xref: /freebsd/sys/kern/kern_exec.c (revision aacbf3fb8482226fcdf6a19af1011f9d65c20f2a)
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 core_dump_can_intr = 1;
155 SYSCTL_INT(_kern, OID_AUTO, core_dump_can_intr, CTLFLAG_RWTUN,
156     &core_dump_can_intr, 0,
157     "Core dumping interruptible with SIGKILL");
158 
159 static int
160 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
161 {
162 	struct proc *p;
163 	int error;
164 
165 	p = curproc;
166 #ifdef SCTL_MASK32
167 	if (req->flags & SCTL_MASK32) {
168 		unsigned int val;
169 		val = (unsigned int)p->p_sysent->sv_psstrings;
170 		error = SYSCTL_OUT(req, &val, sizeof(val));
171 	} else
172 #endif
173 		error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
174 		   sizeof(p->p_sysent->sv_psstrings));
175 	return error;
176 }
177 
178 static int
179 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
180 {
181 	struct proc *p;
182 	int error;
183 
184 	p = curproc;
185 #ifdef SCTL_MASK32
186 	if (req->flags & SCTL_MASK32) {
187 		unsigned int val;
188 		val = (unsigned int)p->p_sysent->sv_usrstack;
189 		error = SYSCTL_OUT(req, &val, sizeof(val));
190 	} else
191 #endif
192 		error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
193 		    sizeof(p->p_sysent->sv_usrstack));
194 	return error;
195 }
196 
197 static int
198 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
199 {
200 	struct proc *p;
201 
202 	p = curproc;
203 	return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
204 	    sizeof(p->p_sysent->sv_stackprot)));
205 }
206 
207 /*
208  * Each of the items is a pointer to a `const struct execsw', hence the
209  * double pointer here.
210  */
211 static const struct execsw **execsw;
212 
213 #ifndef _SYS_SYSPROTO_H_
214 struct execve_args {
215 	char    *fname;
216 	char    **argv;
217 	char    **envv;
218 };
219 #endif
220 
221 int
222 sys_execve(struct thread *td, struct execve_args *uap)
223 {
224 	struct image_args args;
225 	struct vmspace *oldvmspace;
226 	int error;
227 
228 	error = pre_execve(td, &oldvmspace);
229 	if (error != 0)
230 		return (error);
231 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
232 	    uap->argv, uap->envv);
233 	if (error == 0)
234 		error = kern_execve(td, &args, NULL, oldvmspace);
235 	post_execve(td, error, oldvmspace);
236 	AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
237 	return (error);
238 }
239 
240 #ifndef _SYS_SYSPROTO_H_
241 struct fexecve_args {
242 	int	fd;
243 	char	**argv;
244 	char	**envv;
245 };
246 #endif
247 int
248 sys_fexecve(struct thread *td, struct fexecve_args *uap)
249 {
250 	struct image_args args;
251 	struct vmspace *oldvmspace;
252 	int error;
253 
254 	error = pre_execve(td, &oldvmspace);
255 	if (error != 0)
256 		return (error);
257 	error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
258 	    uap->argv, uap->envv);
259 	if (error == 0) {
260 		args.fd = uap->fd;
261 		error = kern_execve(td, &args, NULL, oldvmspace);
262 	}
263 	post_execve(td, error, oldvmspace);
264 	AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
265 	return (error);
266 }
267 
268 #ifndef _SYS_SYSPROTO_H_
269 struct __mac_execve_args {
270 	char	*fname;
271 	char	**argv;
272 	char	**envv;
273 	struct mac	*mac_p;
274 };
275 #endif
276 
277 int
278 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap)
279 {
280 #ifdef MAC
281 	struct image_args args;
282 	struct vmspace *oldvmspace;
283 	int error;
284 
285 	error = pre_execve(td, &oldvmspace);
286 	if (error != 0)
287 		return (error);
288 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
289 	    uap->argv, uap->envv);
290 	if (error == 0)
291 		error = kern_execve(td, &args, uap->mac_p, oldvmspace);
292 	post_execve(td, error, oldvmspace);
293 	AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
294 	return (error);
295 #else
296 	return (ENOSYS);
297 #endif
298 }
299 
300 int
301 pre_execve(struct thread *td, struct vmspace **oldvmspace)
302 {
303 	struct proc *p;
304 	int error;
305 
306 	KASSERT(td == curthread, ("non-current thread %p", td));
307 	error = 0;
308 	p = td->td_proc;
309 	if ((p->p_flag & P_HADTHREADS) != 0) {
310 		PROC_LOCK(p);
311 		if (thread_single(p, SINGLE_BOUNDARY) != 0)
312 			error = ERESTART;
313 		PROC_UNLOCK(p);
314 	}
315 	KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0,
316 	    ("nested execve"));
317 	*oldvmspace = p->p_vmspace;
318 	return (error);
319 }
320 
321 void
322 post_execve(struct thread *td, int error, struct vmspace *oldvmspace)
323 {
324 	struct proc *p;
325 
326 	KASSERT(td == curthread, ("non-current thread %p", td));
327 	p = td->td_proc;
328 	if ((p->p_flag & P_HADTHREADS) != 0) {
329 		PROC_LOCK(p);
330 		/*
331 		 * If success, we upgrade to SINGLE_EXIT state to
332 		 * force other threads to suicide.
333 		 */
334 		if (error == EJUSTRETURN)
335 			thread_single(p, SINGLE_EXIT);
336 		else
337 			thread_single_end(p, SINGLE_BOUNDARY);
338 		PROC_UNLOCK(p);
339 	}
340 	exec_cleanup(td, oldvmspace);
341 }
342 
343 /*
344  * kern_execve() has the astonishing property of not always returning to
345  * the caller.  If sufficiently bad things happen during the call to
346  * do_execve(), it can end up calling exit1(); as a result, callers must
347  * avoid doing anything which they might need to undo (e.g., allocating
348  * memory).
349  */
350 int
351 kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p,
352     struct vmspace *oldvmspace)
353 {
354 
355 	AUDIT_ARG_ARGV(args->begin_argv, args->argc,
356 	    exec_args_get_begin_envv(args) - args->begin_argv);
357 	AUDIT_ARG_ENVV(exec_args_get_begin_envv(args), args->envc,
358 	    args->endp - exec_args_get_begin_envv(args));
359 	return (do_execve(td, args, mac_p, oldvmspace));
360 }
361 
362 static void
363 execve_nosetid(struct image_params *imgp)
364 {
365 	imgp->credential_setid = false;
366 	if (imgp->newcred != NULL) {
367 		crfree(imgp->newcred);
368 		imgp->newcred = NULL;
369 	}
370 }
371 
372 /*
373  * In-kernel implementation of execve().  All arguments are assumed to be
374  * userspace pointers from the passed thread.
375  */
376 static int
377 do_execve(struct thread *td, struct image_args *args, struct mac *mac_p,
378     struct vmspace *oldvmspace)
379 {
380 	struct proc *p = td->td_proc;
381 	struct nameidata nd;
382 	struct ucred *oldcred;
383 	struct uidinfo *euip = NULL;
384 	uintptr_t stack_base;
385 	struct image_params image_params, *imgp;
386 	struct vattr attr;
387 	int (*img_first)(struct image_params *);
388 	struct pargs *oldargs = NULL, *newargs = NULL;
389 	struct sigacts *oldsigacts = NULL, *newsigacts = NULL;
390 #ifdef KTRACE
391 	struct ktr_io_params *kiop;
392 #endif
393 	struct vnode *oldtextvp = NULL, *newtextvp;
394 	int credential_changing;
395 #ifdef MAC
396 	struct label *interpvplabel = NULL;
397 	int will_transition;
398 #endif
399 #ifdef HWPMC_HOOKS
400 	struct pmckern_procexec pe;
401 #endif
402 	int error, i, orig_osrel;
403 	uint32_t orig_fctl0;
404 	Elf_Brandinfo *orig_brandinfo;
405 	static const char fexecv_proc_title[] = "(fexecv)";
406 
407 	imgp = &image_params;
408 #ifdef KTRACE
409 	kiop = NULL;
410 #endif
411 
412 	/*
413 	 * Lock the process and set the P_INEXEC flag to indicate that
414 	 * it should be left alone until we're done here.  This is
415 	 * necessary to avoid race conditions - e.g. in ptrace() -
416 	 * that might allow a local user to illicitly obtain elevated
417 	 * privileges.
