xref: /freebsd/sys/kern/kern_exec.c (revision c1839039b193b48c8eb7520c75487f0bd4340c3b)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
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 	struct pargs *oldargs = NULL, *newargs = NULL;
394 	struct sigacts *oldsigacts = NULL, *newsigacts = NULL;
395 #ifdef KTRACE
396 	struct ktr_io_params *kiop;
397 #endif
398 	struct vnode *oldtextvp, *newtextvp;
399 	struct vnode *oldtextdvp, *newtextdvp;
400 	char *oldbinname, *newbinname;
401 	bool credential_changing;
402 #ifdef MAC
403 	struct label *interpvplabel = NULL;
404 	bool will_transition;
405 #endif
406 #ifdef HWPMC_HOOKS
407 	struct pmckern_procexec pe;
408 #endif
409 	int error, i, orig_osrel;
410 	uint32_t orig_fctl0;
411 	Elf_Brandinfo *orig_brandinfo;
412 	size_t freepath_size;
413 	static const char fexecv_proc_title[] = "(fexecv)";
414 
415 	imgp = &image_params;
416 	oldtextvp = oldtextdvp = NULL;
417 	newtextvp = newtextdvp = NULL;
418 	newbinname = oldbinname = NULL;
419 #ifdef KTRACE
420 	kiop = NULL;
421 #endif
422 
423 	/*
424 	 * Lock the process and set the P_INEXEC flag to indicate that
425 	 * it should be left alone until we're done here.  This is
426 	 * necessary to avoid race conditions - e.g. in ptrace() -
427 	 * that might allow a local user to illicitly obtain elevated
428 	 * privileges.
429 	 */
430 	PROC_LOCK(p);
431 	KASSERT((p->p_flag & P_INEXEC) == 0,
432 	    ("%s(): process already has P_INEXEC flag", __func__));
433 	p->p_flag |= P_INEXEC;
434 	PROC_UNLOCK(p);
435 
436 	/*
437 	 * Initialize part of the common data
438 	 */
439 	bzero(imgp, sizeof(*imgp));
440 	imgp->proc = p;
441 	imgp->attr = &attr;
442 	imgp->args = args;
443 	oldcred = p->p_ucred;
444 	orig_osrel = p->p_osrel;
445 	orig_fctl0 = p->p_fctl0;
446 	orig_brandinfo = p->p_elf_brandinfo;
447 
448 #ifdef MAC
449 	error = mac_execve_enter(imgp, mac_p);
450 	if (error)
451 		goto exec_fail;
452 #endif
453 
454 	SDT_PROBE1(proc, , , exec, args->fname);
455 
456 interpret:
457 	if (args->fname != NULL) {
458 #ifdef CAPABILITY_MODE
459 		/*
460 		 * While capability mode can't reach this point via direct
461 		 * path arguments to execve(), we also don't allow
462 		 * interpreters to be used in capability mode (for now).
463 		 * Catch indirect lookups and return a permissions error.
464 		 */
465 		if (IN_CAPABILITY_MODE(td)) {
466 			error = ECAPMODE;
467 			goto exec_fail;
468 		}
469 #endif
470 
471 		/*
472 		 * Translate the file name. namei() returns a vnode
473 		 * pointer in ni_vp among other things.
474 		 */
475 		NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW |
476 		    AUDITVNODE1 | WANTPARENT, UIO_SYSSPACE,
477 		    args->fname);
478 
479 		error = namei(&nd);
480 		if (error)
481 			goto exec_fail;
482 
483 		newtextvp = nd.ni_vp;
484 		newtextdvp = nd.ni_dvp;
485 		nd.ni_dvp = NULL;
486 		newbinname = malloc(nd.ni_cnd.cn_namelen + 1, M_PARGS,
487 		    M_WAITOK);
488 		memcpy(newbinname, nd.ni_cnd.cn_nameptr, nd.ni_cnd.cn_namelen);
489 		newbinname[nd.ni_cnd.cn_namelen] = '\0';
490 		imgp->vp = newtextvp;
491 
492 		/*
493 		 * Do the best to calculate the full path to the image file.
494 		 */
495 		if (args->fname[0] == '/') {
496 			imgp->execpath = args->fname;
497 		} else {
498 			VOP_UNLOCK(imgp->vp);
499 			freepath_size = MAXPATHLEN;
500 			if (vn_fullpath_hardlink(newtextvp, newtextdvp,
501 			    newbinname, nd.ni_cnd.cn_namelen, &imgp->execpath,
502 			    &imgp->freepath, &freepath_size) != 0)
503 				imgp->execpath = args->fname;
504 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
505 		}
506 	} else if (imgp->interpreter_vp) {
507 		/*
508 		 * An image activator has already provided an open vnode
509 		 */
510 		newtextvp = imgp->interpreter_vp;
511 		imgp->interpreter_vp = NULL;
512 		if (vn_fullpath(newtextvp, &imgp->execpath,
513 		    &imgp->freepath) != 0)
514 			imgp->execpath = args->fname;
515 		vn_lock(newtextvp, LK_SHARED | LK_RETRY);
516 		AUDIT_ARG_VNODE1(newtextvp);
517 		imgp->vp = newtextvp;
518 	} else {
519 		AUDIT_ARG_FD(args->fd);
520 
521 		/*
522 		 * If the descriptors was not opened with O_PATH, then
523 		 * we require that it was opened with O_EXEC or
524 		 * O_RDONLY.  In either case, exec_check_permissions()
525 		 * below checks _current_ file access mode regardless
526 		 * of the permissions additionally checked at the
527 		 * open(2).
528 		 */
529 		error = fgetvp_exec(td, args->fd, &cap_fexecve_rights,
530 		    &newtextvp);
531 		if (error != 0)
532 			goto exec_fail;
533 
534 		if (vn_fullpath(newtextvp, &imgp->execpath,
535 		    &imgp->freepath) != 0)
536 			imgp->execpath = args->fname;
537 		vn_lock(newtextvp, LK_SHARED | LK_RETRY);
538 		AUDIT_ARG_VNODE1(newtextvp);
539 		imgp->vp = newtextvp;
540 	}
541 
542 	/*
543 	 * Check file permissions.  Also 'opens' file and sets its vnode to
544 	 * text mode.
545 	 */
546 	error = exec_check_permissions(imgp);
547 	if (error)
548 		goto exec_fail_dealloc;
549 
550 	imgp->object = imgp->vp->v_object;
551 	if (imgp->object != NULL)
552 		vm_object_reference(imgp->object);
553 
554 	error = exec_map_first_page(imgp);
555 	if (error)
556 		goto exec_fail_dealloc;
557 
558 	imgp->proc->p_osrel = 0;
559 	imgp->proc->p_fctl0 = 0;
560 	imgp->proc->p_elf_brandinfo = NULL;
561 
562 	/*
563 	 * Implement image setuid/setgid.
564 	 *
565 	 * Determine new credentials before attempting image activators
566 	 * so that it can be used by process_exec handlers to determine
567 	 * credential/setid changes.
568 	 *
569 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
570 	 * the process is being traced.
571 	 *
572 	 * We disable setuid/setgid/etc in capability mode on the basis
573 	 * that most setugid applications are not written with that
574 	 * environment in mind, and will therefore almost certainly operate
575 	 * incorrectly. In principle there's no reason that setugid
576 	 * applications might not be useful in capability mode, so we may want
577 	 * to reconsider this conservative design choice in the future.
578 	 *
579 	 * XXXMAC: For the time being, use NOSUID to also prohibit
580 	 * transitions on the file system.
