xref: /freebsd/sys/kern/kern_exec.c (revision fa38579f317d5c2ff2926fab9b12ee6d429bd155)
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, uap->argv, uap->envv);
229 	if (error == 0)
230 		error = kern_execve(td, &args, NULL, oldvmspace);
231 	post_execve(td, error, oldvmspace);
232 	AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
233 	return (error);
234 }
235 
236 #ifndef _SYS_SYSPROTO_H_
237 struct fexecve_args {
238 	int	fd;
239 	char	**argv;
240 	char	**envv;
241 };
242 #endif
243 int
244 sys_fexecve(struct thread *td, struct fexecve_args *uap)
245 {
246 	struct image_args args;
247 	struct vmspace *oldvmspace;
248 	int error;
249 
250 	error = pre_execve(td, &oldvmspace);
251 	if (error != 0)
252 		return (error);
253 	error = exec_copyin_args(&args, NULL, uap->argv, uap->envv);
254 	if (error == 0) {
255 		args.fd = uap->fd;
256 		error = kern_execve(td, &args, NULL, oldvmspace);
257 	}
258 	post_execve(td, error, oldvmspace);
259 	AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
260 	return (error);
261 }
262 
263 #ifndef _SYS_SYSPROTO_H_
264 struct __mac_execve_args {
265 	char	*fname;
266 	char	**argv;
267 	char	**envv;
268 	struct mac	*mac_p;
269 };
270 #endif
271 
272 int
273 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap)
274 {
275 #ifdef MAC
276 	struct image_args args;
277 	struct vmspace *oldvmspace;
278 	int error;
279 
280 	error = pre_execve(td, &oldvmspace);
281 	if (error != 0)
282 		return (error);
283 	error = exec_copyin_args(&args, uap->fname, uap->argv, uap->envv);
284 	if (error == 0)
285 		error = kern_execve(td, &args, uap->mac_p, oldvmspace);
286 	post_execve(td, error, oldvmspace);
287 	AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
288 	return (error);
289 #else
290 	return (ENOSYS);
291 #endif
292 }
293 
294 int
295 pre_execve(struct thread *td, struct vmspace **oldvmspace)
296 {
297 	struct proc *p;
298 	int error;
299 
300 	KASSERT(td == curthread, ("non-current thread %p", td));
301 	error = 0;
302 	p = td->td_proc;
303 	if ((p->p_flag & P_HADTHREADS) != 0) {
304 		PROC_LOCK(p);
305 		if (thread_single(p, SINGLE_BOUNDARY) != 0)
306 			error = ERESTART;
307 		PROC_UNLOCK(p);
308 	}
309 	KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0,
310 	    ("nested execve"));
311 	*oldvmspace = p->p_vmspace;
312 	return (error);
313 }
314 
315 void
316 post_execve(struct thread *td, int error, struct vmspace *oldvmspace)
317 {
318 	struct proc *p;
319 
320 	KASSERT(td == curthread, ("non-current thread %p", td));
321 	p = td->td_proc;
322 	if ((p->p_flag & P_HADTHREADS) != 0) {
323 		PROC_LOCK(p);
324 		/*
325 		 * If success, we upgrade to SINGLE_EXIT state to
326 		 * force other threads to suicide.
327 		 */
328 		if (error == EJUSTRETURN)
329 			thread_single(p, SINGLE_EXIT);
330 		else
331 			thread_single_end(p, SINGLE_BOUNDARY);
332 		PROC_UNLOCK(p);
333 	}
334 	exec_cleanup(td, oldvmspace);
335 }
336 
337 /*
338  * kern_execve() has the astonishing property of not always returning to
339  * the caller.  If sufficiently bad things happen during the call to
340  * do_execve(), it can end up calling exit1(); as a result, callers must
341  * avoid doing anything which they might need to undo (e.g., allocating
342  * memory).
343  */
344 int
345 kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p,
346     struct vmspace *oldvmspace)
347 {
348 
349 	TSEXEC(td->td_proc->p_pid, args->begin_argv);
350 	AUDIT_ARG_ARGV(args->begin_argv, args->argc,
351 	    exec_args_get_begin_envv(args) - args->begin_argv);
352 	AUDIT_ARG_ENVV(exec_args_get_begin_envv(args), args->envc,
353 	    args->endp - exec_args_get_begin_envv(args));
354 #ifdef KTRACE
355 	if (KTRPOINT(td, KTR_ARGS)) {
356 		ktrdata(KTR_ARGS, args->begin_argv,
357 		    exec_args_get_begin_envv(args) - args->begin_argv);
358         }
359 	if (KTRPOINT(td, KTR_ENVS)) {
360 		ktrdata(KTR_ENVS, exec_args_get_begin_envv(args),
361 		    args->endp - exec_args_get_begin_envv(args));
362         }
363 #endif
364 	/* Must have at least one argument. */
365 	if (args->argc == 0) {
366 		exec_free_args(args);
367 		return (EINVAL);
368 	}
369 	return (do_execve(td, args, mac_p, oldvmspace));
370 }
371 
372 static void
373 execve_nosetid(struct image_params *imgp)
374 {
375 	imgp->credential_setid = false;
376 	if (imgp->newcred != NULL) {
377 		crfree(imgp->newcred);
378 		imgp->newcred = NULL;
379 	}
380 }
381 
382 /*
383  * In-kernel implementation of execve().  All arguments are assumed to be
384  * userspace pointers from the passed thread.
385  */
386 static int
387 do_execve(struct thread *td, struct image_args *args, struct mac *mac_p,
388     struct vmspace *oldvmspace)
389 {
390 	struct proc *p = td->td_proc;
391 	struct nameidata nd;
392 	struct ucred *oldcred;
393 	struct uidinfo *euip = NULL;
394 	uintptr_t stack_base;
395 	struct image_params image_params, *imgp;
396 	struct vattr attr;
397 	struct pargs *oldargs = NULL, *newargs = NULL;
398 	struct sigacts *oldsigacts = NULL, *newsigacts = NULL;
399 #ifdef KTRACE
400 	struct ktr_io_params *kiop;
401 #endif
402 	struct vnode *oldtextvp, *newtextvp;
403 	struct vnode *oldtextdvp, *newtextdvp;
404 	char *oldbinname, *newbinname;
405 	bool credential_changing;
406 #ifdef MAC
407 	struct label *interpvplabel = NULL;
408 	bool will_transition;
409 #endif
410 #ifdef HWPMC_HOOKS
411 	struct pmckern_procexec pe;
412 #endif
413 	int error, i, orig_osrel;
414 	uint32_t orig_fctl0;
415 	Elf_Brandinfo *orig_brandinfo;
416 	size_t freepath_size;
417 	static const char fexecv_proc_title[] = "(fexecv)";
418 
419 	imgp = &image_params;
420 	oldtextvp = oldtextdvp = NULL;
421 	newtextvp = newtextdvp = NULL;
422 	newbinname = oldbinname = NULL;
423 #ifdef KTRACE
424 	kiop = NULL;
425 #endif
426 
427 	/*
428 	 * Lock the process and set the P_INEXEC flag to indicate that
429 	 * it should be left alone until we're done here.  This is
430 	 * necessary to avoid race conditions - e.g. in ptrace() -
431 	 * that might allow a local user to illicitly obtain elevated
432 	 * privileges.