418 	 */
419 	PROC_LOCK(p);
420 	KASSERT((p->p_flag & P_INEXEC) == 0,
421 	    ("%s(): process already has P_INEXEC flag", __func__));
422 	p->p_flag |= P_INEXEC;
423 	PROC_UNLOCK(p);
424 
425 	/*
426 	 * Initialize part of the common data
427 	 */
428 	bzero(imgp, sizeof(*imgp));
429 	imgp->proc = p;
430 	imgp->attr = &attr;
431 	imgp->args = args;
432 	oldcred = p->p_ucred;
433 	orig_osrel = p->p_osrel;
434 	orig_fctl0 = p->p_fctl0;
435 	orig_brandinfo = p->p_elf_brandinfo;
436 
437 #ifdef MAC
438 	error = mac_execve_enter(imgp, mac_p);
439 	if (error)
440 		goto exec_fail;
441 #endif
442 
443 	/*
444 	 * Translate the file name. namei() returns a vnode pointer
445 	 *	in ni_vp among other things.
446 	 *
447 	 * XXXAUDIT: It would be desirable to also audit the name of the
448 	 * interpreter if this is an interpreted binary.
449 	 */
450 	if (args->fname != NULL) {
451 		NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW |
452 		    SAVENAME | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
453 	}
454 
455 	SDT_PROBE1(proc, , , exec, args->fname);
456 
457 interpret:
458 	if (args->fname != NULL) {
459 #ifdef CAPABILITY_MODE
460 		/*
461 		 * While capability mode can't reach this point via direct
462 		 * path arguments to execve(), we also don't allow
463 		 * interpreters to be used in capability mode (for now).
464 		 * Catch indirect lookups and return a permissions error.
465 		 */
466 		if (IN_CAPABILITY_MODE(td)) {
467 			error = ECAPMODE;
468 			goto exec_fail;
469 		}
470 #endif
471 		error = namei(&nd);
472 		if (error)
473 			goto exec_fail;
474 
475 		newtextvp = nd.ni_vp;
476 		imgp->vp = newtextvp;
477 	} else {
478 		AUDIT_ARG_FD(args->fd);
479 		/*
480 		 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
481 		 */
482 		error = fgetvp_exec(td, args->fd, &cap_fexecve_rights, &newtextvp);
483 		if (error)
484 			goto exec_fail;
485 		vn_lock(newtextvp, LK_SHARED | LK_RETRY);
486 		AUDIT_ARG_VNODE1(newtextvp);
487 		imgp->vp = newtextvp;
488 	}
489 
490 	/*
491 	 * Check file permissions.  Also 'opens' file and sets its vnode to
492 	 * text mode.
493 	 */
494 	error = exec_check_permissions(imgp);
495 	if (error)
496 		goto exec_fail_dealloc;
497 
498 	imgp->object = imgp->vp->v_object;
499 	if (imgp->object != NULL)
500 		vm_object_reference(imgp->object);
501 
502 	error = exec_map_first_page(imgp);
503 	if (error)
504 		goto exec_fail_dealloc;
505 
506 	imgp->proc->p_osrel = 0;
507 	imgp->proc->p_fctl0 = 0;
508 	imgp->proc->p_elf_brandinfo = NULL;
509 
510 	/*
511 	 * Implement image setuid/setgid.
512 	 *
513 	 * Determine new credentials before attempting image activators
514 	 * so that it can be used by process_exec handlers to determine
515 	 * credential/setid changes.
516 	 *
517 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
518 	 * the process is being traced.
519 	 *
520 	 * We disable setuid/setgid/etc in capability mode on the basis
521 	 * that most setugid applications are not written with that
522 	 * environment in mind, and will therefore almost certainly operate
523 	 * incorrectly. In principle there's no reason that setugid
524 	 * applications might not be useful in capability mode, so we may want
525 	 * to reconsider this conservative design choice in the future.
526 	 *
527 	 * XXXMAC: For the time being, use NOSUID to also prohibit
528 	 * transitions on the file system.
529 	 */
530 	credential_changing = 0;
531 	credential_changing |= (attr.va_mode & S_ISUID) &&
532 	    oldcred->cr_uid != attr.va_uid;
533 	credential_changing |= (attr.va_mode & S_ISGID) &&
534 	    oldcred->cr_gid != attr.va_gid;
535 #ifdef MAC
536 	will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
537 	    interpvplabel, imgp);
538 	credential_changing |= will_transition;
539 #endif
540 
541 	/* Don't inherit PROC_PDEATHSIG_CTL value if setuid/setgid. */
542 	if (credential_changing)
543 		imgp->proc->p_pdeathsig = 0;
544 
545 	if (credential_changing &&
546 #ifdef CAPABILITY_MODE
547 	    ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
548 #endif
549 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
550 	    (p->p_flag & P_TRACED) == 0) {
551 		imgp->credential_setid = true;
552 		VOP_UNLOCK(imgp->vp);
553 		imgp->newcred = crdup(oldcred);
554 		if (attr.va_mode & S_ISUID) {
555 			euip = uifind(attr.va_uid);
556 			change_euid(imgp->newcred, euip);
557 		}
558 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
559 		if (attr.va_mode & S_ISGID)
560 			change_egid(imgp->newcred, attr.va_gid);
561 		/*
562 		 * Implement correct POSIX saved-id behavior.
563 		 *
564 		 * XXXMAC: Note that the current logic will save the
565 		 * uid and gid if a MAC domain transition occurs, even
566 		 * though maybe it shouldn't.
567 		 */
568 		change_svuid(imgp->newcred, imgp->newcred->cr_uid);
569 		change_svgid(imgp->newcred, imgp->newcred->cr_gid);
570 	} else {
571 		/*
572 		 * Implement correct POSIX saved-id behavior.
573 		 *
574 		 * XXX: It's not clear that the existing behavior is
575 		 * POSIX-compliant.  A number of sources indicate that the
576 		 * saved uid/gid should only be updated if the new ruid is
577 		 * not equal to the old ruid, or the new euid is not equal
578 		 * to the old euid and the new euid is not equal to the old
579 		 * ruid.  The FreeBSD code always updates the saved uid/gid.
580 		 * Also, this code uses the new (replaced) euid and egid as
581 		 * the source, which may or may not be the right ones to use.
582 		 */
583 		if (oldcred->cr_svuid != oldcred->cr_uid ||
584 		    oldcred->cr_svgid != oldcred->cr_gid) {
585 			VOP_UNLOCK(imgp->vp);
586 			imgp->newcred = crdup(oldcred);
587 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
588 			change_svuid(imgp->newcred, imgp->newcred->cr_uid);
589 			change_svgid(imgp->newcred, imgp->newcred->cr_gid);
590 		}
591 	}
592 	/* The new credentials are installed into the process later. */
593 
594 	/*
595 	 * Do the best to calculate the full path to the image file.
596 	 */
597 	if (args->fname != NULL && args->fname[0] == '/')
598 		imgp->execpath = args->fname;
599 	else {
600 		VOP_UNLOCK(imgp->vp);
601 		if (vn_fullpath(imgp->vp, &imgp->execpath, &imgp->freepath) != 0)
602 			imgp->execpath = args->fname;
603 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
604 	}
605 
606 	/*
607 	 *	If the current process has a special image activator it
608 	 *	wants to try first, call it.   For example, emulating shell
609 	 *	scripts differently.
610 	 */
611 	error = -1;
612 	if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
613 		error = img_first(imgp);
614 
615 	/*
616 	 *	Loop through the list of image activators, calling each one.
617 	 *	An activator returns -1 if there is no match, 0 on success,
618 	 *	and an error otherwise.
