xref: /freebsd/sys/kern/kern_exec.c (revision 6580f5c38dd5b01aeeaed16b370f1a12423437f0)
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 #include "opt_capsicum.h"
31 #include "opt_hwpmc_hooks.h"
32 #include "opt_ktrace.h"
33 #include "opt_vm.h"
34 
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/acct.h>
38 #include <sys/asan.h>
39 #include <sys/capsicum.h>
40 #include <sys/compressor.h>
41 #include <sys/eventhandler.h>
42 #include <sys/exec.h>
43 #include <sys/fcntl.h>
44 #include <sys/filedesc.h>
45 #include <sys/imgact.h>
46 #include <sys/imgact_elf.h>
47 #include <sys/kernel.h>
48 #include <sys/lock.h>
49 #include <sys/malloc.h>
50 #include <sys/mman.h>
51 #include <sys/mount.h>
52 #include <sys/mutex.h>
53 #include <sys/namei.h>
54 #include <sys/priv.h>
55 #include <sys/proc.h>
56 #include <sys/ptrace.h>
57 #include <sys/reg.h>
58 #include <sys/resourcevar.h>
59 #include <sys/rwlock.h>
60 #include <sys/sched.h>
61 #include <sys/sdt.h>
62 #include <sys/sf_buf.h>
63 #include <sys/shm.h>
64 #include <sys/signalvar.h>
65 #include <sys/smp.h>
66 #include <sys/stat.h>
67 #include <sys/syscallsubr.h>
68 #include <sys/sysctl.h>
69 #include <sys/sysent.h>
70 #include <sys/sysproto.h>
71 #include <sys/timers.h>
72 #include <sys/umtxvar.h>
73 #include <sys/vnode.h>
74 #include <sys/wait.h>
75 #ifdef KTRACE
76 #include <sys/ktrace.h>
77 #endif
78 
79 #include <vm/vm.h>
80 #include <vm/vm_param.h>
81 #include <vm/pmap.h>
82 #include <vm/vm_page.h>
83 #include <vm/vm_map.h>
84 #include <vm/vm_kern.h>
85 #include <vm/vm_extern.h>
86 #include <vm/vm_object.h>
87 #include <vm/vm_pager.h>
88 
89 #ifdef	HWPMC_HOOKS
90 #include <sys/pmckern.h>
91 #endif
92 
93 #include <security/audit/audit.h>
94 #include <security/mac/mac_framework.h>
95 
96 #ifdef KDTRACE_HOOKS
97 #include <sys/dtrace_bsd.h>
98 dtrace_execexit_func_t	dtrace_fasttrap_exec;
99 #endif
100 
101 SDT_PROVIDER_DECLARE(proc);
102 SDT_PROBE_DEFINE1(proc, , , exec, "char *");
103 SDT_PROBE_DEFINE1(proc, , , exec__failure, "int");
104 SDT_PROBE_DEFINE1(proc, , , exec__success, "char *");
105 
106 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
107 
108 int coredump_pack_fileinfo = 1;
109 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_fileinfo, CTLFLAG_RWTUN,
110     &coredump_pack_fileinfo, 0,
111     "Enable file path packing in 'procstat -f' coredump notes");
112 
113 int coredump_pack_vmmapinfo = 1;
114 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_vmmapinfo, CTLFLAG_RWTUN,
115     &coredump_pack_vmmapinfo, 0,
116     "Enable file path packing in 'procstat -v' coredump notes");
117 
118 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
119 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
120 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
121 static int do_execve(struct thread *td, struct image_args *args,
122     struct mac *mac_p, struct vmspace *oldvmspace);
123 
124 /* XXX This should be vm_size_t. */
125 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD|
126     CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_ps_strings, "LU",
127     "Location of process' ps_strings structure");
128 
129 /* XXX This should be vm_size_t. */
130 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD|
131     CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_usrstack, "LU",
132     "Top of process stack");
133 
134 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD|CTLFLAG_MPSAFE,
135     NULL, 0, sysctl_kern_stackprot, "I",
136     "Stack memory permissions");
137 
138 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
139 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
140     &ps_arg_cache_limit, 0,
141     "Process' command line characters cache limit");
142 
143 static int disallow_high_osrel;
144 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW,
145     &disallow_high_osrel, 0,
146     "Disallow execution of binaries built for higher version of the world");
147 
148 static int map_at_zero = 0;
149 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RWTUN, &map_at_zero, 0,
150     "Permit processes to map an object at virtual address 0.");
151 
152 static int core_dump_can_intr = 1;
153 SYSCTL_INT(_kern, OID_AUTO, core_dump_can_intr, CTLFLAG_RWTUN,
154     &core_dump_can_intr, 0,
155     "Core dumping interruptible with SIGKILL");
156 
157 static int
158 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
159 {
160 	struct proc *p;
161 	vm_offset_t ps_strings;
162 
163 	p = curproc;
164 #ifdef SCTL_MASK32
165 	if (req->flags & SCTL_MASK32) {
166 		unsigned int val;
167 		val = (unsigned int)PROC_PS_STRINGS(p);
168 		return (SYSCTL_OUT(req, &val, sizeof(val)));
169 	}
170 #endif
171 	ps_strings = PROC_PS_STRINGS(p);
172 	return (SYSCTL_OUT(req, &ps_strings, sizeof(ps_strings)));
173 }
174 
175 static int
176 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
177 {
178 	struct proc *p;
179 	vm_offset_t val;
180 
181 	p = curproc;
182 #ifdef SCTL_MASK32
183 	if (req->flags & SCTL_MASK32) {
184 		unsigned int val32;
185 
186 		val32 = round_page((unsigned int)p->p_vmspace->vm_stacktop);
187 		return (SYSCTL_OUT(req, &val32, sizeof(val32)));
188 	}
189 #endif
190 	val = round_page(p->p_vmspace->vm_stacktop);
191 	return (SYSCTL_OUT(req, &val, sizeof(val)));
192 }
193 
194 static int
195 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
196 {
197 	struct proc *p;
198 
199 	p = curproc;
200 	return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
201 	    sizeof(p->p_sysent->sv_stackprot)));
202 }
203 
204 /*
205  * Each of the items is a pointer to a `const struct execsw', hence the
206  * double pointer here.