433 	 */
434 	PROC_LOCK(p);
435 	KASSERT((p->p_flag & P_INEXEC) == 0,
436 	    ("%s(): process already has P_INEXEC flag", __func__));
437 	p->p_flag |= P_INEXEC;
438 	PROC_UNLOCK(p);
439 
440 	/*
441 	 * Initialize part of the common data
442 	 */
443 	bzero(imgp, sizeof(*imgp));
444 	imgp->proc = p;
445 	imgp->attr = &attr;
446 	imgp->args = args;
447 	oldcred = p->p_ucred;
448 	orig_osrel = p->p_osrel;
449 	orig_fctl0 = p->p_fctl0;
450 	orig_brandinfo = p->p_elf_brandinfo;
451 
452 #ifdef MAC
453 	error = mac_execve_enter(imgp, mac_p);
454 	if (error)
455 		goto exec_fail;
456 #endif
457 
458 	SDT_PROBE1(proc, , , exec, args->fname);
459 
460 interpret:
461 	if (args->fname != NULL) {
462 #ifdef CAPABILITY_MODE
463 		if (CAP_TRACING(td))
464 			ktrcapfail(CAPFAIL_NAMEI, args->fname);
465 		/*
466 		 * While capability mode can't reach this point via direct
467 		 * path arguments to execve(), we also don't allow
468 		 * interpreters to be used in capability mode (for now).
469 		 * Catch indirect lookups and return a permissions error.
470 		 */
471 		if (IN_CAPABILITY_MODE(td)) {
472 			error = ECAPMODE;
473 			goto exec_fail;
474 		}
475 #endif
476 
477 		/*
478 		 * Translate the file name. namei() returns a vnode
479 		 * pointer in ni_vp among other things.
480 		 */
481 		NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW |
482 		    AUDITVNODE1 | WANTPARENT, UIO_SYSSPACE,
483 		    args->fname);
484 
485 		error = namei(&nd);
486 		if (error)
487 			goto exec_fail;
488 
489 		newtextvp = nd.ni_vp;
490 		newtextdvp = nd.ni_dvp;
491 		nd.ni_dvp = NULL;
492 		newbinname = malloc(nd.ni_cnd.cn_namelen + 1, M_PARGS,
493 		    M_WAITOK);
494 		memcpy(newbinname, nd.ni_cnd.cn_nameptr, nd.ni_cnd.cn_namelen);
495 		newbinname[nd.ni_cnd.cn_namelen] = '\0';
496 		imgp->vp = newtextvp;
497 
498 		/*
499 		 * Do the best to calculate the full path to the image file.
500 		 */
501 		if (args->fname[0] == '/') {
502 			imgp->execpath = args->fname;
503 		} else {
504 			VOP_UNLOCK(imgp->vp);
505 			freepath_size = MAXPATHLEN;
506 			if (vn_fullpath_hardlink(newtextvp, newtextdvp,
507 			    newbinname, nd.ni_cnd.cn_namelen, &imgp->execpath,
508 			    &imgp->freepath, &freepath_size) != 0)
509 				imgp->execpath = args->fname;
510 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
511 		}
512 	} else if (imgp->interpreter_vp) {
513 		/*
514 		 * An image activator has already provided an open vnode
515 		 */
516 		newtextvp = imgp->interpreter_vp;
517 		imgp->interpreter_vp = NULL;
518 		if (vn_fullpath(newtextvp, &imgp->execpath,
519 		    &imgp->freepath) != 0)
520 			imgp->execpath = args->fname;
521 		vn_lock(newtextvp, LK_SHARED | LK_RETRY);
522 		AUDIT_ARG_VNODE1(newtextvp);
523 		imgp->vp = newtextvp;
524 	} else {
525 		AUDIT_ARG_FD(args->fd);
526 
527 		/*
528 		 * If the descriptors was not opened with O_PATH, then
529 		 * we require that it was opened with O_EXEC or
530 		 * O_RDONLY.  In either case, exec_check_permissions()
531 		 * below checks _current_ file access mode regardless
532 		 * of the permissions additionally checked at the
533 		 * open(2).
534 		 */
535 		error = fgetvp_exec(td, args->fd, &cap_fexecve_rights,
536 		    &newtextvp);
537 		if (error != 0)
538 			goto exec_fail;
539 
540 		if (vn_fullpath(newtextvp, &imgp->execpath,
541 		    &imgp->freepath) != 0)
542 			imgp->execpath = args->fname;
543 		vn_lock(newtextvp, LK_SHARED | LK_RETRY);
544 		AUDIT_ARG_VNODE1(newtextvp);
545 		imgp->vp = newtextvp;
546 	}
547 
548 	/*
549 	 * Check file permissions.  Also 'opens' file and sets its vnode to
550 	 * text mode.
551 	 */
552 	error = exec_check_permissions(imgp);
553 	if (error)
554 		goto exec_fail_dealloc;
555 
556 	imgp->object = imgp->vp->v_object;
557 	if (imgp->object != NULL)
558 		vm_object_reference(imgp->object);
559 
560 	error = exec_map_first_page(imgp);
561 	if (error)
562 		goto exec_fail_dealloc;
563 
564 	imgp->proc->p_osrel = 0;
565 	imgp->proc->p_fctl0 = 0;
566 	imgp->proc->p_elf_brandinfo = NULL;
567 
568 	/*
569 	 * Implement image setuid/setgid.
570 	 *
571 	 * Determine new credentials before attempting image activators
572 	 * so that it can be used by process_exec handlers to determine
573 	 * credential/setid changes.
574 	 *
575 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
576 	 * the process is being traced.
577 	 *
578 	 * We disable setuid/setgid/etc in capability mode on the basis
579 	 * that most setugid applications are not written with that
580 	 * environment in mind, and will therefore almost certainly operate
581 	 * incorrectly. In principle there's no reason that setugid
582 	 * applications might not be useful in capability mode, so we may want
583 	 * to reconsider this conservative design choice in the future.
584 	 *
585 	 * XXXMAC: For the time being, use NOSUID to also prohibit
586 	 * transitions on the file system.
587 	 */
588 	credential_changing = false;
589 	credential_changing |= (attr.va_mode & S_ISUID) &&
590 	    oldcred->cr_uid != attr.va_uid;
591 	credential_changing |= (attr.va_mode & S_ISGID) &&
592 	    oldcred->cr_gid != attr.va_gid;
593 #ifdef MAC
594 	will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
595 	    interpvplabel, imgp) != 0;
596 	credential_changing |= will_transition;
597 #endif
598 
599 	/* Don't inherit PROC_PDEATHSIG_CTL value if setuid/setgid. */
600 	if (credential_changing)
601 		imgp->proc->p_pdeathsig = 0;
602 
603 	if (credential_changing &&
604 #ifdef CAPABILITY_MODE
605 	    ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
606 #endif
607 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
608 	    (p->p_flag & P_TRACED) == 0) {
609 		imgp->credential_setid = true;
610 		VOP_UNLOCK(imgp->vp);
611 		imgp->newcred = crdup(oldcred);
612 		if (attr.va_mode & S_ISUID) {
613 			euip = uifind(attr.va_uid);
614 			change_euid(imgp->newcred, euip);
615 		}
616 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
617 		if (attr.va_mode & S_ISGID)
618 			change_egid(imgp->newcred, attr.va_gid);
619 		/*
620 		 * Implement correct POSIX saved-id behavior.
621 		 *
622 		 * XXXMAC: Note that the current logic will save the
623 		 * uid and gid if a MAC domain transition occurs, even
624 		 * though maybe it shouldn't.