619 	 */
620 	for (i = 0; error == -1 && execsw[i]; ++i) {
621 		if (execsw[i]->ex_imgact == NULL ||
622 		    execsw[i]->ex_imgact == img_first) {
623 			continue;
624 		}
625 		error = (*execsw[i]->ex_imgact)(imgp);
626 	}
627 
628 	if (error) {
629 		if (error == -1)
630 			error = ENOEXEC;
631 		goto exec_fail_dealloc;
632 	}
633 
634 	/*
635 	 * Special interpreter operation, cleanup and loop up to try to
636 	 * activate the interpreter.
637 	 */
638 	if (imgp->interpreted) {
639 		exec_unmap_first_page(imgp);
640 		/*
641 		 * The text reference needs to be removed for scripts.
642 		 * There is a short period before we determine that
643 		 * something is a script where text reference is active.
644 		 * The vnode lock is held over this entire period
645 		 * so nothing should illegitimately be blocked.
646 		 */
647 		MPASS(imgp->textset);
648 		VOP_UNSET_TEXT_CHECKED(newtextvp);
649 		imgp->textset = false;
650 		/* free name buffer and old vnode */
651 		if (args->fname != NULL)
652 			NDFREE(&nd, NDF_ONLY_PNBUF);
653 #ifdef MAC
654 		mac_execve_interpreter_enter(newtextvp, &interpvplabel);
655 #endif
656 		if (imgp->opened) {
657 			VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td);
658 			imgp->opened = 0;
659 		}
660 		vput(newtextvp);
661 		vm_object_deallocate(imgp->object);
662 		imgp->object = NULL;
663 		execve_nosetid(imgp);
664 		imgp->execpath = NULL;
665 		free(imgp->freepath, M_TEMP);
666 		imgp->freepath = NULL;
667 		/* set new name to that of the interpreter */
668 		NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW |
669 		    SAVENAME, UIO_SYSSPACE, imgp->interpreter_name, td);
670 		args->fname = imgp->interpreter_name;
671 		goto interpret;
672 	}
673 
674 	/*
675 	 * NB: We unlock the vnode here because it is believed that none
676 	 * of the sv_copyout_strings/sv_fixup operations require the vnode.
677 	 */
678 	VOP_UNLOCK(imgp->vp);
679 
680 	if (disallow_high_osrel &&
681 	    P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
682 		error = ENOEXEC;
683 		uprintf("Osrel %d for image %s too high\n", p->p_osrel,
684 		    imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
685 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
686 		goto exec_fail_dealloc;
687 	}
688 
689 	/*
690 	 * Copy out strings (args and env) and initialize stack base.
691 	 */
692 	error = (*p->p_sysent->sv_copyout_strings)(imgp, &stack_base);
693 	if (error != 0) {
694 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
695 		goto exec_fail_dealloc;
696 	}
697 
698 	/*
699 	 * Stack setup.
700 	 */
701 	error = (*p->p_sysent->sv_fixup)(&stack_base, imgp);
702 	if (error != 0) {
703 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
704 		goto exec_fail_dealloc;
705 	}
706 
707 	/*
708 	 * For security and other reasons, the file descriptor table cannot be
709 	 * shared after an exec.
710 	 */
711 	fdunshare(td);
712 	pdunshare(td);
713 	/* close files on exec */
714 	fdcloseexec(td);
715 
716 	/*
717 	 * Malloc things before we need locks.
718 	 */
719 	i = exec_args_get_begin_envv(imgp->args) - imgp->args->begin_argv;
720 	/* Cache arguments if they fit inside our allowance */
721 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
722 		newargs = pargs_alloc(i);
723 		bcopy(imgp->args->begin_argv, newargs->ar_args, i);
724 	}
725 
726 	/*
727 	 * For security and other reasons, signal handlers cannot
728 	 * be shared after an exec. The new process gets a copy of the old
729 	 * handlers. In execsigs(), the new process will have its signals
730 	 * reset.
731 	 */
732 	if (sigacts_shared(p->p_sigacts)) {
733 		oldsigacts = p->p_sigacts;
734 		newsigacts = sigacts_alloc();
735 		sigacts_copy(newsigacts, oldsigacts);
736 	}
737 
738 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
739 
740 	PROC_LOCK(p);
741 	if (oldsigacts)
742 		p->p_sigacts = newsigacts;
743 	/* Stop profiling */
744 	stopprofclock(p);
745 
746 	/* reset caught signals */
747 	execsigs(p);
748 
749 	/* name this process - nameiexec(p, ndp) */
750 	bzero(p->p_comm, sizeof(p->p_comm));
751 	if (args->fname)
752 		bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
753 		    min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
754 	else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0)
755 		bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
756 	bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
757 #ifdef KTR
758 	sched_clear_tdname(td);
759 #endif
760 
761 	/*
762 	 * mark as execed, wakeup the process that vforked (if any) and tell
763 	 * it that it now has its own resources back
764 	 */
765 	p->p_flag |= P_EXEC;
766 	if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0)
767 		p->p_flag2 &= ~P2_NOTRACE;
768 	if ((p->p_flag2 & P2_STKGAP_DISABLE_EXEC) == 0)
769 		p->p_flag2 &= ~P2_STKGAP_DISABLE;
770 	if (p->p_flag & P_PPWAIT) {
771 		p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
772 		cv_broadcast(&p->p_pwait);
773 		/* STOPs are no longer ignored, arrange for AST */
774 		signotify(td);
775 	}
776 
777 	if ((imgp->sysent->sv_setid_allowed != NULL &&
778 	    !(*imgp->sysent->sv_setid_allowed)(td, imgp)) ||
779 	    (p->p_flag2 & P2_NO_NEW_PRIVS) != 0)
780 		execve_nosetid(imgp);
781 
782 	/*
783 	 * Implement image setuid/setgid installation.
784 	 */
785 	if (imgp->credential_setid) {
786 		/*
787 		 * Turn off syscall tracing for set-id programs, except for
788 		 * root.  Record any set-id flags first to make sure that
789 		 * we do not regain any tracing during a possible block.
790 		 */
791 		setsugid(p);
792 #ifdef KTRACE
793 		kiop = ktrprocexec(p);
794 #endif
795 		/*
796 		 * Close any file descriptors 0..2 that reference procfs,
797 		 * then make sure file descriptors 0..2 are in use.
798 		 *
799 		 * Both fdsetugidsafety() and fdcheckstd() may call functions
800 		 * taking sleepable locks, so temporarily drop our locks.
801 		 */
802 		PROC_UNLOCK(p);
803 		VOP_UNLOCK(imgp->vp);
804 		fdsetugidsafety(td);
805 		error = fdcheckstd(td);
806 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
807 		if (error != 0)
808 			goto exec_fail_dealloc;
809 		PROC_LOCK(p);
810 #ifdef MAC
811 		if (will_transition) {
812 			mac_vnode_execve_transition(oldcred, imgp->newcred,
813 			    imgp->vp, interpvplabel, imgp);
814 		}
815 #endif
816 	} else {
817 		if (oldcred->cr_uid == oldcred->cr_ruid &&
818 		    oldcred->cr_gid == oldcred->cr_rgid)
819 			p->p_flag &= ~P_SUGID;
820 	}
821 	/*
822 	 * Set the new credentials.
823 	 */
824 	if (imgp->newcred != NULL) {
825 		proc_set_cred(p, imgp->newcred);
826 		crfree(oldcred);
827 		oldcred = NULL;
828 	}
829 
830 	/*
831 	 * Store the vp for use in procfs.  This vnode was referenced by namei
832 	 * or fgetvp_exec.
833 	 */
834 	oldtextvp = p->p_textvp;
835 	p->p_textvp = newtextvp;
836 
837 #ifdef KDTRACE_HOOKS
838 	/*
839 	 * Tell the DTrace fasttrap provider about the exec if it
840 	 * has declared an interest.
841 	 */
842 	if (dtrace_fasttrap_exec)
843 		dtrace_fasttrap_exec(p);
844 #endif
845 
846 	/*
847 	 * Notify others that we exec'd, and clear the P_INEXEC flag
848 	 * as we're now a bona fide freshly-execed process.
