xref: /freebsd/sys/kern/kern_exec.c (revision 924226fba12cc9a228c73b956e1b7fa24c60b055)
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 	vm_offset_t ps_strings;
164 
165 	p = curproc;
166 #ifdef SCTL_MASK32
167 	if (req->flags & SCTL_MASK32) {
168 		unsigned int val;
169 		val = (unsigned int)PROC_PS_STRINGS(p);
170 		return (SYSCTL_OUT(req, &val, sizeof(val)));
171 	}
172 #endif
173 	ps_strings = PROC_PS_STRINGS(p);
174 	return (SYSCTL_OUT(req, &ps_strings, sizeof(ps_strings)));
175 }
176 
177 static int
178 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
179 {
180 	struct proc *p;
181 	vm_offset_t val;
182 
183 	p = curproc;
184 #ifdef SCTL_MASK32
185 	if (req->flags & SCTL_MASK32) {
186 		unsigned int val32;
187 
188 		val32 = round_page((unsigned int)p->p_vmspace->vm_stacktop);
189 		return (SYSCTL_OUT(req, &val32, sizeof(val32)));
190 	}
191 #endif
192 	val = round_page(p->p_vmspace->vm_stacktop);
193 	return (SYSCTL_OUT(req, &val, sizeof(val)));
194 }
195 
196 static int
197 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
198 {
199 	struct proc *p;
200 
201 	p = curproc;
202 	return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
203 	    sizeof(p->p_sysent->sv_stackprot)));
204 }
205 
206 /*
207  * Each of the items is a pointer to a `const struct execsw', hence the
208  * double pointer here.
209  */
210 static const struct execsw **execsw;
211 
212 #ifndef _SYS_SYSPROTO_H_
213 struct execve_args {
214 	char    *fname;
215 	char    **argv;
216 	char    **envv;
217 };
218 #endif
219 
220 int
221 sys_execve(struct thread *td, struct execve_args *uap)
222 {
223 	struct image_args args;
224 	struct vmspace *oldvmspace;
225 	int error;
226 
227 	error = pre_execve(td, &oldvmspace);
228 	if (error != 0)
229 		return (error);
230 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
231 	    uap->argv, uap->envv);
232 	if (error == 0)
233 		error = kern_execve(td, &args, NULL, oldvmspace);
234 	post_execve(td, error, oldvmspace);
235 	AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
236 	return (error);
237 }
238 
239 #ifndef _SYS_SYSPROTO_H_
240 struct fexecve_args {
241 	int	fd;
242 	char	**argv;
243 	char	**envv;
244 };
245 #endif
246 int
247 sys_fexecve(struct thread *td, struct fexecve_args *uap)
248 {
249 	struct image_args args;
250 	struct vmspace *oldvmspace;
251 	int error;
252 
253 	error = pre_execve(td, &oldvmspace);
254 	if (error != 0)
255 		return (error);
256 	error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
257 	    uap->argv, uap->envv);
258 	if (error == 0) {
259 		args.fd = uap->fd;
260 		error = kern_execve(td, &args, NULL, oldvmspace);
261 	}
262 	post_execve(td, error, oldvmspace);
263 	AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
264 	return (error);
265 }
266 
267 #ifndef _SYS_SYSPROTO_H_
268 struct __mac_execve_args {
269 	char	*fname;
270 	char	**argv;
271 	char	**envv;
272 	struct mac	*mac_p;
273 };
274 #endif
275 
276 int
277 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap)
278 {
279 #ifdef MAC
280 	struct image_args args;
281 	struct vmspace *oldvmspace;
282 	int error;
283 
284 	error = pre_execve(td, &oldvmspace);
285 	if (error != 0)
286 		return (error);
287 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
288 	    uap->argv, uap->envv);
289 	if (error == 0)
290 		error = kern_execve(td, &args, uap->mac_p, oldvmspace);
291 	post_execve(td, error, oldvmspace);
292 	AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
293 	return (error);
294 #else
295 	return (ENOSYS);
296 #endif
297 }
298 
299 int
300 pre_execve(struct thread *td, struct vmspace **oldvmspace)
301 {
302 	struct proc *p;
303 	int error;
304 
305 	KASSERT(td == curthread, ("non-current thread %p", td));
306 	error = 0;
307 	p = td->td_proc;
308 	if ((p->p_flag & P_HADTHREADS) != 0) {
309 		PROC_LOCK(p);
310 		while (p->p_singlethr > 0) {
311 			error = msleep(&p->p_singlethr, &p->p_mtx,
312 			    PWAIT | PCATCH, "exec1t", 0);
313 			if (error != 0) {
314 				error = ERESTART;
315 				goto unlock;
316 			}
317 		}
318 		if (thread_single(p, SINGLE_BOUNDARY) != 0)
319 			error = ERESTART;
320 unlock:
321 		PROC_UNLOCK(p);
322 	}
323 	KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0,
324 	    ("nested execve"));
325 	*oldvmspace = p->p_vmspace;
326 	return (error);
327 }
328 
329 void
330 post_execve(struct thread *td, int error, struct vmspace *oldvmspace)
331 {
332 	struct proc *p;
333 
334 	KASSERT(td == curthread, ("non-current thread %p", td));
335 	p = td->td_proc;
336 	if ((p->p_flag & P_HADTHREADS) != 0) {
337 		PROC_LOCK(p);
338 		/*
339 		 * If success, we upgrade to SINGLE_EXIT state to
340 		 * force other threads to suicide.
341 		 */
342 		if (error == EJUSTRETURN)
343 			thread_single(p, SINGLE_EXIT);
344 		else
345 			thread_single_end(p, SINGLE_BOUNDARY);
346 		PROC_UNLOCK(p);
347 	}
348 	exec_cleanup(td, oldvmspace);
349 }
350 
351 /*
352  * kern_execve() has the astonishing property of not always returning to
353  * the caller.  If sufficiently bad things happen during the call to
354  * do_execve(), it can end up calling exit1(); as a result, callers must
355  * avoid doing anything which they might need to undo (e.g., allocating
356  * memory).
357  */
358 int
359 kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p,
360     struct vmspace *oldvmspace)
361 {
362 
363 	TSEXEC(td->td_proc->p_pid, args->begin_argv);
364 	AUDIT_ARG_ARGV(args->begin_argv, args->argc,
365 	    exec_args_get_begin_envv(args) - args->begin_argv);
366 	AUDIT_ARG_ENVV(exec_args_get_begin_envv(args), args->envc,
367 	    args->endp - exec_args_get_begin_envv(args));
368 
369 	/* Must have at least one argument. */
370 	if (args->argc == 0) {
371 		exec_free_args(args);
372 		return (EINVAL);
373 	}
374 	return (do_execve(td, args, mac_p, oldvmspace));
375 }
376 
377 static void
378 execve_nosetid(struct image_params *imgp)
379 {
380 	imgp->credential_setid = false;
381 	if (imgp->newcred != NULL) {
382 		crfree(imgp->newcred);
383 		imgp->newcred = NULL;
384 	}
385 }
386 
387 /*
388  * In-kernel implementation of execve().  All arguments are assumed to be
389  * userspace pointers from the passed thread.
390  */
391 static int
392 do_execve(struct thread *td, struct image_args *args, struct mac *mac_p,
393     struct vmspace *oldvmspace)
394 {
395 	struct proc *p = td->td_proc;
396 	struct nameidata nd;
397 	struct ucred *oldcred;
398 	struct uidinfo *euip = NULL;
399 	uintptr_t stack_base;
400 	struct image_params image_params, *imgp;
401 	struct vattr attr;
402 	int (*img_first)(struct image_params *);
403 	struct pargs *oldargs = NULL, *newargs = NULL;
404 	struct sigacts *oldsigacts = NULL, *newsigacts = NULL;
405 #ifdef KTRACE
406 	struct ktr_io_params *kiop;
407 #endif
408 	struct vnode *oldtextvp, *newtextvp;
409 	struct vnode *oldtextdvp, *newtextdvp;
410 	char *oldbinname, *newbinname;
411 	bool credential_changing;
412 #ifdef MAC
413 	struct label *interpvplabel = NULL;
414 	bool will_transition;
415 #endif
416 #ifdef HWPMC_HOOKS
417 	struct pmckern_procexec pe;
418 #endif
419 	int error, i, orig_osrel;
420 	uint32_t orig_fctl0;
421 	Elf_Brandinfo *orig_brandinfo;
422 	size_t freepath_size;
423 	static const char fexecv_proc_title[] = "(fexecv)";
424 
425 	imgp = &image_params;
426 	oldtextvp = oldtextdvp = NULL;
427 	newtextvp = newtextdvp = NULL;
428 	newbinname = oldbinname = NULL;
429 #ifdef KTRACE
430 	kiop = NULL;
431 #endif
432 
433 	/*
434 	 * Lock the process and set the P_INEXEC flag to indicate that
435 	 * it should be left alone until we're done here.  This is
436 	 * necessary to avoid race conditions - e.g. in ptrace() -
437 	 * that might allow a local user to illicitly obtain elevated
438 	 * privileges.