625 		 */
626 		change_svuid(imgp->newcred, imgp->newcred->cr_uid);
627 		change_svgid(imgp->newcred, imgp->newcred->cr_gid);
628 	} else {
629 		/*
630 		 * Implement correct POSIX saved-id behavior.
631 		 *
632 		 * XXX: It's not clear that the existing behavior is
633 		 * POSIX-compliant.  A number of sources indicate that the
634 		 * saved uid/gid should only be updated if the new ruid is
635 		 * not equal to the old ruid, or the new euid is not equal
636 		 * to the old euid and the new euid is not equal to the old
637 		 * ruid.  The FreeBSD code always updates the saved uid/gid.
638 		 * Also, this code uses the new (replaced) euid and egid as
639 		 * the source, which may or may not be the right ones to use.
640 		 */
641 		if (oldcred->cr_svuid != oldcred->cr_uid ||
642 		    oldcred->cr_svgid != oldcred->cr_gid) {
643 			VOP_UNLOCK(imgp->vp);
644 			imgp->newcred = crdup(oldcred);
645 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
646 			change_svuid(imgp->newcred, imgp->newcred->cr_uid);
647 			change_svgid(imgp->newcred, imgp->newcred->cr_gid);
648 		}
649 	}
650 	/* The new credentials are installed into the process later. */
651 
652 	/*
653 	 *	Loop through the list of image activators, calling each one.
654 	 *	An activator returns -1 if there is no match, 0 on success,
655 	 *	and an error otherwise.
656 	 */
657 	error = -1;
658 	for (i = 0; error == -1 && execsw[i]; ++i) {
659 		if (execsw[i]->ex_imgact == NULL)
660 			continue;
661 		error = (*execsw[i]->ex_imgact)(imgp);
662 	}
663 
664 	if (error) {
665 		if (error == -1)
666 			error = ENOEXEC;
667 		goto exec_fail_dealloc;
668 	}
669 
670 	/*
671 	 * Special interpreter operation, cleanup and loop up to try to
672 	 * activate the interpreter.
673 	 */
674 	if (imgp->interpreted) {
675 		exec_unmap_first_page(imgp);
676 		/*
677 		 * The text reference needs to be removed for scripts.
678 		 * There is a short period before we determine that
679 		 * something is a script where text reference is active.
680 		 * The vnode lock is held over this entire period
681 		 * so nothing should illegitimately be blocked.
682 		 */
683 		MPASS(imgp->textset);
684 		VOP_UNSET_TEXT_CHECKED(newtextvp);
685 		imgp->textset = false;
686 		/* free name buffer and old vnode */
687 #ifdef MAC
688 		mac_execve_interpreter_enter(newtextvp, &interpvplabel);
689 #endif
690 		if (imgp->opened) {
691 			VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td);
692 			imgp->opened = false;
693 		}
694 		vput(newtextvp);
695 		imgp->vp = newtextvp = NULL;
696 		if (args->fname != NULL) {
697 			if (newtextdvp != NULL) {
698 				vrele(newtextdvp);
699 				newtextdvp = NULL;
700 			}
701 			NDFREE_PNBUF(&nd);
702 			free(newbinname, M_PARGS);
703 			newbinname = NULL;
704 		}
705 		vm_object_deallocate(imgp->object);
706 		imgp->object = NULL;
707 		execve_nosetid(imgp);
708 		imgp->execpath = NULL;
709 		free(imgp->freepath, M_TEMP);
710 		imgp->freepath = NULL;
711 		/* set new name to that of the interpreter */
712 		if (imgp->interpreter_vp) {
713 			args->fname = NULL;
714 		} else {
715 			args->fname = imgp->interpreter_name;
716 		}
717 		goto interpret;
718 	}
719 
720 	/*
721 	 * NB: We unlock the vnode here because it is believed that none
722 	 * of the sv_copyout_strings/sv_fixup operations require the vnode.
723 	 */
724 	VOP_UNLOCK(imgp->vp);
725 
726 	if (disallow_high_osrel &&
727 	    P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
728 		error = ENOEXEC;
729 		uprintf("Osrel %d for image %s too high\n", p->p_osrel,
730 		    imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
731 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
732 		goto exec_fail_dealloc;
733 	}
734 
735 	/*
736 	 * Copy out strings (args and env) and initialize stack base.
737 	 */
738 	error = (*p->p_sysent->sv_copyout_strings)(imgp, &stack_base);
739 	if (error != 0) {
740 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
741 		goto exec_fail_dealloc;
742 	}
743 
744 	/*
745 	 * Stack setup.
746 	 */
747 	error = (*p->p_sysent->sv_fixup)(&stack_base, imgp);
748 	if (error != 0) {
749 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
750 		goto exec_fail_dealloc;
751 	}
752 
753 	/*
754 	 * For security and other reasons, the file descriptor table cannot be
755 	 * shared after an exec.
756 	 */
757 	fdunshare(td);
758 	pdunshare(td);
759 	/* close files on exec */
760 	fdcloseexec(td);
761 
762 	/*
763 	 * Malloc things before we need locks.
764 	 */
765 	i = exec_args_get_begin_envv(imgp->args) - imgp->args->begin_argv;
766 	/* Cache arguments if they fit inside our allowance */
767 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
768 		newargs = pargs_alloc(i);
769 		bcopy(imgp->args->begin_argv, newargs->ar_args, i);
770 	}
771 
772 	/*
773 	 * For security and other reasons, signal handlers cannot
774 	 * be shared after an exec. The new process gets a copy of the old
775 	 * handlers. In execsigs(), the new process will have its signals
776 	 * reset.
777 	 */
778 	if (sigacts_shared(p->p_sigacts)) {
779 		oldsigacts = p->p_sigacts;
780 		newsigacts = sigacts_alloc();
781 		sigacts_copy(newsigacts, oldsigacts);
782 	}
783 
784 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
785 
786 	PROC_LOCK(p);
787 	if (oldsigacts)
788 		p->p_sigacts = newsigacts;
789 	/* Stop profiling */
790 	stopprofclock(p);
791 
792 	/* reset caught signals */
793 	execsigs(p);
794 
795 	/* name this process - nameiexec(p, ndp) */
796 	bzero(p->p_comm, sizeof(p->p_comm));
797 	if (args->fname)
798 		bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
799 		    min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
800 	else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0)
801 		bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
802 	bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
803 #ifdef KTR
804 	sched_clear_tdname(td);
805 #endif
806 
807 	/*
808 	 * mark as execed, wakeup the process that vforked (if any) and tell
809 	 * it that it now has its own resources back
810 	 */
811 	p->p_flag |= P_EXEC;
812 	if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0)
813 		p->p_flag2 &= ~P2_NOTRACE;
814 	if ((p->p_flag2 & P2_STKGAP_DISABLE_EXEC) == 0)
815 		p->p_flag2 &= ~P2_STKGAP_DISABLE;
816 	p->p_flag2 &= ~(P2_MEMBAR_PRIVE | P2_MEMBAR_PRIVE_SYNCORE |
817 	    P2_MEMBAR_GLOBE);
818 	if (p->p_flag & P_PPWAIT) {
819 		p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
820 		cv_broadcast(&p->p_pwait);
821 		/* STOPs are no longer ignored, arrange for AST */
822 		signotify(td);
823 	}
824 
825 	if ((imgp->sysent->sv_setid_allowed != NULL &&
826 	    !(*imgp->sysent->sv_setid_allowed)(td, imgp)) ||
827 	    (p->p_flag2 & P2_NO_NEW_PRIVS) != 0)
828 		execve_nosetid(imgp);
829 
830 	/*
831 	 * Implement image setuid/setgid installation.