581 	 */
582 	credential_changing = false;
583 	credential_changing |= (attr.va_mode & S_ISUID) &&
584 	    oldcred->cr_uid != attr.va_uid;
585 	credential_changing |= (attr.va_mode & S_ISGID) &&
586 	    oldcred->cr_gid != attr.va_gid;
587 #ifdef MAC
588 	will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
589 	    interpvplabel, imgp) != 0;
590 	credential_changing |= will_transition;
591 #endif
592 
593 	/* Don't inherit PROC_PDEATHSIG_CTL value if setuid/setgid. */
594 	if (credential_changing)
595 		imgp->proc->p_pdeathsig = 0;
596 
597 	if (credential_changing &&
598 #ifdef CAPABILITY_MODE
599 	    ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
600 #endif
601 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
602 	    (p->p_flag & P_TRACED) == 0) {
603 		imgp->credential_setid = true;
604 		VOP_UNLOCK(imgp->vp);
605 		imgp->newcred = crdup(oldcred);
606 		if (attr.va_mode & S_ISUID) {
607 			euip = uifind(attr.va_uid);
608 			change_euid(imgp->newcred, euip);
609 		}
610 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
611 		if (attr.va_mode & S_ISGID)
612 			change_egid(imgp->newcred, attr.va_gid);
613 		/*
614 		 * Implement correct POSIX saved-id behavior.
615 		 *
616 		 * XXXMAC: Note that the current logic will save the
617 		 * uid and gid if a MAC domain transition occurs, even
618 		 * though maybe it shouldn't.
619 		 */
620 		change_svuid(imgp->newcred, imgp->newcred->cr_uid);
621 		change_svgid(imgp->newcred, imgp->newcred->cr_gid);
622 	} else {
623 		/*
624 		 * Implement correct POSIX saved-id behavior.
625 		 *
626 		 * XXX: It's not clear that the existing behavior is
627 		 * POSIX-compliant.  A number of sources indicate that the
628 		 * saved uid/gid should only be updated if the new ruid is
629 		 * not equal to the old ruid, or the new euid is not equal
630 		 * to the old euid and the new euid is not equal to the old
631 		 * ruid.  The FreeBSD code always updates the saved uid/gid.
632 		 * Also, this code uses the new (replaced) euid and egid as
633 		 * the source, which may or may not be the right ones to use.
634 		 */
635 		if (oldcred->cr_svuid != oldcred->cr_uid ||
636 		    oldcred->cr_svgid != oldcred->cr_gid) {
637 			VOP_UNLOCK(imgp->vp);
638 			imgp->newcred = crdup(oldcred);
639 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
640 			change_svuid(imgp->newcred, imgp->newcred->cr_uid);
641 			change_svgid(imgp->newcred, imgp->newcred->cr_gid);
642 		}
643 	}
644 	/* The new credentials are installed into the process later. */
645 
646 	/*
647 	 *	Loop through the list of image activators, calling each one.
648 	 *	An activator returns -1 if there is no match, 0 on success,
649 	 *	and an error otherwise.
650 	 */
651 	error = -1;
652 	for (i = 0; error == -1 && execsw[i]; ++i) {
653 		if (execsw[i]->ex_imgact == NULL)
654 			continue;
655 		error = (*execsw[i]->ex_imgact)(imgp);
656 	}
657 
658 	if (error) {
659 		if (error == -1)
660 			error = ENOEXEC;
661 		goto exec_fail_dealloc;
662 	}
663 
664 	/*
665 	 * Special interpreter operation, cleanup and loop up to try to
666 	 * activate the interpreter.
667 	 */
668 	if (imgp->interpreted) {
669 		exec_unmap_first_page(imgp);
670 		/*
671 		 * The text reference needs to be removed for scripts.
672 		 * There is a short period before we determine that
673 		 * something is a script where text reference is active.
674 		 * The vnode lock is held over this entire period
675 		 * so nothing should illegitimately be blocked.
676 		 */
677 		MPASS(imgp->textset);
678 		VOP_UNSET_TEXT_CHECKED(newtextvp);
679 		imgp->textset = false;
680 		/* free name buffer and old vnode */
681 #ifdef MAC
682 		mac_execve_interpreter_enter(newtextvp, &interpvplabel);
683 #endif
684 		if (imgp->opened) {
685 			VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td);
686 			imgp->opened = false;
687 		}
688 		vput(newtextvp);
689 		imgp->vp = newtextvp = NULL;
690 		if (args->fname != NULL) {
691 			if (newtextdvp != NULL) {
692 				vrele(newtextdvp);
693 				newtextdvp = NULL;
694 			}
695 			NDFREE_PNBUF(&nd);
696 			free(newbinname, M_PARGS);
697 			newbinname = NULL;
698 		}
699 		vm_object_deallocate(imgp->object);
700 		imgp->object = NULL;
701 		execve_nosetid(imgp);
702 		imgp->execpath = NULL;
703 		free(imgp->freepath, M_TEMP);
704 		imgp->freepath = NULL;
705 		/* set new name to that of the interpreter */
706 		if (imgp->interpreter_vp) {
707 			args->fname = NULL;
708 		} else {
709 			args->fname = imgp->interpreter_name;
710 		}
711 		goto interpret;
712 	}
713 
714 	/*
715 	 * NB: We unlock the vnode here because it is believed that none
716 	 * of the sv_copyout_strings/sv_fixup operations require the vnode.
717 	 */
718 	VOP_UNLOCK(imgp->vp);
719 
720 	if (disallow_high_osrel &&
721 	    P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
722 		error = ENOEXEC;
723 		uprintf("Osrel %d for image %s too high\n", p->p_osrel,
724 		    imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
725 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
726 		goto exec_fail_dealloc;
727 	}
728 
729 	/*
730 	 * Copy out strings (args and env) and initialize stack base.
731 	 */
732 	error = (*p->p_sysent->sv_copyout_strings)(imgp, &stack_base);
733 	if (error != 0) {
734 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
735 		goto exec_fail_dealloc;
736 	}
737 
738 	/*
739 	 * Stack setup.
740 	 */
741 	error = (*p->p_sysent->sv_fixup)(&stack_base, imgp);
742 	if (error != 0) {
743 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
744 		goto exec_fail_dealloc;
745 	}
746 
747 	/*
748 	 * For security and other reasons, the file descriptor table cannot be
749 	 * shared after an exec.
750 	 */
751 	fdunshare(td);
752 	pdunshare(td);
753 	/* close files on exec */
754 	fdcloseexec(td);
755 
756 	/*
757 	 * Malloc things before we need locks.
758 	 */
759 	i = exec_args_get_begin_envv(imgp->args) - imgp->args->begin_argv;
760 	/* Cache arguments if they fit inside our allowance */
761 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
762 		newargs = pargs_alloc(i);
763 		bcopy(imgp->args->begin_argv, newargs->ar_args, i);
764 	}
765 
766 	/*
767 	 * For security and other reasons, signal handlers cannot
768 	 * be shared after an exec. The new process gets a copy of the old
769 	 * handlers. In execsigs(), the new process will have its signals
770 	 * reset.
771 	 */
772 	if (sigacts_shared(p->p_sigacts)) {
773 		oldsigacts = p->p_sigacts;
774 		newsigacts = sigacts_alloc();
775 		sigacts_copy(newsigacts, oldsigacts);
776 	}
777 
778 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
779 
780 	PROC_LOCK(p);
781 	if (oldsigacts)
782 		p->p_sigacts = newsigacts;
783 	/* Stop profiling */
784 	stopprofclock(p);
785 
786 	/* reset caught signals */
787 	execsigs(p);
788 
789 	/* name this process - nameiexec(p, ndp) */
790 	bzero(p->p_comm, sizeof(p->p_comm));
791 	if (args->fname)
792 		bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
793 		    min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
794 	else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0)
795 		bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
796 	bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
797 #ifdef KTR
798 	sched_clear_tdname(td);
799 #endif
800 
801 	/*
802 	 * mark as execed, wakeup the process that vforked (if any) and tell
803 	 * it that it now has its own resources back
804 	 */
805 	p->p_flag |= P_EXEC;
806 	if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0)
807 		p->p_flag2 &= ~P2_NOTRACE;
808 	if ((p->p_flag2 & P2_STKGAP_DISABLE_EXEC) == 0)
809 		p->p_flag2 &= ~P2_STKGAP_DISABLE;
810 	if (p->p_flag & P_PPWAIT) {
811 		p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
812 		cv_broadcast(&p->p_pwait);
813 		/* STOPs are no longer ignored, arrange for AST */
814 		signotify(td);
815 	}
816 
817 	if ((imgp->sysent->sv_setid_allowed != NULL &&
818 	    !(*imgp->sysent->sv_setid_allowed)(td, imgp)) ||
819 	    (p->p_flag2 & P2_NO_NEW_PRIVS) != 0)
820 		execve_nosetid(imgp);
821 
822 	/*
823 	 * Implement image setuid/setgid installation.