439 	 */
440 	PROC_LOCK(p);
441 	KASSERT((p->p_flag & P_INEXEC) == 0,
442 	    ("%s(): process already has P_INEXEC flag", __func__));
443 	p->p_flag |= P_INEXEC;
444 	PROC_UNLOCK(p);
445 
446 	/*
447 	 * Initialize part of the common data
448 	 */
449 	bzero(imgp, sizeof(*imgp));
450 	imgp->proc = p;
451 	imgp->attr = &attr;
452 	imgp->args = args;
453 	oldcred = p->p_ucred;
454 	orig_osrel = p->p_osrel;
455 	orig_fctl0 = p->p_fctl0;
456 	orig_brandinfo = p->p_elf_brandinfo;
457 
458 #ifdef MAC
459 	error = mac_execve_enter(imgp, mac_p);
460 	if (error)
461 		goto exec_fail;
462 #endif
463 
464 	SDT_PROBE1(proc, , , exec, args->fname);
465 
466 interpret:
467 	if (args->fname != NULL) {
468 #ifdef CAPABILITY_MODE
469 		/*
470 		 * While capability mode can't reach this point via direct
471 		 * path arguments to execve(), we also don't allow
472 		 * interpreters to be used in capability mode (for now).
473 		 * Catch indirect lookups and return a permissions error.
474 		 */
475 		if (IN_CAPABILITY_MODE(td)) {
476 			error = ECAPMODE;
477 			goto exec_fail;
478 		}
479 #endif
480 
481 		/*
482 		 * Translate the file name. namei() returns a vnode
483 		 * pointer in ni_vp among other things.
484 		 */
485 		NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW |
486 		    SAVENAME | AUDITVNODE1 | WANTPARENT, UIO_SYSSPACE,
487 		    args->fname);
488 
489 		error = namei(&nd);
490 		if (error)
491 			goto exec_fail;
492 
493 		newtextvp = nd.ni_vp;
494 		newtextdvp = nd.ni_dvp;
495 		nd.ni_dvp = NULL;
496 		newbinname = malloc(nd.ni_cnd.cn_namelen + 1, M_PARGS,
497 		    M_WAITOK);
498 		memcpy(newbinname, nd.ni_cnd.cn_nameptr, nd.ni_cnd.cn_namelen);
499 		newbinname[nd.ni_cnd.cn_namelen] = '\0';
500 		imgp->vp = newtextvp;
501 
502 		/*
503 		 * Do the best to calculate the full path to the image file.
504 		 */
505 		if (args->fname[0] == '/') {
506 			imgp->execpath = args->fname;
507 		} else {
508 			VOP_UNLOCK(imgp->vp);
509 			freepath_size = MAXPATHLEN;
510 			if (vn_fullpath_hardlink(newtextvp, newtextdvp,
511 			    newbinname, nd.ni_cnd.cn_namelen, &imgp->execpath,
512 			    &imgp->freepath, &freepath_size) != 0)
513 				imgp->execpath = args->fname;
514 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
515 		}
516 	} else {
517 		AUDIT_ARG_FD(args->fd);
518 
519 		/*
520 		 * If the descriptors was not opened with O_PATH, then
521 		 * we require that it was opened with O_EXEC or
522 		 * O_RDONLY.  In either case, exec_check_permissions()
523 		 * below checks _current_ file access mode regardless
524 		 * of the permissions additionally checked at the
525 		 * open(2).
526 		 */
527 		error = fgetvp_exec(td, args->fd, &cap_fexecve_rights,
528 		    &newtextvp);
529 		if (error != 0)
530 			goto exec_fail;
531 
532 		if (vn_fullpath(newtextvp, &imgp->execpath,
533 		    &imgp->freepath) != 0)
534 			imgp->execpath = args->fname;
535 		vn_lock(newtextvp, LK_SHARED | LK_RETRY);
536 		AUDIT_ARG_VNODE1(newtextvp);
537 		imgp->vp = newtextvp;
538 	}
539 
540 	/*
541 	 * Check file permissions.  Also 'opens' file and sets its vnode to
542 	 * text mode.
543 	 */
544 	error = exec_check_permissions(imgp);
545 	if (error)
546 		goto exec_fail_dealloc;
547 
548 	imgp->object = imgp->vp->v_object;
549 	if (imgp->object != NULL)
550 		vm_object_reference(imgp->object);
551 
552 	error = exec_map_first_page(imgp);
553 	if (error)
554 		goto exec_fail_dealloc;
555 
556 	imgp->proc->p_osrel = 0;
557 	imgp->proc->p_fctl0 = 0;
558 	imgp->proc->p_elf_brandinfo = NULL;
559 
560 	/*
561 	 * Implement image setuid/setgid.
562 	 *
563 	 * Determine new credentials before attempting image activators
564 	 * so that it can be used by process_exec handlers to determine
565 	 * credential/setid changes.
566 	 *
567 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
568 	 * the process is being traced.
569 	 *
570 	 * We disable setuid/setgid/etc in capability mode on the basis
571 	 * that most setugid applications are not written with that
572 	 * environment in mind, and will therefore almost certainly operate
573 	 * incorrectly. In principle there's no reason that setugid
574 	 * applications might not be useful in capability mode, so we may want
575 	 * to reconsider this conservative design choice in the future.
576 	 *
577 	 * XXXMAC: For the time being, use NOSUID to also prohibit
578 	 * transitions on the file system.
579 	 */
580 	credential_changing = false;
581 	credential_changing |= (attr.va_mode & S_ISUID) &&
582 	    oldcred->cr_uid != attr.va_uid;
583 	credential_changing |= (attr.va_mode & S_ISGID) &&
584 	    oldcred->cr_gid != attr.va_gid;
585 #ifdef MAC
586 	will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
587 	    interpvplabel, imgp) != 0;
588 	credential_changing |= will_transition;
589 #endif
590 
591 	/* Don't inherit PROC_PDEATHSIG_CTL value if setuid/setgid. */
592 	if (credential_changing)
593 		imgp->proc->p_pdeathsig = 0;
594 
595 	if (credential_changing &&
596 #ifdef CAPABILITY_MODE
597 	    ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
598 #endif
599 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
600 	    (p->p_flag & P_TRACED) == 0) {
601 		imgp->credential_setid = true;
602 		VOP_UNLOCK(imgp->vp);
603 		imgp->newcred = crdup(oldcred);
604 		if (attr.va_mode & S_ISUID) {
605 			euip = uifind(attr.va_uid);
606 			change_euid(imgp->newcred, euip);
607 		}
608 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
609 		if (attr.va_mode & S_ISGID)
610 			change_egid(imgp->newcred, attr.va_gid);
611 		/*
612 		 * Implement correct POSIX saved-id behavior.
613 		 *
614 		 * XXXMAC: Note that the current logic will save the
615 		 * uid and gid if a MAC domain transition occurs, even
616 		 * though maybe it shouldn't.
617 		 */
618 		change_svuid(imgp->newcred, imgp->newcred->cr_uid);
619 		change_svgid(imgp->newcred, imgp->newcred->cr_gid);
620 	} else {
621 		/*
622 		 * Implement correct POSIX saved-id behavior.
623 		 *
624 		 * XXX: It's not clear that the existing behavior is
625 		 * POSIX-compliant.  A number of sources indicate that the
626 		 * saved uid/gid should only be updated if the new ruid is
627 		 * not equal to the old ruid, or the new euid is not equal
628 		 * to the old euid and the new euid is not equal to the old
629 		 * ruid.  The FreeBSD code always updates the saved uid/gid.
630 		 * Also, this code uses the new (replaced) euid and egid as
631 		 * the source, which may or may not be the right ones to use.
632 		 */
633 		if (oldcred->cr_svuid != oldcred->cr_uid ||
634 		    oldcred->cr_svgid != oldcred->cr_gid) {
635 			VOP_UNLOCK(imgp->vp);
636 			imgp->newcred = crdup(oldcred);
637 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
638 			change_svuid(imgp->newcred, imgp->newcred->cr_uid);
639 			change_svgid(imgp->newcred, imgp->newcred->cr_gid);
640 		}
641 	}
642 	/* The new credentials are installed into the process later. */
643 
644 	/*
645 	 *	If the current process has a special image activator it
646 	 *	wants to try first, call it.   For example, emulating shell
647 	 *	scripts differently.
