xref: /freebsd/sys/kern/kern_exec.c (revision cb894f746c07001dd5aebfafca596374ec335964)
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 		if (thread_single(p, SINGLE_BOUNDARY) != 0)
311 			error = ERESTART;
312 		PROC_UNLOCK(p);
313 	}
314 	KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0,
315 	    ("nested execve"));
316 	*oldvmspace = p->p_vmspace;
317 	return (error);
318 }
319 
320 void
321 post_execve(struct thread *td, int error, struct vmspace *oldvmspace)
322 {
323 	struct proc *p;
324 
325 	KASSERT(td == curthread, ("non-current thread %p", td));
326 	p = td->td_proc;
327 	if ((p->p_flag & P_HADTHREADS) != 0) {
328 		PROC_LOCK(p);
329 		/*
330 		 * If success, we upgrade to SINGLE_EXIT state to
331 		 * force other threads to suicide.
332 		 */
333 		if (error == EJUSTRETURN)
334 			thread_single(p, SINGLE_EXIT);
335 		else
336 			thread_single_end(p, SINGLE_BOUNDARY);
337 		PROC_UNLOCK(p);
338 	}
339 	exec_cleanup(td, oldvmspace);
340 }
341 
342 /*
343  * kern_execve() has the astonishing property of not always returning to
344  * the caller.  If sufficiently bad things happen during the call to
345  * do_execve(), it can end up calling exit1(); as a result, callers must
346  * avoid doing anything which they might need to undo (e.g., allocating
347  * memory).
348  */
349 int
350 kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p,
351     struct vmspace *oldvmspace)
352 {
353 
354 	TSEXEC(td->td_proc->p_pid, args->begin_argv);
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 
360 	/* Must have at least one argument. */
361 	if (args->argc == 0) {
362 		exec_free_args(args);
363 		return (EINVAL);
364 	}
365 	return (do_execve(td, args, mac_p, oldvmspace));
366 }
367 
368 static void
369 execve_nosetid(struct image_params *imgp)
370 {
371 	imgp->credential_setid = false;
372 	if (imgp->newcred != NULL) {
373 		crfree(imgp->newcred);
374 		imgp->newcred = NULL;
375 	}
376 }
377 
378 /*
379  * In-kernel implementation of execve().  All arguments are assumed to be
380  * userspace pointers from the passed thread.
381  */
382 static int
383 do_execve(struct thread *td, struct image_args *args, struct mac *mac_p,
384     struct vmspace *oldvmspace)
385 {
386 	struct proc *p = td->td_proc;
387 	struct nameidata nd;
388 	struct ucred *oldcred;
389 	struct uidinfo *euip = NULL;
390 	uintptr_t stack_base;
391 	struct image_params image_params, *imgp;
392 	struct vattr attr;
393 	int (*img_first)(struct image_params *);
394 	struct pargs *oldargs = NULL, *newargs = NULL;
395 	struct sigacts *oldsigacts = NULL, *newsigacts = NULL;
396 #ifdef KTRACE
397 	struct ktr_io_params *kiop;
398 #endif
399 	struct vnode *oldtextvp, *newtextvp;
400 	struct vnode *oldtextdvp, *newtextdvp;
401 	char *oldbinname, *newbinname;
402 	bool credential_changing;
403 #ifdef MAC
404 	struct label *interpvplabel = NULL;
405 	bool will_transition;
406 #endif
407 #ifdef HWPMC_HOOKS
408 	struct pmckern_procexec pe;
409 #endif
410 	int error, i, orig_osrel;
411 	uint32_t orig_fctl0;
412 	Elf_Brandinfo *orig_brandinfo;
413 	size_t freepath_size;
414 	static const char fexecv_proc_title[] = "(fexecv)";
415 
416 	imgp = &image_params;
417 	oldtextvp = oldtextdvp = NULL;
418 	newtextvp = newtextdvp = NULL;
419 	newbinname = oldbinname = NULL;
420 #ifdef KTRACE
421 	kiop = NULL;
422 #endif
423 
424 	/*
425 	 * Lock the process and set the P_INEXEC flag to indicate that
426 	 * it should be left alone until we're done here.  This is
427 	 * necessary to avoid race conditions - e.g. in ptrace() -
428 	 * that might allow a local user to illicitly obtain elevated
429 	 * privileges.
430 	 */
431 	PROC_LOCK(p);
432 	KASSERT((p->p_flag & P_INEXEC) == 0,
433 	    ("%s(): process already has P_INEXEC flag", __func__));
434 	p->p_flag |= P_INEXEC;
435 	PROC_UNLOCK(p);
436 
437 	/*
438 	 * Initialize part of the common data
439 	 */
440 	bzero(imgp, sizeof(*imgp));
441 	imgp->proc = p;
442 	imgp->attr = &attr;
443 	imgp->args = args;
444 	oldcred = p->p_ucred;
445 	orig_osrel = p->p_osrel;
446 	orig_fctl0 = p->p_fctl0;
447 	orig_brandinfo = p->p_elf_brandinfo;
448 
449 #ifdef MAC
450 	error = mac_execve_enter(imgp, mac_p);
451 	if (error)
452 		goto exec_fail;
453 #endif
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 
472 		/*
473 		 * Translate the file name. namei() returns a vnode
474 		 * pointer in ni_vp among other things.
475 		 */
476 		NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW |
477 		    AUDITVNODE1 | WANTPARENT, UIO_SYSSPACE,
478 		    args->fname);
479 
480 		error = namei(&nd);
481 		if (error)
482 			goto exec_fail;
483 
484 		newtextvp = nd.ni_vp;
485 		newtextdvp = nd.ni_dvp;
486 		nd.ni_dvp = NULL;
487 		newbinname = malloc(nd.ni_cnd.cn_namelen + 1, M_PARGS,
488 		    M_WAITOK);
489 		memcpy(newbinname, nd.ni_cnd.cn_nameptr, nd.ni_cnd.cn_namelen);
490 		newbinname[nd.ni_cnd.cn_namelen] = '\0';
491 		imgp->vp = newtextvp;
492 
493 		/*
494 		 * Do the best to calculate the full path to the image file.
495 		 */
496 		if (args->fname[0] == '/') {
497 			imgp->execpath = args->fname;
498 		} else {
499 			VOP_UNLOCK(imgp->vp);
500 			freepath_size = MAXPATHLEN;
501 			if (vn_fullpath_hardlink(newtextvp, newtextdvp,
502 			    newbinname, nd.ni_cnd.cn_namelen, &imgp->execpath,
503 			    &imgp->freepath, &freepath_size) != 0)
504 				imgp->execpath = args->fname;
505 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
506 		}
507 	} else if (imgp->interpreter_vp) {
508 		/*
509 		 * An image activator has already provided an open vnode
510 		 */
511 		newtextvp = imgp->interpreter_vp;
512 		imgp->interpreter_vp = NULL;
513 		if (vn_fullpath(newtextvp, &imgp->execpath,
514 		    &imgp->freepath) != 0)
515 			imgp->execpath = args->fname;
516 		vn_lock(newtextvp, LK_SHARED | LK_RETRY);
517 		AUDIT_ARG_VNODE1(newtextvp);
518 		imgp->vp = newtextvp;
519 	} else {
520 		AUDIT_ARG_FD(args->fd);
521 
522 		/*
523 		 * If the descriptors was not opened with O_PATH, then
524 		 * we require that it was opened with O_EXEC or
525 		 * O_RDONLY.  In either case, exec_check_permissions()
526 		 * below checks _current_ file access mode regardless
527 		 * of the permissions additionally checked at the
528 		 * open(2).
529 		 */
530 		error = fgetvp_exec(td, args->fd, &cap_fexecve_rights,
531 		    &newtextvp);
532 		if (error != 0)
533 			goto exec_fail;
534 
535 		if (vn_fullpath(newtextvp, &imgp->execpath,
536 		    &imgp->freepath) != 0)
537 			imgp->execpath = args->fname;
538 		vn_lock(newtextvp, LK_SHARED | LK_RETRY);
539 		AUDIT_ARG_VNODE1(newtextvp);
540 		imgp->vp = newtextvp;
541 	}
542 
543 	/*
544 	 * Check file permissions.  Also 'opens' file and sets its vnode to
545 	 * text mode.