207  */
208 static const struct execsw **execsw;
209 
210 #ifndef _SYS_SYSPROTO_H_
211 struct execve_args {
212 	char    *fname;
213 	char    **argv;
214 	char    **envv;
215 };
216 #endif
217 
218 int
219 sys_execve(struct thread *td, struct execve_args *uap)
220 {
221 	struct image_args args;
222 	struct vmspace *oldvmspace;
223 	int error;
224 
225 	error = pre_execve(td, &oldvmspace);
226 	if (error != 0)
227 		return (error);
228 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
229 	    uap->argv, uap->envv);
230 	if (error == 0)
231 		error = kern_execve(td, &args, NULL, oldvmspace);
232 	post_execve(td, error, oldvmspace);
233 	AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
234 	return (error);
235 }
236 
237 #ifndef _SYS_SYSPROTO_H_
238 struct fexecve_args {
239 	int	fd;
240 	char	**argv;
241 	char	**envv;
242 };
243 #endif
244 int
245 sys_fexecve(struct thread *td, struct fexecve_args *uap)
246 {
247 	struct image_args args;
248 	struct vmspace *oldvmspace;
249 	int error;
250 
251 	error = pre_execve(td, &oldvmspace);
252 	if (error != 0)
253 		return (error);
254 	error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
255 	    uap->argv, uap->envv);
256 	if (error == 0) {
257 		args.fd = uap->fd;
258 		error = kern_execve(td, &args, NULL, oldvmspace);
259 	}
260 	post_execve(td, error, oldvmspace);
261 	AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
262 	return (error);
263 }
264 
265 #ifndef _SYS_SYSPROTO_H_
266 struct __mac_execve_args {
267 	char	*fname;
268 	char	**argv;
269 	char	**envv;
270 	struct mac	*mac_p;
271 };
272 #endif
273 
274 int
275 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap)
276 {
277 #ifdef MAC
278 	struct image_args args;
279 	struct vmspace *oldvmspace;
280 	int error;
281 
282 	error = pre_execve(td, &oldvmspace);
283 	if (error != 0)
284 		return (error);
285 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
286 	    uap->argv, uap->envv);
287 	if (error == 0)
288 		error = kern_execve(td, &args, uap->mac_p, oldvmspace);
289 	post_execve(td, error, oldvmspace);
290 	AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
291 	return (error);
292 #else
293 	return (ENOSYS);
294 #endif
295 }
296 
297 int
298 pre_execve(struct thread *td, struct vmspace **oldvmspace)
299 {
300 	struct proc *p;
301 	int error;
302 
303 	KASSERT(td == curthread, ("non-current thread %p", td));
304 	error = 0;
305 	p = td->td_proc;
306 	if ((p->p_flag & P_HADTHREADS) != 0) {
307 		PROC_LOCK(p);
308 		if (thread_single(p, SINGLE_BOUNDARY) != 0)
309 			error = ERESTART;
310 		PROC_UNLOCK(p);
311 	}
312 	KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0,
313 	    ("nested execve"));
314 	*oldvmspace = p->p_vmspace;
315 	return (error);
316 }
317 
318 void
319 post_execve(struct thread *td, int error, struct vmspace *oldvmspace)
320 {
321 	struct proc *p;
322 
323 	KASSERT(td == curthread, ("non-current thread %p", td));
324 	p = td->td_proc;
325 	if ((p->p_flag & P_HADTHREADS) != 0) {
326 		PROC_LOCK(p);
327 		/*
328 		 * If success, we upgrade to SINGLE_EXIT state to
329 		 * force other threads to suicide.
330 		 */
331 		if (error == EJUSTRETURN)
332 			thread_single(p, SINGLE_EXIT);
333 		else
334 			thread_single_end(p, SINGLE_BOUNDARY);
335 		PROC_UNLOCK(p);
336 	}
337 	exec_cleanup(td, oldvmspace);
338 }
339 
340 /*
341  * kern_execve() has the astonishing property of not always returning to
342  * the caller.  If sufficiently bad things happen during the call to
343  * do_execve(), it can end up calling exit1(); as a result, callers must
344  * avoid doing anything which they might need to undo (e.g., allocating
345  * memory).
346  */
347 int
348 kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p,
349     struct vmspace *oldvmspace)
350 {
351 
352 	TSEXEC(td->td_proc->p_pid, args->begin_argv);
353 	AUDIT_ARG_ARGV(args->begin_argv, args->argc,
354 	    exec_args_get_begin_envv(args) - args->begin_argv);
355 	AUDIT_ARG_ENVV(exec_args_get_begin_envv(args), args->envc,
356 	    args->endp - exec_args_get_begin_envv(args));
357 
358 	/* Must have at least one argument. */
359 	if (args->argc == 0) {
360 		exec_free_args(args);
361 		return (EINVAL);
362 	}
363 	return (do_execve(td, args, mac_p, oldvmspace));
364 }
365 
366 static void
367 execve_nosetid(struct image_params *imgp)
368 {
369 	imgp->credential_setid = false;
370 	if (imgp->newcred != NULL) {
371 		crfree(imgp->newcred);
372 		imgp->newcred = NULL;
373 	}
374 }
375 
376 /*
377  * In-kernel implementation of execve().  All arguments are assumed to be
378  * userspace pointers from the passed thread.
379  */
380 static int
381 do_execve(struct thread *td, struct image_args *args, struct mac *mac_p,
382     struct vmspace *oldvmspace)
383 {
384 	struct proc *p = td->td_proc;
385 	struct nameidata nd;
386 	struct ucred *oldcred;
387 	struct uidinfo *euip = NULL;
388 	uintptr_t stack_base;
389 	struct image_params image_params, *imgp;
390 	struct vattr attr;
391 	struct pargs *oldargs = NULL, *newargs = NULL;
392 	struct sigacts *oldsigacts = NULL, *newsigacts = NULL;
393 #ifdef KTRACE
394 	struct ktr_io_params *kiop;
395 #endif
396 	struct vnode *oldtextvp, *newtextvp;
397 	struct vnode *oldtextdvp, *newtextdvp;
398 	char *oldbinname, *newbinname;
399 	bool credential_changing;
400 #ifdef MAC
401 	struct label *interpvplabel = NULL;
402 	bool will_transition;
403 #endif
404 #ifdef HWPMC_HOOKS
405 	struct pmckern_procexec pe;
406 #endif
407 	int error, i, orig_osrel;
408 	uint32_t orig_fctl0;
409 	Elf_Brandinfo *orig_brandinfo;
410 	size_t freepath_size;
411 	static const char fexecv_proc_title[] = "(fexecv)";
412 
413 	imgp = &image_params;
414 	oldtextvp = oldtextdvp = NULL;
415 	newtextvp = newtextdvp = NULL;
416 	newbinname = oldbinname = NULL;
417 #ifdef KTRACE
418 	kiop = NULL;
419 #endif
420 
421 	/*
422 	 * Lock the process and set the P_INEXEC flag to indicate that
423 	 * it should be left alone until we're done here.  This is
424 	 * necessary to avoid race conditions - e.g. in ptrace() -
425 	 * that might allow a local user to illicitly obtain elevated
426 	 * privileges.
427 	 */
428 	PROC_LOCK(p);
429 	KASSERT((p->p_flag & P_INEXEC) == 0,
430 	    ("%s(): process already has P_INEXEC flag", __func__));
431 	p->p_flag |= P_INEXEC;
432 	PROC_UNLOCK(p);
433 
434 	/*
435 	 * Initialize part of the common data
436 	 */
437 	bzero(imgp, sizeof(*imgp));
438 	imgp->proc = p;
439 	imgp->attr = &attr;
440 	imgp->args = args;
441 	oldcred = p->p_ucred;
442 	orig_osrel = p->p_osrel;
443 	orig_fctl0 = p->p_fctl0;
444 	orig_brandinfo = p->p_elf_brandinfo;
445 
446 #ifdef MAC
447 	error = mac_execve_enter(imgp, mac_p);
448 	if (error)
449 		goto exec_fail;
450 #endif
451 
452 	SDT_PROBE1(proc, , , exec, args->fname);
453 
454 interpret:
455 	if (args->fname != NULL) {
456 #ifdef CAPABILITY_MODE
457 		if (CAP_TRACING(td))
458 			ktrcapfail(CAPFAIL_NAMEI, args->fname);
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 	p->p_flag2 &= ~(P2_MEMBAR_PRIVE | P2_MEMBAR_PRIVE_SYNCORE |
811 	    P2_MEMBAR_GLOBE);
812 	if (p->p_flag & P_PPWAIT) {
813 		p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
814 		cv_broadcast(&p->p_pwait);
815 		/* STOPs are no longer ignored, arrange for AST */
816 		signotify(td);
817 	}
818 
819 	if ((imgp->sysent->sv_setid_allowed != NULL &&
820 	    !(*imgp->sysent->sv_setid_allowed)(td, imgp)) ||
821 	    (p->p_flag2 & P2_NO_NEW_PRIVS) != 0)
822 		execve_nosetid(imgp);
823 
824 	/*
825 	 * Implement image setuid/setgid installation.
826 	 */
827 	if (imgp->credential_setid) {
828 		/*
829 		 * Turn off syscall tracing for set-id programs, except for
830 		 * root.  Record any set-id flags first to make sure that
831 		 * we do not regain any tracing during a possible block.