648 	 */
649 	error = -1;
650 	if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
651 		error = img_first(imgp);
652 
653 	/*
654 	 *	Loop through the list of image activators, calling each one.
655 	 *	An activator returns -1 if there is no match, 0 on success,
656 	 *	and an error otherwise.
657 	 */
658 	for (i = 0; error == -1 && execsw[i]; ++i) {
659 		if (execsw[i]->ex_imgact == NULL ||
660 		    execsw[i]->ex_imgact == img_first) {
661 			continue;
662 		}
663 		error = (*execsw[i]->ex_imgact)(imgp);
664 	}
665 
666 	if (error) {
667 		if (error == -1)
668 			error = ENOEXEC;
669 		goto exec_fail_dealloc;
670 	}
671 
672 	/*
673 	 * Special interpreter operation, cleanup and loop up to try to
674 	 * activate the interpreter.
675 	 */
676 	if (imgp->interpreted) {
677 		exec_unmap_first_page(imgp);
678 		/*
679 		 * The text reference needs to be removed for scripts.
680 		 * There is a short period before we determine that
681 		 * something is a script where text reference is active.
682 		 * The vnode lock is held over this entire period
683 		 * so nothing should illegitimately be blocked.
684 		 */
685 		MPASS(imgp->textset);
686 		VOP_UNSET_TEXT_CHECKED(newtextvp);
687 		imgp->textset = false;
688 		/* free name buffer and old vnode */
689 #ifdef MAC
690 		mac_execve_interpreter_enter(newtextvp, &interpvplabel);
691 #endif
692 		if (imgp->opened) {
693 			VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td);
694 			imgp->opened = false;
695 		}
696 		vput(newtextvp);
697 		imgp->vp = newtextvp = NULL;
698 		if (args->fname != NULL) {
699 			if (newtextdvp != NULL) {
700 				vrele(newtextdvp);
701 				newtextdvp = NULL;
702 			}
703 			NDFREE_PNBUF(&nd);
704 			free(newbinname, M_PARGS);
705 			newbinname = NULL;
706 		}
707 		vm_object_deallocate(imgp->object);
708 		imgp->object = NULL;
709 		execve_nosetid(imgp);
710 		imgp->execpath = NULL;
711 		free(imgp->freepath, M_TEMP);
712 		imgp->freepath = NULL;
713 		/* set new name to that of the interpreter */
714 		args->fname = imgp->interpreter_name;
715 		goto interpret;
716 	}
717 
718 	/*
719 	 * NB: We unlock the vnode here because it is believed that none
720 	 * of the sv_copyout_strings/sv_fixup operations require the vnode.
721 	 */
722 	VOP_UNLOCK(imgp->vp);
723 
724 	if (disallow_high_osrel &&
725 	    P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
726 		error = ENOEXEC;
727 		uprintf("Osrel %d for image %s too high\n", p->p_osrel,
728 		    imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
729 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
730 		goto exec_fail_dealloc;
731 	}
732 
733 	/*
734 	 * Copy out strings (args and env) and initialize stack base.
735 	 */
736 	error = (*p->p_sysent->sv_copyout_strings)(imgp, &stack_base);
737 	if (error != 0) {
738 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
739 		goto exec_fail_dealloc;
740 	}
741 
742 	/*
743 	 * Stack setup.
744 	 */
745 	error = (*p->p_sysent->sv_fixup)(&stack_base, imgp);
746 	if (error != 0) {
747 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
748 		goto exec_fail_dealloc;
749 	}
750 
751 	/*
752 	 * For security and other reasons, the file descriptor table cannot be
753 	 * shared after an exec.
754 	 */
755 	fdunshare(td);
756 	pdunshare(td);
757 	/* close files on exec */
758 	fdcloseexec(td);
759 
760 	/*
761 	 * Malloc things before we need locks.
762 	 */
763 	i = exec_args_get_begin_envv(imgp->args) - imgp->args->begin_argv;
764 	/* Cache arguments if they fit inside our allowance */
765 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
766 		newargs = pargs_alloc(i);
767 		bcopy(imgp->args->begin_argv, newargs->ar_args, i);
768 	}
769 
770 	/*
771 	 * For security and other reasons, signal handlers cannot
772 	 * be shared after an exec. The new process gets a copy of the old
773 	 * handlers. In execsigs(), the new process will have its signals
774 	 * reset.
775 	 */
776 	if (sigacts_shared(p->p_sigacts)) {
777 		oldsigacts = p->p_sigacts;
778 		newsigacts = sigacts_alloc();
779 		sigacts_copy(newsigacts, oldsigacts);
780 	}
781 
782 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
783 
784 	PROC_LOCK(p);
785 	if (oldsigacts)
786 		p->p_sigacts = newsigacts;
787 	/* Stop profiling */
788 	stopprofclock(p);
789 
790 	/* reset caught signals */
791 	execsigs(p);
792 
793 	/* name this process - nameiexec(p, ndp) */
794 	bzero(p->p_comm, sizeof(p->p_comm));
795 	if (args->fname)
796 		bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
797 		    min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
798 	else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0)
799 		bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
800 	bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
801 #ifdef KTR
802 	sched_clear_tdname(td);
803 #endif
804 
805 	/*
806 	 * mark as execed, wakeup the process that vforked (if any) and tell
807 	 * it that it now has its own resources back
808 	 */
809 	p->p_flag |= P_EXEC;
810 	if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0)
811 		p->p_flag2 &= ~P2_NOTRACE;
812 	if ((p->p_flag2 & P2_STKGAP_DISABLE_EXEC) == 0)
813 		p->p_flag2 &= ~P2_STKGAP_DISABLE;
814 	if (p->p_flag & P_PPWAIT) {
815 		p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
816 		cv_broadcast(&p->p_pwait);
817 		/* STOPs are no longer ignored, arrange for AST */
818 		signotify(td);
819 	}
820 
821 	if ((imgp->sysent->sv_setid_allowed != NULL &&
822 	    !(*imgp->sysent->sv_setid_allowed)(td, imgp)) ||
823 	    (p->p_flag2 & P2_NO_NEW_PRIVS) != 0)
824 		execve_nosetid(imgp);
825 
826 	/*
827 	 * Implement image setuid/setgid installation.
828 	 */
829 	if (imgp->credential_setid) {
830 		/*
831 		 * Turn off syscall tracing for set-id programs, except for
832 		 * root.  Record any set-id flags first to make sure that
833 		 * we do not regain any tracing during a possible block.
834 		 */
835 		setsugid(p);
836 #ifdef KTRACE
837 		kiop = ktrprocexec(p);
838 #endif
839 		/*
840 		 * Close any file descriptors 0..2 that reference procfs,
841 		 * then make sure file descriptors 0..2 are in use.
842 		 *
843 		 * Both fdsetugidsafety() and fdcheckstd() may call functions
844 		 * taking sleepable locks, so temporarily drop our locks.
845 		 */
846 		PROC_UNLOCK(p);
847 		VOP_UNLOCK(imgp->vp);
848 		fdsetugidsafety(td);
849 		error = fdcheckstd(td);
850 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
851 		if (error != 0)
852 			goto exec_fail_dealloc;
853 		PROC_LOCK(p);
854 #ifdef MAC
855 		if (will_transition) {
856 			mac_vnode_execve_transition(oldcred, imgp->newcred,
857 			    imgp->vp, interpvplabel, imgp);
858 		}
859 #endif
860 	} else {
861 		if (oldcred->cr_uid == oldcred->cr_ruid &&
862 		    oldcred->cr_gid == oldcred->cr_rgid)
863 			p->p_flag &= ~P_SUGID;
864 	}
865 	/*
866 	 * Set the new credentials.
867 	 */
868 	if (imgp->newcred != NULL) {
869 		proc_set_cred(p, imgp->newcred);
870 		crfree(oldcred);
871 		oldcred = NULL;
872 	}
873 
874 	/*
875 	 * Store the vp for use in kern.proc.pathname.  This vnode was
876 	 * referenced by namei() or by fexecve variant of fname handling.
877 	 */
878 	oldtextvp = p->p_textvp;
879 	p->p_textvp = newtextvp;
880 	oldtextdvp = p->p_textdvp;
881 	p->p_textdvp = newtextdvp;
882 	newtextdvp = NULL;
883 	oldbinname = p->p_binname;
884 	p->p_binname = newbinname;
885 	newbinname = NULL;
886 
887 #ifdef KDTRACE_HOOKS
888 	/*
889 	 * Tell the DTrace fasttrap provider about the exec if it
890 	 * has declared an interest.