546 	 */
547 	error = exec_check_permissions(imgp);
548 	if (error)
549 		goto exec_fail_dealloc;
550 
551 	imgp->object = imgp->vp->v_object;
552 	if (imgp->object != NULL)
553 		vm_object_reference(imgp->object);
554 
555 	error = exec_map_first_page(imgp);
556 	if (error)
557 		goto exec_fail_dealloc;
558 
559 	imgp->proc->p_osrel = 0;
560 	imgp->proc->p_fctl0 = 0;
561 	imgp->proc->p_elf_brandinfo = NULL;
562 
563 	/*
564 	 * Implement image setuid/setgid.
565 	 *
566 	 * Determine new credentials before attempting image activators
567 	 * so that it can be used by process_exec handlers to determine
568 	 * credential/setid changes.
569 	 *
570 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
571 	 * the process is being traced.
572 	 *
573 	 * We disable setuid/setgid/etc in capability mode on the basis
574 	 * that most setugid applications are not written with that
575 	 * environment in mind, and will therefore almost certainly operate
576 	 * incorrectly. In principle there's no reason that setugid
577 	 * applications might not be useful in capability mode, so we may want
578 	 * to reconsider this conservative design choice in the future.
579 	 *
580 	 * XXXMAC: For the time being, use NOSUID to also prohibit
581 	 * transitions on the file system.
582 	 */
583 	credential_changing = false;
584 	credential_changing |= (attr.va_mode & S_ISUID) &&
585 	    oldcred->cr_uid != attr.va_uid;
586 	credential_changing |= (attr.va_mode & S_ISGID) &&
587 	    oldcred->cr_gid != attr.va_gid;
588 #ifdef MAC
589 	will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
590 	    interpvplabel, imgp) != 0;
591 	credential_changing |= will_transition;
592 #endif
593 
594 	/* Don't inherit PROC_PDEATHSIG_CTL value if setuid/setgid. */
595 	if (credential_changing)
596 		imgp->proc->p_pdeathsig = 0;
597 
598 	if (credential_changing &&
599 #ifdef CAPABILITY_MODE
600 	    ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
601 #endif
602 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
603 	    (p->p_flag & P_TRACED) == 0) {
604 		imgp->credential_setid = true;
605 		VOP_UNLOCK(imgp->vp);
606 		imgp->newcred = crdup(oldcred);
607 		if (attr.va_mode & S_ISUID) {
608 			euip = uifind(attr.va_uid);
609 			change_euid(imgp->newcred, euip);
610 		}
611 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
612 		if (attr.va_mode & S_ISGID)
613 			change_egid(imgp->newcred, attr.va_gid);
614 		/*
615 		 * Implement correct POSIX saved-id behavior.
616 		 *
617 		 * XXXMAC: Note that the current logic will save the
618 		 * uid and gid if a MAC domain transition occurs, even
619 		 * though maybe it shouldn't.
620 		 */
621 		change_svuid(imgp->newcred, imgp->newcred->cr_uid);
622 		change_svgid(imgp->newcred, imgp->newcred->cr_gid);
623 	} else {
624 		/*
625 		 * Implement correct POSIX saved-id behavior.
626 		 *
627 		 * XXX: It's not clear that the existing behavior is
628 		 * POSIX-compliant.  A number of sources indicate that the
629 		 * saved uid/gid should only be updated if the new ruid is
630 		 * not equal to the old ruid, or the new euid is not equal
631 		 * to the old euid and the new euid is not equal to the old
632 		 * ruid.  The FreeBSD code always updates the saved uid/gid.
633 		 * Also, this code uses the new (replaced) euid and egid as
634 		 * the source, which may or may not be the right ones to use.
635 		 */
636 		if (oldcred->cr_svuid != oldcred->cr_uid ||
637 		    oldcred->cr_svgid != oldcred->cr_gid) {
638 			VOP_UNLOCK(imgp->vp);
639 			imgp->newcred = crdup(oldcred);
640 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
641 			change_svuid(imgp->newcred, imgp->newcred->cr_uid);
642 			change_svgid(imgp->newcred, imgp->newcred->cr_gid);
643 		}
644 	}
645 	/* The new credentials are installed into the process later. */
646 
647 	/*
648 	 *	If the current process has a special image activator it
649 	 *	wants to try first, call it.   For example, emulating shell
650 	 *	scripts differently.
651 	 */
652 	error = -1;
653 	if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
654 		error = img_first(imgp);
655 
656 	/*
657 	 *	Loop through the list of image activators, calling each one.
658 	 *	An activator returns -1 if there is no match, 0 on success,
659 	 *	and an error otherwise.
660 	 */
661 	for (i = 0; error == -1 && execsw[i]; ++i) {
662 		if (execsw[i]->ex_imgact == NULL ||
663 		    execsw[i]->ex_imgact == img_first) {
664 			continue;
665 		}
666 		error = (*execsw[i]->ex_imgact)(imgp);
667 	}
668 
669 	if (error) {
670 		if (error == -1)
671 			error = ENOEXEC;
672 		goto exec_fail_dealloc;
673 	}
674 
675 	/*
676 	 * Special interpreter operation, cleanup and loop up to try to
677 	 * activate the interpreter.
678 	 */
679 	if (imgp->interpreted) {
680 		exec_unmap_first_page(imgp);
681 		/*
682 		 * The text reference needs to be removed for scripts.
683 		 * There is a short period before we determine that
684 		 * something is a script where text reference is active.
685 		 * The vnode lock is held over this entire period
686 		 * so nothing should illegitimately be blocked.
687 		 */
688 		MPASS(imgp->textset);
689 		VOP_UNSET_TEXT_CHECKED(newtextvp);
690 		imgp->textset = false;
691 		/* free name buffer and old vnode */
692 #ifdef MAC
693 		mac_execve_interpreter_enter(newtextvp, &interpvplabel);
694 #endif
695 		if (imgp->opened) {
696 			VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td);
697 			imgp->opened = false;
698 		}
699 		vput(newtextvp);
700 		imgp->vp = newtextvp = NULL;
701 		if (args->fname != NULL) {
702 			if (newtextdvp != NULL) {
703 				vrele(newtextdvp);
704 				newtextdvp = NULL;
705 			}
706 			NDFREE_PNBUF(&nd);
707 			free(newbinname, M_PARGS);
708 			newbinname = NULL;
709 		}
710 		vm_object_deallocate(imgp->object);
711 		imgp->object = NULL;
712 		execve_nosetid(imgp);
713 		imgp->execpath = NULL;
714 		free(imgp->freepath, M_TEMP);
715 		imgp->freepath = NULL;
716 		/* set new name to that of the interpreter */
717 		if (imgp->interpreter_vp) {
718 			args->fname = NULL;
719 		} else {
720 			args->fname = imgp->interpreter_name;
721 		}
722 		goto interpret;
723 	}
724 
725 	/*
726 	 * NB: We unlock the vnode here because it is believed that none
727 	 * of the sv_copyout_strings/sv_fixup operations require the vnode.
728 	 */
729 	VOP_UNLOCK(imgp->vp);
730 
731 	if (disallow_high_osrel &&
732 	    P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
733 		error = ENOEXEC;
734 		uprintf("Osrel %d for image %s too high\n", p->p_osrel,
735 		    imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
736 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
737 		goto exec_fail_dealloc;
738 	}
739 
740 	/*
741 	 * Copy out strings (args and env) and initialize stack base.
742 	 */
743 	error = (*p->p_sysent->sv_copyout_strings)(imgp, &stack_base);
744 	if (error != 0) {
745 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
746 		goto exec_fail_dealloc;
747 	}
748 
749 	/*
750 	 * Stack setup.
751 	 */
752 	error = (*p->p_sysent->sv_fixup)(&stack_base, imgp);
753 	if (error != 0) {
754 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
755 		goto exec_fail_dealloc;
756 	}
757 
758 	/*
759 	 * For security and other reasons, the file descriptor table cannot be
760 	 * shared after an exec.
761 	 */
762 	fdunshare(td);
763 	pdunshare(td);
764 	/* close files on exec */
765 	fdcloseexec(td);
766 
767 	/*
768 	 * Malloc things before we need locks.