832 	 */
833 	if (imgp->credential_setid) {
834 		/*
835 		 * Turn off syscall tracing for set-id programs, except for
836 		 * root.  Record any set-id flags first to make sure that
837 		 * we do not regain any tracing during a possible block.
838 		 */
839 		setsugid(p);
840 #ifdef KTRACE
841 		kiop = ktrprocexec(p);
842 #endif
843 		/*
844 		 * Close any file descriptors 0..2 that reference procfs,
845 		 * then make sure file descriptors 0..2 are in use.
846 		 *
847 		 * Both fdsetugidsafety() and fdcheckstd() may call functions
848 		 * taking sleepable locks, so temporarily drop our locks.
849 		 */
850 		PROC_UNLOCK(p);
851 		VOP_UNLOCK(imgp->vp);
852 		fdsetugidsafety(td);
853 		error = fdcheckstd(td);
854 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
855 		if (error != 0)
856 			goto exec_fail_dealloc;
857 		PROC_LOCK(p);
858 #ifdef MAC
859 		if (will_transition) {
860 			mac_vnode_execve_transition(oldcred, imgp->newcred,
861 			    imgp->vp, interpvplabel, imgp);
862 		}
863 #endif
864 	} else {
865 		if (oldcred->cr_uid == oldcred->cr_ruid &&
866 		    oldcred->cr_gid == oldcred->cr_rgid)
867 			p->p_flag &= ~P_SUGID;
868 	}
869 	/*
870 	 * Set the new credentials.
871 	 */
872 	if (imgp->newcred != NULL) {
873 		proc_set_cred(p, imgp->newcred);
874 		crfree(oldcred);
875 		oldcred = NULL;
876 	}
877 
878 	/*
879 	 * Store the vp for use in kern.proc.pathname.  This vnode was
880 	 * referenced by namei() or by fexecve variant of fname handling.
881 	 */
882 	oldtextvp = p->p_textvp;
883 	p->p_textvp = newtextvp;
884 	oldtextdvp = p->p_textdvp;
885 	p->p_textdvp = newtextdvp;
886 	newtextdvp = NULL;
887 	oldbinname = p->p_binname;
888 	p->p_binname = newbinname;
889 	newbinname = NULL;
890 
891 #ifdef KDTRACE_HOOKS
892 	/*
893 	 * Tell the DTrace fasttrap provider about the exec if it
894 	 * has declared an interest.
895 	 */
896 	if (dtrace_fasttrap_exec)
897 		dtrace_fasttrap_exec(p);
898 #endif
899 
900 	/*
901 	 * Notify others that we exec'd, and clear the P_INEXEC flag
902 	 * as we're now a bona fide freshly-execed process.
903 	 */
904 	KNOTE_LOCKED(p->p_klist, NOTE_EXEC);
905 	p->p_flag &= ~P_INEXEC;
906 
907 	/* clear "fork but no exec" flag, as we _are_ execing */
908 	p->p_acflag &= ~AFORK;
909 
910 	/*
911 	 * Free any previous argument cache and replace it with
912 	 * the new argument cache, if any.
913 	 */
914 	oldargs = p->p_args;
915 	p->p_args = newargs;
916 	newargs = NULL;
917 
918 	PROC_UNLOCK(p);
919 
920 #ifdef	HWPMC_HOOKS
921 	/*
922 	 * Check if system-wide sampling is in effect or if the
923 	 * current process is using PMCs.  If so, do exec() time
924 	 * processing.  This processing needs to happen AFTER the
925 	 * P_INEXEC flag is cleared.
926 	 */
927 	if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
928 		VOP_UNLOCK(imgp->vp);
929 		pe.pm_credentialschanged = credential_changing;
930 		pe.pm_baseaddr = imgp->reloc_base;
931 		pe.pm_dynaddr = imgp->et_dyn_addr;
932 
933 		PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
934 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
935 	}
936 #endif
937 
938 	/* Set values passed into the program in registers. */
939 	(*p->p_sysent->sv_setregs)(td, imgp, stack_base);
940 
941 	VOP_MMAPPED(imgp->vp);
942 
943 	SDT_PROBE1(proc, , , exec__success, args->fname);
944 
945 exec_fail_dealloc:
946 	if (error != 0) {
947 		p->p_osrel = orig_osrel;
948 		p->p_fctl0 = orig_fctl0;
949 		p->p_elf_brandinfo = orig_brandinfo;
950 	}
951 
952 	if (imgp->firstpage != NULL)
953 		exec_unmap_first_page(imgp);
954 
955 	if (imgp->vp != NULL) {
956 		if (imgp->opened)
957 			VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
958 		if (imgp->textset)
959 			VOP_UNSET_TEXT_CHECKED(imgp->vp);
960 		if (error != 0)
961 			vput(imgp->vp);
962 		else
963 			VOP_UNLOCK(imgp->vp);
964 		if (args->fname != NULL)
965 			NDFREE_PNBUF(&nd);
966 		if (newtextdvp != NULL)
967 			vrele(newtextdvp);
968 		free(newbinname, M_PARGS);
969 	}
970 
971 	if (imgp->object != NULL)
972 		vm_object_deallocate(imgp->object);
973 
974 	free(imgp->freepath, M_TEMP);
975 
976 	if (error == 0) {
977 		if (p->p_ptevents & PTRACE_EXEC) {
978 			PROC_LOCK(p);
979 			if (p->p_ptevents & PTRACE_EXEC)
980 				td->td_dbgflags |= TDB_EXEC;
981 			PROC_UNLOCK(p);
982 		}
983 	} else {
984 exec_fail:
985 		/* we're done here, clear P_INEXEC */
986 		PROC_LOCK(p);
987 		p->p_flag &= ~P_INEXEC;
988 		PROC_UNLOCK(p);
989 
990 		SDT_PROBE1(proc, , , exec__failure, error);
991 	}
992 
993 	if (imgp->newcred != NULL && oldcred != NULL)
994 		crfree(imgp->newcred);
995 
996 #ifdef MAC
997 	mac_execve_exit(imgp);
998 	mac_execve_interpreter_exit(interpvplabel);
999 #endif
1000 	exec_free_args(args);
1001 
1002 	/*
1003 	 * Handle deferred decrement of ref counts.
1004 	 */
1005 	if (oldtextvp != NULL)
1006 		vrele(oldtextvp);
1007 	if (oldtextdvp != NULL)
1008 		vrele(oldtextdvp);
1009 	free(oldbinname, M_PARGS);
1010 #ifdef KTRACE
1011 	ktr_io_params_free(kiop);
1012 #endif
1013 	pargs_drop(oldargs);
1014 	pargs_drop(newargs);
1015 	if (oldsigacts != NULL)
1016 		sigacts_free(oldsigacts);
1017 	if (euip != NULL)
1018 		uifree(euip);
1019 
1020 	if (error && imgp->vmspace_destroyed) {
1021 		/* sorry, no more process anymore. exit gracefully */
1022 		exec_cleanup(td, oldvmspace);
1023 		exit1(td, 0, SIGABRT);
1024 		/* NOT REACHED */
1025 	}
1026 
1027 #ifdef KTRACE
1028 	if (error == 0)
1029 		ktrprocctor(p);
1030 #endif
1031 
1032 	/*
1033 	 * We don't want cpu_set_syscall_retval() to overwrite any of
1034 	 * the register values put in place by exec_setregs().