849 	 */
850 	KNOTE_LOCKED(p->p_klist, NOTE_EXEC);
851 	p->p_flag &= ~P_INEXEC;
852 
853 	/* clear "fork but no exec" flag, as we _are_ execing */
854 	p->p_acflag &= ~AFORK;
855 
856 	/*
857 	 * Free any previous argument cache and replace it with
858 	 * the new argument cache, if any.
859 	 */
860 	oldargs = p->p_args;
861 	p->p_args = newargs;
862 	newargs = NULL;
863 
864 	PROC_UNLOCK(p);
865 
866 #ifdef	HWPMC_HOOKS
867 	/*
868 	 * Check if system-wide sampling is in effect or if the
869 	 * current process is using PMCs.  If so, do exec() time
870 	 * processing.  This processing needs to happen AFTER the
871 	 * P_INEXEC flag is cleared.
872 	 */
873 	if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
874 		VOP_UNLOCK(imgp->vp);
875 		pe.pm_credentialschanged = credential_changing;
876 		pe.pm_entryaddr = imgp->entry_addr;
877 
878 		PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
879 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
880 	}
881 #endif
882 
883 	/* Set values passed into the program in registers. */
884 	(*p->p_sysent->sv_setregs)(td, imgp, stack_base);
885 
886 	VOP_MMAPPED(imgp->vp);
887 
888 	SDT_PROBE1(proc, , , exec__success, args->fname);
889 
890 exec_fail_dealloc:
891 	if (error != 0) {
892 		p->p_osrel = orig_osrel;
893 		p->p_fctl0 = orig_fctl0;
894 		p->p_elf_brandinfo = orig_brandinfo;
895 	}
896 
897 	if (imgp->firstpage != NULL)
898 		exec_unmap_first_page(imgp);
899 
900 	if (imgp->vp != NULL) {
901 		if (args->fname)
902 			NDFREE(&nd, NDF_ONLY_PNBUF);
903 		if (imgp->opened)
904 			VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
905 		if (imgp->textset)
906 			VOP_UNSET_TEXT_CHECKED(imgp->vp);
907 		if (error != 0)
908 			vput(imgp->vp);
909 		else
910 			VOP_UNLOCK(imgp->vp);
911 	}
912 
913 	if (imgp->object != NULL)
914 		vm_object_deallocate(imgp->object);
915 
916 	free(imgp->freepath, M_TEMP);
917 
918 	if (error == 0) {
919 		if (p->p_ptevents & PTRACE_EXEC) {
920 			PROC_LOCK(p);
921 			if (p->p_ptevents & PTRACE_EXEC)
922 				td->td_dbgflags |= TDB_EXEC;
923 			PROC_UNLOCK(p);
924 		}
925 	} else {
926 exec_fail:
927 		/* we're done here, clear P_INEXEC */
928 		PROC_LOCK(p);
929 		p->p_flag &= ~P_INEXEC;
930 		PROC_UNLOCK(p);
931 
932 		SDT_PROBE1(proc, , , exec__failure, error);
933 	}
934 
935 	if (imgp->newcred != NULL && oldcred != NULL)
936 		crfree(imgp->newcred);
937 
938 #ifdef MAC
939 	mac_execve_exit(imgp);
940 	mac_execve_interpreter_exit(interpvplabel);
941 #endif
942 	exec_free_args(args);
943 
944 	/*
945 	 * Handle deferred decrement of ref counts.
946 	 */
947 	if (oldtextvp != NULL)
948 		vrele(oldtextvp);
949 #ifdef KTRACE
950 	ktr_io_params_free(kiop);
951 #endif
952 	pargs_drop(oldargs);
953 	pargs_drop(newargs);
954 	if (oldsigacts != NULL)
955 		sigacts_free(oldsigacts);
956 	if (euip != NULL)
957 		uifree(euip);
958 
959 	if (error && imgp->vmspace_destroyed) {
960 		/* sorry, no more process anymore. exit gracefully */
961 		exec_cleanup(td, oldvmspace);
962 		exit1(td, 0, SIGABRT);
963 		/* NOT REACHED */
964 	}
965 
966 #ifdef KTRACE
967 	if (error == 0)
968 		ktrprocctor(p);
969 #endif
970 
971 	/*
972 	 * We don't want cpu_set_syscall_retval() to overwrite any of
973 	 * the register values put in place by exec_setregs().
974 	 * Implementations of cpu_set_syscall_retval() will leave
975 	 * registers unmodified when returning EJUSTRETURN.
976 	 */
977 	return (error == 0 ? EJUSTRETURN : error);
978 }
979 
980 void
981 exec_cleanup(struct thread *td, struct vmspace *oldvmspace)
982 {
983 	if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
984 		KASSERT(td->td_proc->p_vmspace != oldvmspace,
985 		    ("oldvmspace still used"));
986 		vmspace_free(oldvmspace);
987 		td->td_pflags &= ~TDP_EXECVMSPC;
988 	}
989 }
990 
991 int
992 exec_map_first_page(struct image_params *imgp)
993 {
994 	vm_object_t object;
995 	vm_page_t m;
996 	int error;
997 
998 	if (imgp->firstpage != NULL)
999 		exec_unmap_first_page(imgp);
1000 
1001 	object = imgp->vp->v_object;
1002 	if (object == NULL)
1003 		return (EACCES);
1004 #if VM_NRESERVLEVEL > 0
1005 	if ((object->flags & OBJ_COLORED) == 0) {
1006 		VM_OBJECT_WLOCK(object);
1007 		vm_object_color(object, 0);
1008 		VM_OBJECT_WUNLOCK(object);
1009 	}
1010 #endif
1011 	error = vm_page_grab_valid_unlocked(&m, object, 0,
1012 	    VM_ALLOC_COUNT(VM_INITIAL_PAGEIN) |
1013             VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED);
1014 
1015 	if (error != VM_PAGER_OK)
1016 		return (EIO);
1017 	imgp->firstpage = sf_buf_alloc(m, 0);
1018 	imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
1019 
1020 	return (0);
1021 }
1022 
1023 void
1024 exec_unmap_first_page(struct image_params *imgp)
1025 {
1026 	vm_page_t m;
1027 
1028 	if (imgp->firstpage != NULL) {
1029 		m = sf_buf_page(imgp->firstpage);
1030 		sf_buf_free(imgp->firstpage);
1031 		imgp->firstpage = NULL;
1032 		vm_page_unwire(m, PQ_ACTIVE);
1033 	}
1034 }
1035 
1036 void
1037 exec_onexec_old(struct thread *td)
1038 {
1039 	sigfastblock_clear(td);
1040 	umtx_exec(td->td_proc);
1041 }
1042 
1043 /*
1044  * Destroy old address space, and allocate a new stack.
1045  *	The new stack is only sgrowsiz large because it is grown
1046  *	automatically on a page fault.
1047  */
1048 int
1049 exec_new_vmspace(struct image_params *imgp, struct sysentvec *sv)
1050 {
1051 	int error;
1052 	struct proc *p = imgp->proc;
1053 	struct vmspace *vmspace = p->p_vmspace;
1054 	struct thread *td = curthread;
1055 	vm_object_t obj;
1056 	struct rlimit rlim_stack;
1057 	vm_offset_t sv_minuser, stack_addr;
1058 	vm_map_t map;
1059 	vm_prot_t stack_prot;
1060 	u_long ssiz;
1061 
1062 	imgp->vmspace_destroyed = 1;
1063 	imgp->sysent = sv;
1064 
1065 	if (p->p_sysent->sv_onexec_old != NULL)
1066 		p->p_sysent->sv_onexec_old(td);
1067 	itimers_exec(p);
1068 
1069 	EVENTHANDLER_DIRECT_INVOKE(process_exec, p, imgp);
1070 
1071 	/*
1072 	 * Blow away entire process VM, if address space not shared,
1073 	 * otherwise, create a new VM space so that other threads are
1074 	 * not disrupted
1075 	 */
1076 	map = &vmspace->vm_map;
1077 	if (map_at_zero)
1078 		sv_minuser = sv->sv_minuser;
1079 	else
1080 		sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1081 	if (refcount_load(&vmspace->vm_refcnt) == 1 &&
1082 	    vm_map_min(map) == sv_minuser &&
1083 	    vm_map_max(map) == sv->sv_maxuser &&
1084 	    cpu_exec_vmspace_reuse(p, map)) {
1085 		shmexit(vmspace);
1086 		pmap_remove_pages(vmspace_pmap(vmspace));
1087 		vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1088 		/*
1089 		 * An exec terminates mlockall(MCL_FUTURE).