769 	 */
770 	i = exec_args_get_begin_envv(imgp->args) - imgp->args->begin_argv;
771 	/* Cache arguments if they fit inside our allowance */
772 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
773 		newargs = pargs_alloc(i);
774 		bcopy(imgp->args->begin_argv, newargs->ar_args, i);
775 	}
776 
777 	/*
778 	 * For security and other reasons, signal handlers cannot
779 	 * be shared after an exec. The new process gets a copy of the old
780 	 * handlers. In execsigs(), the new process will have its signals
781 	 * reset.
782 	 */
783 	if (sigacts_shared(p->p_sigacts)) {
784 		oldsigacts = p->p_sigacts;
785 		newsigacts = sigacts_alloc();
786 		sigacts_copy(newsigacts, oldsigacts);
787 	}
788 
789 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
790 
791 	PROC_LOCK(p);
792 	if (oldsigacts)
793 		p->p_sigacts = newsigacts;
794 	/* Stop profiling */
795 	stopprofclock(p);
796 
797 	/* reset caught signals */
798 	execsigs(p);
799 
800 	/* name this process - nameiexec(p, ndp) */
801 	bzero(p->p_comm, sizeof(p->p_comm));
802 	if (args->fname)
803 		bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
804 		    min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
805 	else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0)
806 		bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
807 	bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
808 #ifdef KTR
809 	sched_clear_tdname(td);
810 #endif
811 
812 	/*
813 	 * mark as execed, wakeup the process that vforked (if any) and tell
814 	 * it that it now has its own resources back
815 	 */
816 	p->p_flag |= P_EXEC;
817 	if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0)
818 		p->p_flag2 &= ~P2_NOTRACE;
819 	if ((p->p_flag2 & P2_STKGAP_DISABLE_EXEC) == 0)
820 		p->p_flag2 &= ~P2_STKGAP_DISABLE;
821 	if (p->p_flag & P_PPWAIT) {
822 		p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
823 		cv_broadcast(&p->p_pwait);
824 		/* STOPs are no longer ignored, arrange for AST */
825 		signotify(td);
826 	}
827 
828 	if ((imgp->sysent->sv_setid_allowed != NULL &&
829 	    !(*imgp->sysent->sv_setid_allowed)(td, imgp)) ||
830 	    (p->p_flag2 & P2_NO_NEW_PRIVS) != 0)
831 		execve_nosetid(imgp);
832 
833 	/*
834 	 * Implement image setuid/setgid installation.
835 	 */
836 	if (imgp->credential_setid) {
837 		/*
838 		 * Turn off syscall tracing for set-id programs, except for
839 		 * root.  Record any set-id flags first to make sure that
840 		 * we do not regain any tracing during a possible block.
841 		 */
842 		setsugid(p);
843 #ifdef KTRACE
844 		kiop = ktrprocexec(p);
845 #endif
846 		/*
847 		 * Close any file descriptors 0..2 that reference procfs,
848 		 * then make sure file descriptors 0..2 are in use.
849 		 *
850 		 * Both fdsetugidsafety() and fdcheckstd() may call functions
851 		 * taking sleepable locks, so temporarily drop our locks.
852 		 */
853 		PROC_UNLOCK(p);
854 		VOP_UNLOCK(imgp->vp);
855 		fdsetugidsafety(td);
856 		error = fdcheckstd(td);
857 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
858 		if (error != 0)
859 			goto exec_fail_dealloc;
860 		PROC_LOCK(p);
861 #ifdef MAC
862 		if (will_transition) {
863 			mac_vnode_execve_transition(oldcred, imgp->newcred,
864 			    imgp->vp, interpvplabel, imgp);
865 		}
866 #endif
867 	} else {
868 		if (oldcred->cr_uid == oldcred->cr_ruid &&
869 		    oldcred->cr_gid == oldcred->cr_rgid)
870 			p->p_flag &= ~P_SUGID;
871 	}
872 	/*
873 	 * Set the new credentials.
874 	 */
875 	if (imgp->newcred != NULL) {
876 		proc_set_cred(p, imgp->newcred);
877 		crfree(oldcred);
878 		oldcred = NULL;
879 	}
880 
881 	/*
882 	 * Store the vp for use in kern.proc.pathname.  This vnode was
883 	 * referenced by namei() or by fexecve variant of fname handling.
884 	 */
885 	oldtextvp = p->p_textvp;
886 	p->p_textvp = newtextvp;
887 	oldtextdvp = p->p_textdvp;
888 	p->p_textdvp = newtextdvp;
889 	newtextdvp = NULL;
890 	oldbinname = p->p_binname;
891 	p->p_binname = newbinname;
892 	newbinname = NULL;
893 
894 #ifdef KDTRACE_HOOKS
895 	/*
896 	 * Tell the DTrace fasttrap provider about the exec if it
897 	 * has declared an interest.
898 	 */
899 	if (dtrace_fasttrap_exec)
900 		dtrace_fasttrap_exec(p);
901 #endif
902 
903 	/*
904 	 * Notify others that we exec'd, and clear the P_INEXEC flag
905 	 * as we're now a bona fide freshly-execed process.
906 	 */
907 	KNOTE_LOCKED(p->p_klist, NOTE_EXEC);
908 	p->p_flag &= ~P_INEXEC;
909 
910 	/* clear "fork but no exec" flag, as we _are_ execing */
911 	p->p_acflag &= ~AFORK;
912 
913 	/*
914 	 * Free any previous argument cache and replace it with
915 	 * the new argument cache, if any.
916 	 */
917 	oldargs = p->p_args;
918 	p->p_args = newargs;
919 	newargs = NULL;
920 
921 	PROC_UNLOCK(p);
922 
923 #ifdef	HWPMC_HOOKS
924 	/*
925 	 * Check if system-wide sampling is in effect or if the
926 	 * current process is using PMCs.  If so, do exec() time
927 	 * processing.  This processing needs to happen AFTER the
928 	 * P_INEXEC flag is cleared.
929 	 */
930 	if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
931 		VOP_UNLOCK(imgp->vp);
932 		pe.pm_credentialschanged = credential_changing;
933 		pe.pm_entryaddr = imgp->entry_addr;
934 
935 		PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
936 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
937 	}
938 #endif
939 
940 	/* Set values passed into the program in registers. */
941 	(*p->p_sysent->sv_setregs)(td, imgp, stack_base);
942 
943 	VOP_MMAPPED(imgp->vp);
944 
945 	SDT_PROBE1(proc, , , exec__success, args->fname);
946 
947 exec_fail_dealloc:
948 	if (error != 0) {
949 		p->p_osrel = orig_osrel;
950 		p->p_fctl0 = orig_fctl0;
951 		p->p_elf_brandinfo = orig_brandinfo;
952 	}
953 
954 	if (imgp->firstpage != NULL)
955 		exec_unmap_first_page(imgp);
956 
957 	if (imgp->vp != NULL) {
958 		if (imgp->opened)
959 			VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
960 		if (imgp->textset)
961 			VOP_UNSET_TEXT_CHECKED(imgp->vp);
962 		if (error != 0)
963 			vput(imgp->vp);
964 		else
965 			VOP_UNLOCK(imgp->vp);
966 		if (args->fname != NULL)
967 			NDFREE_PNBUF(&nd);
968 		if (newtextdvp != NULL)
969 			vrele(newtextdvp);
970 		free(newbinname, M_PARGS);
971 	}
972 
973 	if (imgp->object != NULL)
974 		vm_object_deallocate(imgp->object);
975 
976 	free(imgp->freepath, M_TEMP);
977 
978 	if (error == 0) {
979 		if (p->p_ptevents & PTRACE_EXEC) {
980 			PROC_LOCK(p);
981 			if (p->p_ptevents & PTRACE_EXEC)
982 				td->td_dbgflags |= TDB_EXEC;
983 			PROC_UNLOCK(p);
984 		}
985 	} else {
986 exec_fail:
987 		/* we're done here, clear P_INEXEC */
988 		PROC_LOCK(p);
989 		p->p_flag &= ~P_INEXEC;
990 		PROC_UNLOCK(p);
991 
992 		SDT_PROBE1(proc, , , exec__failure, error);
993 	}
994 
995 	if (imgp->newcred != NULL && oldcred != NULL)
996 		crfree(imgp->newcred);
997 
998 #ifdef MAC
999 	mac_execve_exit(imgp);
1000 	mac_execve_interpreter_exit(interpvplabel);
1001 #endif
1002 	exec_free_args(args);
1003 
1004 	/*
1005 	 * Handle deferred decrement of ref counts.