1035 	 * Implementations of cpu_set_syscall_retval() will leave
1036 	 * registers unmodified when returning EJUSTRETURN.
1037 	 */
1038 	return (error == 0 ? EJUSTRETURN : error);
1039 }
1040 
1041 void
1042 exec_cleanup(struct thread *td, struct vmspace *oldvmspace)
1043 {
1044 	if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
1045 		KASSERT(td->td_proc->p_vmspace != oldvmspace,
1046 		    ("oldvmspace still used"));
1047 		vmspace_free(oldvmspace);
1048 		td->td_pflags &= ~TDP_EXECVMSPC;
1049 	}
1050 }
1051 
1052 int
1053 exec_map_first_page(struct image_params *imgp)
1054 {
1055 	vm_object_t object;
1056 	vm_page_t m;
1057 	int error;
1058 
1059 	if (imgp->firstpage != NULL)
1060 		exec_unmap_first_page(imgp);
1061 
1062 	object = imgp->vp->v_object;
1063 	if (object == NULL)
1064 		return (EACCES);
1065 #if VM_NRESERVLEVEL > 0
1066 	if ((object->flags & OBJ_COLORED) == 0) {
1067 		VM_OBJECT_WLOCK(object);
1068 		vm_object_color(object, 0);
1069 		VM_OBJECT_WUNLOCK(object);
1070 	}
1071 #endif
1072 	error = vm_page_grab_valid_unlocked(&m, object, 0,
1073 	    VM_ALLOC_COUNT(VM_INITIAL_PAGEIN) |
1074 	    VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED);
1075 
1076 	if (error != VM_PAGER_OK)
1077 		return (EIO);
1078 	imgp->firstpage = sf_buf_alloc(m, 0);
1079 	imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
1080 
1081 	return (0);
1082 }
1083 
1084 void
1085 exec_unmap_first_page(struct image_params *imgp)
1086 {
1087 	vm_page_t m;
1088 
1089 	if (imgp->firstpage != NULL) {
1090 		m = sf_buf_page(imgp->firstpage);
1091 		sf_buf_free(imgp->firstpage);
1092 		imgp->firstpage = NULL;
1093 		vm_page_unwire(m, PQ_ACTIVE);
1094 	}
1095 }
1096 
1097 void
1098 exec_onexec_old(struct thread *td)
1099 {
1100 	sigfastblock_clear(td);
1101 	umtx_exec(td->td_proc);
1102 }
1103 
1104 /*
1105  * This is an optimization which removes the unmanaged shared page
1106  * mapping. In combination with pmap_remove_pages(), which cleans all
1107  * managed mappings in the process' vmspace pmap, no work will be left
1108  * for pmap_remove(min, max).
1109  */
1110 void
1111 exec_free_abi_mappings(struct proc *p)
1112 {
1113 	struct vmspace *vmspace;
1114 
1115 	vmspace = p->p_vmspace;
1116 	if (refcount_load(&vmspace->vm_refcnt) != 1)
1117 		return;
1118 
1119 	if (!PROC_HAS_SHP(p))
1120 		return;
1121 
1122 	pmap_remove(vmspace_pmap(vmspace), vmspace->vm_shp_base,
1123 	    vmspace->vm_shp_base + p->p_sysent->sv_shared_page_len);
1124 }
1125 
1126 /*
1127  * Run down the current address space and install a new one.
1128  */
1129 int
1130 exec_new_vmspace(struct image_params *imgp, struct sysentvec *sv)
1131 {
1132 	int error;
1133 	struct proc *p = imgp->proc;
1134 	struct vmspace *vmspace = p->p_vmspace;
1135 	struct thread *td = curthread;
1136 	vm_offset_t sv_minuser;
1137 	vm_map_t map;
1138 
1139 	imgp->vmspace_destroyed = true;
1140 	imgp->sysent = sv;
1141 
1142 	if (p->p_sysent->sv_onexec_old != NULL)
1143 		p->p_sysent->sv_onexec_old(td);
1144 	itimers_exec(p);
1145 
1146 	EVENTHANDLER_DIRECT_INVOKE(process_exec, p, imgp);
1147 
1148 	/*
1149 	 * Blow away entire process VM, if address space not shared,
1150 	 * otherwise, create a new VM space so that other threads are
1151 	 * not disrupted
1152 	 */
1153 	map = &vmspace->vm_map;
1154 	if (map_at_zero)
1155 		sv_minuser = sv->sv_minuser;
1156 	else
1157 		sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1158 	if (refcount_load(&vmspace->vm_refcnt) == 1 &&
1159 	    vm_map_min(map) == sv_minuser &&
1160 	    vm_map_max(map) == sv->sv_maxuser &&
1161 	    cpu_exec_vmspace_reuse(p, map)) {
1162 		exec_free_abi_mappings(p);
1163 		shmexit(vmspace);
1164 		pmap_remove_pages(vmspace_pmap(vmspace));
1165 		vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1166 		/*
1167 		 * An exec terminates mlockall(MCL_FUTURE).
1168 		 * ASLR and W^X states must be re-evaluated.
1169 		 */
1170 		vm_map_lock(map);
1171 		vm_map_modflags(map, 0, MAP_WIREFUTURE | MAP_ASLR |
1172 		    MAP_ASLR_IGNSTART | MAP_ASLR_STACK | MAP_WXORX);
1173 		vm_map_unlock(map);
1174 	} else {
1175 		error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1176 		if (error)
1177 			return (error);
1178 		vmspace = p->p_vmspace;
1179 		map = &vmspace->vm_map;
1180 	}
1181 	map->flags |= imgp->map_flags;
1182 
1183 	return (sv->sv_onexec != NULL ? sv->sv_onexec(p, imgp) : 0);
1184 }
1185 
1186 /*
1187  * Compute the stack size limit and map the main process stack.
1188  * Map the shared page.
1189  */
1190 int
1191 exec_map_stack(struct image_params *imgp)
1192 {
1193 	struct rlimit rlim_stack;
1194 	struct sysentvec *sv;
1195 	struct proc *p;
1196 	vm_map_t map;
1197 	struct vmspace *vmspace;
1198 	vm_offset_t stack_addr, stack_top;
1199 	vm_offset_t sharedpage_addr;
1200 	u_long ssiz;
1201 	int error, find_space, stack_off;
1202 	vm_prot_t stack_prot;
1203 	vm_object_t obj;
1204 
1205 	p = imgp->proc;
1206 	sv = p->p_sysent;
1207 
1208 	if (imgp->stack_sz != 0) {
1209 		ssiz = trunc_page(imgp->stack_sz);
1210 		PROC_LOCK(p);
1211 		lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack);
1212 		PROC_UNLOCK(p);
1213 		if (ssiz > rlim_stack.rlim_max)
1214 			ssiz = rlim_stack.rlim_max;
1215 		if (ssiz > rlim_stack.rlim_cur) {
1216 			rlim_stack.rlim_cur = ssiz;
1217 			kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack);
1218 		}
1219 	} else if (sv->sv_maxssiz != NULL) {
1220 		ssiz = *sv->sv_maxssiz;
1221 	} else {
1222 		ssiz = maxssiz;
1223 	}
1224 
1225 	vmspace = p->p_vmspace;
1226 	map = &vmspace->vm_map;
1227 
1228 	stack_prot = sv->sv_shared_page_obj != NULL && imgp->stack_prot != 0 ?