891 	 */
892 	if (dtrace_fasttrap_exec)
893 		dtrace_fasttrap_exec(p);
894 #endif
895 
896 	/*
897 	 * Notify others that we exec'd, and clear the P_INEXEC flag
898 	 * as we're now a bona fide freshly-execed process.
899 	 */
900 	KNOTE_LOCKED(p->p_klist, NOTE_EXEC);
901 	p->p_flag &= ~P_INEXEC;
902 
903 	/* clear "fork but no exec" flag, as we _are_ execing */
904 	p->p_acflag &= ~AFORK;
905 
906 	/*
907 	 * Free any previous argument cache and replace it with
908 	 * the new argument cache, if any.
909 	 */
910 	oldargs = p->p_args;
911 	p->p_args = newargs;
912 	newargs = NULL;
913 
914 	PROC_UNLOCK(p);
915 
916 #ifdef	HWPMC_HOOKS
917 	/*
918 	 * Check if system-wide sampling is in effect or if the
919 	 * current process is using PMCs.  If so, do exec() time
920 	 * processing.  This processing needs to happen AFTER the
921 	 * P_INEXEC flag is cleared.
922 	 */
923 	if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
924 		VOP_UNLOCK(imgp->vp);
925 		pe.pm_credentialschanged = credential_changing;
926 		pe.pm_entryaddr = imgp->entry_addr;
927 
928 		PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
929 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
930 	}
931 #endif
932 
933 	/* Set values passed into the program in registers. */
934 	(*p->p_sysent->sv_setregs)(td, imgp, stack_base);
935 
936 	VOP_MMAPPED(imgp->vp);
937 
938 	SDT_PROBE1(proc, , , exec__success, args->fname);
939 
940 exec_fail_dealloc:
941 	if (error != 0) {
942 		p->p_osrel = orig_osrel;
943 		p->p_fctl0 = orig_fctl0;
944 		p->p_elf_brandinfo = orig_brandinfo;
945 	}
946 
947 	if (imgp->firstpage != NULL)
948 		exec_unmap_first_page(imgp);
949 
950 	if (imgp->vp != NULL) {
951 		if (imgp->opened)
952 			VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
953 		if (imgp->textset)
954 			VOP_UNSET_TEXT_CHECKED(imgp->vp);
955 		if (error != 0)
956 			vput(imgp->vp);
957 		else
958 			VOP_UNLOCK(imgp->vp);
959 		if (args->fname != NULL)
960 			NDFREE_PNBUF(&nd);
961 		if (newtextdvp != NULL)
962 			vrele(newtextdvp);
963 		free(newbinname, M_PARGS);
964 	}
965 
966 	if (imgp->object != NULL)
967 		vm_object_deallocate(imgp->object);
968 
969 	free(imgp->freepath, M_TEMP);
970 
971 	if (error == 0) {
972 		if (p->p_ptevents & PTRACE_EXEC) {
973 			PROC_LOCK(p);
974 			if (p->p_ptevents & PTRACE_EXEC)
975 				td->td_dbgflags |= TDB_EXEC;
976 			PROC_UNLOCK(p);
977 		}
978 	} else {
979 exec_fail:
980 		/* we're done here, clear P_INEXEC */
981 		PROC_LOCK(p);
982 		p->p_flag &= ~P_INEXEC;
983 		PROC_UNLOCK(p);
984 
985 		SDT_PROBE1(proc, , , exec__failure, error);
986 	}
987 
988 	if (imgp->newcred != NULL && oldcred != NULL)
989 		crfree(imgp->newcred);
990 
991 #ifdef MAC
992 	mac_execve_exit(imgp);
993 	mac_execve_interpreter_exit(interpvplabel);
994 #endif
995 	exec_free_args(args);
996 
997 	/*
998 	 * Handle deferred decrement of ref counts.
999 	 */
1000 	if (oldtextvp != NULL)
1001 		vrele(oldtextvp);
1002 	if (oldtextdvp != NULL)
1003 		vrele(oldtextdvp);
1004 	free(oldbinname, M_PARGS);
1005 #ifdef KTRACE
1006 	ktr_io_params_free(kiop);
1007 #endif
1008 	pargs_drop(oldargs);
1009 	pargs_drop(newargs);
1010 	if (oldsigacts != NULL)
1011 		sigacts_free(oldsigacts);
1012 	if (euip != NULL)
1013 		uifree(euip);
1014 
1015 	if (error && imgp->vmspace_destroyed) {
1016 		/* sorry, no more process anymore. exit gracefully */
1017 		exec_cleanup(td, oldvmspace);
1018 		exit1(td, 0, SIGABRT);
1019 		/* NOT REACHED */
1020 	}
1021 
1022 #ifdef KTRACE
1023 	if (error == 0)
1024 		ktrprocctor(p);
1025 #endif
1026 
1027 	/*
1028 	 * We don't want cpu_set_syscall_retval() to overwrite any of
1029 	 * the register values put in place by exec_setregs().
1030 	 * Implementations of cpu_set_syscall_retval() will leave
1031 	 * registers unmodified when returning EJUSTRETURN.
1032 	 */
1033 	return (error == 0 ? EJUSTRETURN : error);
1034 }
1035 
1036 void
1037 exec_cleanup(struct thread *td, struct vmspace *oldvmspace)
1038 {
1039 	if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
1040 		KASSERT(td->td_proc->p_vmspace != oldvmspace,
1041 		    ("oldvmspace still used"));
1042 		vmspace_free(oldvmspace);
1043 		td->td_pflags &= ~TDP_EXECVMSPC;
1044 	}
1045 }
1046 
1047 int
1048 exec_map_first_page(struct image_params *imgp)
1049 {
1050 	vm_object_t object;
1051 	vm_page_t m;
1052 	int error;
1053 
1054 	if (imgp->firstpage != NULL)
1055 		exec_unmap_first_page(imgp);
1056 
1057 	object = imgp->vp->v_object;
1058 	if (object == NULL)
1059 		return (EACCES);
1060 #if VM_NRESERVLEVEL > 0
1061 	if ((object->flags & OBJ_COLORED) == 0) {
1062 		VM_OBJECT_WLOCK(object);
1063 		vm_object_color(object, 0);
1064 		VM_OBJECT_WUNLOCK(object);
1065 	}
1066 #endif
1067 	error = vm_page_grab_valid_unlocked(&m, object, 0,
1068 	    VM_ALLOC_COUNT(VM_INITIAL_PAGEIN) |
1069 	    VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED);
1070 
1071 	if (error != VM_PAGER_OK)
1072 		return (EIO);
1073 	imgp->firstpage = sf_buf_alloc(m, 0);
1074 	imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
1075 
1076 	return (0);
1077 }
1078 
1079 void
1080 exec_unmap_first_page(struct image_params *imgp)
1081 {
1082 	vm_page_t m;
1083 
1084 	if (imgp->firstpage != NULL) {
1085 		m = sf_buf_page(imgp->firstpage);
1086 		sf_buf_free(imgp->firstpage);
1087 		imgp->firstpage = NULL;
1088 		vm_page_unwire(m, PQ_ACTIVE);
1089 	}
1090 }
1091 
1092 void
1093 exec_onexec_old(struct thread *td)
1094 {
1095 	sigfastblock_clear(td);
1096 	umtx_exec(td->td_proc);
1097 }
1098 
1099 /*
1100  * This is an optimization which removes the unmanaged shared page
1101  * mapping. In combination with pmap_remove_pages(), which cleans all
1102  * managed mappings in the process' vmspace pmap, no work will be left
1103  * for pmap_remove(min, max).
1104  */
1105 void
1106 exec_free_abi_mappings(struct proc *p)
1107 {
1108 	struct vmspace *vmspace;
1109 	struct sysentvec *sv;
1110 
1111 	vmspace = p->p_vmspace;
1112 	if (refcount_load(&vmspace->vm_refcnt) != 1)
1113 		return;
1114 
1115 	sv = p->p_sysent;
1116 	if (sv->sv_shared_page_obj == NULL)
1117 		return;
1118 
1119 	pmap_remove(vmspace_pmap(vmspace), sv->sv_shared_page_base,
1120 	    sv->sv_shared_page_base + sv->sv_shared_page_len);
1121 }
1122 
1123 /*
1124  * Run down the current address space and install a new one.  Map the shared
1125  * page.