824 	 */
825 	if (imgp->credential_setid) {
826 		/*
827 		 * Turn off syscall tracing for set-id programs, except for
828 		 * root.  Record any set-id flags first to make sure that
829 		 * we do not regain any tracing during a possible block.
830 		 */
831 		setsugid(p);
832 #ifdef KTRACE
833 		kiop = ktrprocexec(p);
834 #endif
835 		/*
836 		 * Close any file descriptors 0..2 that reference procfs,
837 		 * then make sure file descriptors 0..2 are in use.
838 		 *
839 		 * Both fdsetugidsafety() and fdcheckstd() may call functions
840 		 * taking sleepable locks, so temporarily drop our locks.
841 		 */
842 		PROC_UNLOCK(p);
843 		VOP_UNLOCK(imgp->vp);
844 		fdsetugidsafety(td);
845 		error = fdcheckstd(td);
846 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
847 		if (error != 0)
848 			goto exec_fail_dealloc;
849 		PROC_LOCK(p);
850 #ifdef MAC
851 		if (will_transition) {
852 			mac_vnode_execve_transition(oldcred, imgp->newcred,
853 			    imgp->vp, interpvplabel, imgp);
854 		}
855 #endif
856 	} else {
857 		if (oldcred->cr_uid == oldcred->cr_ruid &&
858 		    oldcred->cr_gid == oldcred->cr_rgid)
859 			p->p_flag &= ~P_SUGID;
860 	}
861 	/*
862 	 * Set the new credentials.
863 	 */
864 	if (imgp->newcred != NULL) {
865 		proc_set_cred(p, imgp->newcred);
866 		crfree(oldcred);
867 		oldcred = NULL;
868 	}
869 
870 	/*
871 	 * Store the vp for use in kern.proc.pathname.  This vnode was
872 	 * referenced by namei() or by fexecve variant of fname handling.
873 	 */
874 	oldtextvp = p->p_textvp;
875 	p->p_textvp = newtextvp;
876 	oldtextdvp = p->p_textdvp;
877 	p->p_textdvp = newtextdvp;
878 	newtextdvp = NULL;
879 	oldbinname = p->p_binname;
880 	p->p_binname = newbinname;
881 	newbinname = NULL;
882 
883 #ifdef KDTRACE_HOOKS
884 	/*
885 	 * Tell the DTrace fasttrap provider about the exec if it
886 	 * has declared an interest.
887 	 */
888 	if (dtrace_fasttrap_exec)
889 		dtrace_fasttrap_exec(p);
890 #endif
891 
892 	/*
893 	 * Notify others that we exec'd, and clear the P_INEXEC flag
894 	 * as we're now a bona fide freshly-execed process.
895 	 */
896 	KNOTE_LOCKED(p->p_klist, NOTE_EXEC);
897 	p->p_flag &= ~P_INEXEC;
898 
899 	/* clear "fork but no exec" flag, as we _are_ execing */
900 	p->p_acflag &= ~AFORK;
901 
902 	/*
903 	 * Free any previous argument cache and replace it with
904 	 * the new argument cache, if any.
905 	 */
906 	oldargs = p->p_args;
907 	p->p_args = newargs;
908 	newargs = NULL;
909 
910 	PROC_UNLOCK(p);
911 
912 #ifdef	HWPMC_HOOKS
913 	/*
914 	 * Check if system-wide sampling is in effect or if the
915 	 * current process is using PMCs.  If so, do exec() time
916 	 * processing.  This processing needs to happen AFTER the
917 	 * P_INEXEC flag is cleared.
918 	 */
919 	if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
920 		VOP_UNLOCK(imgp->vp);
921 		pe.pm_credentialschanged = credential_changing;
922 		pe.pm_baseaddr = imgp->reloc_base;
923 		pe.pm_dynaddr = imgp->et_dyn_addr;
924 
925 		PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
926 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
927 	}
928 #endif
929 
930 	/* Set values passed into the program in registers. */
931 	(*p->p_sysent->sv_setregs)(td, imgp, stack_base);
932 
933 	VOP_MMAPPED(imgp->vp);
934 
935 	SDT_PROBE1(proc, , , exec__success, args->fname);
936 
937 exec_fail_dealloc:
938 	if (error != 0) {
939 		p->p_osrel = orig_osrel;
940 		p->p_fctl0 = orig_fctl0;
941 		p->p_elf_brandinfo = orig_brandinfo;
942 	}
943 
944 	if (imgp->firstpage != NULL)
945 		exec_unmap_first_page(imgp);
946 
947 	if (imgp->vp != NULL) {
948 		if (imgp->opened)
949 			VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
950 		if (imgp->textset)
951 			VOP_UNSET_TEXT_CHECKED(imgp->vp);
952 		if (error != 0)
953 			vput(imgp->vp);
954 		else
955 			VOP_UNLOCK(imgp->vp);
956 		if (args->fname != NULL)
957 			NDFREE_PNBUF(&nd);
958 		if (newtextdvp != NULL)
959 			vrele(newtextdvp);
960 		free(newbinname, M_PARGS);
961 	}
962 
963 	if (imgp->object != NULL)
964 		vm_object_deallocate(imgp->object);
965 
966 	free(imgp->freepath, M_TEMP);
967 
968 	if (error == 0) {
969 		if (p->p_ptevents & PTRACE_EXEC) {
970 			PROC_LOCK(p);
971 			if (p->p_ptevents & PTRACE_EXEC)
972 				td->td_dbgflags |= TDB_EXEC;
973 			PROC_UNLOCK(p);
974 		}
975 	} else {
976 exec_fail:
977 		/* we're done here, clear P_INEXEC */
978 		PROC_LOCK(p);
979 		p->p_flag &= ~P_INEXEC;
980 		PROC_UNLOCK(p);
981 
982 		SDT_PROBE1(proc, , , exec__failure, error);
983 	}
984 
985 	if (imgp->newcred != NULL && oldcred != NULL)
986 		crfree(imgp->newcred);
987 
988 #ifdef MAC
989 	mac_execve_exit(imgp);
990 	mac_execve_interpreter_exit(interpvplabel);
991 #endif
992 	exec_free_args(args);
993 
994 	/*
995 	 * Handle deferred decrement of ref counts.
996 	 */
997 	if (oldtextvp != NULL)
998 		vrele(oldtextvp);
999 	if (oldtextdvp != NULL)
1000 		vrele(oldtextdvp);
1001 	free(oldbinname, M_PARGS);
1002 #ifdef KTRACE
1003 	ktr_io_params_free(kiop);
1004 #endif
1005 	pargs_drop(oldargs);
1006 	pargs_drop(newargs);
1007 	if (oldsigacts != NULL)
1008 		sigacts_free(oldsigacts);
1009 	if (euip != NULL)
1010 		uifree(euip);
1011 
1012 	if (error && imgp->vmspace_destroyed) {
1013 		/* sorry, no more process anymore. exit gracefully */
1014 		exec_cleanup(td, oldvmspace);
1015 		exit1(td, 0, SIGABRT);
1016 		/* NOT REACHED */
1017 	}
1018 
1019 #ifdef KTRACE
1020 	if (error == 0)
1021 		ktrprocctor(p);
1022 #endif
1023 
1024 	/*
1025 	 * We don't want cpu_set_syscall_retval() to overwrite any of
1026 	 * the register values put in place by exec_setregs().
1027 	 * Implementations of cpu_set_syscall_retval() will leave
1028 	 * registers unmodified when returning EJUSTRETURN.