832 		 */
833 		setsugid(p);
834 #ifdef KTRACE
835 		kiop = ktrprocexec(p);
836 #endif
837 		/*
838 		 * Close any file descriptors 0..2 that reference procfs,
839 		 * then make sure file descriptors 0..2 are in use.
840 		 *
841 		 * Both fdsetugidsafety() and fdcheckstd() may call functions
842 		 * taking sleepable locks, so temporarily drop our locks.
843 		 */
844 		PROC_UNLOCK(p);
845 		VOP_UNLOCK(imgp->vp);
846 		fdsetugidsafety(td);
847 		error = fdcheckstd(td);
848 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
849 		if (error != 0)
850 			goto exec_fail_dealloc;
851 		PROC_LOCK(p);
852 #ifdef MAC
853 		if (will_transition) {
854 			mac_vnode_execve_transition(oldcred, imgp->newcred,
855 			    imgp->vp, interpvplabel, imgp);
856 		}
857 #endif
858 	} else {
859 		if (oldcred->cr_uid == oldcred->cr_ruid &&
860 		    oldcred->cr_gid == oldcred->cr_rgid)
861 			p->p_flag &= ~P_SUGID;
862 	}
863 	/*
864 	 * Set the new credentials.
865 	 */
866 	if (imgp->newcred != NULL) {
867 		proc_set_cred(p, imgp->newcred);
868 		crfree(oldcred);
869 		oldcred = NULL;
870 	}
871 
872 	/*
873 	 * Store the vp for use in kern.proc.pathname.  This vnode was
874 	 * referenced by namei() or by fexecve variant of fname handling.
875 	 */
876 	oldtextvp = p->p_textvp;
877 	p->p_textvp = newtextvp;
878 	oldtextdvp = p->p_textdvp;
879 	p->p_textdvp = newtextdvp;
880 	newtextdvp = NULL;
881 	oldbinname = p->p_binname;
882 	p->p_binname = newbinname;
883 	newbinname = NULL;
884 
885 #ifdef KDTRACE_HOOKS
886 	/*
887 	 * Tell the DTrace fasttrap provider about the exec if it
888 	 * has declared an interest.
889 	 */
890 	if (dtrace_fasttrap_exec)
891 		dtrace_fasttrap_exec(p);
892 #endif
893 
894 	/*
895 	 * Notify others that we exec'd, and clear the P_INEXEC flag
896 	 * as we're now a bona fide freshly-execed process.
897 	 */
898 	KNOTE_LOCKED(p->p_klist, NOTE_EXEC);
899 	p->p_flag &= ~P_INEXEC;
900 
901 	/* clear "fork but no exec" flag, as we _are_ execing */
902 	p->p_acflag &= ~AFORK;
903 
904 	/*
905 	 * Free any previous argument cache and replace it with
906 	 * the new argument cache, if any.
907 	 */
908 	oldargs = p->p_args;
909 	p->p_args = newargs;
910 	newargs = NULL;
911 
912 	PROC_UNLOCK(p);
913 
914 #ifdef	HWPMC_HOOKS
915 	/*
916 	 * Check if system-wide sampling is in effect or if the
917 	 * current process is using PMCs.  If so, do exec() time
918 	 * processing.  This processing needs to happen AFTER the
919 	 * P_INEXEC flag is cleared.
920 	 */
921 	if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
922 		VOP_UNLOCK(imgp->vp);
923 		pe.pm_credentialschanged = credential_changing;
924 		pe.pm_baseaddr = imgp->reloc_base;
925 		pe.pm_dynaddr = imgp->et_dyn_addr;
926 
927 		PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
928 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
929 	}
930 #endif
931 
932 	/* Set values passed into the program in registers. */
933 	(*p->p_sysent->sv_setregs)(td, imgp, stack_base);
934 
935 	VOP_MMAPPED(imgp->vp);
936 
937 	SDT_PROBE1(proc, , , exec__success, args->fname);
938 
939 exec_fail_dealloc:
940 	if (error != 0) {
941 		p->p_osrel = orig_osrel;
942 		p->p_fctl0 = orig_fctl0;
943 		p->p_elf_brandinfo = orig_brandinfo;
944 	}
945 
946 	if (imgp->firstpage != NULL)
947 		exec_unmap_first_page(imgp);
948 
949 	if (imgp->vp != NULL) {
950 		if (imgp->opened)
951 			VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
952 		if (imgp->textset)
953 			VOP_UNSET_TEXT_CHECKED(imgp->vp);
954 		if (error != 0)
955 			vput(imgp->vp);
956 		else
957 			VOP_UNLOCK(imgp->vp);
958 		if (args->fname != NULL)
959 			NDFREE_PNBUF(&nd);
960 		if (newtextdvp != NULL)
961 			vrele(newtextdvp);
962 		free(newbinname, M_PARGS);
963 	}
964 
965 	if (imgp->object != NULL)
966 		vm_object_deallocate(imgp->object);
967 
968 	free(imgp->freepath, M_TEMP);
969 
970 	if (error == 0) {
971 		if (p->p_ptevents & PTRACE_EXEC) {
972 			PROC_LOCK(p);
973 			if (p->p_ptevents & PTRACE_EXEC)
974 				td->td_dbgflags |= TDB_EXEC;
975 			PROC_UNLOCK(p);
976 		}
977 	} else {
978 exec_fail:
979 		/* we're done here, clear P_INEXEC */
980 		PROC_LOCK(p);
981 		p->p_flag &= ~P_INEXEC;
982 		PROC_UNLOCK(p);
983 
984 		SDT_PROBE1(proc, , , exec__failure, error);
985 	}
986 
987 	if (imgp->newcred != NULL && oldcred != NULL)
988 		crfree(imgp->newcred);
989 
990 #ifdef MAC
991 	mac_execve_exit(imgp);
992 	mac_execve_interpreter_exit(interpvplabel);
993 #endif
994 	exec_free_args(args);
995 
996 	/*
997 	 * Handle deferred decrement of ref counts.
998 	 */
999 	if (oldtextvp != NULL)
1000 		vrele(oldtextvp);
1001 	if (oldtextdvp != NULL)
1002 		vrele(oldtextdvp);
1003 	free(oldbinname, M_PARGS);
1004 #ifdef KTRACE
1005 	ktr_io_params_free(kiop);
1006 #endif
1007 	pargs_drop(oldargs);
1008 	pargs_drop(newargs);
1009 	if (oldsigacts != NULL)
1010 		sigacts_free(oldsigacts);
1011 	if (euip != NULL)
1012 		uifree(euip);
1013 
1014 	if (error && imgp->vmspace_destroyed) {
1015 		/* sorry, no more process anymore. exit gracefully */
1016 		exec_cleanup(td, oldvmspace);
1017 		exit1(td, 0, SIGABRT);
1018 		/* NOT REACHED */
1019 	}
1020 
1021 #ifdef KTRACE
1022 	if (error == 0)
1023 		ktrprocctor(p);
1024 #endif
1025 
1026 	/*
1027 	 * We don't want cpu_set_syscall_retval() to overwrite any of
1028 	 * the register values put in place by exec_setregs().
1029 	 * Implementations of cpu_set_syscall_retval() will leave
1030 	 * registers unmodified when returning EJUSTRETURN.