1126  */
1127 int
1128 exec_new_vmspace(struct image_params *imgp, struct sysentvec *sv)
1129 {
1130 	int error;
1131 	struct proc *p = imgp->proc;
1132 	struct vmspace *vmspace = p->p_vmspace;
1133 	struct thread *td = curthread;
1134 	vm_object_t obj;
1135 	vm_offset_t sv_minuser;
1136 	vm_map_t map;
1137 
1138 	imgp->vmspace_destroyed = true;
1139 	imgp->sysent = sv;
1140 
1141 	if (p->p_sysent->sv_onexec_old != NULL)
1142 		p->p_sysent->sv_onexec_old(td);
1143 	itimers_exec(p);
1144 
1145 	EVENTHANDLER_DIRECT_INVOKE(process_exec, p, imgp);
1146 
1147 	/*
1148 	 * Blow away entire process VM, if address space not shared,
1149 	 * otherwise, create a new VM space so that other threads are
1150 	 * not disrupted
1151 	 */
1152 	map = &vmspace->vm_map;
1153 	if (map_at_zero)
1154 		sv_minuser = sv->sv_minuser;
1155 	else
1156 		sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1157 	if (refcount_load(&vmspace->vm_refcnt) == 1 &&
1158 	    vm_map_min(map) == sv_minuser &&
1159 	    vm_map_max(map) == sv->sv_maxuser &&
1160 	    cpu_exec_vmspace_reuse(p, map)) {
1161 		exec_free_abi_mappings(p);
1162 		shmexit(vmspace);
1163 		pmap_remove_pages(vmspace_pmap(vmspace));
1164 		vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1165 		/*
1166 		 * An exec terminates mlockall(MCL_FUTURE).
1167 		 * ASLR and W^X states must be re-evaluated.
1168 		 */
1169 		vm_map_lock(map);
1170 		vm_map_modflags(map, 0, MAP_WIREFUTURE | MAP_ASLR |
1171 		    MAP_ASLR_IGNSTART | MAP_ASLR_STACK | MAP_WXORX);
1172 		vm_map_unlock(map);
1173 	} else {
1174 		error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1175 		if (error)
1176 			return (error);
1177 		vmspace = p->p_vmspace;
1178 		map = &vmspace->vm_map;
1179 	}
1180 	map->flags |= imgp->map_flags;
1181 
1182 	/* Map a shared page */
1183 	obj = sv->sv_shared_page_obj;
1184 	if (obj != NULL) {
1185 		vm_object_reference(obj);
1186 		error = vm_map_fixed(map, obj, 0,
1187 		    sv->sv_shared_page_base, sv->sv_shared_page_len,
1188 		    VM_PROT_READ | VM_PROT_EXECUTE,
1189 		    VM_PROT_READ | VM_PROT_EXECUTE,
1190 		    MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1191 		if (error != KERN_SUCCESS) {
1192 			vm_object_deallocate(obj);
1193 			return (vm_mmap_to_errno(error));
1194 		}
1195 	}
1196 
1197 	return (sv->sv_onexec != NULL ? sv->sv_onexec(p, imgp) : 0);
1198 }
1199 
1200 /*
1201  * Compute the stack size limit and map the main process stack.
1202  */
1203 int
1204 exec_map_stack(struct image_params *imgp)
1205 {
1206 	struct rlimit rlim_stack;
1207 	struct sysentvec *sv;
1208 	struct proc *p;
1209 	vm_map_t map;
1210 	struct vmspace *vmspace;
1211 	vm_offset_t stack_addr, stack_top;
1212 	u_long ssiz;
1213 	int error, find_space, stack_off;
1214 	vm_prot_t stack_prot;
1215 
1216 	p = imgp->proc;
1217 	sv = p->p_sysent;
1218 
1219 	if (imgp->stack_sz != 0) {
1220 		ssiz = trunc_page(imgp->stack_sz);
1221 		PROC_LOCK(p);
1222 		lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack);
1223 		PROC_UNLOCK(p);
1224 		if (ssiz > rlim_stack.rlim_max)
1225 			ssiz = rlim_stack.rlim_max;
1226 		if (ssiz > rlim_stack.rlim_cur) {
1227 			rlim_stack.rlim_cur = ssiz;
1228 			kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack);
1229 		}
1230 	} else if (sv->sv_maxssiz != NULL) {
1231 		ssiz = *sv->sv_maxssiz;
1232 	} else {
1233 		ssiz = maxssiz;
1234 	}
1235 
1236 	vmspace = p->p_vmspace;
1237 	map = &vmspace->vm_map;
1238 
1239 	stack_prot = sv->sv_shared_page_obj != NULL && imgp->stack_prot != 0 ?
1240 	    imgp->stack_prot : sv->sv_stackprot;
1241 	if ((map->flags & MAP_ASLR_STACK) != 0) {
1242 		stack_addr = round_page((vm_offset_t)p->p_vmspace->vm_daddr +
1243 		    lim_max(curthread, RLIMIT_DATA));
1244 		find_space = VMFS_ANY_SPACE;
1245 	} else {
1246 		stack_addr = sv->sv_usrstack - ssiz;
1247 		find_space = VMFS_NO_SPACE;
1248 	}
1249 	error = vm_map_find(map, NULL, 0, &stack_addr, (vm_size_t)ssiz,
1250 	    sv->sv_usrstack, find_space, stack_prot, VM_PROT_ALL,
1251 	    MAP_STACK_GROWS_DOWN);
1252 	if (error != KERN_SUCCESS) {
1253 		uprintf("exec_new_vmspace: mapping stack size %#jx prot %#x "
1254 		    "failed, mach error %d errno %d\n", (uintmax_t)ssiz,
1255 		    stack_prot, error, vm_mmap_to_errno(error));
1256 		return (vm_mmap_to_errno(error));
1257 	}
1258 
1259 	stack_top = stack_addr + ssiz;
1260 	if ((map->flags & MAP_ASLR_STACK) != 0) {
1261 		/* Randomize within the first page of the stack. */
1262 		arc4rand(&stack_off, sizeof(stack_off), 0);
1263 		stack_top -= rounddown2(stack_off & PAGE_MASK, sizeof(void *));
1264 	}
1265 
1266 	/*
1267 	 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they
1268 	 * are still used to enforce the stack rlimit on the process stack.
1269 	 */
1270 	vmspace->vm_maxsaddr = (char *)stack_addr;
1271 	vmspace->vm_stacktop = stack_top;
1272 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1273 
1274 	return (0);
1275 }
1276 
1277 /*
1278  * Copy out argument and environment strings from the old process address
1279  * space into the temporary string buffer.
1280  */
1281 int
1282 exec_copyin_args(struct image_args *args, const char *fname,
1283     enum uio_seg segflg, char **argv, char **envv)
1284 {
1285 	u_long arg, env;
1286 	int error;
1287 
1288 	bzero(args, sizeof(*args));
1289 	if (argv == NULL)
1290 		return (EFAULT);
1291 
1292 	/*
1293 	 * Allocate demand-paged memory for the file name, argument, and
1294 	 * environment strings.
1295 	 */
1296 	error = exec_alloc_args(args);
1297 	if (error != 0)
1298 		return (error);
1299 
1300 	/*
1301 	 * Copy the file name.