1229 	    imgp->stack_prot : sv->sv_stackprot;
1230 	if ((map->flags & MAP_ASLR_STACK) != 0) {
1231 		stack_addr = round_page((vm_offset_t)p->p_vmspace->vm_daddr +
1232 		    lim_max(curthread, RLIMIT_DATA));
1233 		find_space = VMFS_ANY_SPACE;
1234 	} else {
1235 		stack_addr = sv->sv_usrstack - ssiz;
1236 		find_space = VMFS_NO_SPACE;
1237 	}
1238 	error = vm_map_find(map, NULL, 0, &stack_addr, (vm_size_t)ssiz,
1239 	    sv->sv_usrstack, find_space, stack_prot, VM_PROT_ALL,
1240 	    MAP_STACK_AREA);
1241 	if (error != KERN_SUCCESS) {
1242 		uprintf("exec_new_vmspace: mapping stack size %#jx prot %#x "
1243 		    "failed, mach error %d errno %d\n", (uintmax_t)ssiz,
1244 		    stack_prot, error, vm_mmap_to_errno(error));
1245 		return (vm_mmap_to_errno(error));
1246 	}
1247 
1248 	stack_top = stack_addr + ssiz;
1249 	if ((map->flags & MAP_ASLR_STACK) != 0) {
1250 		/* Randomize within the first page of the stack. */
1251 		arc4rand(&stack_off, sizeof(stack_off), 0);
1252 		stack_top -= rounddown2(stack_off & PAGE_MASK, sizeof(void *));
1253 	}
1254 
1255 	/* Map a shared page */
1256 	obj = sv->sv_shared_page_obj;
1257 	if (obj == NULL) {
1258 		sharedpage_addr = 0;
1259 		goto out;
1260 	}
1261 
1262 	/*
1263 	 * If randomization is disabled then the shared page will
1264 	 * be mapped at address specified in sysentvec.
1265 	 * Otherwise any address above .data section can be selected.
1266 	 * Same logic is used for stack address randomization.
1267 	 * If the address randomization is applied map a guard page
1268 	 * at the top of UVA.
1269 	 */
1270 	vm_object_reference(obj);
1271 	if ((imgp->imgp_flags & IMGP_ASLR_SHARED_PAGE) != 0) {
1272 		sharedpage_addr = round_page((vm_offset_t)p->p_vmspace->vm_daddr +
1273 		    lim_max(curthread, RLIMIT_DATA));
1274 
1275 		error = vm_map_fixed(map, NULL, 0,
1276 		    sv->sv_maxuser - PAGE_SIZE, PAGE_SIZE,
1277 		    VM_PROT_NONE, VM_PROT_NONE, MAP_CREATE_GUARD);
1278 		if (error != KERN_SUCCESS) {
1279 			/*
1280 			 * This is not fatal, so let's just print a warning
1281 			 * and continue.
1282 			 */
1283 			uprintf("%s: Mapping guard page at the top of UVA failed"
1284 			    " mach error %d errno %d",
1285 			    __func__, error, vm_mmap_to_errno(error));
1286 		}
1287 
1288 		error = vm_map_find(map, obj, 0,
1289 		    &sharedpage_addr, sv->sv_shared_page_len,
1290 		    sv->sv_maxuser, VMFS_ANY_SPACE,
1291 		    VM_PROT_READ | VM_PROT_EXECUTE,
1292 		    VM_PROT_READ | VM_PROT_EXECUTE,
1293 		    MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1294 	} else {
1295 		sharedpage_addr = sv->sv_shared_page_base;
1296 		vm_map_fixed(map, obj, 0,
1297 		    sharedpage_addr, sv->sv_shared_page_len,
1298 		    VM_PROT_READ | VM_PROT_EXECUTE,
1299 		    VM_PROT_READ | VM_PROT_EXECUTE,
1300 		    MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1301 	}
1302 	if (error != KERN_SUCCESS) {
1303 		uprintf("%s: mapping shared page at addr: %p"
1304 		    "failed, mach error %d errno %d\n", __func__,
1305 		    (void *)sharedpage_addr, error, vm_mmap_to_errno(error));
1306 		vm_object_deallocate(obj);
1307 		return (vm_mmap_to_errno(error));
1308 	}
1309 out:
1310 	/*
1311 	 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they
1312 	 * are still used to enforce the stack rlimit on the process stack.
1313 	 */
1314 	vmspace->vm_maxsaddr = (char *)stack_addr;
1315 	vmspace->vm_stacktop = stack_top;
1316 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1317 	vmspace->vm_shp_base = sharedpage_addr;
1318 
1319 	return (0);
1320 }
1321 
1322 /*
1323  * Copy out argument and environment strings from the old process address
1324  * space into the temporary string buffer.
1325  */
1326 int
1327 exec_copyin_args(struct image_args *args, const char *fname,
1328     char **argv, char **envv)
1329 {
1330 	u_long arg, env;
1331 	int error;
1332 
1333 	bzero(args, sizeof(*args));
1334 	if (argv == NULL)
1335 		return (EFAULT);
1336 
1337 	/*
1338 	 * Allocate demand-paged memory for the file name, argument, and
1339 	 * environment strings.
1340 	 */
1341 	error = exec_alloc_args(args);
1342 	if (error != 0)
1343 		return (error);
1344 
1345 	/*
1346 	 * Copy the file name.