1031 	 */
1032 	return (error == 0 ? EJUSTRETURN : error);
1033 }
1034 
1035 void
1036 exec_cleanup(struct thread *td, struct vmspace *oldvmspace)
1037 {
1038 	if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
1039 		KASSERT(td->td_proc->p_vmspace != oldvmspace,
1040 		    ("oldvmspace still used"));
1041 		vmspace_free(oldvmspace);
1042 		td->td_pflags &= ~TDP_EXECVMSPC;
1043 	}
1044 }
1045 
1046 int
1047 exec_map_first_page(struct image_params *imgp)
1048 {
1049 	vm_object_t object;
1050 	vm_page_t m;
1051 	int error;
1052 
1053 	if (imgp->firstpage != NULL)
1054 		exec_unmap_first_page(imgp);
1055 
1056 	object = imgp->vp->v_object;
1057 	if (object == NULL)
1058 		return (EACCES);
1059 #if VM_NRESERVLEVEL > 0
1060 	if ((object->flags & OBJ_COLORED) == 0) {
1061 		VM_OBJECT_WLOCK(object);
1062 		vm_object_color(object, 0);
1063 		VM_OBJECT_WUNLOCK(object);
1064 	}
1065 #endif
1066 	error = vm_page_grab_valid_unlocked(&m, object, 0,
1067 	    VM_ALLOC_COUNT(VM_INITIAL_PAGEIN) |
1068 	    VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED);
1069 
1070 	if (error != VM_PAGER_OK)
1071 		return (EIO);
1072 	imgp->firstpage = sf_buf_alloc(m, 0);
1073 	imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
1074 
1075 	return (0);
1076 }
1077 
1078 void
1079 exec_unmap_first_page(struct image_params *imgp)
1080 {
1081 	vm_page_t m;
1082 
1083 	if (imgp->firstpage != NULL) {
1084 		m = sf_buf_page(imgp->firstpage);
1085 		sf_buf_free(imgp->firstpage);
1086 		imgp->firstpage = NULL;
1087 		vm_page_unwire(m, PQ_ACTIVE);
1088 	}
1089 }
1090 
1091 void
1092 exec_onexec_old(struct thread *td)
1093 {
1094 	sigfastblock_clear(td);
1095 	umtx_exec(td->td_proc);
1096 }
1097 
1098 /*
1099  * This is an optimization which removes the unmanaged shared page
1100  * mapping. In combination with pmap_remove_pages(), which cleans all
1101  * managed mappings in the process' vmspace pmap, no work will be left
1102  * for pmap_remove(min, max).
1103  */
1104 void
1105 exec_free_abi_mappings(struct proc *p)
1106 {
1107 	struct vmspace *vmspace;
1108 
1109 	vmspace = p->p_vmspace;
1110 	if (refcount_load(&vmspace->vm_refcnt) != 1)
1111 		return;
1112 
1113 	if (!PROC_HAS_SHP(p))
1114 		return;
1115 
1116 	pmap_remove(vmspace_pmap(vmspace), vmspace->vm_shp_base,
1117 	    vmspace->vm_shp_base + p->p_sysent->sv_shared_page_len);
1118 }
1119 
1120 /*
1121  * Run down the current address space and install a new one.
1122  */
1123 int
1124 exec_new_vmspace(struct image_params *imgp, struct sysentvec *sv)
1125 {
1126 	int error;
1127 	struct proc *p = imgp->proc;
1128 	struct vmspace *vmspace = p->p_vmspace;
1129 	struct thread *td = curthread;
1130 	vm_offset_t sv_minuser;
1131 	vm_map_t map;
1132 
1133 	imgp->vmspace_destroyed = true;
1134 	imgp->sysent = sv;
1135 
1136 	if (p->p_sysent->sv_onexec_old != NULL)
1137 		p->p_sysent->sv_onexec_old(td);
1138 	itimers_exec(p);
1139 
1140 	EVENTHANDLER_DIRECT_INVOKE(process_exec, p, imgp);
1141 
1142 	/*
1143 	 * Blow away entire process VM, if address space not shared,
1144 	 * otherwise, create a new VM space so that other threads are
1145 	 * not disrupted
1146 	 */
1147 	map = &vmspace->vm_map;
1148 	if (map_at_zero)
1149 		sv_minuser = sv->sv_minuser;
1150 	else
1151 		sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1152 	if (refcount_load(&vmspace->vm_refcnt) == 1 &&
1153 	    vm_map_min(map) == sv_minuser &&
1154 	    vm_map_max(map) == sv->sv_maxuser &&
1155 	    cpu_exec_vmspace_reuse(p, map)) {
1156 		exec_free_abi_mappings(p);
1157 		shmexit(vmspace);
1158 		pmap_remove_pages(vmspace_pmap(vmspace));
1159 		vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1160 		/*
1161 		 * An exec terminates mlockall(MCL_FUTURE).
1162 		 * ASLR and W^X states must be re-evaluated.
1163 		 */
1164 		vm_map_lock(map);
1165 		vm_map_modflags(map, 0, MAP_WIREFUTURE | MAP_ASLR |
1166 		    MAP_ASLR_IGNSTART | MAP_ASLR_STACK | MAP_WXORX);
1167 		vm_map_unlock(map);
1168 	} else {
1169 		error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1170 		if (error)
1171 			return (error);
1172 		vmspace = p->p_vmspace;
1173 		map = &vmspace->vm_map;
1174 	}
1175 	map->flags |= imgp->map_flags;
1176 
1177 	return (sv->sv_onexec != NULL ? sv->sv_onexec(p, imgp) : 0);
1178 }
1179 
1180 /*
1181  * Compute the stack size limit and map the main process stack.
1182  * Map the shared page.
1183  */
1184 int
1185 exec_map_stack(struct image_params *imgp)
1186 {
1187 	struct rlimit rlim_stack;
1188 	struct sysentvec *sv;
1189 	struct proc *p;
1190 	vm_map_t map;
1191 	struct vmspace *vmspace;
1192 	vm_offset_t stack_addr, stack_top;
1193 	vm_offset_t sharedpage_addr;
1194 	u_long ssiz;
1195 	int error, find_space, stack_off;
1196 	vm_prot_t stack_prot;
1197 	vm_object_t obj;
1198 
1199 	p = imgp->proc;
1200 	sv = p->p_sysent;
1201 
1202 	if (imgp->stack_sz != 0) {
1203 		ssiz = trunc_page(imgp->stack_sz);
1204 		PROC_LOCK(p);
1205 		lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack);
1206 		PROC_UNLOCK(p);
1207 		if (ssiz > rlim_stack.rlim_max)
1208 			ssiz = rlim_stack.rlim_max;
1209 		if (ssiz > rlim_stack.rlim_cur) {
1210 			rlim_stack.rlim_cur = ssiz;
1211 			kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack);
1212 		}
1213 	} else if (sv->sv_maxssiz != NULL) {
1214 		ssiz = *sv->sv_maxssiz;
1215 	} else {
1216 		ssiz = maxssiz;
1217 	}
1218 
1219 	vmspace = p->p_vmspace;
1220 	map = &vmspace->vm_map;
1221 
1222 	stack_prot = sv->sv_shared_page_obj != NULL && imgp->stack_prot != 0 ?
1223 	    imgp->stack_prot : sv->sv_stackprot;
1224 	if ((map->flags & MAP_ASLR_STACK) != 0) {
1225 		stack_addr = round_page((vm_offset_t)p->p_vmspace->vm_daddr +
1226 		    lim_max(curthread, RLIMIT_DATA));
1227 		find_space = VMFS_ANY_SPACE;
1228 	} else {
1229 		stack_addr = sv->sv_usrstack - ssiz;
1230 		find_space = VMFS_NO_SPACE;
1231 	}
1232 	error = vm_map_find(map, NULL, 0, &stack_addr, (vm_size_t)ssiz,
1233 	    sv->sv_usrstack, find_space, stack_prot, VM_PROT_ALL,
1234 	    MAP_STACK_GROWS_DOWN);
1235 	if (error != KERN_SUCCESS) {
1236 		uprintf("exec_new_vmspace: mapping stack size %#jx prot %#x "
1237 		    "failed, mach error %d errno %d\n", (uintmax_t)ssiz,
1238 		    stack_prot, error, vm_mmap_to_errno(error));
1239 		return (vm_mmap_to_errno(error));
1240 	}
1241 
1242 	stack_top = stack_addr + ssiz;
1243 	if ((map->flags & MAP_ASLR_STACK) != 0) {
1244 		/* Randomize within the first page of the stack. */
1245 		arc4rand(&stack_off, sizeof(stack_off), 0);
1246 		stack_top -= rounddown2(stack_off & PAGE_MASK, sizeof(void *));
1247 	}
1248 
1249 	/* Map a shared page */
1250 	obj = sv->sv_shared_page_obj;
1251 	if (obj == NULL) {
1252 		sharedpage_addr = 0;
1253 		goto out;
1254 	}
1255 
1256 	/*
1257 	 * If randomization is disabled then the shared page will
1258 	 * be mapped at address specified in sysentvec.