1006 	 */
1007 	if (oldtextvp != NULL)
1008 		vrele(oldtextvp);
1009 	if (oldtextdvp != NULL)
1010 		vrele(oldtextdvp);
1011 	free(oldbinname, M_PARGS);
1012 #ifdef KTRACE
1013 	ktr_io_params_free(kiop);
1014 #endif
1015 	pargs_drop(oldargs);
1016 	pargs_drop(newargs);
1017 	if (oldsigacts != NULL)
1018 		sigacts_free(oldsigacts);
1019 	if (euip != NULL)
1020 		uifree(euip);
1021 
1022 	if (error && imgp->vmspace_destroyed) {
1023 		/* sorry, no more process anymore. exit gracefully */
1024 		exec_cleanup(td, oldvmspace);
1025 		exit1(td, 0, SIGABRT);
1026 		/* NOT REACHED */
1027 	}
1028 
1029 #ifdef KTRACE
1030 	if (error == 0)
1031 		ktrprocctor(p);
1032 #endif
1033 
1034 	/*
1035 	 * We don't want cpu_set_syscall_retval() to overwrite any of
1036 	 * the register values put in place by exec_setregs().
1037 	 * Implementations of cpu_set_syscall_retval() will leave
1038 	 * registers unmodified when returning EJUSTRETURN.
1039 	 */
1040 	return (error == 0 ? EJUSTRETURN : error);
1041 }
1042 
1043 void
1044 exec_cleanup(struct thread *td, struct vmspace *oldvmspace)
1045 {
1046 	if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
1047 		KASSERT(td->td_proc->p_vmspace != oldvmspace,
1048 		    ("oldvmspace still used"));
1049 		vmspace_free(oldvmspace);
1050 		td->td_pflags &= ~TDP_EXECVMSPC;
1051 	}
1052 }
1053 
1054 int
1055 exec_map_first_page(struct image_params *imgp)
1056 {
1057 	vm_object_t object;
1058 	vm_page_t m;
1059 	int error;
1060 
1061 	if (imgp->firstpage != NULL)
1062 		exec_unmap_first_page(imgp);
1063 
1064 	object = imgp->vp->v_object;
1065 	if (object == NULL)
1066 		return (EACCES);
1067 #if VM_NRESERVLEVEL > 0
1068 	if ((object->flags & OBJ_COLORED) == 0) {
1069 		VM_OBJECT_WLOCK(object);
1070 		vm_object_color(object, 0);
1071 		VM_OBJECT_WUNLOCK(object);
1072 	}
1073 #endif
1074 	error = vm_page_grab_valid_unlocked(&m, object, 0,
1075 	    VM_ALLOC_COUNT(VM_INITIAL_PAGEIN) |
1076 	    VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED);
1077 
1078 	if (error != VM_PAGER_OK)
1079 		return (EIO);
1080 	imgp->firstpage = sf_buf_alloc(m, 0);
1081 	imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
1082 
1083 	return (0);
1084 }
1085 
1086 void
1087 exec_unmap_first_page(struct image_params *imgp)
1088 {
1089 	vm_page_t m;
1090 
1091 	if (imgp->firstpage != NULL) {
1092 		m = sf_buf_page(imgp->firstpage);
1093 		sf_buf_free(imgp->firstpage);
1094 		imgp->firstpage = NULL;
1095 		vm_page_unwire(m, PQ_ACTIVE);
1096 	}
1097 }
1098 
1099 void
1100 exec_onexec_old(struct thread *td)
1101 {
1102 	sigfastblock_clear(td);
1103 	umtx_exec(td->td_proc);
1104 }
1105 
1106 /*
1107  * This is an optimization which removes the unmanaged shared page
1108  * mapping. In combination with pmap_remove_pages(), which cleans all
1109  * managed mappings in the process' vmspace pmap, no work will be left
1110  * for pmap_remove(min, max).
1111  */
1112 void
1113 exec_free_abi_mappings(struct proc *p)
1114 {
1115 	struct vmspace *vmspace;
1116 
1117 	vmspace = p->p_vmspace;
1118 	if (refcount_load(&vmspace->vm_refcnt) != 1)
1119 		return;
1120 
1121 	if (!PROC_HAS_SHP(p))
1122 		return;
1123 
1124 	pmap_remove(vmspace_pmap(vmspace), vmspace->vm_shp_base,
1125 	    vmspace->vm_shp_base + p->p_sysent->sv_shared_page_len);
1126 }
1127 
1128 /*
1129  * Run down the current address space and install a new one.
1130  */
1131 int
1132 exec_new_vmspace(struct image_params *imgp, struct sysentvec *sv)
1133 {
1134 	int error;
1135 	struct proc *p = imgp->proc;
1136 	struct vmspace *vmspace = p->p_vmspace;
1137 	struct thread *td = curthread;
1138 	vm_offset_t sv_minuser;
1139 	vm_map_t map;
1140 
1141 	imgp->vmspace_destroyed = true;
1142 	imgp->sysent = sv;
1143 
1144 	if (p->p_sysent->sv_onexec_old != NULL)
1145 		p->p_sysent->sv_onexec_old(td);
1146 	itimers_exec(p);
1147 
1148 	EVENTHANDLER_DIRECT_INVOKE(process_exec, p, imgp);
1149 
1150 	/*
1151 	 * Blow away entire process VM, if address space not shared,
1152 	 * otherwise, create a new VM space so that other threads are
1153 	 * not disrupted
1154 	 */
1155 	map = &vmspace->vm_map;
1156 	if (map_at_zero)
1157 		sv_minuser = sv->sv_minuser;
1158 	else
1159 		sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1160 	if (refcount_load(&vmspace->vm_refcnt) == 1 &&
1161 	    vm_map_min(map) == sv_minuser &&
1162 	    vm_map_max(map) == sv->sv_maxuser &&
1163 	    cpu_exec_vmspace_reuse(p, map)) {
1164 		exec_free_abi_mappings(p);
1165 		shmexit(vmspace);
1166 		pmap_remove_pages(vmspace_pmap(vmspace));
1167 		vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1168 		/*
1169 		 * An exec terminates mlockall(MCL_FUTURE).
1170 		 * ASLR and W^X states must be re-evaluated.
1171 		 */
1172 		vm_map_lock(map);
1173 		vm_map_modflags(map, 0, MAP_WIREFUTURE | MAP_ASLR |
1174 		    MAP_ASLR_IGNSTART | MAP_ASLR_STACK | MAP_WXORX);
1175 		vm_map_unlock(map);
1176 	} else {
1177 		error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1178 		if (error)
1179 			return (error);
1180 		vmspace = p->p_vmspace;
1181 		map = &vmspace->vm_map;
1182 	}
1183 	map->flags |= imgp->map_flags;
1184 
1185 	return (sv->sv_onexec != NULL ? sv->sv_onexec(p, imgp) : 0);
1186 }
1187 
1188 /*
1189  * Compute the stack size limit and map the main process stack.
1190  * Map the shared page.
1191  */
1192 int
1193 exec_map_stack(struct image_params *imgp)
1194 {
1195 	struct rlimit rlim_stack;
1196 	struct sysentvec *sv;
1197 	struct proc *p;
1198 	vm_map_t map;
1199 	struct vmspace *vmspace;
1200 	vm_offset_t stack_addr, stack_top;
1201 	vm_offset_t sharedpage_addr;
1202 	u_long ssiz;
1203 	int error, find_space, stack_off;
1204 	vm_prot_t stack_prot;
1205 	vm_object_t obj;
1206 
1207 	p = imgp->proc;
1208 	sv = p->p_sysent;
1209 
1210 	if (imgp->stack_sz != 0) {
1211 		ssiz = trunc_page(imgp->stack_sz);
1212 		PROC_LOCK(p);
1213 		lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack);
1214 		PROC_UNLOCK(p);
1215 		if (ssiz > rlim_stack.rlim_max)
1216 			ssiz = rlim_stack.rlim_max;
1217 		if (ssiz > rlim_stack.rlim_cur) {
1218 			rlim_stack.rlim_cur = ssiz;
1219 			kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack);
1220 		}
1221 	} else if (sv->sv_maxssiz != NULL) {
1222 		ssiz = *sv->sv_maxssiz;
1223 	} else {
1224 		ssiz = maxssiz;
1225 	}
1226 
1227 	vmspace = p->p_vmspace;
1228 	map = &vmspace->vm_map;
1229 
1230 	stack_prot = sv->sv_shared_page_obj != NULL && imgp->stack_prot != 0 ?