1302 	 */
1303 	error = exec_args_add_fname(args, fname, segflg);
1304 	if (error != 0)
1305 		goto err_exit;
1306 
1307 	/*
1308 	 * extract arguments first
1309 	 */
1310 	for (;;) {
1311 		error = fueword(argv++, &arg);
1312 		if (error == -1) {
1313 			error = EFAULT;
1314 			goto err_exit;
1315 		}
1316 		if (arg == 0)
1317 			break;
1318 		error = exec_args_add_arg(args, (char *)(uintptr_t)arg,
1319 		    UIO_USERSPACE);
1320 		if (error != 0)
1321 			goto err_exit;
1322 	}
1323 
1324 	/*
1325 	 * extract environment strings
1326 	 */
1327 	if (envv) {
1328 		for (;;) {
1329 			error = fueword(envv++, &env);
1330 			if (error == -1) {
1331 				error = EFAULT;
1332 				goto err_exit;
1333 			}
1334 			if (env == 0)
1335 				break;
1336 			error = exec_args_add_env(args,
1337 			    (char *)(uintptr_t)env, UIO_USERSPACE);
1338 			if (error != 0)
1339 				goto err_exit;
1340 		}
1341 	}
1342 
1343 	return (0);
1344 
1345 err_exit:
1346 	exec_free_args(args);
1347 	return (error);
1348 }
1349 
1350 struct exec_args_kva {
1351 	vm_offset_t addr;
1352 	u_int gen;
1353 	SLIST_ENTRY(exec_args_kva) next;
1354 };
1355 
1356 DPCPU_DEFINE_STATIC(struct exec_args_kva *, exec_args_kva);
1357 
1358 static SLIST_HEAD(, exec_args_kva) exec_args_kva_freelist;
1359 static struct mtx exec_args_kva_mtx;
1360 static u_int exec_args_gen;
1361 
1362 static void
1363 exec_prealloc_args_kva(void *arg __unused)
1364 {
1365 	struct exec_args_kva *argkva;
1366 	u_int i;
1367 
1368 	SLIST_INIT(&exec_args_kva_freelist);
1369 	mtx_init(&exec_args_kva_mtx, "exec args kva", NULL, MTX_DEF);
1370 	for (i = 0; i < exec_map_entries; i++) {
1371 		argkva = malloc(sizeof(*argkva), M_PARGS, M_WAITOK);
1372 		argkva->addr = kmap_alloc_wait(exec_map, exec_map_entry_size);
1373 		argkva->gen = exec_args_gen;
1374 		SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1375 	}
1376 }
1377 SYSINIT(exec_args_kva, SI_SUB_EXEC, SI_ORDER_ANY, exec_prealloc_args_kva, NULL);
1378 
1379 static vm_offset_t
1380 exec_alloc_args_kva(void **cookie)
1381 {
1382 	struct exec_args_kva *argkva;
1383 
1384 	argkva = (void *)atomic_readandclear_ptr(
1385 	    (uintptr_t *)DPCPU_PTR(exec_args_kva));
1386 	if (argkva == NULL) {
1387 		mtx_lock(&exec_args_kva_mtx);
1388 		while ((argkva = SLIST_FIRST(&exec_args_kva_freelist)) == NULL)
1389 			(void)mtx_sleep(&exec_args_kva_freelist,
1390 			    &exec_args_kva_mtx, 0, "execkva", 0);
1391 		SLIST_REMOVE_HEAD(&exec_args_kva_freelist, next);
1392 		mtx_unlock(&exec_args_kva_mtx);
1393 	}
1394 	kasan_mark((void *)argkva->addr, exec_map_entry_size,
1395 	    exec_map_entry_size, 0);
1396 	*(struct exec_args_kva **)cookie = argkva;
1397 	return (argkva->addr);
1398 }
1399 
1400 static void
1401 exec_release_args_kva(struct exec_args_kva *argkva, u_int gen)
1402 {
1403 	vm_offset_t base;
1404 
1405 	base = argkva->addr;
1406 	kasan_mark((void *)argkva->addr, 0, exec_map_entry_size,
1407 	    KASAN_EXEC_ARGS_FREED);
1408 	if (argkva->gen != gen) {
1409 		(void)vm_map_madvise(exec_map, base, base + exec_map_entry_size,
1410 		    MADV_FREE);
1411 		argkva->gen = gen;
1412 	}
1413 	if (!atomic_cmpset_ptr((uintptr_t *)DPCPU_PTR(exec_args_kva),
1414 	    (uintptr_t)NULL, (uintptr_t)argkva)) {
1415 		mtx_lock(&exec_args_kva_mtx);
1416 		SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1417 		wakeup_one(&exec_args_kva_freelist);
1418 		mtx_unlock(&exec_args_kva_mtx);
1419 	}
1420 }
1421 
1422 static void
1423 exec_free_args_kva(void *cookie)
1424 {
1425 
1426 	exec_release_args_kva(cookie, exec_args_gen);
1427 }
1428 
1429 static void
1430 exec_args_kva_lowmem(void *arg __unused)
1431 {
1432 	SLIST_HEAD(, exec_args_kva) head;
1433 	struct exec_args_kva *argkva;
1434 	u_int gen;
1435 	int i;
1436 
1437 	gen = atomic_fetchadd_int(&exec_args_gen, 1) + 1;
1438 
1439 	/*
1440 	 * Force an madvise of each KVA range. Any currently allocated ranges
1441 	 * will have MADV_FREE applied once they are freed.
1442 	 */
1443 	SLIST_INIT(&head);
1444 	mtx_lock(&exec_args_kva_mtx);
1445 	SLIST_SWAP(&head, &exec_args_kva_freelist, exec_args_kva);
1446 	mtx_unlock(&exec_args_kva_mtx);
1447 	while ((argkva = SLIST_FIRST(&head)) != NULL) {
1448 		SLIST_REMOVE_HEAD(&head, next);
1449 		exec_release_args_kva(argkva, gen);
1450 	}
1451 
1452 	CPU_FOREACH(i) {
1453 		argkva = (void *)atomic_readandclear_ptr(
1454 		    (uintptr_t *)DPCPU_ID_PTR(i, exec_args_kva));
1455 		if (argkva != NULL)
1456 			exec_release_args_kva(argkva, gen);
1457 	}
1458 }
1459 EVENTHANDLER_DEFINE(vm_lowmem, exec_args_kva_lowmem, NULL,
1460     EVENTHANDLER_PRI_ANY);
1461 
1462 /*
1463  * Allocate temporary demand-paged, zero-filled memory for the file name,
1464  * argument, and environment strings.
1465  */
1466 int
1467 exec_alloc_args(struct image_args *args)
1468 {
1469 
1470 	args->buf = (char *)exec_alloc_args_kva(&args->bufkva);
1471 	return (0);
1472 }
1473 
1474 void
1475 exec_free_args(struct image_args *args)
1476 {
1477 
1478 	if (args->buf != NULL) {
1479 		exec_free_args_kva(args->bufkva);
1480 		args->buf = NULL;
1481 	}
1482 	if (args->fname_buf != NULL) {
1483 		free(args->fname_buf, M_TEMP);
1484 		args->fname_buf = NULL;
1485 	}
1486 }
1487 
1488 /*
1489  * A set to functions to fill struct image args.
1490  *
1491  * NOTE: exec_args_add_fname() must be called (possibly with a NULL
1492  * fname) before the other functions.  All exec_args_add_arg() calls must
1493  * be made before any exec_args_add_env() calls.  exec_args_adjust_args()
1494  * may be called any time after exec_args_add_fname().
1495  *
1496  * exec_args_add_fname() - install path to be executed
1497  * exec_args_add_arg() - append an argument string
1498  * exec_args_add_env() - append an env string
1499  * exec_args_adjust_args() - adjust location of the argument list to
1500  *                           allow new arguments to be prepended
1501  */
1502 int
1503 exec_args_add_fname(struct image_args *args, const char *fname,
1504     enum uio_seg segflg)
1505 {
1506 	int error;
1507 	size_t length;
1508 
1509 	KASSERT(args->fname == NULL, ("fname already appended"));
1510 	KASSERT(args->endp == NULL, ("already appending to args"));
1511 
1512 	if (fname != NULL) {
1513 		args->fname = args->buf;
1514 		error = segflg == UIO_SYSSPACE ?
1515 		    copystr(fname, args->fname, PATH_MAX, &length) :
1516 		    copyinstr(fname, args->fname, PATH_MAX, &length);
1517 		if (error != 0)
1518 			return (error == ENAMETOOLONG ? E2BIG : error);
1519 	} else
1520 		length = 0;
1521 
1522 	/* Set up for _arg_*()/_env_*() */
1523 	args->endp = args->buf + length;
1524 	/* begin_argv must be set and kept updated */
1525 	args->begin_argv = args->endp;
1526 	KASSERT(exec_map_entry_size - length >= ARG_MAX,
1527 	    ("too little space remaining for arguments %zu < %zu",
1528 	    exec_map_entry_size - length, (size_t)ARG_MAX));
1529 	args->stringspace = ARG_MAX;
1530 
1531 	return (0);
1532 }
1533 
1534 static int
1535 exec_args_add_str(struct image_args *args, const char *str,
1536     enum uio_seg segflg, int *countp)
1537 {
1538 	int error;
1539 	size_t length;
1540 
1541 	KASSERT(args->endp != NULL, ("endp not initialized"));
1542 	KASSERT(args->begin_argv != NULL, ("begin_argp not initialized"));
1543 
1544 	error = (segflg == UIO_SYSSPACE) ?