1259 	 * Otherwise any address above .data section can be selected.
1260 	 * Same logic is used for stack address randomization.
1261 	 * If the address randomization is applied map a guard page
1262 	 * at the top of UVA.
1263 	 */
1264 	vm_object_reference(obj);
1265 	if ((imgp->imgp_flags & IMGP_ASLR_SHARED_PAGE) != 0) {
1266 		sharedpage_addr = round_page((vm_offset_t)p->p_vmspace->vm_daddr +
1267 		    lim_max(curthread, RLIMIT_DATA));
1268 
1269 		error = vm_map_fixed(map, NULL, 0,
1270 		    sv->sv_maxuser - PAGE_SIZE, PAGE_SIZE,
1271 		    VM_PROT_NONE, VM_PROT_NONE, MAP_CREATE_GUARD);
1272 		if (error != KERN_SUCCESS) {
1273 			/*
1274 			 * This is not fatal, so let's just print a warning
1275 			 * and continue.
1276 			 */
1277 			uprintf("%s: Mapping guard page at the top of UVA failed"
1278 			    " mach error %d errno %d",
1279 			    __func__, error, vm_mmap_to_errno(error));
1280 		}
1281 
1282 		error = vm_map_find(map, obj, 0,
1283 		    &sharedpage_addr, sv->sv_shared_page_len,
1284 		    sv->sv_maxuser, VMFS_ANY_SPACE,
1285 		    VM_PROT_READ | VM_PROT_EXECUTE,
1286 		    VM_PROT_READ | VM_PROT_EXECUTE,
1287 		    MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1288 	} else {
1289 		sharedpage_addr = sv->sv_shared_page_base;
1290 		vm_map_fixed(map, obj, 0,
1291 		    sharedpage_addr, sv->sv_shared_page_len,
1292 		    VM_PROT_READ | VM_PROT_EXECUTE,
1293 		    VM_PROT_READ | VM_PROT_EXECUTE,
1294 		    MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1295 	}
1296 	if (error != KERN_SUCCESS) {
1297 		uprintf("%s: mapping shared page at addr: %p"
1298 		    "failed, mach error %d errno %d\n", __func__,
1299 		    (void *)sharedpage_addr, error, vm_mmap_to_errno(error));
1300 		vm_object_deallocate(obj);
1301 		return (vm_mmap_to_errno(error));
1302 	}
1303 out:
1304 	/*
1305 	 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they
1306 	 * are still used to enforce the stack rlimit on the process stack.
1307 	 */
1308 	vmspace->vm_maxsaddr = (char *)stack_addr;
1309 	vmspace->vm_stacktop = stack_top;
1310 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1311 	vmspace->vm_shp_base = sharedpage_addr;
1312 
1313 	return (0);
1314 }
1315 
1316 /*
1317  * Copy out argument and environment strings from the old process address
1318  * space into the temporary string buffer.
1319  */
1320 int
1321 exec_copyin_args(struct image_args *args, const char *fname,
1322     enum uio_seg segflg, char **argv, char **envv)
1323 {
1324 	u_long arg, env;
1325 	int error;
1326 
1327 	bzero(args, sizeof(*args));
1328 	if (argv == NULL)
1329 		return (EFAULT);
1330 
1331 	/*
1332 	 * Allocate demand-paged memory for the file name, argument, and
1333 	 * environment strings.
1334 	 */
1335 	error = exec_alloc_args(args);
1336 	if (error != 0)
1337 		return (error);
1338 
1339 	/*
1340 	 * Copy the file name.
1341 	 */
1342 	error = exec_args_add_fname(args, fname, segflg);
1343 	if (error != 0)
1344 		goto err_exit;
1345 
1346 	/*
1347 	 * extract arguments first
1348 	 */
1349 	for (;;) {
1350 		error = fueword(argv++, &arg);
1351 		if (error == -1) {
1352 			error = EFAULT;
1353 			goto err_exit;
1354 		}
1355 		if (arg == 0)
1356 			break;
1357 		error = exec_args_add_arg(args, (char *)(uintptr_t)arg,
1358 		    UIO_USERSPACE);
1359 		if (error != 0)
1360 			goto err_exit;
1361 	}
1362 
1363 	/*
1364 	 * extract environment strings
1365 	 */
1366 	if (envv) {
1367 		for (;;) {
1368 			error = fueword(envv++, &env);
1369 			if (error == -1) {
1370 				error = EFAULT;
1371 				goto err_exit;
1372 			}
1373 			if (env == 0)
1374 				break;
1375 			error = exec_args_add_env(args,
1376 			    (char *)(uintptr_t)env, UIO_USERSPACE);
1377 			if (error != 0)
1378 				goto err_exit;
1379 		}
1380 	}
1381 
1382 	return (0);
1383 
1384 err_exit:
1385 	exec_free_args(args);
1386 	return (error);
1387 }
1388 
1389 struct exec_args_kva {
1390 	vm_offset_t addr;
1391 	u_int gen;
1392 	SLIST_ENTRY(exec_args_kva) next;
1393 };
1394 
1395 DPCPU_DEFINE_STATIC(struct exec_args_kva *, exec_args_kva);
1396 
1397 static SLIST_HEAD(, exec_args_kva) exec_args_kva_freelist;
1398 static struct mtx exec_args_kva_mtx;
1399 static u_int exec_args_gen;
1400 
1401 static void
1402 exec_prealloc_args_kva(void *arg __unused)
1403 {
1404 	struct exec_args_kva *argkva;
1405 	u_int i;
1406 
1407 	SLIST_INIT(&exec_args_kva_freelist);
1408 	mtx_init(&exec_args_kva_mtx, "exec args kva", NULL, MTX_DEF);
1409 	for (i = 0; i < exec_map_entries; i++) {
1410 		argkva = malloc(sizeof(*argkva), M_PARGS, M_WAITOK);
1411 		argkva->addr = kmap_alloc_wait(exec_map, exec_map_entry_size);
1412 		argkva->gen = exec_args_gen;
1413 		SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1414 	}
1415 }
1416 SYSINIT(exec_args_kva, SI_SUB_EXEC, SI_ORDER_ANY, exec_prealloc_args_kva, NULL);
1417 
1418 static vm_offset_t
1419 exec_alloc_args_kva(void **cookie)
1420 {
1421 	struct exec_args_kva *argkva;
1422 
1423 	argkva = (void *)atomic_readandclear_ptr(
1424 	    (uintptr_t *)DPCPU_PTR(exec_args_kva));
1425 	if (argkva == NULL) {
1426 		mtx_lock(&exec_args_kva_mtx);
1427 		while ((argkva = SLIST_FIRST(&exec_args_kva_freelist)) == NULL)
1428 			(void)mtx_sleep(&exec_args_kva_freelist,
1429 			    &exec_args_kva_mtx, 0, "execkva", 0);
1430 		SLIST_REMOVE_HEAD(&exec_args_kva_freelist, next);
1431 		mtx_unlock(&exec_args_kva_mtx);
1432 	}
1433 	kasan_mark((void *)argkva->addr, exec_map_entry_size,
1434 	    exec_map_entry_size, 0);
1435 	*(struct exec_args_kva **)cookie = argkva;
1436 	return (argkva->addr);
1437 }
1438 
1439 static void
1440 exec_release_args_kva(struct exec_args_kva *argkva, u_int gen)
1441 {
1442 	vm_offset_t base;
1443 
1444 	base = argkva->addr;
1445 	kasan_mark((void *)argkva->addr, 0, exec_map_entry_size,
1446 	    KASAN_EXEC_ARGS_FREED);
1447 	if (argkva->gen != gen) {
1448 		(void)vm_map_madvise(exec_map, base, base + exec_map_entry_size,
1449 		    MADV_FREE);
1450 		argkva->gen = gen;
1451 	}
1452 	if (!atomic_cmpset_ptr((uintptr_t *)DPCPU_PTR(exec_args_kva),
1453 	    (uintptr_t)NULL, (uintptr_t)argkva)) {
1454 		mtx_lock(&exec_args_kva_mtx);
1455 		SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1456 		wakeup_one(&exec_args_kva_freelist);
1457 		mtx_unlock(&exec_args_kva_mtx);
1458 	}
1459 }
1460 
1461 static void
1462 exec_free_args_kva(void *cookie)
1463 {
1464 
1465 	exec_release_args_kva(cookie, exec_args_gen);
1466 }
1467 
1468 static void
1469 exec_args_kva_lowmem(void *arg __unused)
1470 {
1471 	SLIST_HEAD(, exec_args_kva) head;
1472 	struct exec_args_kva *argkva;
1473 	u_int gen;
1474 	int i;
1475 
1476 	gen = atomic_fetchadd_int(&exec_args_gen, 1) + 1;
1477 
1478 	/*
1479 	 * Force an madvise of each KVA range. Any currently allocated ranges
1480 	 * will have MADV_FREE applied once they are freed.