1231 	    imgp->stack_prot : sv->sv_stackprot;
1232 	if ((map->flags & MAP_ASLR_STACK) != 0) {
1233 		stack_addr = round_page((vm_offset_t)p->p_vmspace->vm_daddr +
1234 		    lim_max(curthread, RLIMIT_DATA));
1235 		find_space = VMFS_ANY_SPACE;
1236 	} else {
1237 		stack_addr = sv->sv_usrstack - ssiz;
1238 		find_space = VMFS_NO_SPACE;
1239 	}
1240 	error = vm_map_find(map, NULL, 0, &stack_addr, (vm_size_t)ssiz,
1241 	    sv->sv_usrstack, find_space, stack_prot, VM_PROT_ALL,
1242 	    MAP_STACK_GROWS_DOWN);
1243 	if (error != KERN_SUCCESS) {
1244 		uprintf("exec_new_vmspace: mapping stack size %#jx prot %#x "
1245 		    "failed, mach error %d errno %d\n", (uintmax_t)ssiz,
1246 		    stack_prot, error, vm_mmap_to_errno(error));
1247 		return (vm_mmap_to_errno(error));
1248 	}
1249 
1250 	stack_top = stack_addr + ssiz;
1251 	if ((map->flags & MAP_ASLR_STACK) != 0) {
1252 		/* Randomize within the first page of the stack. */
1253 		arc4rand(&stack_off, sizeof(stack_off), 0);
1254 		stack_top -= rounddown2(stack_off & PAGE_MASK, sizeof(void *));
1255 	}
1256 
1257 	/* Map a shared page */
1258 	obj = sv->sv_shared_page_obj;
1259 	if (obj == NULL) {
1260 		sharedpage_addr = 0;
1261 		goto out;
1262 	}
1263 
1264 	/*
1265 	 * If randomization is disabled then the shared page will
1266 	 * be mapped at address specified in sysentvec.
1267 	 * Otherwise any address above .data section can be selected.
1268 	 * Same logic is used for stack address randomization.
1269 	 * If the address randomization is applied map a guard page
1270 	 * at the top of UVA.
1271 	 */
1272 	vm_object_reference(obj);
1273 	if ((imgp->imgp_flags & IMGP_ASLR_SHARED_PAGE) != 0) {
1274 		sharedpage_addr = round_page((vm_offset_t)p->p_vmspace->vm_daddr +
1275 		    lim_max(curthread, RLIMIT_DATA));
1276 
1277 		error = vm_map_fixed(map, NULL, 0,
1278 		    sv->sv_maxuser - PAGE_SIZE, PAGE_SIZE,
1279 		    VM_PROT_NONE, VM_PROT_NONE, MAP_CREATE_GUARD);
1280 		if (error != KERN_SUCCESS) {
1281 			/*
1282 			 * This is not fatal, so let's just print a warning
1283 			 * and continue.
1284 			 */
1285 			uprintf("%s: Mapping guard page at the top of UVA failed"
1286 			    " mach error %d errno %d",
1287 			    __func__, error, vm_mmap_to_errno(error));
1288 		}
1289 
1290 		error = vm_map_find(map, obj, 0,
1291 		    &sharedpage_addr, sv->sv_shared_page_len,
1292 		    sv->sv_maxuser, VMFS_ANY_SPACE,
1293 		    VM_PROT_READ | VM_PROT_EXECUTE,
1294 		    VM_PROT_READ | VM_PROT_EXECUTE,
1295 		    MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1296 	} else {
1297 		sharedpage_addr = sv->sv_shared_page_base;
1298 		vm_map_fixed(map, obj, 0,
1299 		    sharedpage_addr, sv->sv_shared_page_len,
1300 		    VM_PROT_READ | VM_PROT_EXECUTE,
1301 		    VM_PROT_READ | VM_PROT_EXECUTE,
1302 		    MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1303 	}
1304 	if (error != KERN_SUCCESS) {
1305 		uprintf("%s: mapping shared page at addr: %p"
1306 		    "failed, mach error %d errno %d\n", __func__,
1307 		    (void *)sharedpage_addr, error, vm_mmap_to_errno(error));
1308 		vm_object_deallocate(obj);
1309 		return (vm_mmap_to_errno(error));
1310 	}
1311 out:
1312 	/*
1313 	 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they
1314 	 * are still used to enforce the stack rlimit on the process stack.
1315 	 */
1316 	vmspace->vm_maxsaddr = (char *)stack_addr;
1317 	vmspace->vm_stacktop = stack_top;
1318 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1319 	vmspace->vm_shp_base = sharedpage_addr;
1320 
1321 	return (0);
1322 }
1323 
1324 /*
1325  * Copy out argument and environment strings from the old process address
1326  * space into the temporary string buffer.
1327  */
1328 int
1329 exec_copyin_args(struct image_args *args, const char *fname,
1330     enum uio_seg segflg, char **argv, char **envv)
1331 {
1332 	u_long arg, env;
1333 	int error;
1334 
1335 	bzero(args, sizeof(*args));
1336 	if (argv == NULL)
1337 		return (EFAULT);
1338 
1339 	/*
1340 	 * Allocate demand-paged memory for the file name, argument, and
1341 	 * environment strings.
1342 	 */
1343 	error = exec_alloc_args(args);
1344 	if (error != 0)
1345 		return (error);
1346 
1347 	/*
1348 	 * Copy the file name.
1349 	 */
1350 	error = exec_args_add_fname(args, fname, segflg);
1351 	if (error != 0)
1352 		goto err_exit;
1353 
1354 	/*
1355 	 * extract arguments first
1356 	 */
1357 	for (;;) {
1358 		error = fueword(argv++, &arg);
1359 		if (error == -1) {
1360 			error = EFAULT;
1361 			goto err_exit;
1362 		}
1363 		if (arg == 0)
1364 			break;
1365 		error = exec_args_add_arg(args, (char *)(uintptr_t)arg,
1366 		    UIO_USERSPACE);
1367 		if (error != 0)
1368 			goto err_exit;
1369 	}
1370 
1371 	/*
1372 	 * extract environment strings
1373 	 */
1374 	if (envv) {
1375 		for (;;) {
1376 			error = fueword(envv++, &env);
1377 			if (error == -1) {
1378 				error = EFAULT;
1379 				goto err_exit;
1380 			}
1381 			if (env == 0)
1382 				break;
1383 			error = exec_args_add_env(args,
1384 			    (char *)(uintptr_t)env, UIO_USERSPACE);
1385 			if (error != 0)
1386 				goto err_exit;
1387 		}
1388 	}
1389 
1390 	return (0);
1391 
1392 err_exit:
1393 	exec_free_args(args);
1394 	return (error);
1395 }
1396 
1397 struct exec_args_kva {
1398 	vm_offset_t addr;
1399 	u_int gen;
1400 	SLIST_ENTRY(exec_args_kva) next;
1401 };
1402 
1403 DPCPU_DEFINE_STATIC(struct exec_args_kva *, exec_args_kva);
1404 
1405 static SLIST_HEAD(, exec_args_kva) exec_args_kva_freelist;
1406 static struct mtx exec_args_kva_mtx;
1407 static u_int exec_args_gen;
1408 
1409 static void
1410 exec_prealloc_args_kva(void *arg __unused)
1411 {
1412 	struct exec_args_kva *argkva;
1413 	u_int i;
1414 
1415 	SLIST_INIT(&exec_args_kva_freelist);
1416 	mtx_init(&exec_args_kva_mtx, "exec args kva", NULL, MTX_DEF);
1417 	for (i = 0; i < exec_map_entries; i++) {
1418 		argkva = malloc(sizeof(*argkva), M_PARGS, M_WAITOK);
1419 		argkva->addr = kmap_alloc_wait(exec_map, exec_map_entry_size);
1420 		argkva->gen = exec_args_gen;
1421 		SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1422 	}
1423 }
1424 SYSINIT(exec_args_kva, SI_SUB_EXEC, SI_ORDER_ANY, exec_prealloc_args_kva, NULL);
1425 
1426 static vm_offset_t
1427 exec_alloc_args_kva(void **cookie)
1428 {
1429 	struct exec_args_kva *argkva;
1430 
1431 	argkva = (void *)atomic_readandclear_ptr(
1432 	    (uintptr_t *)DPCPU_PTR(exec_args_kva));
1433 	if (argkva == NULL) {
1434 		mtx_lock(&exec_args_kva_mtx);
1435 		while ((argkva = SLIST_FIRST(&exec_args_kva_freelist)) == NULL)
1436 			(void)mtx_sleep(&exec_args_kva_freelist,
1437 			    &exec_args_kva_mtx, 0, "execkva", 0);
1438 		SLIST_REMOVE_HEAD(&exec_args_kva_freelist, next);
1439 		mtx_unlock(&exec_args_kva_mtx);