1545 	    copystr(str, args->endp, args->stringspace, &length) :
1546 	    copyinstr(str, args->endp, args->stringspace, &length);
1547 	if (error != 0)
1548 		return (error == ENAMETOOLONG ? E2BIG : error);
1549 	args->stringspace -= length;
1550 	args->endp += length;
1551 	(*countp)++;
1552 
1553 	return (0);
1554 }
1555 
1556 int
1557 exec_args_add_arg(struct image_args *args, const char *argp,
1558     enum uio_seg segflg)
1559 {
1560 
1561 	KASSERT(args->envc == 0, ("appending args after env"));
1562 
1563 	return (exec_args_add_str(args, argp, segflg, &args->argc));
1564 }
1565 
1566 int
1567 exec_args_add_env(struct image_args *args, const char *envp,
1568     enum uio_seg segflg)
1569 {
1570 
1571 	if (args->envc == 0)
1572 		args->begin_envv = args->endp;
1573 
1574 	return (exec_args_add_str(args, envp, segflg, &args->envc));
1575 }
1576 
1577 int
1578 exec_args_adjust_args(struct image_args *args, size_t consume, ssize_t extend)
1579 {
1580 	ssize_t offset;
1581 
1582 	KASSERT(args->endp != NULL, ("endp not initialized"));
1583 	KASSERT(args->begin_argv != NULL, ("begin_argp not initialized"));
1584 
1585 	offset = extend - consume;
1586 	if (args->stringspace < offset)
1587 		return (E2BIG);
1588 	memmove(args->begin_argv + extend, args->begin_argv + consume,
1589 	    args->endp - args->begin_argv + consume);
1590 	if (args->envc > 0)
1591 		args->begin_envv += offset;
1592 	args->endp += offset;
1593 	args->stringspace -= offset;
1594 	return (0);
1595 }
1596 
1597 char *
1598 exec_args_get_begin_envv(struct image_args *args)
1599 {
1600 
1601 	KASSERT(args->endp != NULL, ("endp not initialized"));
1602 
1603 	if (args->envc > 0)
1604 		return (args->begin_envv);
1605 	return (args->endp);
1606 }
1607 
1608 /*
1609  * Copy strings out to the new process address space, constructing new arg
1610  * and env vector tables. Return a pointer to the base so that it can be used
1611  * as the initial stack pointer.
1612  */
1613 int
1614 exec_copyout_strings(struct image_params *imgp, uintptr_t *stack_base)
1615 {
1616 	int argc, envc;
1617 	char **vectp;
1618 	char *stringp;
1619 	uintptr_t destp, ustringp;
1620 	struct ps_strings *arginfo;
1621 	struct proc *p;
1622 	struct sysentvec *sysent;
1623 	size_t execpath_len;
1624 	int error, szsigcode;
1625 	char canary[sizeof(long) * 8];
1626 
1627 	p = imgp->proc;
1628 	sysent = p->p_sysent;
1629 
1630 	destp =	PROC_PS_STRINGS(p);
1631 	arginfo = imgp->ps_strings = (void *)destp;
1632 
1633 	/*
1634 	 * Install sigcode.
1635 	 */
1636 	if (sysent->sv_sigcode_base == 0 && sysent->sv_szsigcode != NULL) {
1637 		szsigcode = *(sysent->sv_szsigcode);
1638 		destp -= szsigcode;
1639 		destp = rounddown2(destp, sizeof(void *));
1640 		error = copyout(sysent->sv_sigcode, (void *)destp, szsigcode);
1641 		if (error != 0)
1642 			return (error);
1643 	}
1644 
1645 	/*
1646 	 * Copy the image path for the rtld.
1647 	 */
1648 	if (imgp->execpath != NULL && imgp->auxargs != NULL) {
1649 		execpath_len = strlen(imgp->execpath) + 1;
1650 		destp -= execpath_len;
1651 		destp = rounddown2(destp, sizeof(void *));
1652 		imgp->execpathp = (void *)destp;
1653 		error = copyout(imgp->execpath, imgp->execpathp, execpath_len);
1654 		if (error != 0)
1655 			return (error);
1656 	}
1657 
1658 	/*
1659 	 * Prepare the canary for SSP.
1660 	 */
1661 	arc4rand(canary, sizeof(canary), 0);
1662 	destp -= sizeof(canary);
1663 	imgp->canary = (void *)destp;
1664 	error = copyout(canary, imgp->canary, sizeof(canary));
1665 	if (error != 0)
1666 		return (error);
1667 	imgp->canarylen = sizeof(canary);
1668 
1669 	/*
1670 	 * Prepare the pagesizes array.
1671 	 */
1672 	imgp->pagesizeslen = sizeof(pagesizes[0]) * MAXPAGESIZES;
1673 	destp -= imgp->pagesizeslen;
1674 	destp = rounddown2(destp, sizeof(void *));
1675 	imgp->pagesizes = (void *)destp;
1676 	error = copyout(pagesizes, imgp->pagesizes, imgp->pagesizeslen);
1677 	if (error != 0)
1678 		return (error);
1679 
1680 	/*
1681 	 * Allocate room for the argument and environment strings.
1682 	 */
1683 	destp -= ARG_MAX - imgp->args->stringspace;
1684 	destp = rounddown2(destp, sizeof(void *));
1685 	ustringp = destp;
1686 
1687 	if (imgp->auxargs) {
1688 		/*
1689 		 * Allocate room on the stack for the ELF auxargs
1690 		 * array.  It has up to AT_COUNT entries.
1691 		 */
1692 		destp -= AT_COUNT * sizeof(Elf_Auxinfo);
1693 		destp = rounddown2(destp, sizeof(void *));
1694 	}
1695 
1696 	vectp = (char **)destp;
1697 
1698 	/*
1699 	 * Allocate room for the argv[] and env vectors including the
1700 	 * terminating NULL pointers.
1701 	 */
1702 	vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
1703 
1704 	/*
1705 	 * vectp also becomes our initial stack base
1706 	 */
1707 	*stack_base = (uintptr_t)vectp;
1708 
1709 	stringp = imgp->args->begin_argv;
1710 	argc = imgp->args->argc;
1711 	envc = imgp->args->envc;
1712 
1713 	/*
1714 	 * Copy out strings - arguments and environment.
1715 	 */
1716 	error = copyout(stringp, (void *)ustringp,
1717 	    ARG_MAX - imgp->args->stringspace);
1718 	if (error != 0)
1719 		return (error);
1720 
1721 	/*
1722 	 * Fill in "ps_strings" struct for ps, w, etc.
1723 	 */
1724 	imgp->argv = vectp;
1725 	if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 ||
1726 	    suword32(&arginfo->ps_nargvstr, argc) != 0)
1727 		return (EFAULT);
1728 
1729 	/*
1730 	 * Fill in argument portion of vector table.
1731 	 */
1732 	for (; argc > 0; --argc) {
1733 		if (suword(vectp++, ustringp) != 0)
1734 			return (EFAULT);
1735 		while (*stringp++ != 0)
1736 			ustringp++;
1737 		ustringp++;
1738 	}
1739 
1740 	/* a null vector table pointer separates the argp's from the envp's */
1741 	if (suword(vectp++, 0) != 0)
1742 		return (EFAULT);
1743 
1744 	imgp->envv = vectp;
1745 	if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 ||
1746 	    suword32(&arginfo->ps_nenvstr, envc) != 0)
1747 		return (EFAULT);
1748 
1749 	/*
1750 	 * Fill in environment portion of vector table.
1751 	 */
1752 	for (; envc > 0; --envc) {
1753 		if (suword(vectp++, ustringp) != 0)
1754 			return (EFAULT);
1755 		while (*stringp++ != 0)
1756 			ustringp++;
1757 		ustringp++;
1758 	}
1759 
1760 	/* end of vector table is a null pointer */
1761 	if (suword(vectp, 0) != 0)
1762 		return (EFAULT);
1763 
1764 	if (imgp->auxargs) {
1765 		vectp++;
1766 		error = imgp->sysent->sv_copyout_auxargs(imgp,
1767 		    (uintptr_t)vectp);
1768 		if (error != 0)
1769 			return (error);
1770 	}
1771 
1772 	return (0);
1773 }
1774 
1775 /*
1776  * Check permissions of file to execute.
1777  *	Called with imgp->vp locked.
1778  *	Return 0 for success or error code on failure.
1779  */
1780 int
1781 exec_check_permissions(struct image_params *imgp)
1782 {
1783 	struct vnode *vp = imgp->vp;
1784 	struct vattr *attr = imgp->attr;
1785 	struct thread *td;
1786 	int error;
1787 
1788 	td = curthread;
1789 
1790 	/* Get file attributes */
1791 	error = VOP_GETATTR(vp, attr, td->td_ucred);
1792 	if (error)
1793 		return (error);
1794 
1795 #ifdef MAC
1796 	error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1797 	if (error)
1798 		return (error);
1799 #endif
1800 
1801 	/*
1802 	 * 1) Check if file execution is disabled for the filesystem that
1803 	 *    this file resides on.