1347 	 */
1348 	error = exec_args_add_fname(args, fname, UIO_USERSPACE);
1349 	if (error != 0)
1350 		goto err_exit;
1351 
1352 	/*
1353 	 * extract arguments first
1354 	 */
1355 	for (;;) {
1356 		error = fueword(argv++, &arg);
1357 		if (error == -1) {
1358 			error = EFAULT;
1359 			goto err_exit;
1360 		}
1361 		if (arg == 0)
1362 			break;
1363 		error = exec_args_add_arg(args, (char *)(uintptr_t)arg,
1364 		    UIO_USERSPACE);
1365 		if (error != 0)
1366 			goto err_exit;
1367 	}
1368 
1369 	/*
1370 	 * extract environment strings
1371 	 */
1372 	if (envv) {
1373 		for (;;) {
1374 			error = fueword(envv++, &env);
1375 			if (error == -1) {
1376 				error = EFAULT;
1377 				goto err_exit;
1378 			}
1379 			if (env == 0)
1380 				break;
1381 			error = exec_args_add_env(args,
1382 			    (char *)(uintptr_t)env, UIO_USERSPACE);
1383 			if (error != 0)
1384 				goto err_exit;
1385 		}
1386 	}
1387 
1388 	return (0);
1389 
1390 err_exit:
1391 	exec_free_args(args);
1392 	return (error);
1393 }
1394 
1395 struct exec_args_kva {
1396 	vm_offset_t addr;
1397 	u_int gen;
1398 	SLIST_ENTRY(exec_args_kva) next;
1399 };
1400 
1401 DPCPU_DEFINE_STATIC(struct exec_args_kva *, exec_args_kva);
1402 
1403 static SLIST_HEAD(, exec_args_kva) exec_args_kva_freelist;
1404 static struct mtx exec_args_kva_mtx;
1405 static u_int exec_args_gen;
1406 
1407 static void
1408 exec_prealloc_args_kva(void *arg __unused)
1409 {
1410 	struct exec_args_kva *argkva;
1411 	u_int i;
1412 
1413 	SLIST_INIT(&exec_args_kva_freelist);
1414 	mtx_init(&exec_args_kva_mtx, "exec args kva", NULL, MTX_DEF);
1415 	for (i = 0; i < exec_map_entries; i++) {
1416 		argkva = malloc(sizeof(*argkva), M_PARGS, M_WAITOK);
1417 		argkva->addr = kmap_alloc_wait(exec_map, exec_map_entry_size);
1418 		argkva->gen = exec_args_gen;
1419 		SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1420 	}
1421 }
1422 SYSINIT(exec_args_kva, SI_SUB_EXEC, SI_ORDER_ANY, exec_prealloc_args_kva, NULL);
1423 
1424 static vm_offset_t
1425 exec_alloc_args_kva(void **cookie)
1426 {
1427 	struct exec_args_kva *argkva;
1428 
1429 	argkva = (void *)atomic_readandclear_ptr(
1430 	    (uintptr_t *)DPCPU_PTR(exec_args_kva));
1431 	if (argkva == NULL) {
1432 		mtx_lock(&exec_args_kva_mtx);
1433 		while ((argkva = SLIST_FIRST(&exec_args_kva_freelist)) == NULL)
1434 			(void)mtx_sleep(&exec_args_kva_freelist,
1435 			    &exec_args_kva_mtx, 0, "execkva", 0);
1436 		SLIST_REMOVE_HEAD(&exec_args_kva_freelist, next);
1437 		mtx_unlock(&exec_args_kva_mtx);
1438 	}
1439 	kasan_mark((void *)argkva->addr, exec_map_entry_size,
1440 	    exec_map_entry_size, 0);
1441 	*(struct exec_args_kva **)cookie = argkva;
1442 	return (argkva->addr);
1443 }
1444 
1445 static void
1446 exec_release_args_kva(struct exec_args_kva *argkva, u_int gen)
1447 {
1448 	vm_offset_t base;
1449 
1450 	base = argkva->addr;
1451 	kasan_mark((void *)argkva->addr, 0, exec_map_entry_size,
1452 	    KASAN_EXEC_ARGS_FREED);
1453 	if (argkva->gen != gen) {
1454 		(void)vm_map_madvise(exec_map, base, base + exec_map_entry_size,
1455 		    MADV_FREE);
1456 		argkva->gen = gen;
1457 	}
1458 	if (!atomic_cmpset_ptr((uintptr_t *)DPCPU_PTR(exec_args_kva),
1459 	    (uintptr_t)NULL, (uintptr_t)argkva)) {
1460 		mtx_lock(&exec_args_kva_mtx);
1461 		SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1462 		wakeup_one(&exec_args_kva_freelist);
1463 		mtx_unlock(&exec_args_kva_mtx);
1464 	}
1465 }
1466 
1467 static void
1468 exec_free_args_kva(void *cookie)
1469 {
1470 
1471 	exec_release_args_kva(cookie, exec_args_gen);
1472 }
1473 
1474 static void
1475 exec_args_kva_lowmem(void *arg __unused, int flags __unused)
1476 {
1477 	SLIST_HEAD(, exec_args_kva) head;
1478 	struct exec_args_kva *argkva;
1479 	u_int gen;
1480 	int i;
1481 
1482 	gen = atomic_fetchadd_int(&exec_args_gen, 1) + 1;
1483 
1484 	/*
1485 	 * Force an madvise of each KVA range. Any currently allocated ranges
1486 	 * will have MADV_FREE applied once they are freed.
1487 	 */
1488 	SLIST_INIT(&head);
1489 	mtx_lock(&exec_args_kva_mtx);
1490 	SLIST_SWAP(&head, &exec_args_kva_freelist, exec_args_kva);
1491 	mtx_unlock(&exec_args_kva_mtx);
1492 	while ((argkva = SLIST_FIRST(&head)) != NULL) {
1493 		SLIST_REMOVE_HEAD(&head, next);
1494 		exec_release_args_kva(argkva, gen);
1495 	}
1496 
1497 	CPU_FOREACH(i) {
1498 		argkva = (void *)atomic_readandclear_ptr(
1499 		    (uintptr_t *)DPCPU_ID_PTR(i, exec_args_kva));
1500 		if (argkva != NULL)
1501 			exec_release_args_kva(argkva, gen);
1502 	}
1503 }
1504 EVENTHANDLER_DEFINE(vm_lowmem, exec_args_kva_lowmem, NULL,
1505     EVENTHANDLER_PRI_ANY);
1506 
1507 /*
1508  * Allocate temporary demand-paged, zero-filled memory for the file name,
1509  * argument, and environment strings.
1510  */
1511 int
1512 exec_alloc_args(struct image_args *args)
1513 {
1514 
1515 	args->buf = (char *)exec_alloc_args_kva(&args->bufkva);
1516 	return (0);
1517 }
1518 
1519 void
1520 exec_free_args(struct image_args *args)
1521 {
1522 
1523 	if (args->buf != NULL) {
1524 		exec_free_args_kva(args->bufkva);
1525 		args->buf = NULL;
1526 	}
1527 	if (args->fname_buf != NULL) {
1528 		free(args->fname_buf, M_TEMP);
1529 		args->fname_buf = NULL;
1530 	}
1531 }
1532 
1533 /*
1534  * A set to functions to fill struct image args.
1535  *
1536  * NOTE: exec_args_add_fname() must be called (possibly with a NULL
1537  * fname) before the other functions.  All exec_args_add_arg() calls must
1538  * be made before any exec_args_add_env() calls.  exec_args_adjust_args()
1539  * may be called any time after exec_args_add_fname().
1540  *
1541  * exec_args_add_fname() - install path to be executed
1542  * exec_args_add_arg() - append an argument string
1543  * exec_args_add_env() - append an env string
1544  * exec_args_adjust_args() - adjust location of the argument list to
1545  *                           allow new arguments to be prepended
1546  */
1547 int
1548 exec_args_add_fname(struct image_args *args, const char *fname,
1549     enum uio_seg segflg)
1550 {
1551 	int error;
1552 	size_t length;
1553 
1554 	KASSERT(args->fname == NULL, ("fname already appended"));
1555 	KASSERT(args->endp == NULL, ("already appending to args"));
1556 
1557 	if (fname != NULL) {
1558 		args->fname = args->buf;
1559 		error = segflg == UIO_SYSSPACE ?
1560 		    copystr(fname, args->fname, PATH_MAX, &length) :
1561 		    copyinstr(fname, args->fname, PATH_MAX, &length);
1562 		if (error != 0)
1563 			return (error == ENAMETOOLONG ? E2BIG : error);
1564 	} else
1565 		length = 0;
1566 
1567 	/* Set up for _arg_*()/_env_*() */
1568 	args->endp = args->buf + length;
1569 	/* begin_argv must be set and kept updated */
1570 	args->begin_argv = args->endp;
1571 	KASSERT(exec_map_entry_size - length >= ARG_MAX,
1572 	    ("too little space remaining for arguments %zu < %zu",
1573 	    exec_map_entry_size - length, (size_t)ARG_MAX));
1574 	args->stringspace = ARG_MAX;
1575 
1576 	return (0);
1577 }
1578 
1579 static int
1580 exec_args_add_str(struct image_args *args, const char *str,
1581     enum uio_seg segflg, int *countp)
1582 {
1583 	int error;
1584 	size_t length;
1585 
1586 	KASSERT(args->endp != NULL, ("endp not initialized"));
1587 	KASSERT(args->begin_argv != NULL, ("begin_argp not initialized"));
1588 
1589 	error = (segflg == UIO_SYSSPACE) ?