1029 	 */
1030 	return (error == 0 ? EJUSTRETURN : error);
1031 }
1032 
1033 void
1034 exec_cleanup(struct thread *td, struct vmspace *oldvmspace)
1035 {
1036 	if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
1037 		KASSERT(td->td_proc->p_vmspace != oldvmspace,
1038 		    ("oldvmspace still used"));
1039 		vmspace_free(oldvmspace);
1040 		td->td_pflags &= ~TDP_EXECVMSPC;
1041 	}
1042 }
1043 
1044 int
1045 exec_map_first_page(struct image_params *imgp)
1046 {
1047 	vm_object_t object;
1048 	vm_page_t m;
1049 	int error;
1050 
1051 	if (imgp->firstpage != NULL)
1052 		exec_unmap_first_page(imgp);
1053 
1054 	object = imgp->vp->v_object;
1055 	if (object == NULL)
1056 		return (EACCES);
1057 #if VM_NRESERVLEVEL > 0
1058 	if ((object->flags & OBJ_COLORED) == 0) {
1059 		VM_OBJECT_WLOCK(object);
1060 		vm_object_color(object, 0);
1061 		VM_OBJECT_WUNLOCK(object);
1062 	}
1063 #endif
1064 	error = vm_page_grab_valid_unlocked(&m, object, 0,
1065 	    VM_ALLOC_COUNT(VM_INITIAL_PAGEIN) |
1066 	    VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED);
1067 
1068 	if (error != VM_PAGER_OK)
1069 		return (EIO);
1070 	imgp->firstpage = sf_buf_alloc(m, 0);
1071 	imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
1072 
1073 	return (0);
1074 }
1075 
1076 void
1077 exec_unmap_first_page(struct image_params *imgp)
1078 {
1079 	vm_page_t m;
1080 
1081 	if (imgp->firstpage != NULL) {
1082 		m = sf_buf_page(imgp->firstpage);
1083 		sf_buf_free(imgp->firstpage);
1084 		imgp->firstpage = NULL;
1085 		vm_page_unwire(m, PQ_ACTIVE);
1086 	}
1087 }
1088 
1089 void
1090 exec_onexec_old(struct thread *td)
1091 {
1092 	sigfastblock_clear(td);
1093 	umtx_exec(td->td_proc);
1094 }
1095 
1096 /*
1097  * This is an optimization which removes the unmanaged shared page
1098  * mapping. In combination with pmap_remove_pages(), which cleans all
1099  * managed mappings in the process' vmspace pmap, no work will be left
1100  * for pmap_remove(min, max).
1101  */
1102 void
1103 exec_free_abi_mappings(struct proc *p)
1104 {
1105 	struct vmspace *vmspace;
1106 
1107 	vmspace = p->p_vmspace;
1108 	if (refcount_load(&vmspace->vm_refcnt) != 1)
1109 		return;
1110 
1111 	if (!PROC_HAS_SHP(p))
1112 		return;
1113 
1114 	pmap_remove(vmspace_pmap(vmspace), vmspace->vm_shp_base,
1115 	    vmspace->vm_shp_base + p->p_sysent->sv_shared_page_len);
1116 }
1117 
1118 /*
1119  * Run down the current address space and install a new one.
1120  */
1121 int
1122 exec_new_vmspace(struct image_params *imgp, struct sysentvec *sv)
1123 {
1124 	int error;
1125 	struct proc *p = imgp->proc;
1126 	struct vmspace *vmspace = p->p_vmspace;
1127 	struct thread *td = curthread;
1128 	vm_offset_t sv_minuser;
1129 	vm_map_t map;
1130 
1131 	imgp->vmspace_destroyed = true;
1132 	imgp->sysent = sv;
1133 
1134 	if (p->p_sysent->sv_onexec_old != NULL)
1135 		p->p_sysent->sv_onexec_old(td);
1136 	itimers_exec(p);
1137 
1138 	EVENTHANDLER_DIRECT_INVOKE(process_exec, p, imgp);
1139 
1140 	/*
1141 	 * Blow away entire process VM, if address space not shared,
1142 	 * otherwise, create a new VM space so that other threads are
1143 	 * not disrupted
1144 	 */
1145 	map = &vmspace->vm_map;
1146 	if (map_at_zero)
1147 		sv_minuser = sv->sv_minuser;
1148 	else
1149 		sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1150 	if (refcount_load(&vmspace->vm_refcnt) == 1 &&
1151 	    vm_map_min(map) == sv_minuser &&
1152 	    vm_map_max(map) == sv->sv_maxuser &&
1153 	    cpu_exec_vmspace_reuse(p, map)) {
1154 		exec_free_abi_mappings(p);
1155 		shmexit(vmspace);
1156 		pmap_remove_pages(vmspace_pmap(vmspace));
1157 		vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1158 		/*
1159 		 * An exec terminates mlockall(MCL_FUTURE).
1160 		 * ASLR and W^X states must be re-evaluated.
1161 		 */
1162 		vm_map_lock(map);
1163 		vm_map_modflags(map, 0, MAP_WIREFUTURE | MAP_ASLR |
1164 		    MAP_ASLR_IGNSTART | MAP_ASLR_STACK | MAP_WXORX);
1165 		vm_map_unlock(map);
1166 	} else {
1167 		error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1168 		if (error)
1169 			return (error);
1170 		vmspace = p->p_vmspace;
1171 		map = &vmspace->vm_map;
1172 	}
1173 	map->flags |= imgp->map_flags;
1174 
1175 	return (sv->sv_onexec != NULL ? sv->sv_onexec(p, imgp) : 0);
1176 }
1177 
1178 /*
1179  * Compute the stack size limit and map the main process stack.
1180  * Map the shared page.
1181  */
1182 int
1183 exec_map_stack(struct image_params *imgp)
1184 {
1185 	struct rlimit rlim_stack;
1186 	struct sysentvec *sv;
1187 	struct proc *p;
1188 	vm_map_t map;
1189 	struct vmspace *vmspace;
1190 	vm_offset_t stack_addr, stack_top;
1191 	vm_offset_t sharedpage_addr;
1192 	u_long ssiz;
1193 	int error, find_space, stack_off;
1194 	vm_prot_t stack_prot;
1195 	vm_object_t obj;
1196 
1197 	p = imgp->proc;
1198 	sv = p->p_sysent;
1199 
1200 	if (imgp->stack_sz != 0) {
1201 		ssiz = trunc_page(imgp->stack_sz);
1202 		PROC_LOCK(p);
1203 		lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack);
1204 		PROC_UNLOCK(p);
1205 		if (ssiz > rlim_stack.rlim_max)
1206 			ssiz = rlim_stack.rlim_max;
1207 		if (ssiz > rlim_stack.rlim_cur) {
1208 			rlim_stack.rlim_cur = ssiz;
1209 			kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack);
1210 		}
1211 	} else if (sv->sv_maxssiz != NULL) {
1212 		ssiz = *sv->sv_maxssiz;
1213 	} else {
1214 		ssiz = maxssiz;
1215 	}
1216 
1217 	vmspace = p->p_vmspace;
1218 	map = &vmspace->vm_map;
1219 
1220 	stack_prot = sv->sv_shared_page_obj != NULL && imgp->stack_prot != 0 ?