1481 	 */
1482 	SLIST_INIT(&head);
1483 	mtx_lock(&exec_args_kva_mtx);
1484 	SLIST_SWAP(&head, &exec_args_kva_freelist, exec_args_kva);
1485 	mtx_unlock(&exec_args_kva_mtx);
1486 	while ((argkva = SLIST_FIRST(&head)) != NULL) {
1487 		SLIST_REMOVE_HEAD(&head, next);
1488 		exec_release_args_kva(argkva, gen);
1489 	}
1490 
1491 	CPU_FOREACH(i) {
1492 		argkva = (void *)atomic_readandclear_ptr(
1493 		    (uintptr_t *)DPCPU_ID_PTR(i, exec_args_kva));
1494 		if (argkva != NULL)
1495 			exec_release_args_kva(argkva, gen);
1496 	}
1497 }
1498 EVENTHANDLER_DEFINE(vm_lowmem, exec_args_kva_lowmem, NULL,
1499     EVENTHANDLER_PRI_ANY);
1500 
1501 /*
1502  * Allocate temporary demand-paged, zero-filled memory for the file name,
1503  * argument, and environment strings.
1504  */
1505 int
1506 exec_alloc_args(struct image_args *args)
1507 {
1508 
1509 	args->buf = (char *)exec_alloc_args_kva(&args->bufkva);
1510 	return (0);
1511 }
1512 
1513 void
1514 exec_free_args(struct image_args *args)
1515 {
1516 
1517 	if (args->buf != NULL) {
1518 		exec_free_args_kva(args->bufkva);
1519 		args->buf = NULL;
1520 	}
1521 	if (args->fname_buf != NULL) {
1522 		free(args->fname_buf, M_TEMP);
1523 		args->fname_buf = NULL;
1524 	}
1525 }
1526 
1527 /*
1528  * A set to functions to fill struct image args.
1529  *
1530  * NOTE: exec_args_add_fname() must be called (possibly with a NULL
1531  * fname) before the other functions.  All exec_args_add_arg() calls must
1532  * be made before any exec_args_add_env() calls.  exec_args_adjust_args()
1533  * may be called any time after exec_args_add_fname().
1534  *
1535  * exec_args_add_fname() - install path to be executed
1536  * exec_args_add_arg() - append an argument string
1537  * exec_args_add_env() - append an env string
1538  * exec_args_adjust_args() - adjust location of the argument list to
1539  *                           allow new arguments to be prepended
1540  */
1541 int
1542 exec_args_add_fname(struct image_args *args, const char *fname,
1543     enum uio_seg segflg)
1544 {
1545 	int error;
1546 	size_t length;
1547 
1548 	KASSERT(args->fname == NULL, ("fname already appended"));
1549 	KASSERT(args->endp == NULL, ("already appending to args"));
1550 
1551 	if (fname != NULL) {
1552 		args->fname = args->buf;
1553 		error = segflg == UIO_SYSSPACE ?
1554 		    copystr(fname, args->fname, PATH_MAX, &length) :
1555 		    copyinstr(fname, args->fname, PATH_MAX, &length);
1556 		if (error != 0)
1557 			return (error == ENAMETOOLONG ? E2BIG : error);
1558 	} else
1559 		length = 0;
1560 
1561 	/* Set up for _arg_*()/_env_*() */
1562 	args->endp = args->buf + length;
1563 	/* begin_argv must be set and kept updated */
1564 	args->begin_argv = args->endp;
1565 	KASSERT(exec_map_entry_size - length >= ARG_MAX,
1566 	    ("too little space remaining for arguments %zu < %zu",
1567 	    exec_map_entry_size - length, (size_t)ARG_MAX));
1568 	args->stringspace = ARG_MAX;
1569 
1570 	return (0);
1571 }
1572 
1573 static int
1574 exec_args_add_str(struct image_args *args, const char *str,
1575     enum uio_seg segflg, int *countp)
1576 {
1577 	int error;
1578 	size_t length;
1579 
1580 	KASSERT(args->endp != NULL, ("endp not initialized"));
1581 	KASSERT(args->begin_argv != NULL, ("begin_argp not initialized"));
1582 
1583 	error = (segflg == UIO_SYSSPACE) ?
1584 	    copystr(str, args->endp, args->stringspace, &length) :
1585 	    copyinstr(str, args->endp, args->stringspace, &length);
1586 	if (error != 0)
1587 		return (error == ENAMETOOLONG ? E2BIG : error);
1588 	args->stringspace -= length;
1589 	args->endp += length;
1590 	(*countp)++;
1591 
1592 	return (0);
1593 }
1594 
1595 int
1596 exec_args_add_arg(struct image_args *args, const char *argp,
1597     enum uio_seg segflg)
1598 {
1599 
1600 	KASSERT(args->envc == 0, ("appending args after env"));
1601 
1602 	return (exec_args_add_str(args, argp, segflg, &args->argc));
1603 }
1604 
1605 int
1606 exec_args_add_env(struct image_args *args, const char *envp,
1607     enum uio_seg segflg)
1608 {
1609 
1610 	if (args->envc == 0)
1611 		args->begin_envv = args->endp;
1612 
1613 	return (exec_args_add_str(args, envp, segflg, &args->envc));
1614 }
1615 
1616 int
1617 exec_args_adjust_args(struct image_args *args, size_t consume, ssize_t extend)
1618 {
1619 	ssize_t offset;
1620 
1621 	KASSERT(args->endp != NULL, ("endp not initialized"));
1622 	KASSERT(args->begin_argv != NULL, ("begin_argp not initialized"));
1623 
1624 	offset = extend - consume;
1625 	if (args->stringspace < offset)
1626 		return (E2BIG);
1627 	memmove(args->begin_argv + extend, args->begin_argv + consume,
1628 	    args->endp - args->begin_argv + consume);
1629 	if (args->envc > 0)
1630 		args->begin_envv += offset;
1631 	args->endp += offset;
1632 	args->stringspace -= offset;
1633 	return (0);
1634 }
1635 
1636 char *
1637 exec_args_get_begin_envv(struct image_args *args)
1638 {
1639 
1640 	KASSERT(args->endp != NULL, ("endp not initialized"));
1641 
1642 	if (args->envc > 0)
1643 		return (args->begin_envv);
1644 	return (args->endp);
1645 }
1646 
1647 /*
1648  * Copy strings out to the new process address space, constructing new arg
1649  * and env vector tables. Return a pointer to the base so that it can be used
1650  * as the initial stack pointer.