1440 	}
1441 	kasan_mark((void *)argkva->addr, exec_map_entry_size,
1442 	    exec_map_entry_size, 0);
1443 	*(struct exec_args_kva **)cookie = argkva;
1444 	return (argkva->addr);
1445 }
1446 
1447 static void
1448 exec_release_args_kva(struct exec_args_kva *argkva, u_int gen)
1449 {
1450 	vm_offset_t base;
1451 
1452 	base = argkva->addr;
1453 	kasan_mark((void *)argkva->addr, 0, exec_map_entry_size,
1454 	    KASAN_EXEC_ARGS_FREED);
1455 	if (argkva->gen != gen) {
1456 		(void)vm_map_madvise(exec_map, base, base + exec_map_entry_size,
1457 		    MADV_FREE);
1458 		argkva->gen = gen;
1459 	}
1460 	if (!atomic_cmpset_ptr((uintptr_t *)DPCPU_PTR(exec_args_kva),
1461 	    (uintptr_t)NULL, (uintptr_t)argkva)) {
1462 		mtx_lock(&exec_args_kva_mtx);
1463 		SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1464 		wakeup_one(&exec_args_kva_freelist);
1465 		mtx_unlock(&exec_args_kva_mtx);
1466 	}
1467 }
1468 
1469 static void
1470 exec_free_args_kva(void *cookie)
1471 {
1472 
1473 	exec_release_args_kva(cookie, exec_args_gen);
1474 }
1475 
1476 static void
1477 exec_args_kva_lowmem(void *arg __unused)
1478 {
1479 	SLIST_HEAD(, exec_args_kva) head;
1480 	struct exec_args_kva *argkva;
1481 	u_int gen;
1482 	int i;
1483 
1484 	gen = atomic_fetchadd_int(&exec_args_gen, 1) + 1;
1485 
1486 	/*
1487 	 * Force an madvise of each KVA range. Any currently allocated ranges
1488 	 * will have MADV_FREE applied once they are freed.
1489 	 */
1490 	SLIST_INIT(&head);
1491 	mtx_lock(&exec_args_kva_mtx);
1492 	SLIST_SWAP(&head, &exec_args_kva_freelist, exec_args_kva);
1493 	mtx_unlock(&exec_args_kva_mtx);
1494 	while ((argkva = SLIST_FIRST(&head)) != NULL) {
1495 		SLIST_REMOVE_HEAD(&head, next);
1496 		exec_release_args_kva(argkva, gen);
1497 	}
1498 
1499 	CPU_FOREACH(i) {
1500 		argkva = (void *)atomic_readandclear_ptr(
1501 		    (uintptr_t *)DPCPU_ID_PTR(i, exec_args_kva));
1502 		if (argkva != NULL)
1503 			exec_release_args_kva(argkva, gen);
1504 	}
1505 }
1506 EVENTHANDLER_DEFINE(vm_lowmem, exec_args_kva_lowmem, NULL,
1507     EVENTHANDLER_PRI_ANY);
1508 
1509 /*
1510  * Allocate temporary demand-paged, zero-filled memory for the file name,
1511  * argument, and environment strings.
1512  */
1513 int
1514 exec_alloc_args(struct image_args *args)
1515 {
1516 
1517 	args->buf = (char *)exec_alloc_args_kva(&args->bufkva);
1518 	return (0);
1519 }
1520 
1521 void
1522 exec_free_args(struct image_args *args)
1523 {
1524 
1525 	if (args->buf != NULL) {
1526 		exec_free_args_kva(args->bufkva);
1527 		args->buf = NULL;
1528 	}
1529 	if (args->fname_buf != NULL) {
1530 		free(args->fname_buf, M_TEMP);
1531 		args->fname_buf = NULL;
1532 	}
1533 }
1534 
1535 /*
1536  * A set to functions to fill struct image args.
1537  *
1538  * NOTE: exec_args_add_fname() must be called (possibly with a NULL
1539  * fname) before the other functions.  All exec_args_add_arg() calls must
1540  * be made before any exec_args_add_env() calls.  exec_args_adjust_args()
1541  * may be called any time after exec_args_add_fname().
1542  *
1543  * exec_args_add_fname() - install path to be executed
1544  * exec_args_add_arg() - append an argument string
1545  * exec_args_add_env() - append an env string
1546  * exec_args_adjust_args() - adjust location of the argument list to
1547  *                           allow new arguments to be prepended
1548  */
1549 int
1550 exec_args_add_fname(struct image_args *args, const char *fname,
1551     enum uio_seg segflg)
1552 {
1553 	int error;
1554 	size_t length;
1555 
1556 	KASSERT(args->fname == NULL, ("fname already appended"));
1557 	KASSERT(args->endp == NULL, ("already appending to args"));
1558 
1559 	if (fname != NULL) {
1560 		args->fname = args->buf;
1561 		error = segflg == UIO_SYSSPACE ?
1562 		    copystr(fname, args->fname, PATH_MAX, &length) :
1563 		    copyinstr(fname, args->fname, PATH_MAX, &length);
1564 		if (error != 0)
1565 			return (error == ENAMETOOLONG ? E2BIG : error);
1566 	} else
1567 		length = 0;
1568 
1569 	/* Set up for _arg_*()/_env_*() */
1570 	args->endp = args->buf + length;
1571 	/* begin_argv must be set and kept updated */
1572 	args->begin_argv = args->endp;
1573 	KASSERT(exec_map_entry_size - length >= ARG_MAX,
1574 	    ("too little space remaining for arguments %zu < %zu",
1575 	    exec_map_entry_size - length, (size_t)ARG_MAX));
1576 	args->stringspace = ARG_MAX;
1577 
1578 	return (0);
1579 }
1580 
1581 static int
1582 exec_args_add_str(struct image_args *args, const char *str,
1583     enum uio_seg segflg, int *countp)
1584 {
1585 	int error;
1586 	size_t length;
1587 
1588 	KASSERT(args->endp != NULL, ("endp not initialized"));
1589 	KASSERT(args->begin_argv != NULL, ("begin_argp not initialized"));
1590 
1591 	error = (segflg == UIO_SYSSPACE) ?
1592 	    copystr(str, args->endp, args->stringspace, &length) :
1593 	    copyinstr(str, args->endp, args->stringspace, &length);
1594 	if (error != 0)
1595 		return (error == ENAMETOOLONG ? E2BIG : error);
1596 	args->stringspace -= length;
1597 	args->endp += length;
1598 	(*countp)++;
1599 
1600 	return (0);
1601 }
1602 
1603 int
1604 exec_args_add_arg(struct image_args *args, const char *argp,
1605     enum uio_seg segflg)
1606 {
1607 
1608 	KASSERT(args->envc == 0, ("appending args after env"));
1609 
1610 	return (exec_args_add_str(args, argp, segflg, &args->argc));
1611 }
1612 
1613 int
1614 exec_args_add_env(struct image_args *args, const char *envp,
1615     enum uio_seg segflg)
1616 {
1617 
1618 	if (args->envc == 0)
1619 		args->begin_envv = args->endp;
1620 
1621 	return (exec_args_add_str(args, envp, segflg, &args->envc));
1622 }
1623 
1624 int
1625 exec_args_adjust_args(struct image_args *args, size_t consume, ssize_t extend)
1626 {
1627 	ssize_t offset;
1628 
1629 	KASSERT(args->endp != NULL, ("endp not initialized"));
1630 	KASSERT(args->begin_argv != NULL, ("begin_argp not initialized"));
1631 
1632 	offset = extend - consume;
1633 	if (args->stringspace < offset)
1634 		return (E2BIG);
1635 	memmove(args->begin_argv + extend, args->begin_argv + consume,
1636 	    args->endp - args->begin_argv + consume);
1637 	if (args->envc > 0)
1638 		args->begin_envv += offset;
1639 	args->endp += offset;
1640 	args->stringspace -= offset;
1641 	return (0);
1642 }
1643 
1644 char *
1645 exec_args_get_begin_envv(struct image_args *args)
1646 {
1647 
1648 	KASSERT(args->endp != NULL, ("endp not initialized"));
1649 
1650 	if (args->envc > 0)
1651 		return (args->begin_envv);
1652 	return (args->endp);
1653 }
1654 
1655 /*
1656  * Copy strings out to the new process address space, constructing new arg
1657  * and env vector tables. Return a pointer to the base so that it can be used
1658  * as the initial stack pointer.