1221 	    imgp->stack_prot : sv->sv_stackprot;
1222 	if ((map->flags & MAP_ASLR_STACK) != 0) {
1223 		stack_addr = round_page((vm_offset_t)p->p_vmspace->vm_daddr +
1224 		    lim_max(curthread, RLIMIT_DATA));
1225 		find_space = VMFS_ANY_SPACE;
1226 	} else {
1227 		stack_addr = sv->sv_usrstack - ssiz;
1228 		find_space = VMFS_NO_SPACE;
1229 	}
1230 	error = vm_map_find(map, NULL, 0, &stack_addr, (vm_size_t)ssiz,
1231 	    sv->sv_usrstack, find_space, stack_prot, VM_PROT_ALL,
1232 	    MAP_STACK_GROWS_DOWN);
1233 	if (error != KERN_SUCCESS) {
1234 		uprintf("exec_new_vmspace: mapping stack size %#jx prot %#x "
1235 		    "failed, mach error %d errno %d\n", (uintmax_t)ssiz,
1236 		    stack_prot, error, vm_mmap_to_errno(error));
1237 		return (vm_mmap_to_errno(error));
1238 	}
1239 
1240 	stack_top = stack_addr + ssiz;
1241 	if ((map->flags & MAP_ASLR_STACK) != 0) {
1242 		/* Randomize within the first page of the stack. */
1243 		arc4rand(&stack_off, sizeof(stack_off), 0);
1244 		stack_top -= rounddown2(stack_off & PAGE_MASK, sizeof(void *));
1245 	}
1246 
1247 	/* Map a shared page */
1248 	obj = sv->sv_shared_page_obj;
1249 	if (obj == NULL) {
1250 		sharedpage_addr = 0;
1251 		goto out;
1252 	}
1253 
1254 	/*
1255 	 * If randomization is disabled then the shared page will
1256 	 * be mapped at address specified in sysentvec.
1257 	 * Otherwise any address above .data section can be selected.
1258 	 * Same logic is used for stack address randomization.
1259 	 * If the address randomization is applied map a guard page
1260 	 * at the top of UVA.
1261 	 */
1262 	vm_object_reference(obj);
1263 	if ((imgp->imgp_flags & IMGP_ASLR_SHARED_PAGE) != 0) {
1264 		sharedpage_addr = round_page((vm_offset_t)p->p_vmspace->vm_daddr +
1265 		    lim_max(curthread, RLIMIT_DATA));
1266 
1267 		error = vm_map_fixed(map, NULL, 0,
1268 		    sv->sv_maxuser - PAGE_SIZE, PAGE_SIZE,
1269 		    VM_PROT_NONE, VM_PROT_NONE, MAP_CREATE_GUARD);
1270 		if (error != KERN_SUCCESS) {
1271 			/*
1272 			 * This is not fatal, so let's just print a warning
1273 			 * and continue.
1274 			 */
1275 			uprintf("%s: Mapping guard page at the top of UVA failed"
1276 			    " mach error %d errno %d",
1277 			    __func__, error, vm_mmap_to_errno(error));
1278 		}
1279 
1280 		error = vm_map_find(map, obj, 0,
1281 		    &sharedpage_addr, sv->sv_shared_page_len,
1282 		    sv->sv_maxuser, VMFS_ANY_SPACE,
1283 		    VM_PROT_READ | VM_PROT_EXECUTE,
1284 		    VM_PROT_READ | VM_PROT_EXECUTE,
1285 		    MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1286 	} else {
1287 		sharedpage_addr = sv->sv_shared_page_base;
1288 		vm_map_fixed(map, obj, 0,
1289 		    sharedpage_addr, sv->sv_shared_page_len,
1290 		    VM_PROT_READ | VM_PROT_EXECUTE,
1291 		    VM_PROT_READ | VM_PROT_EXECUTE,
1292 		    MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1293 	}
1294 	if (error != KERN_SUCCESS) {
1295 		uprintf("%s: mapping shared page at addr: %p"
1296 		    "failed, mach error %d errno %d\n", __func__,
1297 		    (void *)sharedpage_addr, error, vm_mmap_to_errno(error));
1298 		vm_object_deallocate(obj);
1299 		return (vm_mmap_to_errno(error));
1300 	}
1301 out:
1302 	/*
1303 	 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they
1304 	 * are still used to enforce the stack rlimit on the process stack.
1305 	 */
1306 	vmspace->vm_maxsaddr = (char *)stack_addr;
1307 	vmspace->vm_stacktop = stack_top;
1308 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1309 	vmspace->vm_shp_base = sharedpage_addr;
1310 
1311 	return (0);
1312 }
1313 
1314 /*
1315  * Copy out argument and environment strings from the old process address
1316  * space into the temporary string buffer.
1317  */
1318 int
1319 exec_copyin_args(struct image_args *args, const char *fname,
1320     enum uio_seg segflg, char **argv, char **envv)
1321 {
1322 	u_long arg, env;
1323 	int error;
1324 
1325 	bzero(args, sizeof(*args));
1326 	if (argv == NULL)
1327 		return (EFAULT);
1328 
1329 	/*
1330 	 * Allocate demand-paged memory for the file name, argument, and
1331 	 * environment strings.
1332 	 */
1333 	error = exec_alloc_args(args);
1334 	if (error != 0)
1335 		return (error);
1336 
1337 	/*
1338 	 * Copy the file name.
1339 	 */
1340 	error = exec_args_add_fname(args, fname, segflg);
1341 	if (error != 0)
1342 		goto err_exit;
1343 
1344 	/*
1345 	 * extract arguments first
1346 	 */
1347 	for (;;) {
1348 		error = fueword(argv++, &arg);
1349 		if (error == -1) {
1350 			error = EFAULT;
1351 			goto err_exit;
1352 		}
1353 		if (arg == 0)
1354 			break;
1355 		error = exec_args_add_arg(args, (char *)(uintptr_t)arg,
1356 		    UIO_USERSPACE);
1357 		if (error != 0)
1358 			goto err_exit;
1359 	}
1360 
1361 	/*
1362 	 * extract environment strings
1363 	 */
1364 	if (envv) {
1365 		for (;;) {
1366 			error = fueword(envv++, &env);
1367 			if (error == -1) {
1368 				error = EFAULT;
1369 				goto err_exit;
1370 			}
1371 			if (env == 0)
1372 				break;
1373 			error = exec_args_add_env(args,
1374 			    (char *)(uintptr_t)env, UIO_USERSPACE);
1375 			if (error != 0)
1376 				goto err_exit;
1377 		}
1378 	}
1379 
1380 	return (0);
1381 
1382 err_exit:
1383 	exec_free_args(args);
1384 	return (error);
1385 }
1386 
1387 struct exec_args_kva {
1388 	vm_offset_t addr;
1389 	u_int gen;
1390 	SLIST_ENTRY(exec_args_kva) next;
1391 };
1392 
1393 DPCPU_DEFINE_STATIC(struct exec_args_kva *, exec_args_kva);
1394 
1395 static SLIST_HEAD(, exec_args_kva) exec_args_kva_freelist;
1396 static struct mtx exec_args_kva_mtx;
1397 static u_int exec_args_gen;
1398 
1399 static void
1400 exec_prealloc_args_kva(void *arg __unused)
1401 {
1402 	struct exec_args_kva *argkva;
1403 	u_int i;
1404 
1405 	SLIST_INIT(&exec_args_kva_freelist);
1406 	mtx_init(&exec_args_kva_mtx, "exec args kva", NULL, MTX_DEF);
1407 	for (i = 0; i < exec_map_entries; i++) {
1408 		argkva = malloc(sizeof(*argkva), M_PARGS, M_WAITOK);
1409 		argkva->addr = kmap_alloc_wait(exec_map, exec_map_entry_size);
1410 		argkva->gen = exec_args_gen;
1411 		SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1412 	}
1413 }
1414 SYSINIT(exec_args_kva, SI_SUB_EXEC, SI_ORDER_ANY, exec_prealloc_args_kva, NULL);
1415 
1416 static vm_offset_t
1417 exec_alloc_args_kva(void **cookie)
1418 {
1419 	struct exec_args_kva *argkva;
1420 
1421 	argkva = (void *)atomic_readandclear_ptr(
1422 	    (uintptr_t *)DPCPU_PTR(exec_args_kva));
1423 	if (argkva == NULL) {
1424 		mtx_lock(&exec_args_kva_mtx);
1425 		while ((argkva = SLIST_FIRST(&exec_args_kva_freelist)) == NULL)
1426 			(void)mtx_sleep(&exec_args_kva_freelist,
1427 			    &exec_args_kva_mtx, 0, "execkva", 0);
1428 		SLIST_REMOVE_HEAD(&exec_args_kva_freelist, next);
1429 		mtx_unlock(&exec_args_kva_mtx);
1430 	}
1431 	kasan_mark((void *)argkva->addr, exec_map_entry_size,
1432 	    exec_map_entry_size, 0);
1433 	*(struct exec_args_kva **)cookie = argkva;
1434 	return (argkva->addr);
1435 }
1436 
1437 static void
1438 exec_release_args_kva(struct exec_args_kva *argkva, u_int gen)
1439 {
1440 	vm_offset_t base;
1441 
1442 	base = argkva->addr;
1443 	kasan_mark((void *)argkva->addr, 0, exec_map_entry_size,
1444 	    KASAN_EXEC_ARGS_FREED);
1445 	if (argkva->gen != gen) {
1446 		(void)vm_map_madvise(exec_map, base, base + exec_map_entry_size,
1447 		    MADV_FREE);
1448 		argkva->gen = gen;
1449 	}
1450 	if (!atomic_cmpset_ptr((uintptr_t *)DPCPU_PTR(exec_args_kva),
1451 	    (uintptr_t)NULL, (uintptr_t)argkva)) {
1452 		mtx_lock(&exec_args_kva_mtx);
1453 		SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1454 		wakeup_one(&exec_args_kva_freelist);
1455 		mtx_unlock(&exec_args_kva_mtx);
1456 	}
1457 }
1458 
1459 static void
1460 exec_free_args_kva(void *cookie)
1461 {
1462 
1463 	exec_release_args_kva(cookie, exec_args_gen);
1464 }
1465 
1466 static void
1467 exec_args_kva_lowmem(void *arg __unused)
1468 {
1469 	SLIST_HEAD(, exec_args_kva) head;
1470 	struct exec_args_kva *argkva;
1471 	u_int gen;
1472 	int i;
1473 
1474 	gen = atomic_fetchadd_int(&exec_args_gen, 1) + 1;
1475 
1476 	/*
1477 	 * Force an madvise of each KVA range. Any currently allocated ranges
1478 	 * will have MADV_FREE applied once they are freed.