1090 		 * ASLR and W^X states must be re-evaluated.
1091 		 */
1092 		vm_map_lock(map);
1093 		vm_map_modflags(map, 0, MAP_WIREFUTURE | MAP_ASLR |
1094 		    MAP_ASLR_IGNSTART | MAP_WXORX);
1095 		vm_map_unlock(map);
1096 	} else {
1097 		error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1098 		if (error)
1099 			return (error);
1100 		vmspace = p->p_vmspace;
1101 		map = &vmspace->vm_map;
1102 	}
1103 	map->flags |= imgp->map_flags;
1104 
1105 	/* Map a shared page */
1106 	obj = sv->sv_shared_page_obj;
1107 	if (obj != NULL) {
1108 		vm_object_reference(obj);
1109 		error = vm_map_fixed(map, obj, 0,
1110 		    sv->sv_shared_page_base, sv->sv_shared_page_len,
1111 		    VM_PROT_READ | VM_PROT_EXECUTE,
1112 		    VM_PROT_READ | VM_PROT_EXECUTE,
1113 		    MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1114 		if (error != KERN_SUCCESS) {
1115 			vm_object_deallocate(obj);
1116 			return (vm_mmap_to_errno(error));
1117 		}
1118 	}
1119 
1120 	/* Allocate a new stack */
1121 	if (imgp->stack_sz != 0) {
1122 		ssiz = trunc_page(imgp->stack_sz);
1123 		PROC_LOCK(p);
1124 		lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack);
1125 		PROC_UNLOCK(p);
1126 		if (ssiz > rlim_stack.rlim_max)
1127 			ssiz = rlim_stack.rlim_max;
1128 		if (ssiz > rlim_stack.rlim_cur) {
1129 			rlim_stack.rlim_cur = ssiz;
1130 			kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack);
1131 		}
1132 	} else if (sv->sv_maxssiz != NULL) {
1133 		ssiz = *sv->sv_maxssiz;
1134 	} else {
1135 		ssiz = maxssiz;
1136 	}
1137 	imgp->eff_stack_sz = lim_cur(curthread, RLIMIT_STACK);
1138 	if (ssiz < imgp->eff_stack_sz)
1139 		imgp->eff_stack_sz = ssiz;
1140 	stack_addr = sv->sv_usrstack - ssiz;
1141 	stack_prot = obj != NULL && imgp->stack_prot != 0 ?
1142 	    imgp->stack_prot : sv->sv_stackprot;
1143 	error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz, stack_prot,
1144 	    VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1145 	if (error != KERN_SUCCESS) {
1146 		uprintf("exec_new_vmspace: mapping stack size %#jx prot %#x "
1147 		    "failed mach error %d errno %d\n", (uintmax_t)ssiz,
1148 		    stack_prot, error, vm_mmap_to_errno(error));
1149 		return (vm_mmap_to_errno(error));
1150 	}
1151 
1152 	/*
1153 	 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they
1154 	 * are still used to enforce the stack rlimit on the process stack.
1155 	 */
1156 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1157 	vmspace->vm_maxsaddr = (char *)stack_addr;
1158 
1159 	return (sv->sv_onexec != NULL ? sv->sv_onexec(p, imgp) : 0);
1160 }
1161 
1162 /*
1163  * Copy out argument and environment strings from the old process address
1164  * space into the temporary string buffer.
1165  */
1166 int
1167 exec_copyin_args(struct image_args *args, const char *fname,
1168     enum uio_seg segflg, char **argv, char **envv)
1169 {
1170 	u_long arg, env;
1171 	int error;
1172 
1173 	bzero(args, sizeof(*args));
1174 	if (argv == NULL)
1175 		return (EFAULT);
1176 
1177 	/*
1178 	 * Allocate demand-paged memory for the file name, argument, and
1179 	 * environment strings.
1180 	 */
1181 	error = exec_alloc_args(args);
1182 	if (error != 0)
1183 		return (error);
1184 
1185 	/*
1186 	 * Copy the file name.
1187 	 */
1188 	error = exec_args_add_fname(args, fname, segflg);
1189 	if (error != 0)
1190 		goto err_exit;
1191 
1192 	/*
1193 	 * extract arguments first
1194 	 */
1195 	for (;;) {
1196 		error = fueword(argv++, &arg);
1197 		if (error == -1) {
1198 			error = EFAULT;
1199 			goto err_exit;
1200 		}
1201 		if (arg == 0)
1202 			break;
1203 		error = exec_args_add_arg(args, (char *)(uintptr_t)arg,
1204 		    UIO_USERSPACE);
1205 		if (error != 0)
1206 			goto err_exit;
1207 	}
1208 
1209 	/*
1210 	 * extract environment strings
1211 	 */
1212 	if (envv) {
1213 		for (;;) {
1214 			error = fueword(envv++, &env);
1215 			if (error == -1) {
1216 				error = EFAULT;
1217 				goto err_exit;
1218 			}
1219 			if (env == 0)
1220 				break;
1221 			error = exec_args_add_env(args,
1222 			    (char *)(uintptr_t)env, UIO_USERSPACE);
1223 			if (error != 0)
1224 				goto err_exit;
1225 		}
1226 	}
1227 
1228 	return (0);
1229 
1230 err_exit:
1231 	exec_free_args(args);
1232 	return (error);
1233 }
1234 
1235 struct exec_args_kva {
1236 	vm_offset_t addr;
1237 	u_int gen;
1238 	SLIST_ENTRY(exec_args_kva) next;
1239 };
1240 
1241 DPCPU_DEFINE_STATIC(struct exec_args_kva *, exec_args_kva);
1242 
1243 static SLIST_HEAD(, exec_args_kva) exec_args_kva_freelist;
1244 static struct mtx exec_args_kva_mtx;
1245 static u_int exec_args_gen;
1246 
1247 static void
1248 exec_prealloc_args_kva(void *arg __unused)
1249 {
1250 	struct exec_args_kva *argkva;
1251 	u_int i;
1252 
1253 	SLIST_INIT(&exec_args_kva_freelist);
1254 	mtx_init(&exec_args_kva_mtx, "exec args kva", NULL, MTX_DEF);
1255 	for (i = 0; i < exec_map_entries; i++) {
1256 		argkva = malloc(sizeof(*argkva), M_PARGS, M_WAITOK);
1257 		argkva->addr = kmap_alloc_wait(exec_map, exec_map_entry_size);
1258 		argkva->gen = exec_args_gen;
1259 		SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1260 	}
1261 }
1262 SYSINIT(exec_args_kva, SI_SUB_EXEC, SI_ORDER_ANY, exec_prealloc_args_kva, NULL);
1263 
1264 static vm_offset_t
1265 exec_alloc_args_kva(void **cookie)
1266 {
1267 	struct exec_args_kva *argkva;
1268 
1269 	argkva = (void *)atomic_readandclear_ptr(
1270 	    (uintptr_t *)DPCPU_PTR(exec_args_kva));
1271 	if (argkva == NULL) {
1272 		mtx_lock(&exec_args_kva_mtx);
1273 		while ((argkva = SLIST_FIRST(&exec_args_kva_freelist)) == NULL)
1274 			(void)mtx_sleep(&exec_args_kva_freelist,
1275 			    &exec_args_kva_mtx, 0, "execkva", 0);
1276 		SLIST_REMOVE_HEAD(&exec_args_kva_freelist, next);
1277 		mtx_unlock(&exec_args_kva_mtx);
1278 	}
1279 	kasan_mark((void *)argkva->addr, exec_map_entry_size,
1280 	    exec_map_entry_size, 0);
1281 	*(struct exec_args_kva **)cookie = argkva;
1282 	return (argkva->addr);
1283 }
1284 
1285 static void
1286 exec_release_args_kva(struct exec_args_kva *argkva, u_int gen)
1287 {
1288 	vm_offset_t base;
1289 
1290 	base = argkva->addr;
1291 	kasan_mark((void *)argkva->addr, 0, exec_map_entry_size,
1292 	    KASAN_EXEC_ARGS_FREED);
1293 	if (argkva->gen != gen) {
1294 		(void)vm_map_madvise(exec_map, base, base + exec_map_entry_size,
1295 		    MADV_FREE);
1296 		argkva->gen = gen;
1297 	}
1298 	if (!atomic_cmpset_ptr((uintptr_t *)DPCPU_PTR(exec_args_kva),
1299 	    (uintptr_t)NULL, (uintptr_t)argkva)) {
1300 		mtx_lock(&exec_args_kva_mtx);
1301 		SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1302 		wakeup_one(&exec_args_kva_freelist);
1303 		mtx_unlock(&exec_args_kva_mtx);
1304 	}
1305 }
1306 
1307 static void
1308 exec_free_args_kva(void *cookie)
1309 {
1310 
1311 	exec_release_args_kva(cookie, exec_args_gen);
1312 }
1313 
1314 static void
1315 exec_args_kva_lowmem(void *arg __unused)
1316 {
1317 	SLIST_HEAD(, exec_args_kva) head;
1318 	struct exec_args_kva *argkva;
1319 	u_int gen;
1320 	int i;
1321 
1322 	gen = atomic_fetchadd_int(&exec_args_gen, 1) + 1;
1323 
1324 	/*
1325 	 * Force an madvise of each KVA range. Any currently allocated ranges
1326 	 * will have MADV_FREE applied once they are freed.