1651  */
1652 int
1653 exec_copyout_strings(struct image_params *imgp, uintptr_t *stack_base)
1654 {
1655 	int argc, envc;
1656 	char **vectp;
1657 	char *stringp;
1658 	uintptr_t destp, ustringp;
1659 	struct ps_strings *arginfo;
1660 	struct proc *p;
1661 	struct sysentvec *sysent;
1662 	size_t execpath_len;
1663 	int error, szsigcode;
1664 	char canary[sizeof(long) * 8];
1665 
1666 	p = imgp->proc;
1667 	sysent = p->p_sysent;
1668 
1669 	destp =	PROC_PS_STRINGS(p);
1670 	arginfo = imgp->ps_strings = (void *)destp;
1671 
1672 	/*
1673 	 * Install sigcode.
1674 	 */
1675 	if (sysent->sv_shared_page_base == 0 && sysent->sv_szsigcode != NULL) {
1676 		szsigcode = *(sysent->sv_szsigcode);
1677 		destp -= szsigcode;
1678 		destp = rounddown2(destp, sizeof(void *));
1679 		error = copyout(sysent->sv_sigcode, (void *)destp, szsigcode);
1680 		if (error != 0)
1681 			return (error);
1682 	}
1683 
1684 	/*
1685 	 * Copy the image path for the rtld.
1686 	 */
1687 	if (imgp->execpath != NULL && imgp->auxargs != NULL) {
1688 		execpath_len = strlen(imgp->execpath) + 1;
1689 		destp -= execpath_len;
1690 		destp = rounddown2(destp, sizeof(void *));
1691 		imgp->execpathp = (void *)destp;
1692 		error = copyout(imgp->execpath, imgp->execpathp, execpath_len);
1693 		if (error != 0)
1694 			return (error);
1695 	}
1696 
1697 	/*
1698 	 * Prepare the canary for SSP.
1699 	 */
1700 	arc4rand(canary, sizeof(canary), 0);
1701 	destp -= sizeof(canary);
1702 	imgp->canary = (void *)destp;
1703 	error = copyout(canary, imgp->canary, sizeof(canary));
1704 	if (error != 0)
1705 		return (error);
1706 	imgp->canarylen = sizeof(canary);
1707 
1708 	/*
1709 	 * Prepare the pagesizes array.
1710 	 */
1711 	imgp->pagesizeslen = sizeof(pagesizes[0]) * MAXPAGESIZES;
1712 	destp -= imgp->pagesizeslen;
1713 	destp = rounddown2(destp, sizeof(void *));
1714 	imgp->pagesizes = (void *)destp;
1715 	error = copyout(pagesizes, imgp->pagesizes, imgp->pagesizeslen);
1716 	if (error != 0)
1717 		return (error);
1718 
1719 	/*
1720 	 * Allocate room for the argument and environment strings.
1721 	 */
1722 	destp -= ARG_MAX - imgp->args->stringspace;
1723 	destp = rounddown2(destp, sizeof(void *));
1724 	ustringp = destp;
1725 
1726 	if (imgp->auxargs) {
1727 		/*
1728 		 * Allocate room on the stack for the ELF auxargs
1729 		 * array.  It has up to AT_COUNT entries.
1730 		 */
1731 		destp -= AT_COUNT * sizeof(Elf_Auxinfo);
1732 		destp = rounddown2(destp, sizeof(void *));
1733 	}
1734 
1735 	vectp = (char **)destp;
1736 
1737 	/*
1738 	 * Allocate room for the argv[] and env vectors including the
1739 	 * terminating NULL pointers.
1740 	 */
1741 	vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
1742 
1743 	/*
1744 	 * vectp also becomes our initial stack base
1745 	 */
1746 	*stack_base = (uintptr_t)vectp;
1747 
1748 	stringp = imgp->args->begin_argv;
1749 	argc = imgp->args->argc;
1750 	envc = imgp->args->envc;
1751 
1752 	/*
1753 	 * Copy out strings - arguments and environment.
1754 	 */
1755 	error = copyout(stringp, (void *)ustringp,
1756 	    ARG_MAX - imgp->args->stringspace);
1757 	if (error != 0)
1758 		return (error);
1759 
1760 	/*
1761 	 * Fill in "ps_strings" struct for ps, w, etc.
1762 	 */
1763 	imgp->argv = vectp;
1764 	if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 ||
1765 	    suword32(&arginfo->ps_nargvstr, argc) != 0)
1766 		return (EFAULT);
1767 
1768 	/*
1769 	 * Fill in argument portion of vector table.
1770 	 */
1771 	for (; argc > 0; --argc) {
1772 		if (suword(vectp++, ustringp) != 0)
1773 			return (EFAULT);
1774 		while (*stringp++ != 0)
1775 			ustringp++;
1776 		ustringp++;
1777 	}
1778 
1779 	/* a null vector table pointer separates the argp's from the envp's */
1780 	if (suword(vectp++, 0) != 0)
1781 		return (EFAULT);
1782 
1783 	imgp->envv = vectp;
1784 	if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 ||
1785 	    suword32(&arginfo->ps_nenvstr, envc) != 0)
1786 		return (EFAULT);
1787 
1788 	/*
1789 	 * Fill in environment portion of vector table.
1790 	 */
1791 	for (; envc > 0; --envc) {
1792 		if (suword(vectp++, ustringp) != 0)
1793 			return (EFAULT);
1794 		while (*stringp++ != 0)
1795 			ustringp++;
1796 		ustringp++;
1797 	}
1798 
1799 	/* end of vector table is a null pointer */
1800 	if (suword(vectp, 0) != 0)
1801 		return (EFAULT);
1802 
1803 	if (imgp->auxargs) {
1804 		vectp++;
1805 		error = imgp->sysent->sv_copyout_auxargs(imgp,
1806 		    (uintptr_t)vectp);
1807 		if (error != 0)
1808 			return (error);
1809 	}
1810 
1811 	return (0);
1812 }
1813 
1814 /*
1815  * Check permissions of file to execute.
1816  *	Called with imgp->vp locked.
1817  *	Return 0 for success or error code on failure.
1818  */
1819 int
1820 exec_check_permissions(struct image_params *imgp)
1821 {
1822 	struct vnode *vp = imgp->vp;
1823 	struct vattr *attr = imgp->attr;
1824 	struct thread *td;
1825 	int error;
1826 
1827 	td = curthread;
1828 
1829 	/* Get file attributes */
1830 	error = VOP_GETATTR(vp, attr, td->td_ucred);
1831 	if (error)
1832 		return (error);
1833 
1834 #ifdef MAC
1835 	error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1836 	if (error)
1837 		return (error);
1838 #endif
1839 
1840 	/*
1841 	 * 1) Check if file execution is disabled for the filesystem that
1842 	 *    this file resides on.
1843 	 * 2) Ensure that at least one execute bit is on. Otherwise, a
1844 	 *    privileged user will always succeed, and we don't want this
1845 	 *    to happen unless the file really is executable.
1846 	 * 3) Ensure that the file is a regular file.
1847 	 */
1848 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1849 	    (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1850 	    (attr->va_type != VREG))
1851 		return (EACCES);
1852 
1853 	/*
1854 	 * Zero length files can't be exec'd
1855 	 */
1856 	if (attr->va_size == 0)
1857 		return (ENOEXEC);
1858 
1859 	/*
1860 	 *  Check for execute permission to file based on current credentials.
1861 	 */
1862 	error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1863 	if (error)
1864 		return (error);
1865 
1866 	/*
1867 	 * Check number of open-for-writes on the file and deny execution
1868 	 * if there are any.