1479 	 */
1480 	SLIST_INIT(&head);
1481 	mtx_lock(&exec_args_kva_mtx);
1482 	SLIST_SWAP(&head, &exec_args_kva_freelist, exec_args_kva);
1483 	mtx_unlock(&exec_args_kva_mtx);
1484 	while ((argkva = SLIST_FIRST(&head)) != NULL) {
1485 		SLIST_REMOVE_HEAD(&head, next);
1486 		exec_release_args_kva(argkva, gen);
1487 	}
1488 
1489 	CPU_FOREACH(i) {
1490 		argkva = (void *)atomic_readandclear_ptr(
1491 		    (uintptr_t *)DPCPU_ID_PTR(i, exec_args_kva));
1492 		if (argkva != NULL)
1493 			exec_release_args_kva(argkva, gen);
1494 	}
1495 }
1496 EVENTHANDLER_DEFINE(vm_lowmem, exec_args_kva_lowmem, NULL,
1497     EVENTHANDLER_PRI_ANY);
1498 
1499 /*
1500  * Allocate temporary demand-paged, zero-filled memory for the file name,
1501  * argument, and environment strings.
1502  */
1503 int
1504 exec_alloc_args(struct image_args *args)
1505 {
1506 
1507 	args->buf = (char *)exec_alloc_args_kva(&args->bufkva);
1508 	return (0);
1509 }
1510 
1511 void
1512 exec_free_args(struct image_args *args)
1513 {
1514 
1515 	if (args->buf != NULL) {
1516 		exec_free_args_kva(args->bufkva);
1517 		args->buf = NULL;
1518 	}
1519 	if (args->fname_buf != NULL) {
1520 		free(args->fname_buf, M_TEMP);
1521 		args->fname_buf = NULL;
1522 	}
1523 }
1524 
1525 /*
1526  * A set to functions to fill struct image args.
1527  *
1528  * NOTE: exec_args_add_fname() must be called (possibly with a NULL
1529  * fname) before the other functions.  All exec_args_add_arg() calls must
1530  * be made before any exec_args_add_env() calls.  exec_args_adjust_args()
1531  * may be called any time after exec_args_add_fname().
1532  *
1533  * exec_args_add_fname() - install path to be executed
1534  * exec_args_add_arg() - append an argument string
1535  * exec_args_add_env() - append an env string
1536  * exec_args_adjust_args() - adjust location of the argument list to
1537  *                           allow new arguments to be prepended
1538  */
1539 int
1540 exec_args_add_fname(struct image_args *args, const char *fname,
1541     enum uio_seg segflg)
1542 {
1543 	int error;
1544 	size_t length;
1545 
1546 	KASSERT(args->fname == NULL, ("fname already appended"));
1547 	KASSERT(args->endp == NULL, ("already appending to args"));
1548 
1549 	if (fname != NULL) {
1550 		args->fname = args->buf;
1551 		error = segflg == UIO_SYSSPACE ?
1552 		    copystr(fname, args->fname, PATH_MAX, &length) :
1553 		    copyinstr(fname, args->fname, PATH_MAX, &length);
1554 		if (error != 0)
1555 			return (error == ENAMETOOLONG ? E2BIG : error);
1556 	} else
1557 		length = 0;
1558 
1559 	/* Set up for _arg_*()/_env_*() */
1560 	args->endp = args->buf + length;
1561 	/* begin_argv must be set and kept updated */
1562 	args->begin_argv = args->endp;
1563 	KASSERT(exec_map_entry_size - length >= ARG_MAX,
1564 	    ("too little space remaining for arguments %zu < %zu",
1565 	    exec_map_entry_size - length, (size_t)ARG_MAX));
1566 	args->stringspace = ARG_MAX;
1567 
1568 	return (0);
1569 }
1570 
1571 static int
1572 exec_args_add_str(struct image_args *args, const char *str,
1573     enum uio_seg segflg, int *countp)
1574 {
1575 	int error;
1576 	size_t length;
1577 
1578 	KASSERT(args->endp != NULL, ("endp not initialized"));
1579 	KASSERT(args->begin_argv != NULL, ("begin_argp not initialized"));
1580 
1581 	error = (segflg == UIO_SYSSPACE) ?
1582 	    copystr(str, args->endp, args->stringspace, &length) :
1583 	    copyinstr(str, args->endp, args->stringspace, &length);
1584 	if (error != 0)
1585 		return (error == ENAMETOOLONG ? E2BIG : error);
1586 	args->stringspace -= length;
1587 	args->endp += length;
1588 	(*countp)++;
1589 
1590 	return (0);
1591 }
1592 
1593 int
1594 exec_args_add_arg(struct image_args *args, const char *argp,
1595     enum uio_seg segflg)
1596 {
1597 
1598 	KASSERT(args->envc == 0, ("appending args after env"));
1599 
1600 	return (exec_args_add_str(args, argp, segflg, &args->argc));
1601 }
1602 
1603 int
1604 exec_args_add_env(struct image_args *args, const char *envp,
1605     enum uio_seg segflg)
1606 {
1607 
1608 	if (args->envc == 0)
1609 		args->begin_envv = args->endp;
1610 
1611 	return (exec_args_add_str(args, envp, segflg, &args->envc));
1612 }
1613 
1614 int
1615 exec_args_adjust_args(struct image_args *args, size_t consume, ssize_t extend)
1616 {
1617 	ssize_t offset;
1618 
1619 	KASSERT(args->endp != NULL, ("endp not initialized"));
1620 	KASSERT(args->begin_argv != NULL, ("begin_argp not initialized"));
1621 
1622 	offset = extend - consume;
1623 	if (args->stringspace < offset)
1624 		return (E2BIG);
1625 	memmove(args->begin_argv + extend, args->begin_argv + consume,
1626 	    args->endp - args->begin_argv + consume);
1627 	if (args->envc > 0)
1628 		args->begin_envv += offset;
1629 	args->endp += offset;
1630 	args->stringspace -= offset;
1631 	return (0);
1632 }
1633 
1634 char *
1635 exec_args_get_begin_envv(struct image_args *args)
1636 {
1637 
1638 	KASSERT(args->endp != NULL, ("endp not initialized"));
1639 
1640 	if (args->envc > 0)
1641 		return (args->begin_envv);
1642 	return (args->endp);
1643 }
1644 
1645 /*
1646  * Copy strings out to the new process address space, constructing new arg
1647  * and env vector tables. Return a pointer to the base so that it can be used
1648  * as the initial stack pointer.