1327 	 */
1328 	SLIST_INIT(&head);
1329 	mtx_lock(&exec_args_kva_mtx);
1330 	SLIST_SWAP(&head, &exec_args_kva_freelist, exec_args_kva);
1331 	mtx_unlock(&exec_args_kva_mtx);
1332 	while ((argkva = SLIST_FIRST(&head)) != NULL) {
1333 		SLIST_REMOVE_HEAD(&head, next);
1334 		exec_release_args_kva(argkva, gen);
1335 	}
1336 
1337 	CPU_FOREACH(i) {
1338 		argkva = (void *)atomic_readandclear_ptr(
1339 		    (uintptr_t *)DPCPU_ID_PTR(i, exec_args_kva));
1340 		if (argkva != NULL)
1341 			exec_release_args_kva(argkva, gen);
1342 	}
1343 }
1344 EVENTHANDLER_DEFINE(vm_lowmem, exec_args_kva_lowmem, NULL,
1345     EVENTHANDLER_PRI_ANY);
1346 
1347 /*
1348  * Allocate temporary demand-paged, zero-filled memory for the file name,
1349  * argument, and environment strings.
1350  */
1351 int
1352 exec_alloc_args(struct image_args *args)
1353 {
1354 
1355 	args->buf = (char *)exec_alloc_args_kva(&args->bufkva);
1356 	return (0);
1357 }
1358 
1359 void
1360 exec_free_args(struct image_args *args)
1361 {
1362 
1363 	if (args->buf != NULL) {
1364 		exec_free_args_kva(args->bufkva);
1365 		args->buf = NULL;
1366 	}
1367 	if (args->fname_buf != NULL) {
1368 		free(args->fname_buf, M_TEMP);
1369 		args->fname_buf = NULL;
1370 	}
1371 }
1372 
1373 /*
1374  * A set to functions to fill struct image args.
1375  *
1376  * NOTE: exec_args_add_fname() must be called (possibly with a NULL
1377  * fname) before the other functions.  All exec_args_add_arg() calls must
1378  * be made before any exec_args_add_env() calls.  exec_args_adjust_args()
1379  * may be called any time after exec_args_add_fname().
1380  *
1381  * exec_args_add_fname() - install path to be executed
1382  * exec_args_add_arg() - append an argument string
1383  * exec_args_add_env() - append an env string
1384  * exec_args_adjust_args() - adjust location of the argument list to
1385  *                           allow new arguments to be prepended
1386  */
1387 int
1388 exec_args_add_fname(struct image_args *args, const char *fname,
1389     enum uio_seg segflg)
1390 {
1391 	int error;
1392 	size_t length;
1393 
1394 	KASSERT(args->fname == NULL, ("fname already appended"));
1395 	KASSERT(args->endp == NULL, ("already appending to args"));
1396 
1397 	if (fname != NULL) {
1398 		args->fname = args->buf;
1399 		error = segflg == UIO_SYSSPACE ?
1400 		    copystr(fname, args->fname, PATH_MAX, &length) :
1401 		    copyinstr(fname, args->fname, PATH_MAX, &length);
1402 		if (error != 0)
1403 			return (error == ENAMETOOLONG ? E2BIG : error);
1404 	} else
1405 		length = 0;
1406 
1407 	/* Set up for _arg_*()/_env_*() */
1408 	args->endp = args->buf + length;
1409 	/* begin_argv must be set and kept updated */
1410 	args->begin_argv = args->endp;
1411 	KASSERT(exec_map_entry_size - length >= ARG_MAX,
1412 	    ("too little space remaining for arguments %zu < %zu",
1413 	    exec_map_entry_size - length, (size_t)ARG_MAX));
1414 	args->stringspace = ARG_MAX;
1415 
1416 	return (0);
1417 }
1418 
1419 static int
1420 exec_args_add_str(struct image_args *args, const char *str,
1421     enum uio_seg segflg, int *countp)
1422 {
1423 	int error;
1424 	size_t length;
1425 
1426 	KASSERT(args->endp != NULL, ("endp not initialized"));
1427 	KASSERT(args->begin_argv != NULL, ("begin_argp not initialized"));
1428 
1429 	error = (segflg == UIO_SYSSPACE) ?
1430 	    copystr(str, args->endp, args->stringspace, &length) :
1431 	    copyinstr(str, args->endp, args->stringspace, &length);
1432 	if (error != 0)
1433 		return (error == ENAMETOOLONG ? E2BIG : error);
1434 	args->stringspace -= length;
1435 	args->endp += length;
1436 	(*countp)++;
1437 
1438 	return (0);
1439 }
1440 
1441 int
1442 exec_args_add_arg(struct image_args *args, const char *argp,
1443     enum uio_seg segflg)
1444 {
1445 
1446 	KASSERT(args->envc == 0, ("appending args after env"));
1447 
1448 	return (exec_args_add_str(args, argp, segflg, &args->argc));
1449 }
1450 
1451 int
1452 exec_args_add_env(struct image_args *args, const char *envp,
1453     enum uio_seg segflg)
1454 {
1455 
1456 	if (args->envc == 0)
1457 		args->begin_envv = args->endp;
1458 
1459 	return (exec_args_add_str(args, envp, segflg, &args->envc));
1460 }
1461 
1462 int
1463 exec_args_adjust_args(struct image_args *args, size_t consume, ssize_t extend)
1464 {
1465 	ssize_t offset;
1466 
1467 	KASSERT(args->endp != NULL, ("endp not initialized"));
1468 	KASSERT(args->begin_argv != NULL, ("begin_argp not initialized"));
1469 
1470 	offset = extend - consume;
1471 	if (args->stringspace < offset)
1472 		return (E2BIG);
1473 	memmove(args->begin_argv + extend, args->begin_argv + consume,
1474 	    args->endp - args->begin_argv + consume);
1475 	if (args->envc > 0)
1476 		args->begin_envv += offset;
1477 	args->endp += offset;
1478 	args->stringspace -= offset;
1479 	return (0);
1480 }
1481 
1482 char *
1483 exec_args_get_begin_envv(struct image_args *args)
1484 {
1485 
1486 	KASSERT(args->endp != NULL, ("endp not initialized"));
1487 
1488 	if (args->envc > 0)
1489 		return (args->begin_envv);
1490 	return (args->endp);
1491 }
1492 
1493 void
1494 exec_stackgap(struct image_params *imgp, uintptr_t *dp)
1495 {
1496 	if (imgp->sysent->sv_stackgap == NULL ||
1497 	    (imgp->proc->p_fctl0 & (NT_FREEBSD_FCTL_ASLR_DISABLE |
1498 	    NT_FREEBSD_FCTL_ASG_DISABLE)) != 0 ||
1499 	    (imgp->map_flags & MAP_ASLR) == 0)
1500 		return;
1501 	imgp->sysent->sv_stackgap(imgp, dp);
1502 }
1503 
1504 /*
1505  * Copy strings out to the new process address space, constructing new arg
1506  * and env vector tables. Return a pointer to the base so that it can be used
1507  * as the initial stack pointer.
