privileged user will always succeed, and we don't want this
1786 	 *    to happen unless the file really is executable.
1787 	 * 3) Ensure that the file is a regular file.
1788 	 */
1789 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1790 	    (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1791 	    (attr->va_type != VREG))
1792 		return (EACCES);
1793 
1794 	/*
1795 	 * Zero length files can't be exec'd
1796 	 */
1797 	if (attr->va_size == 0)
1798 		return (ENOEXEC);
1799 
1800 	/*
1801 	 *  Check for execute permission to file based on current credentials.
1802 	 */
1803 	error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1804 	if (error)
1805 		return (error);
1806 
1807 	/*
1808 	 * Check number of open-for-writes on the file and deny execution
1809 	 * if there are any.
1810 	 *
1811 	 * Add a text reference now so no one can write to the
1812 	 * executable while we're activating it.
1813 	 *
1814 	 * Remember if this was set before and unset it in case this is not
1815 	 * actually an executable image.
1816 	 */
1817 	error = VOP_SET_TEXT(vp);
1818 	if (error != 0)
1819 		return (error);
1820 	imgp->textset = true;
1821 
1822 	/*
1823 	 * Call filesystem specific open routine (which does nothing in the
1824 	 * general case).
1825 	 */
1826 	error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1827 	if (error == 0)
1828 		imgp->opened = 1;
1829 	return (error);
1830 }
1831 
1832 /*
1833  * Exec handler registration
1834  */
1835 int
1836 exec_register(const struct execsw *execsw_arg)
1837 {
1838 	const struct execsw **es, **xs, **newexecsw;
1839 	u_int count = 2;	/* New slot and trailing NULL */
1840 
1841 	if (execsw)
1842 		for (es = execsw; *es; es++)
1843 			count++;
1844 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1845 	xs = newexecsw;
1846 	if (execsw)
1847 		for (es = execsw; *es; es++)
1848 			*xs++ = *es;
1849 	*xs++ = execsw_arg;
1850 	*xs = NULL;
1851 	if (execsw)
1852 		free(execsw, M_TEMP);
1853 	execsw = newexecsw;
1854 	return (0);
1855 }
1856 
1857 int
1858 exec_unregister(const struct execsw *execsw_arg)
1859 {
1860 	const struct execsw **es, **xs, **newexecsw;
1861 	int count = 1;
1862 
1863 	if (execsw == NULL)
1864 		panic("unregister with no handlers left?\n");
1865 
1866 	for (es = execsw; *es; es++) {
1867 		if (*es == execsw_arg)
1868 			break;
1869 	}
1870 	if (*es == NULL)
1871 		return (ENOENT);
1872 	for (es = execsw; *es; es++)
1873 		if (*es != execsw_arg)
1874 			count++;
1875 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1876 	xs = newexecsw;
1877 	for (es = execsw; *es; es++)
1878 		if (*es != execsw_arg)
1879 			*xs++ = *es;
1880 	*xs = NULL;
1881 	if (execsw)
1882 		free(execsw, M_TEMP);
1883 	execsw = newexecsw;
1884 	return (0);
1885 }
1886 
1887 /*
1888  * Write out a core segment to the compression stream.
1889  */
1890 static int
1891 compress_chunk(struct coredump_params *cp, char *base, char *buf, u_int len)
1892 {
1893 	u_int chunk_len;
1894 	int error;
1895 
1896 	while (len > 0) {
1897 		chunk_len = MIN(len, CORE_BUF_SIZE);
1898 
1899 		/*
1900 		 * We can get EFAULT error here.
1901 		 * In that case zero out the current chunk of the segment.