1649  */
1650 int
1651 exec_copyout_strings(struct image_params *imgp, uintptr_t *stack_base)
1652 {
1653 	int argc, envc;
1654 	char **vectp;
1655 	char *stringp;
1656 	uintptr_t destp, ustringp;
1657 	struct ps_strings *arginfo;
1658 	struct proc *p;
1659 	struct sysentvec *sysent;
1660 	size_t execpath_len;
1661 	int error, szsigcode;
1662 	char canary[sizeof(long) * 8];
1663 
1664 	p = imgp->proc;
1665 	sysent = p->p_sysent;
1666 
1667 	destp =	PROC_PS_STRINGS(p);
1668 	arginfo = imgp->ps_strings = (void *)destp;
1669 
1670 	/*
1671 	 * Install sigcode.
1672 	 */
1673 	if (sysent->sv_shared_page_base == 0 && sysent->sv_szsigcode != NULL) {
1674 		szsigcode = *(sysent->sv_szsigcode);
1675 		destp -= szsigcode;
1676 		destp = rounddown2(destp, sizeof(void *));
1677 		error = copyout(sysent->sv_sigcode, (void *)destp, szsigcode);
1678 		if (error != 0)
1679 			return (error);
1680 	}
1681 
1682 	/*
1683 	 * Copy the image path for the rtld.
1684 	 */
1685 	if (imgp->execpath != NULL && imgp->auxargs != NULL) {
1686 		execpath_len = strlen(imgp->execpath) + 1;
1687 		destp -= execpath_len;
1688 		destp = rounddown2(destp, sizeof(void *));
1689 		imgp->execpathp = (void *)destp;
1690 		error = copyout(imgp->execpath, imgp->execpathp, execpath_len);
1691 		if (error != 0)
1692 			return (error);
1693 	}
1694 
1695 	/*
1696 	 * Prepare the canary for SSP.
1697 	 */
1698 	arc4rand(canary, sizeof(canary), 0);
1699 	destp -= sizeof(canary);
1700 	imgp->canary = (void *)destp;
1701 	error = copyout(canary, imgp->canary, sizeof(canary));
1702 	if (error != 0)
1703 		return (error);
1704 	imgp->canarylen = sizeof(canary);
1705 
1706 	/*
1707 	 * Prepare the pagesizes array.
1708 	 */
1709 	imgp->pagesizeslen = sizeof(pagesizes[0]) * MAXPAGESIZES;
1710 	destp -= imgp->pagesizeslen;
1711 	destp = rounddown2(destp, sizeof(void *));
1712 	imgp->pagesizes = (void *)destp;
1713 	error = copyout(pagesizes, imgp->pagesizes, imgp->pagesizeslen);
1714 	if (error != 0)
1715 		return (error);
1716 
1717 	/*
1718 	 * Allocate room for the argument and environment strings.
1719 	 */
1720 	destp -= ARG_MAX - imgp->args->stringspace;
1721 	destp = rounddown2(destp, sizeof(void *));
1722 	ustringp = destp;
1723 
1724 	if (imgp->auxargs) {
1725 		/*
1726 		 * Allocate room on the stack for the ELF auxargs
1727 		 * array.  It has up to AT_COUNT entries.
1728 		 */
1729 		destp -= AT_COUNT * sizeof(Elf_Auxinfo);
1730 		destp = rounddown2(destp, sizeof(void *));
1731 	}
1732 
1733 	vectp = (char **)destp;
1734 
1735 	/*
1736 	 * Allocate room for the argv[] and env vectors including the
1737 	 * terminating NULL pointers.
1738 	 */
1739 	vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
1740 
1741 	/*
1742 	 * vectp also becomes our initial stack base
1743 	 */
1744 	*stack_base = (uintptr_t)vectp;
1745 
1746 	stringp = imgp->args->begin_argv;
1747 	argc = imgp->args->argc;
1748 	envc = imgp->args->envc;
1749 
1750 	/*
1751 	 * Copy out strings - arguments and environment.
1752 	 */
1753 	error = copyout(stringp, (void *)ustringp,
1754 	    ARG_MAX - imgp->args->stringspace);
1755 	if (error != 0)
1756 		return (error);
1757 
1758 	/*
1759 	 * Fill in "ps_strings" struct for ps, w, etc.
1760 	 */
1761 	imgp->argv = vectp;
1762 	if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 ||
1763 	    suword32(&arginfo->ps_nargvstr, argc) != 0)
1764 		return (EFAULT);
1765 
1766 	/*
1767 	 * Fill in argument portion of vector table.
1768 	 */
1769 	for (; argc > 0; --argc) {
1770 		if (suword(vectp++, ustringp) != 0)
1771 			return (EFAULT);
1772 		while (*stringp++ != 0)
1773 			ustringp++;
1774 		ustringp++;
1775 	}
1776 
1777 	/* a null vector table pointer separates the argp's from the envp's */
1778 	if (suword(vectp++, 0) != 0)
1779 		return (EFAULT);
1780 
1781 	imgp->envv = vectp;
1782 	if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 ||
1783 	    suword32(&arginfo->ps_nenvstr, envc) != 0)
1784 		return (EFAULT);
1785 
1786 	/*
1787 	 * Fill in environment portion of vector table.
1788 	 */
1789 	for (; envc > 0; --envc) {
1790 		if (suword(vectp++, ustringp) != 0)
1791 			return (EFAULT);
1792 		while (*stringp++ != 0)
1793 			ustringp++;
1794 		ustringp++;
1795 	}
1796 
1797 	/* end of vector table is a null pointer */
1798 	if (suword(vectp, 0) != 0)
1799 		return (EFAULT);
1800 
1801 	if (imgp->auxargs) {
1802 		vectp++;
1803 		error = imgp->sysent->sv_copyout_auxargs(imgp,
1804 		    (uintptr_t)vectp);
1805 		if (error != 0)
1806 			return (error);
1807 	}
1808 
1809 	return (0);
1810 }
1811 
1812 /*
1813  * Check permissions of file to execute.
1814  *	Called with imgp->vp locked.
1815  *	Return 0 for success or error code on failure.
1816  */
1817 int
1818 exec_check_permissions(struct image_params *imgp)
1819 {
1820 	struct vnode *vp = imgp->vp;
1821 	struct vattr *attr = imgp->attr;
1822 	struct thread *td;
1823 	int error;
1824 
1825 	td = curthread;
1826 
1827 	/* Get file attributes */
1828 	error = VOP_GETATTR(vp, attr, td->td_ucred);
1829 	if (error)
1830 		return (error);
1831 
1832 #ifdef MAC
1833 	error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1834 	if (error)
1835 		return (error);
1836 #endif
1837 
1838 	/*
1839 	 * 1) Check if file execution is disabled for the filesystem that
1840 	 *    this file resides on.
1841 	 * 2) Ensure that at least one execute bit is on. Otherwise, a
1842 	 *    privileged user will always succeed, and we don't want this
1843 	 *    to happen unless the file really is executable.
1844 	 * 3) Ensure that the file is a regular file.
1845 	 */
1846 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1847 	    (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1848 	    (attr->va_type != VREG))
1849 		return (EACCES);
1850 
1851 	/*
1852 	 * Zero length files can't be exec'd
1853 	 */
1854 	if (attr->va_size == 0)
1855 		return (ENOEXEC);
1856 
1857 	/*
1858 	 *  Check for execute permission to file based on current credentials.