1508  */
1509 int
1510 exec_copyout_strings(struct image_params *imgp, uintptr_t *stack_base)
1511 {
1512 	int argc, envc;
1513 	char **vectp;
1514 	char *stringp;
1515 	uintptr_t destp, ustringp;
1516 	struct ps_strings *arginfo;
1517 	struct proc *p;
1518 	size_t execpath_len;
1519 	int error, szsigcode, szps;
1520 	char canary[sizeof(long) * 8];
1521 
1522 	szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1523 	/*
1524 	 * Calculate string base and vector table pointers.
1525 	 * Also deal with signal trampoline code for this exec type.
1526 	 */
1527 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
1528 		execpath_len = strlen(imgp->execpath) + 1;
1529 	else
1530 		execpath_len = 0;
1531 	p = imgp->proc;
1532 	szsigcode = 0;
1533 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1534 	imgp->ps_strings = arginfo;
1535 	if (p->p_sysent->sv_sigcode_base == 0) {
1536 		if (p->p_sysent->sv_szsigcode != NULL)
1537 			szsigcode = *(p->p_sysent->sv_szsigcode);
1538 	}
1539 	destp =	(uintptr_t)arginfo;
1540 
1541 	/*
1542 	 * install sigcode
1543 	 */
1544 	if (szsigcode != 0) {
1545 		destp -= szsigcode;
1546 		destp = rounddown2(destp, sizeof(void *));
1547 		error = copyout(p->p_sysent->sv_sigcode, (void *)destp,
1548 		    szsigcode);
1549 		if (error != 0)
1550 			return (error);
1551 	}
1552 
1553 	/*
1554 	 * Copy the image path for the rtld.
1555 	 */
1556 	if (execpath_len != 0) {
1557 		destp -= execpath_len;
1558 		destp = rounddown2(destp, sizeof(void *));
1559 		imgp->execpathp = (void *)destp;
1560 		error = copyout(imgp->execpath, imgp->execpathp, execpath_len);
1561 		if (error != 0)
1562 			return (error);
1563 	}
1564 
1565 	/*
1566 	 * Prepare the canary for SSP.
1567 	 */
1568 	arc4rand(canary, sizeof(canary), 0);
1569 	destp -= sizeof(canary);
1570 	imgp->canary = (void *)destp;
1571 	error = copyout(canary, imgp->canary, sizeof(canary));
1572 	if (error != 0)
1573 		return (error);
1574 	imgp->canarylen = sizeof(canary);
1575 
1576 	/*
1577 	 * Prepare the pagesizes array.
1578 	 */
1579 	destp -= szps;
1580 	destp = rounddown2(destp, sizeof(void *));
1581 	imgp->pagesizes = (void *)destp;
1582 	error = copyout(pagesizes, imgp->pagesizes, szps);
1583 	if (error != 0)
1584 		return (error);
1585 	imgp->pagesizeslen = szps;
1586 
1587 	/*
1588 	 * Allocate room for the argument and environment strings.
1589 	 */
1590 	destp -= ARG_MAX - imgp->args->stringspace;
1591 	destp = rounddown2(destp, sizeof(void *));
1592 	ustringp = destp;
1593 
1594 	exec_stackgap(imgp, &destp);
1595 
1596 	if (imgp->auxargs) {
1597 		/*
1598 		 * Allocate room on the stack for the ELF auxargs
1599 		 * array.  It has up to AT_COUNT entries.
1600 		 */
1601 		destp -= AT_COUNT * sizeof(Elf_Auxinfo);
1602 		destp = rounddown2(destp, sizeof(void *));
1603 	}
1604 
1605 	vectp = (char **)destp;
1606 
1607 	/*
1608 	 * Allocate room for the argv[] and env vectors including the
1609 	 * terminating NULL pointers.
1610 	 */
1611 	vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
1612 
1613 	/*
1614 	 * vectp also becomes our initial stack base
1615 	 */
1616 	*stack_base = (uintptr_t)vectp;
1617 
1618 	stringp = imgp->args->begin_argv;
1619 	argc = imgp->args->argc;
1620 	envc = imgp->args->envc;
1621 
1622 	/*
1623 	 * Copy out strings - arguments and environment.
1624 	 */
1625 	error = copyout(stringp, (void *)ustringp,
1626 	    ARG_MAX - imgp->args->stringspace);
1627 	if (error != 0)
1628 		return (error);
1629 
1630 	/*
1631 	 * Fill in "ps_strings" struct for ps, w, etc.
1632 	 */
1633 	imgp->argv = vectp;
1634 	if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 ||
1635 	    suword32(&arginfo->ps_nargvstr, argc) != 0)
1636 		return (EFAULT);
1637 
1638 	/*
1639 	 * Fill in argument portion of vector table.
1640 	 */
1641 	for (; argc > 0; --argc) {
1642 		if (suword(vectp++, ustringp) != 0)
1643 			return (EFAULT);
1644 		while (*stringp++ != 0)
1645 			ustringp++;
1646 		ustringp++;
1647 	}
1648 
1649 	/* a null vector table pointer separates the argp's from the envp's */
1650 	if (suword(vectp++, 0) != 0)
1651 		return (EFAULT);
1652 
1653 	imgp->envv = vectp;
1654 	if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 ||
1655 	    suword32(&arginfo->ps_nenvstr, envc) != 0)
1656 		return (EFAULT);
1657 
1658 	/*
1659 	 * Fill in environment portion of vector table.
1660 	 */
1661 	for (; envc > 0; --envc) {
1662 		if (suword(vectp++, ustringp) != 0)
1663 			return (EFAULT);
1664 		while (*stringp++ != 0)
1665 			ustringp++;
1666 		ustringp++;
1667 	}
1668 
1669 	/* end of vector table is a null pointer */
1670 	if (suword(vectp, 0) != 0)
1671 		return (EFAULT);
1672 
1673 	if (imgp->auxargs) {
1674 		vectp++;
1675 		error = imgp->sysent->sv_copyout_auxargs(imgp,
1676 		    (uintptr_t)vectp);
1677 		if (error != 0)
1678 			return (error);
1679 	}
1680 
1681 	return (0);
1682 }
1683 
1684 /*
1685  * Check permissions of file to execute.
1686  *	Called with imgp->vp locked.
1687  *	Return 0 for success or error code on failure.
1688  */
1689 int
1690 exec_check_permissions(struct image_params *imgp)
1691 {
1692 	struct vnode *vp = imgp->vp;
1693 	struct vattr *attr = imgp->attr;
1694 	struct thread *td;
1695 	int error;
1696 
1697 	td = curthread;
1698 
1699 	/* Get file attributes */
1700 	error = VOP_GETATTR(vp, attr, td->td_ucred);
1701 	if (error)
1702 		return (error);
1703 
1704 #ifdef MAC
1705 	error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1706 	if (error)
1707 		return (error);
1708 #endif
1709 
1710 	/*
1711 	 * 1) Check if file execution is disabled for the filesystem that
1712 	 *    this file resides on.
1713 	 * 2) Ensure that at least one execute bit is on. Otherwise, a
1714 	 *    privileged user will always succeed, and we don't want this
1715 	 *    to happen unless the file really is executable.
1716 	 * 3) Ensure that the file is a regular file.
1717 	 */
1718 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1719 	    (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1720 	    (attr->va_type != VREG))
1721 		return (EACCES);
1722 
1723 	/*
1724 	 * Zero length files can't be exec'd
1725 	 */
1726 	if (attr->va_size == 0)
1727 		return (ENOEXEC);
1728 
1729 	/*
1730 	 *  Check for execute permission to file based on current credentials.