1902 		 */
1903 		error = copyin(base, buf, chunk_len);
1904 		if (error != 0)
1905 			bzero(buf, chunk_len);
1906 		error = compressor_write(cp->comp, buf, chunk_len);
1907 		if (error != 0)
1908 			break;
1909 		base += chunk_len;
1910 		len -= chunk_len;
1911 	}
1912 	return (error);
1913 }
1914 
1915 int
1916 core_write(struct coredump_params *cp, const void *base, size_t len,
1917     off_t offset, enum uio_seg seg, size_t *resid)
1918 {
1919 
1920 	return (vn_rdwr_inchunks(UIO_WRITE, cp->vp, __DECONST(void *, base),
1921 	    len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1922 	    cp->active_cred, cp->file_cred, resid, cp->td));
1923 }
1924 
1925 int
1926 core_output(char *base, size_t len, off_t offset, struct coredump_params *cp,
1927     void *tmpbuf)
1928 {
1929 	vm_map_t map;
1930 	struct mount *mp;
1931 	size_t resid, runlen;
1932 	int error;
1933 	bool success;
1934 
1935 	KASSERT((uintptr_t)base % PAGE_SIZE == 0,
1936 	    ("%s: user address %p is not page-aligned", __func__, base));
1937 
1938 	if (cp->comp != NULL)
1939 		return (compress_chunk(cp, base, tmpbuf, len));
1940 
1941 	map = &cp->td->td_proc->p_vmspace->vm_map;
1942 	for (; len > 0; base += runlen, offset += runlen, len -= runlen) {
1943 		/*
1944 		 * Attempt to page in all virtual pages in the range.  If a
1945 		 * virtual page is not backed by the pager, it is represented as
1946 		 * a hole in the file.  This can occur with zero-filled
1947 		 * anonymous memory or truncated files, for example.
1948 		 */
1949 		for (runlen = 0; runlen < len; runlen += PAGE_SIZE) {
1950 			error = vm_fault(map, (uintptr_t)base + runlen,
1951 			    VM_PROT_READ, VM_FAULT_NOFILL, NULL);
1952 			if (runlen == 0)
1953 				success = error == KERN_SUCCESS;
1954 			else if ((error == KERN_SUCCESS) != success)
1955 				break;
1956 		}
1957 
1958 		if (success) {
1959 			error = core_write(cp, base, runlen, offset,
1960 			    UIO_USERSPACE, &resid);
1961 			if (error != 0) {
1962 				if (error != EFAULT)
1963 					break;
1964 
1965 				/*
1966 				 * EFAULT may be returned if the user mapping
1967 				 * could not be accessed, e.g., because a mapped
1968 				 * file has been truncated.  Skip the page if no
1969 				 * progress was made, to protect against a
1970 				 * hypothetical scenario where vm_fault() was
1971 				 * successful but core_write() returns EFAULT
1972 				 * anyway.
1973 				 */
1974 				runlen -= resid;
1975 				if (runlen == 0) {
1976 					success = false;
1977 					runlen = PAGE_SIZE;
1978 				}
1979 			}
1980 		}
1981 		if (!success) {
1982 			error = vn_start_write(cp->vp, &mp, V_WAIT);
1983 			if (error != 0)
1984 				break;
1985 			vn_lock(cp->vp, LK_EXCLUSIVE | LK_RETRY);
1986 			error = vn_truncate_locked(cp->vp, offset + runlen,
1987 			    false, cp->td->td_ucred);
1988 			VOP_UNLOCK(cp->vp);
1989 			vn_finished_write(mp);
1990 			if (error != 0)
1991 				break;
1992 		}
1993 	}
1994 	return (error);
1995 }
1996 
1997 /*
1998  * Drain into a core file.
1999  */
2000 int
2001 sbuf_drain_core_output(void *arg, const char *data, int len)
2002 {
2003 	struct coredump_params *cp;
2004 	struct proc *p;
2005 	int error, locked;
2006 
2007 	cp = arg;
2008 	p = cp->td->td_proc;
2009 
2010 	/*
2011 	 * Some kern_proc out routines that print to this sbuf may
2012 	 * call us with the process lock held. Draining with the
2013 	 * non-sleepable lock held is unsafe. The lock is needed for
2014 	 * those routines when dumping a live process. In our case we
2015 	 * can safely release the lock before draining and acquire
2016 	 * again after.
2017 	 */
2018 	locked = PROC_LOCKED(p);
2019 	if (locked)
2020 		PROC_UNLOCK(p);
2021 	if (cp->comp != NULL)
2022 		error = compressor_write(cp->comp, __DECONST(char *, data), len);
2023 	else
2024 		error = core_write(cp, __DECONST(void *, data), len, cp->offset,
2025 		    UIO_SYSSPACE, NULL);
2026 	if (locked)
2027 		PROC_LOCK(p);
2028 	if (error != 0)
2029 		return (-error);
2030 	cp->offset += len;
2031 	return (len);
2032 }
2033