1804 	 * 2) Ensure that at least one execute bit is on. Otherwise, a
1805 	 *    privileged user will always succeed, and we don't want this
1806 	 *    to happen unless the file really is executable.
1807 	 * 3) Ensure that the file is a regular file.
1808 	 */
1809 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1810 	    (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1811 	    (attr->va_type != VREG))
1812 		return (EACCES);
1813 
1814 	/*
1815 	 * Zero length files can't be exec'd
1816 	 */
1817 	if (attr->va_size == 0)
1818 		return (ENOEXEC);
1819 
1820 	/*
1821 	 *  Check for execute permission to file based on current credentials.
1822 	 */
1823 	error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1824 	if (error)
1825 		return (error);
1826 
1827 	/*
1828 	 * Check number of open-for-writes on the file and deny execution
1829 	 * if there are any.
1830 	 *
1831 	 * Add a text reference now so no one can write to the
1832 	 * executable while we're activating it.
1833 	 *
1834 	 * Remember if this was set before and unset it in case this is not
1835 	 * actually an executable image.
1836 	 */
1837 	error = VOP_SET_TEXT(vp);
1838 	if (error != 0)
1839 		return (error);
1840 	imgp->textset = true;
1841 
1842 	/*
1843 	 * Call filesystem specific open routine (which does nothing in the
1844 	 * general case).
1845 	 */
1846 	error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1847 	if (error == 0)
1848 		imgp->opened = true;
1849 	return (error);
1850 }
1851 
1852 /*
1853  * Exec handler registration
1854  */
1855 int
1856 exec_register(const struct execsw *execsw_arg)
1857 {
1858 	const struct execsw **es, **xs, **newexecsw;
1859 	u_int count = 2;	/* New slot and trailing NULL */
1860 
1861 	if (execsw)
1862 		for (es = execsw; *es; es++)
1863 			count++;
1864 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1865 	xs = newexecsw;
1866 	if (execsw)
1867 		for (es = execsw; *es; es++)
1868 			*xs++ = *es;
1869 	*xs++ = execsw_arg;
1870 	*xs = NULL;
1871 	if (execsw)
1872 		free(execsw, M_TEMP);
1873 	execsw = newexecsw;
1874 	return (0);
1875 }
1876 
1877 int
1878 exec_unregister(const struct execsw *execsw_arg)
1879 {
1880 	const struct execsw **es, **xs, **newexecsw;
1881 	int count = 1;
1882 
1883 	if (execsw == NULL)
1884 		panic("unregister with no handlers left?\n");
1885 
1886 	for (es = execsw; *es; es++) {
1887 		if (*es == execsw_arg)
1888 			break;
1889 	}
1890 	if (*es == NULL)
1891 		return (ENOENT);
1892 	for (es = execsw; *es; es++)
1893 		if (*es != execsw_arg)
1894 			count++;
1895 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1896 	xs = newexecsw;
1897 	for (es = execsw; *es; es++)
1898 		if (*es != execsw_arg)
1899 			*xs++ = *es;
1900 	*xs = NULL;
1901 	if (execsw)
1902 		free(execsw, M_TEMP);
1903 	execsw = newexecsw;
1904 	return (0);
1905 }
1906 
1907 /*
1908  * Write out a core segment to the compression stream.
1909  */
1910 static int
1911 compress_chunk(struct coredump_params *cp, char *base, char *buf, size_t len)
1912 {
1913 	size_t chunk_len;
1914 	int error;
1915 
1916 	while (len > 0) {
1917 		chunk_len = MIN(len, CORE_BUF_SIZE);
1918 
1919 		/*
1920 		 * We can get EFAULT error here.
1921 		 * In that case zero out the current chunk of the segment.
1922 		 */
1923 		error = copyin(base, buf, chunk_len);
1924 		if (error != 0)
1925 			bzero(buf, chunk_len);
1926 		error = compressor_write(cp->comp, buf, chunk_len);
1927 		if (error != 0)
1928 			break;
1929 		base += chunk_len;
1930 		len -= chunk_len;
1931 	}
1932 	return (error);
1933 }
1934 
1935 int
1936 core_write(struct coredump_params *cp, const void *base, size_t len,
1937     off_t offset, enum uio_seg seg, size_t *resid)
1938 {
1939 
1940 	return (vn_rdwr_inchunks(UIO_WRITE, cp->vp, __DECONST(void *, base),
1941 	    len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1942 	    cp->active_cred, cp->file_cred, resid, cp->td));
1943 }
1944 
1945 int
1946 core_output(char *base, size_t len, off_t offset, struct coredump_params *cp,
1947     void *tmpbuf)
1948 {
1949 	vm_map_t map;
1950 	struct mount *mp;
1951 	size_t resid, runlen;
1952 	int error;
1953 	bool success;
1954 
1955 	KASSERT((uintptr_t)base % PAGE_SIZE == 0,
1956 	    ("%s: user address %p is not page-aligned", __func__, base));
1957 
1958 	if (cp->comp != NULL)
1959 		return (compress_chunk(cp, base, tmpbuf, len));
1960 
1961 	map = &cp->td->td_proc->p_vmspace->vm_map;
1962 	for (; len > 0; base += runlen, offset += runlen, len -= runlen) {
1963 		/*
1964 		 * Attempt to page in all virtual pages in the range.  If a
1965 		 * virtual page is not backed by the pager, it is represented as
1966 		 * a hole in the file.  This can occur with zero-filled
1967 		 * anonymous memory or truncated files, for example.
1968 		 */
1969 		for (runlen = 0; runlen < len; runlen += PAGE_SIZE) {
1970 			if (core_dump_can_intr && curproc_sigkilled())
1971 				return (EINTR);
1972 			error = vm_fault(map, (uintptr_t)base + runlen,
1973 			    VM_PROT_READ, VM_FAULT_NOFILL, NULL);
1974 			if (runlen == 0)
1975 				success = error == KERN_SUCCESS;
1976 			else if ((error == KERN_SUCCESS) != success)
1977 				break;
1978 		}
1979 
1980 		if (success) {
1981 			error = core_write(cp, base, runlen, offset,
1982 			    UIO_USERSPACE, &resid);
1983 			if (error != 0) {
1984 				if (error != EFAULT)
1985 					break;
1986 
1987 				/*
1988 				 * EFAULT may be returned if the user mapping
1989 				 * could not be accessed, e.g., because a mapped
1990 				 * file has been truncated.  Skip the page if no
1991 				 * progress was made, to protect against a
1992 				 * hypothetical scenario where vm_fault() was
1993 				 * successful but core_write() returns EFAULT
1994 				 * anyway.
1995 				 */
1996 				runlen -= resid;
1997 				if (runlen == 0) {
1998 					success = false;
1999 					runlen = PAGE_SIZE;
2000 				}
2001 			}
2002 		}
2003 		if (!success) {
2004 			error = vn_start_write(cp->vp, &mp, V_WAIT);
2005 			if (error != 0)
2006 				break;
2007 			vn_lock(cp->vp, LK_EXCLUSIVE | LK_RETRY);
2008 			error = vn_truncate_locked(cp->vp, offset + runlen,
2009 			    false, cp->td->td_ucred);
2010 			VOP_UNLOCK(cp->vp);
2011 			vn_finished_write(mp);
2012 			if (error != 0)
2013 				break;
2014 		}
2015 	}
2016 	return (error);
2017 }
2018 
2019 /*
2020  * Drain into a core file.
2021  */
2022 int
2023 sbuf_drain_core_output(void *arg, const char *data, int len)
2024 {
2025 	struct coredump_params *cp;
2026 	struct proc *p;
2027 	int error, locked;
2028 
2029 	cp = arg;
2030 	p = cp->td->td_proc;
2031 
2032 	/*
2033 	 * Some kern_proc out routines that print to this sbuf may
2034 	 * call us with the process lock held. Draining with the
2035 	 * non-sleepable lock held is unsafe. The lock is needed for
2036 	 * those routines when dumping a live process. In our case we
2037 	 * can safely release the lock before draining and acquire
2038 	 * again after.
2039 	 */
2040 	locked = PROC_LOCKED(p);
2041 	if (locked)
2042 		PROC_UNLOCK(p);
2043 	if (cp->comp != NULL)
2044 		error = compressor_write(cp->comp, __DECONST(char *, data),
2045 		    len);
2046 	else
2047 		error = core_write(cp, __DECONST(void *, data), len, cp->offset,
2048 		    UIO_SYSSPACE, NULL);
2049 	if (locked)
2050 		PROC_LOCK(p);
2051 	if (error != 0)
2052 		return (-error);
2053 	cp->offset += len;
2054 	return (len);
2055 }
2056