1731 	 */
1732 	error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1733 	if (error)
1734 		return (error);
1735 
1736 	/*
1737 	 * Check number of open-for-writes on the file and deny execution
1738 	 * if there are any.
1739 	 *
1740 	 * Add a text reference now so no one can write to the
1741 	 * executable while we're activating it.
1742 	 *
1743 	 * Remember if this was set before and unset it in case this is not
1744 	 * actually an executable image.
1745 	 */
1746 	error = VOP_SET_TEXT(vp);
1747 	if (error != 0)
1748 		return (error);
1749 	imgp->textset = true;
1750 
1751 	/*
1752 	 * Call filesystem specific open routine (which does nothing in the
1753 	 * general case).
1754 	 */
1755 	error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1756 	if (error == 0)
1757 		imgp->opened = 1;
1758 	return (error);
1759 }
1760 
1761 /*
1762  * Exec handler registration
1763  */
1764 int
1765 exec_register(const struct execsw *execsw_arg)
1766 {
1767 	const struct execsw **es, **xs, **newexecsw;
1768 	u_int count = 2;	/* New slot and trailing NULL */
1769 
1770 	if (execsw)
1771 		for (es = execsw; *es; es++)
1772 			count++;
1773 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1774 	xs = newexecsw;
1775 	if (execsw)
1776 		for (es = execsw; *es; es++)
1777 			*xs++ = *es;
1778 	*xs++ = execsw_arg;
1779 	*xs = NULL;
1780 	if (execsw)
1781 		free(execsw, M_TEMP);
1782 	execsw = newexecsw;
1783 	return (0);
1784 }
1785 
1786 int
1787 exec_unregister(const struct execsw *execsw_arg)
1788 {
1789 	const struct execsw **es, **xs, **newexecsw;
1790 	int count = 1;
1791 
1792 	if (execsw == NULL)
1793 		panic("unregister with no handlers left?\n");
1794 
1795 	for (es = execsw; *es; es++) {
1796 		if (*es == execsw_arg)
1797 			break;
1798 	}
1799 	if (*es == NULL)
1800 		return (ENOENT);
1801 	for (es = execsw; *es; es++)
1802 		if (*es != execsw_arg)
1803 			count++;
1804 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1805 	xs = newexecsw;
1806 	for (es = execsw; *es; es++)
1807 		if (*es != execsw_arg)
1808 			*xs++ = *es;
1809 	*xs = NULL;
1810 	if (execsw)
1811 		free(execsw, M_TEMP);
1812 	execsw = newexecsw;
1813 	return (0);
1814 }
1815 
1816 /*
1817  * Write out a core segment to the compression stream.
1818  */
1819 static int
1820 compress_chunk(struct coredump_params *cp, char *base, char *buf, size_t len)
1821 {
1822 	size_t chunk_len;
1823 	int error;
1824 
1825 	while (len > 0) {
1826 		chunk_len = MIN(len, CORE_BUF_SIZE);
1827 
1828 		/*
1829 		 * We can get EFAULT error here.
1830 		 * In that case zero out the current chunk of the segment.
1831 		 */
1832 		error = copyin(base, buf, chunk_len);
1833 		if (error != 0)
1834 			bzero(buf, chunk_len);
1835 		error = compressor_write(cp->comp, buf, chunk_len);
1836 		if (error != 0)
1837 			break;
1838 		base += chunk_len;
1839 		len -= chunk_len;
1840 	}
1841 	return (error);
1842 }
1843 
1844 int
1845 core_write(struct coredump_params *cp, const void *base, size_t len,
1846     off_t offset, enum uio_seg seg, size_t *resid)
1847 {
1848 
1849 	return (vn_rdwr_inchunks(UIO_WRITE, cp->vp, __DECONST(void *, base),
1850 	    len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1851 	    cp->active_cred, cp->file_cred, resid, cp->td));
1852 }
1853 
1854 int
1855 core_output(char *base, size_t len, off_t offset, struct coredump_params *cp,
1856     void *tmpbuf)
1857 {
1858 	vm_map_t map;
1859 	struct mount *mp;
1860 	size_t resid, runlen;
1861 	int error;
1862 	bool success;
1863 
1864 	KASSERT((uintptr_t)base % PAGE_SIZE == 0,
1865 	    ("%s: user address %p is not page-aligned", __func__, base));
1866 
1867 	if (cp->comp != NULL)
1868 		return (compress_chunk(cp, base, tmpbuf, len));
1869 
1870 	map = &cp->td->td_proc->p_vmspace->vm_map;
1871 	for (; len > 0; base += runlen, offset += runlen, len -= runlen) {
1872 		/*
1873 		 * Attempt to page in all virtual pages in the range.  If a
1874 		 * virtual page is not backed by the pager, it is represented as
1875 		 * a hole in the file.  This can occur with zero-filled
1876 		 * anonymous memory or truncated files, for example.
1877 		 */
1878 		for (runlen = 0; runlen < len; runlen += PAGE_SIZE) {
1879 			if (core_dump_can_intr && curproc_sigkilled())
1880 				return (EINTR);
1881 			error = vm_fault(map, (uintptr_t)base + runlen,
1882 			    VM_PROT_READ, VM_FAULT_NOFILL, NULL);
1883 			if (runlen == 0)
1884 				success = error == KERN_SUCCESS;
1885 			else if ((error == KERN_SUCCESS) != success)
1886 				break;
1887 		}
1888 
1889 		if (success) {
1890 			error = core_write(cp, base, runlen, offset,
1891 			    UIO_USERSPACE, &resid);
1892 			if (error != 0) {
1893 				if (error != EFAULT)
1894 					break;
1895 
1896 				/*
1897 				 * EFAULT may be returned if the user mapping
1898 				 * could not be accessed, e.g., because a mapped
1899 				 * file has been truncated.  Skip the page if no
1900 				 * progress was made, to protect against a
1901 				 * hypothetical scenario where vm_fault() was
1902 				 * successful but core_write() returns EFAULT
1903 				 * anyway.
1904 				 */
1905 				runlen -= resid;
1906 				if (runlen == 0) {
1907 					success = false;
1908 					runlen = PAGE_SIZE;
1909 				}
1910 			}
1911 		}
1912 		if (!success) {
1913 			error = vn_start_write(cp->vp, &mp, V_WAIT);
1914 			if (error != 0)
1915 				break;
1916 			vn_lock(cp->vp, LK_EXCLUSIVE | LK_RETRY);
1917 			error = vn_truncate_locked(cp->vp, offset + runlen,
1918 			    false, cp->td->td_ucred);
1919 			VOP_UNLOCK(cp->vp);
1920 			vn_finished_write(mp);
1921 			if (error != 0)
1922 				break;
1923 		}
1924 	}
1925 	return (error);
1926 }
1927 
1928 /*
1929  * Drain into a core file.
1930  */
1931 int
1932 sbuf_drain_core_output(void *arg, const char *data, int len)
1933 {
1934 	struct coredump_params *cp;
1935 	struct proc *p;
1936 	int error, locked;
1937 
1938 	cp = arg;
1939 	p = cp->td->td_proc;
1940 
1941 	/*
1942 	 * Some kern_proc out routines that print to this sbuf may
1943 	 * call us with the process lock held. Draining with the
1944 	 * non-sleepable lock held is unsafe. The lock is needed for
1945 	 * those routines when dumping a live process. In our case we
1946 	 * can safely release the lock before draining and acquire
1947 	 * again after.
1948 	 */
1949 	locked = PROC_LOCKED(p);
1950 	if (locked)
1951 		PROC_UNLOCK(p);
1952 	if (cp->comp != NULL)
1953 		error = compressor_write(cp->comp, __DECONST(char *, data), len);
1954 	else
1955 		error = core_write(cp, __DECONST(void *, data), len, cp->offset,
1956 		    UIO_SYSSPACE, NULL);
1957 	if (locked)
1958 		PROC_LOCK(p);
1959 	if (error != 0)
1960 		return (-error);
1961 	cp->offset += len;
1962 	return (len);
1963 }
1964