1659  */
1660 int
1661 exec_copyout_strings(struct image_params *imgp, uintptr_t *stack_base)
1662 {
1663 	int argc, envc;
1664 	char **vectp;
1665 	char *stringp;
1666 	uintptr_t destp, ustringp;
1667 	struct ps_strings *arginfo;
1668 	struct proc *p;
1669 	struct sysentvec *sysent;
1670 	size_t execpath_len;
1671 	int error, szsigcode;
1672 	char canary[sizeof(long) * 8];
1673 
1674 	p = imgp->proc;
1675 	sysent = p->p_sysent;
1676 
1677 	destp =	PROC_PS_STRINGS(p);
1678 	arginfo = imgp->ps_strings = (void *)destp;
1679 
1680 	/*
1681 	 * Install sigcode.
1682 	 */
1683 	if (sysent->sv_shared_page_base == 0 && sysent->sv_szsigcode != NULL) {
1684 		szsigcode = *(sysent->sv_szsigcode);
1685 		destp -= szsigcode;
1686 		destp = rounddown2(destp, sizeof(void *));
1687 		error = copyout(sysent->sv_sigcode, (void *)destp, szsigcode);
1688 		if (error != 0)
1689 			return (error);
1690 	}
1691 
1692 	/*
1693 	 * Copy the image path for the rtld.
1694 	 */
1695 	if (imgp->execpath != NULL && imgp->auxargs != NULL) {
1696 		execpath_len = strlen(imgp->execpath) + 1;
1697 		destp -= execpath_len;
1698 		destp = rounddown2(destp, sizeof(void *));
1699 		imgp->execpathp = (void *)destp;
1700 		error = copyout(imgp->execpath, imgp->execpathp, execpath_len);
1701 		if (error != 0)
1702 			return (error);
1703 	}
1704 
1705 	/*
1706 	 * Prepare the canary for SSP.
1707 	 */
1708 	arc4rand(canary, sizeof(canary), 0);
1709 	destp -= sizeof(canary);
1710 	imgp->canary = (void *)destp;
1711 	error = copyout(canary, imgp->canary, sizeof(canary));
1712 	if (error != 0)
1713 		return (error);
1714 	imgp->canarylen = sizeof(canary);
1715 
1716 	/*
1717 	 * Prepare the pagesizes array.
1718 	 */
1719 	imgp->pagesizeslen = sizeof(pagesizes[0]) * MAXPAGESIZES;
1720 	destp -= imgp->pagesizeslen;
1721 	destp = rounddown2(destp, sizeof(void *));
1722 	imgp->pagesizes = (void *)destp;
1723 	error = copyout(pagesizes, imgp->pagesizes, imgp->pagesizeslen);
1724 	if (error != 0)
1725 		return (error);
1726 
1727 	/*
1728 	 * Allocate room for the argument and environment strings.
1729 	 */
1730 	destp -= ARG_MAX - imgp->args->stringspace;
1731 	destp = rounddown2(destp, sizeof(void *));
1732 	ustringp = destp;
1733 
1734 	if (imgp->auxargs) {
1735 		/*
1736 		 * Allocate room on the stack for the ELF auxargs
1737 		 * array.  It has up to AT_COUNT entries.
1738 		 */
1739 		destp -= AT_COUNT * sizeof(Elf_Auxinfo);
1740 		destp = rounddown2(destp, sizeof(void *));
1741 	}
1742 
1743 	vectp = (char **)destp;
1744 
1745 	/*
1746 	 * Allocate room for the argv[] and env vectors including the
1747 	 * terminating NULL pointers.
1748 	 */
1749 	vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
1750 
1751 	/*
1752 	 * vectp also becomes our initial stack base
1753 	 */
1754 	*stack_base = (uintptr_t)vectp;
1755 
1756 	stringp = imgp->args->begin_argv;
1757 	argc = imgp->args->argc;
1758 	envc = imgp->args->envc;
1759 
1760 	/*
1761 	 * Copy out strings - arguments and environment.
1762 	 */
1763 	error = copyout(stringp, (void *)ustringp,
1764 	    ARG_MAX - imgp->args->stringspace);
1765 	if (error != 0)
1766 		return (error);
1767 
1768 	/*
1769 	 * Fill in "ps_strings" struct for ps, w, etc.
1770 	 */
1771 	imgp->argv = vectp;
1772 	if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 ||
1773 	    suword32(&arginfo->ps_nargvstr, argc) != 0)
1774 		return (EFAULT);
1775 
1776 	/*
1777 	 * Fill in argument portion of vector table.
1778 	 */
1779 	for (; argc > 0; --argc) {
1780 		if (suword(vectp++, ustringp) != 0)
1781 			return (EFAULT);
1782 		while (*stringp++ != 0)
1783 			ustringp++;
1784 		ustringp++;
1785 	}
1786 
1787 	/* a null vector table pointer separates the argp's from the envp's */
1788 	if (suword(vectp++, 0) != 0)
1789 		return (EFAULT);
1790 
1791 	imgp->envv = vectp;
1792 	if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 ||
1793 	    suword32(&arginfo->ps_nenvstr, envc) != 0)
1794 		return (EFAULT);
1795 
1796 	/*
1797 	 * Fill in environment portion of vector table.
1798 	 */
1799 	for (; envc > 0; --envc) {
1800 		if (suword(vectp++, ustringp) != 0)
1801 			return (EFAULT);
1802 		while (*stringp++ != 0)
1803 			ustringp++;
1804 		ustringp++;
1805 	}
1806 
1807 	/* end of vector table is a null pointer */
1808 	if (suword(vectp, 0) != 0)
1809 		return (EFAULT);
1810 
1811 	if (imgp->auxargs) {
1812 		vectp++;
1813 		error = imgp->sysent->sv_copyout_auxargs(imgp,
1814 		    (uintptr_t)vectp);
1815 		if (error != 0)
1816 			return (error);
1817 	}
1818 
1819 	return (0);
1820 }
1821 
1822 /*
1823  * Check permissions of file to execute.
1824  *	Called with imgp->vp locked.
1825  *	Return 0 for success or error code on failure.
1826  */
1827 int
1828 exec_check_permissions(struct image_params *imgp)
1829 {
1830 	struct vnode *vp = imgp->vp;
1831 	struct vattr *attr = imgp->attr;
1832 	struct thread *td;
1833 	int error;
1834 
1835 	td = curthread;
1836 
1837 	/* Get file attributes */
1838 	error = VOP_GETATTR(vp, attr, td->td_ucred);
1839 	if (error)
1840 		return (error);
1841 
1842 #ifdef MAC
1843 	error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1844 	if (error)
1845 		return (error);
1846 #endif
1847 
1848 	/*
1849 	 * 1) Check if file execution is disabled for the filesystem that
1850 	 *    this file resides on.
1851 	 * 2) Ensure that at least one execute bit is on. Otherwise, a
1852 	 *    privileged user will always succeed, and we don't want this
1853 	 *    to happen unless the file really is executable.
1854 	 * 3) Ensure that the file is a regular file.