1869 	 *
1870 	 * Add a text reference now so no one can write to the
1871 	 * executable while we're activating it.
1872 	 *
1873 	 * Remember if this was set before and unset it in case this is not
1874 	 * actually an executable image.
1875 	 */
1876 	error = VOP_SET_TEXT(vp);
1877 	if (error != 0)
1878 		return (error);
1879 	imgp->textset = true;
1880 
1881 	/*
1882 	 * Call filesystem specific open routine (which does nothing in the
1883 	 * general case).
1884 	 */
1885 	error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1886 	if (error == 0)
1887 		imgp->opened = true;
1888 	return (error);
1889 }
1890 
1891 /*
1892  * Exec handler registration
1893  */
1894 int
1895 exec_register(const struct execsw *execsw_arg)
1896 {
1897 	const struct execsw **es, **xs, **newexecsw;
1898 	u_int count = 2;	/* New slot and trailing NULL */
1899 
1900 	if (execsw)
1901 		for (es = execsw; *es; es++)
1902 			count++;
1903 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1904 	xs = newexecsw;
1905 	if (execsw)
1906 		for (es = execsw; *es; es++)
1907 			*xs++ = *es;
1908 	*xs++ = execsw_arg;
1909 	*xs = NULL;
1910 	if (execsw)
1911 		free(execsw, M_TEMP);
1912 	execsw = newexecsw;
1913 	return (0);
1914 }
1915 
1916 int
1917 exec_unregister(const struct execsw *execsw_arg)
1918 {
1919 	const struct execsw **es, **xs, **newexecsw;
1920 	int count = 1;
1921 
1922 	if (execsw == NULL)
1923 		panic("unregister with no handlers left?\n");
1924 
1925 	for (es = execsw; *es; es++) {
1926 		if (*es == execsw_arg)
1927 			break;
1928 	}
1929 	if (*es == NULL)
1930 		return (ENOENT);
1931 	for (es = execsw; *es; es++)
1932 		if (*es != execsw_arg)
1933 			count++;
1934 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1935 	xs = newexecsw;
1936 	for (es = execsw; *es; es++)
1937 		if (*es != execsw_arg)
1938 			*xs++ = *es;
1939 	*xs = NULL;
1940 	if (execsw)
1941 		free(execsw, M_TEMP);
1942 	execsw = newexecsw;
1943 	return (0);
1944 }
1945 
1946 /*
1947  * Write out a core segment to the compression stream.
1948  */
1949 static int
1950 compress_chunk(struct coredump_params *cp, char *base, char *buf, size_t len)
1951 {
1952 	size_t chunk_len;
1953 	int error;
1954 
1955 	while (len > 0) {
1956 		chunk_len = MIN(len, CORE_BUF_SIZE);
1957 
1958 		/*
1959 		 * We can get EFAULT error here.
1960 		 * In that case zero out the current chunk of the segment.
1961 		 */
1962 		error = copyin(base, buf, chunk_len);
1963 		if (error != 0)
1964 			bzero(buf, chunk_len);
1965 		error = compressor_write(cp->comp, buf, chunk_len);
1966 		if (error != 0)
1967 			break;
1968 		base += chunk_len;
1969 		len -= chunk_len;
1970 	}
1971 	return (error);
1972 }
1973 
1974 int
1975 core_write(struct coredump_params *cp, const void *base, size_t len,
1976     off_t offset, enum uio_seg seg, size_t *resid)
1977 {
1978 
1979 	return (vn_rdwr_inchunks(UIO_WRITE, cp->vp, __DECONST(void *, base),
1980 	    len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1981 	    cp->active_cred, cp->file_cred, resid, cp->td));
1982 }
1983 
1984 int
1985 core_output(char *base, size_t len, off_t offset, struct coredump_params *cp,
1986     void *tmpbuf)
1987 {
1988 	vm_map_t map;
1989 	struct mount *mp;
1990 	size_t resid, runlen;
1991 	int error;
1992 	bool success;
1993 
1994 	KASSERT((uintptr_t)base % PAGE_SIZE == 0,
1995 	    ("%s: user address %p is not page-aligned", __func__, base));
1996 
1997 	if (cp->comp != NULL)
1998 		return (compress_chunk(cp, base, tmpbuf, len));
1999 
2000 	map = &cp->td->td_proc->p_vmspace->vm_map;
2001 	for (; len > 0; base += runlen, offset += runlen, len -= runlen) {
2002 		/*
2003 		 * Attempt to page in all virtual pages in the range.  If a
2004 		 * virtual page is not backed by the pager, it is represented as
2005 		 * a hole in the file.  This can occur with zero-filled
2006 		 * anonymous memory or truncated files, for example.
2007 		 */
2008 		for (runlen = 0; runlen < len; runlen += PAGE_SIZE) {
2009 			if (core_dump_can_intr && curproc_sigkilled())
2010 				return (EINTR);
2011 			error = vm_fault(map, (uintptr_t)base + runlen,
2012 			    VM_PROT_READ, VM_FAULT_NOFILL, NULL);
2013 			if (runlen == 0)
2014 				success = error == KERN_SUCCESS;
2015 			else if ((error == KERN_SUCCESS) != success)
2016 				break;
2017 		}
2018 
2019 		if (success) {
2020 			error = core_write(cp, base, runlen, offset,
2021 			    UIO_USERSPACE, &resid);
2022 			if (error != 0) {
2023 				if (error != EFAULT)
2024 					break;
2025 
2026 				/*
2027 				 * EFAULT may be returned if the user mapping
2028 				 * could not be accessed, e.g., because a mapped
2029 				 * file has been truncated.  Skip the page if no
2030 				 * progress was made, to protect against a
2031 				 * hypothetical scenario where vm_fault() was
2032 				 * successful but core_write() returns EFAULT
2033 				 * anyway.
2034 				 */
2035 				runlen -= resid;
2036 				if (runlen == 0) {
2037 					success = false;
2038 					runlen = PAGE_SIZE;
2039 				}
2040 			}
2041 		}
2042 		if (!success) {
2043 			error = vn_start_write(cp->vp, &mp, V_WAIT);
2044 			if (error != 0)
2045 				break;
2046 			vn_lock(cp->vp, LK_EXCLUSIVE | LK_RETRY);
2047 			error = vn_truncate_locked(cp->vp, offset + runlen,
2048 			    false, cp->td->td_ucred);
2049 			VOP_UNLOCK(cp->vp);
2050 			vn_finished_write(mp);
2051 			if (error != 0)
2052 				break;
2053 		}
2054 	}
2055 	return (error);
2056 }
2057 
2058 /*
2059  * Drain into a core file.
2060  */
2061 int
2062 sbuf_drain_core_output(void *arg, const char *data, int len)
2063 {
2064 	struct coredump_params *cp;
2065 	struct proc *p;
2066 	int error, locked;
2067 
2068 	cp = arg;
2069 	p = cp->td->td_proc;
2070 
2071 	/*
2072 	 * Some kern_proc out routines that print to this sbuf may
2073 	 * call us with the process lock held. Draining with the
2074 	 * non-sleepable lock held is unsafe. The lock is needed for
2075 	 * those routines when dumping a live process. In our case we
2076 	 * can safely release the lock before draining and acquire
2077 	 * again after.
2078 	 */
2079 	locked = PROC_LOCKED(p);
2080 	if (locked)
2081 		PROC_UNLOCK(p);
2082 	if (cp->comp != NULL)
2083 		error = compressor_write(cp->comp, __DECONST(char *, data),
2084 		    len);
2085 	else
2086 		error = core_write(cp, __DECONST(void *, data), len, cp->offset,
2087 		    UIO_SYSSPACE, NULL);
2088 	if (locked)
2089 		PROC_LOCK(p);
2090 	if (error != 0)
2091 		return (-error);
2092 	cp->offset += len;
2093 	return (len);
2094 }
2095