1859 	 */
1860 	error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1861 	if (error)
1862 		return (error);
1863 
1864 	/*
1865 	 * Check number of open-for-writes on the file and deny execution
1866 	 * if there are any.
1867 	 *
1868 	 * Add a text reference now so no one can write to the
1869 	 * executable while we're activating it.
1870 	 *
1871 	 * Remember if this was set before and unset it in case this is not
1872 	 * actually an executable image.
1873 	 */
1874 	error = VOP_SET_TEXT(vp);
1875 	if (error != 0)
1876 		return (error);
1877 	imgp->textset = true;
1878 
1879 	/*
1880 	 * Call filesystem specific open routine (which does nothing in the
1881 	 * general case).
1882 	 */
1883 	error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1884 	if (error == 0)
1885 		imgp->opened = true;
1886 	return (error);
1887 }
1888 
1889 /*
1890  * Exec handler registration
1891  */
1892 int
1893 exec_register(const struct execsw *execsw_arg)
1894 {
1895 	const struct execsw **es, **xs, **newexecsw;
1896 	u_int count = 2;	/* New slot and trailing NULL */
1897 
1898 	if (execsw)
1899 		for (es = execsw; *es; es++)
1900 			count++;
1901 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1902 	xs = newexecsw;
1903 	if (execsw)
1904 		for (es = execsw; *es; es++)
1905 			*xs++ = *es;
1906 	*xs++ = execsw_arg;
1907 	*xs = NULL;
1908 	if (execsw)
1909 		free(execsw, M_TEMP);
1910 	execsw = newexecsw;
1911 	return (0);
1912 }
1913 
1914 int
1915 exec_unregister(const struct execsw *execsw_arg)
1916 {
1917 	const struct execsw **es, **xs, **newexecsw;
1918 	int count = 1;
1919 
1920 	if (execsw == NULL)
1921 		panic("unregister with no handlers left?\n");
1922 
1923 	for (es = execsw; *es; es++) {
1924 		if (*es == execsw_arg)
1925 			break;
1926 	}
1927 	if (*es == NULL)
1928 		return (ENOENT);
1929 	for (es = execsw; *es; es++)
1930 		if (*es != execsw_arg)
1931 			count++;
1932 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1933 	xs = newexecsw;
1934 	for (es = execsw; *es; es++)
1935 		if (*es != execsw_arg)
1936 			*xs++ = *es;
1937 	*xs = NULL;
1938 	if (execsw)
1939 		free(execsw, M_TEMP);
1940 	execsw = newexecsw;
1941 	return (0);
1942 }
1943 
1944 /*
1945  * Write out a core segment to the compression stream.
1946  */
1947 static int
1948 compress_chunk(struct coredump_params *cp, char *base, char *buf, size_t len)
1949 {
1950 	size_t chunk_len;
1951 	int error;
1952 
1953 	while (len > 0) {
1954 		chunk_len = MIN(len, CORE_BUF_SIZE);
1955 
1956 		/*
1957 		 * We can get EFAULT error here.
1958 		 * In that case zero out the current chunk of the segment.
1959 		 */
1960 		error = copyin(base, buf, chunk_len);
1961 		if (error != 0)
1962 			bzero(buf, chunk_len);
1963 		error = compressor_write(cp->comp, buf, chunk_len);
1964 		if (error != 0)
1965 			break;
1966 		base += chunk_len;
1967 		len -= chunk_len;
1968 	}
1969 	return (error);
1970 }
1971 
1972 int
1973 core_write(struct coredump_params *cp, const void *base, size_t len,
1974     off_t offset, enum uio_seg seg, size_t *resid)
1975 {
1976 
1977 	return (vn_rdwr_inchunks(UIO_WRITE, cp->vp, __DECONST(void *, base),
1978 	    len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1979 	    cp->active_cred, cp->file_cred, resid, cp->td));
1980 }
1981 
1982 int
1983 core_output(char *base, size_t len, off_t offset, struct coredump_params *cp,
1984     void *tmpbuf)
1985 {
1986 	vm_map_t map;
1987 	struct mount *mp;
1988 	size_t resid, runlen;
1989 	int error;
1990 	bool success;
1991 
1992 	KASSERT((uintptr_t)base % PAGE_SIZE == 0,
1993 	    ("%s: user address %p is not page-aligned", __func__, base));
1994 
1995 	if (cp->comp != NULL)
1996 		return (compress_chunk(cp, base, tmpbuf, len));
1997 
1998 	map = &cp->td->td_proc->p_vmspace->vm_map;
1999 	for (; len > 0; base += runlen, offset += runlen, len -= runlen) {
2000 		/*
2001 		 * Attempt to page in all virtual pages in the range.  If a
2002 		 * virtual page is not backed by the pager, it is represented as
2003 		 * a hole in the file.  This can occur with zero-filled
2004 		 * anonymous memory or truncated files, for example.
2005 		 */
2006 		for (runlen = 0; runlen < len; runlen += PAGE_SIZE) {
2007 			if (core_dump_can_intr && curproc_sigkilled())
2008 				return (EINTR);
2009 			error = vm_fault(map, (uintptr_t)base + runlen,
2010 			    VM_PROT_READ, VM_FAULT_NOFILL, NULL);
2011 			if (runlen == 0)
2012 				success = error == KERN_SUCCESS;
2013 			else if ((error == KERN_SUCCESS) != success)
2014 				break;
2015 		}
2016 
2017 		if (success) {
2018 			error = core_write(cp, base, runlen, offset,
2019 			    UIO_USERSPACE, &resid);
2020 			if (error != 0) {
2021 				if (error != EFAULT)
2022 					break;
2023 
2024 				/*
2025 				 * EFAULT may be returned if the user mapping
2026 				 * could not be accessed, e.g., because a mapped
2027 				 * file has been truncated.  Skip the page if no
2028 				 * progress was made, to protect against a
2029 				 * hypothetical scenario where vm_fault() was
2030 				 * successful but core_write() returns EFAULT
2031 				 * anyway.
2032 				 */
2033 				runlen -= resid;
2034 				if (runlen == 0) {
2035 					success = false;
2036 					runlen = PAGE_SIZE;
2037 				}
2038 			}
2039 		}
2040 		if (!success) {
2041 			error = vn_start_write(cp->vp, &mp, V_WAIT);
2042 			if (error != 0)
2043 				break;
2044 			vn_lock(cp->vp, LK_EXCLUSIVE | LK_RETRY);
2045 			error = vn_truncate_locked(cp->vp, offset + runlen,
2046 			    false, cp->td->td_ucred);
2047 			VOP_UNLOCK(cp->vp);
2048 			vn_finished_write(mp);
2049 			if (error != 0)
2050 				break;
2051 		}
2052 	}
2053 	return (error);
2054 }
2055 
2056 /*
2057  * Drain into a core file.
2058  */
2059 int
2060 sbuf_drain_core_output(void *arg, const char *data, int len)
2061 {
2062 	struct coredump_params *cp;
2063 	struct proc *p;
2064 	int error, locked;
2065 
2066 	cp = arg;
2067 	p = cp->td->td_proc;
2068 
2069 	/*
2070 	 * Some kern_proc out routines that print to this sbuf may
2071 	 * call us with the process lock held. Draining with the
2072 	 * non-sleepable lock held is unsafe. The lock is needed for
2073 	 * those routines when dumping a live process. In our case we
2074 	 * can safely release the lock before draining and acquire
2075 	 * again after.
2076 	 */
2077 	locked = PROC_LOCKED(p);
2078 	if (locked)
2079 		PROC_UNLOCK(p);
2080 	if (cp->comp != NULL)
2081 		error = compressor_write(cp->comp, __DECONST(char *, data),
2082 		    len);
2083 	else
2084 		error = core_write(cp, __DECONST(void *, data), len, cp->offset,
2085 		    UIO_SYSSPACE, NULL);
2086 	if (locked)
2087 		PROC_LOCK(p);
2088 	if (error != 0)
2089 		return (-error);
2090 	cp->offset += len;
2091 	return (len);
2092 }
2093