1855 	 */
1856 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1857 	    (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1858 	    (attr->va_type != VREG))
1859 		return (EACCES);
1860 
1861 	/*
1862 	 * Zero length files can't be exec'd
1863 	 */
1864 	if (attr->va_size == 0)
1865 		return (ENOEXEC);
1866 
1867 	/*
1868 	 *  Check for execute permission to file based on current credentials.
1869 	 */
1870 	error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1871 	if (error)
1872 		return (error);
1873 
1874 	/*
1875 	 * Check number of open-for-writes on the file and deny execution
1876 	 * if there are any.
1877 	 *
1878 	 * Add a text reference now so no one can write to the
1879 	 * executable while we're activating it.
1880 	 *
1881 	 * Remember if this was set before and unset it in case this is not
1882 	 * actually an executable image.
1883 	 */
1884 	error = VOP_SET_TEXT(vp);
1885 	if (error != 0)
1886 		return (error);
1887 	imgp->textset = true;
1888 
1889 	/*
1890 	 * Call filesystem specific open routine (which does nothing in the
1891 	 * general case).
1892 	 */
1893 	error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1894 	if (error == 0)
1895 		imgp->opened = true;
1896 	return (error);
1897 }
1898 
1899 /*
1900  * Exec handler registration
1901  */
1902 int
1903 exec_register(const struct execsw *execsw_arg)
1904 {
1905 	const struct execsw **es, **xs, **newexecsw;
1906 	u_int count = 2;	/* New slot and trailing NULL */
1907 
1908 	if (execsw)
1909 		for (es = execsw; *es; es++)
1910 			count++;
1911 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1912 	xs = newexecsw;
1913 	if (execsw)
1914 		for (es = execsw; *es; es++)
1915 			*xs++ = *es;
1916 	*xs++ = execsw_arg;
1917 	*xs = NULL;
1918 	if (execsw)
1919 		free(execsw, M_TEMP);
1920 	execsw = newexecsw;
1921 	return (0);
1922 }
1923 
1924 int
1925 exec_unregister(const struct execsw *execsw_arg)
1926 {
1927 	const struct execsw **es, **xs, **newexecsw;
1928 	int count = 1;
1929 
1930 	if (execsw == NULL)
1931 		panic("unregister with no handlers left?\n");
1932 
1933 	for (es = execsw; *es; es++) {
1934 		if (*es == execsw_arg)
1935 			break;
1936 	}
1937 	if (*es == NULL)
1938 		return (ENOENT);
1939 	for (es = execsw; *es; es++)
1940 		if (*es != execsw_arg)
1941 			count++;
1942 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1943 	xs = newexecsw;
1944 	for (es = execsw; *es; es++)
1945 		if (*es != execsw_arg)
1946 			*xs++ = *es;
1947 	*xs = NULL;
1948 	if (execsw)
1949 		free(execsw, M_TEMP);
1950 	execsw = newexecsw;
1951 	return (0);
1952 }
1953 
1954 /*
1955  * Write out a core segment to the compression stream.
1956  */
1957 static int
1958 compress_chunk(struct coredump_params *cp, char *base, char *buf, size_t len)
1959 {
1960 	size_t chunk_len;
1961 	int error;
1962 
1963 	while (len > 0) {
1964 		chunk_len = MIN(len, CORE_BUF_SIZE);
1965 
1966 		/*
1967 		 * We can get EFAULT error here.
1968 		 * In that case zero out the current chunk of the segment.
1969 		 */
1970 		error = copyin(base, buf, chunk_len);
1971 		if (error != 0)
1972 			bzero(buf, chunk_len);
1973 		error = compressor_write(cp->comp, buf, chunk_len);
1974 		if (error != 0)
1975 			break;
1976 		base += chunk_len;
1977 		len -= chunk_len;
1978 	}
1979 	return (error);
1980 }
1981 
1982 int
1983 core_write(struct coredump_params *cp, const void *base, size_t len,
1984     off_t offset, enum uio_seg seg, size_t *resid)
1985 {
1986 
1987 	return (vn_rdwr_inchunks(UIO_WRITE, cp->vp, __DECONST(void *, base),
1988 	    len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1989 	    cp->active_cred, cp->file_cred, resid, cp->td));
1990 }
1991 
1992 int
1993 core_output(char *base, size_t len, off_t offset, struct coredump_params *cp,
1994     void *tmpbuf)
1995 {
1996 	vm_map_t map;
1997 	struct mount *mp;
1998 	size_t resid, runlen;
1999 	int error;
2000 	bool success;
2001 
2002 	KASSERT((uintptr_t)base % PAGE_SIZE == 0,
2003 	    ("%s: user address %p is not page-aligned", __func__, base));
2004 
2005 	if (cp->comp != NULL)
2006 		return (compress_chunk(cp, base, tmpbuf, len));
2007 
2008 	map = &cp->td->td_proc->p_vmspace->vm_map;
2009 	for (; len > 0; base += runlen, offset += runlen, len -= runlen) {
2010 		/*
2011 		 * Attempt to page in all virtual pages in the range.  If a
2012 		 * virtual page is not backed by the pager, it is represented as
2013 		 * a hole in the file.  This can occur with zero-filled
2014 		 * anonymous memory or truncated files, for example.
2015 		 */
2016 		for (runlen = 0; runlen < len; runlen += PAGE_SIZE) {
2017 			if (core_dump_can_intr && curproc_sigkilled())
2018 				return (EINTR);
2019 			error = vm_fault(map, (uintptr_t)base + runlen,
2020 			    VM_PROT_READ, VM_FAULT_NOFILL, NULL);
2021 			if (runlen == 0)
2022 				success = error == KERN_SUCCESS;
2023 			else if ((error == KERN_SUCCESS) != success)
2024 				break;
2025 		}
2026 
2027 		if (success) {
2028 			error = core_write(cp, base, runlen, offset,
2029 			    UIO_USERSPACE, &resid);
2030 			if (error != 0) {
2031 				if (error != EFAULT)
2032 					break;
2033 
2034 				/*
2035 				 * EFAULT may be returned if the user mapping
2036 				 * could not be accessed, e.g., because a mapped
2037 				 * file has been truncated.  Skip the page if no
2038 				 * progress was made, to protect against a
2039 				 * hypothetical scenario where vm_fault() was
2040 				 * successful but core_write() returns EFAULT
2041 				 * anyway.
2042 				 */
2043 				runlen -= resid;
2044 				if (runlen == 0) {
2045 					success = false;
2046 					runlen = PAGE_SIZE;
2047 				}
2048 			}
2049 		}
2050 		if (!success) {
2051 			error = vn_start_write(cp->vp, &mp, V_WAIT);
2052 			if (error != 0)
2053 				break;
2054 			vn_lock(cp->vp, LK_EXCLUSIVE | LK_RETRY);
2055 			error = vn_truncate_locked(cp->vp, offset + runlen,
2056 			    false, cp->td->td_ucred);
2057 			VOP_UNLOCK(cp->vp);
2058 			vn_finished_write(mp);
2059 			if (error != 0)
2060 				break;
2061 		}
2062 	}
2063 	return (error);
2064 }
2065 
2066 /*
2067  * Drain into a core file.
2068  */
2069 int
2070 sbuf_drain_core_output(void *arg, const char *data, int len)
2071 {
2072 	struct coredump_params *cp;
2073 	struct proc *p;
2074 	int error, locked;
2075 
2076 	cp = arg;
2077 	p = cp->td->td_proc;
2078 
2079 	/*
2080 	 * Some kern_proc out routines that print to this sbuf may
2081 	 * call us with the process lock held. Draining with the
2082 	 * non-sleepable lock held is unsafe. The lock is needed for
2083 	 * those routines when dumping a live process. In our case we
2084 	 * can safely release the lock before draining and acquire
2085 	 * again after.
2086 	 */
2087 	locked = PROC_LOCKED(p);
2088 	if (locked)
2089 		PROC_UNLOCK(p);
2090 	if (cp->comp != NULL)
2091 		error = compressor_write(cp->comp, __DECONST(char *, data),
2092 		    len);
2093 	else
2094 		error = core_write(cp, __DECONST(void *, data), len, cp->offset,
2095 		    UIO_SYSSPACE, NULL);
2096 	if (locked)
2097 		PROC_LOCK(p);
2098 	if (error != 0)
2099 		return (-error);
2100 	cp->offset += len;
2101 	return (len);
2102 }
2103