1590 	    copystr(str, args->endp, args->stringspace, &length) :
1591 	    copyinstr(str, args->endp, args->stringspace, &length);
1592 	if (error != 0)
1593 		return (error == ENAMETOOLONG ? E2BIG : error);
1594 	args->stringspace -= length;
1595 	args->endp += length;
1596 	(*countp)++;
1597 
1598 	return (0);
1599 }
1600 
1601 int
1602 exec_args_add_arg(struct image_args *args, const char *argp,
1603     enum uio_seg segflg)
1604 {
1605 
1606 	KASSERT(args->envc == 0, ("appending args after env"));
1607 
1608 	return (exec_args_add_str(args, argp, segflg, &args->argc));
1609 }
1610 
1611 int
1612 exec_args_add_env(struct image_args *args, const char *envp,
1613     enum uio_seg segflg)
1614 {
1615 
1616 	if (args->envc == 0)
1617 		args->begin_envv = args->endp;
1618 
1619 	return (exec_args_add_str(args, envp, segflg, &args->envc));
1620 }
1621 
1622 int
1623 exec_args_adjust_args(struct image_args *args, size_t consume, ssize_t extend)
1624 {
1625 	ssize_t offset;
1626 
1627 	KASSERT(args->endp != NULL, ("endp not initialized"));
1628 	KASSERT(args->begin_argv != NULL, ("begin_argp not initialized"));
1629 
1630 	offset = extend - consume;
1631 	if (args->stringspace < offset)
1632 		return (E2BIG);
1633 	memmove(args->begin_argv + extend, args->begin_argv + consume,
1634 	    args->endp - args->begin_argv + consume);
1635 	if (args->envc > 0)
1636 		args->begin_envv += offset;
1637 	args->endp += offset;
1638 	args->stringspace -= offset;
1639 	return (0);
1640 }
1641 
1642 char *
1643 exec_args_get_begin_envv(struct image_args *args)
1644 {
1645 
1646 	KASSERT(args->endp != NULL, ("endp not initialized"));
1647 
1648 	if (args->envc > 0)
1649 		return (args->begin_envv);
1650 	return (args->endp);
1651 }
1652 
1653 /*
1654  * Copy strings out to the new process address space, constructing new arg
1655  * and env vector tables. Return a pointer to the base so that it can be used
1656  * as the initial stack pointer.
1657  */
1658 int
1659 exec_copyout_strings(struct image_params *imgp, uintptr_t *stack_base)
1660 {
1661 	int argc, envc;
1662 	char **vectp;
1663 	char *stringp;
1664 	uintptr_t destp, ustringp;
1665 	struct ps_strings *arginfo;
1666 	struct proc *p;
1667 	struct sysentvec *sysent;
1668 	size_t execpath_len;
1669 	int error, szsigcode;
1670 	char canary[sizeof(long) * 8];
1671 
1672 	p = imgp->proc;
1673 	sysent = p->p_sysent;
1674 
1675 	destp =	PROC_PS_STRINGS(p);
1676 	arginfo = imgp->ps_strings = (void *)destp;
1677 
1678 	/*
1679 	 * Install sigcode.
1680 	 */
1681 	if (sysent->sv_shared_page_base == 0 && sysent->sv_szsigcode != NULL) {
1682 		szsigcode = *(sysent->sv_szsigcode);
1683 		destp -= szsigcode;
1684 		destp = rounddown2(destp, sizeof(void *));
1685 		error = copyout(sysent->sv_sigcode, (void *)destp, szsigcode);
1686 		if (error != 0)
1687 			return (error);
1688 	}
1689 
1690 	/*
1691 	 * Copy the image path for the rtld.
1692 	 */
1693 	if (imgp->execpath != NULL && imgp->auxargs != NULL) {
1694 		execpath_len = strlen(imgp->execpath) + 1;
1695 		destp -= execpath_len;
1696 		destp = rounddown2(destp, sizeof(void *));
1697 		imgp->execpathp = (void *)destp;
1698 		error = copyout(imgp->execpath, imgp->execpathp, execpath_len);
1699 		if (error != 0)
1700 			return (error);
1701 	}
1702 
1703 	/*
1704 	 * Prepare the canary for SSP.
1705 	 */
1706 	arc4rand(canary, sizeof(canary), 0);
1707 	destp -= sizeof(canary);
1708 	imgp->canary = (void *)destp;
1709 	error = copyout(canary, imgp->canary, sizeof(canary));
1710 	if (error != 0)
1711 		return (error);
1712 	imgp->canarylen = sizeof(canary);
1713 
1714 	/*
1715 	 * Prepare the pagesizes array.
1716 	 */
1717 	imgp->pagesizeslen = sizeof(pagesizes[0]) * MAXPAGESIZES;
1718 	destp -= imgp->pagesizeslen;
1719 	destp = rounddown2(destp, sizeof(void *));
1720 	imgp->pagesizes = (void *)destp;
1721 	error = copyout(pagesizes, imgp->pagesizes, imgp->pagesizeslen);
1722 	if (error != 0)
1723 		return (error);
1724 
1725 	/*
1726 	 * Allocate room for the argument and environment strings.
1727 	 */
1728 	destp -= ARG_MAX - imgp->args->stringspace;
1729 	destp = rounddown2(destp, sizeof(void *));
1730 	ustringp = destp;
1731 
1732 	if (imgp->auxargs) {
1733 		/*
1734 		 * Allocate room on the stack for the ELF auxargs
1735 		 * array.  It has up to AT_COUNT entries.
1736 		 */
1737 		destp -= AT_COUNT * sizeof(Elf_Auxinfo);
1738 		destp = rounddown2(destp, sizeof(void *));
1739 	}
1740 
1741 	vectp = (char **)destp;
1742 
1743 	/*
1744 	 * Allocate room for the argv[] and env vectors including the
1745 	 * terminating NULL pointers.
1746 	 */
1747 	vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
1748 
1749 	/*
1750 	 * vectp also becomes our initial stack base
1751 	 */
1752 	*stack_base = (uintptr_t)vectp;
1753 
1754 	stringp = imgp->args->begin_argv;
1755 	argc = imgp->args->argc;
1756 	envc = imgp->args->envc;
1757 
1758 	/*
1759 	 * Copy out strings - arguments and environment.
1760 	 */
1761 	error = copyout(stringp, (void *)ustringp,
1762 	    ARG_MAX - imgp->args->stringspace);
1763 	if (error != 0)
1764 		return (error);
1765 
1766 	/*
1767 	 * Fill in "ps_strings" struct for ps, w, etc.
1768 	 */
1769 	imgp->argv = vectp;
1770 	if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 ||
1771 	    suword32(&arginfo->ps_nargvstr, argc) != 0)
1772 		return (EFAULT);
1773 
1774 	/*
1775 	 * Fill in argument portion of vector table.
1776 	 */
1777 	for (; argc > 0; --argc) {
1778 		if (suword(vectp++, ustringp) != 0)
1779 			return (EFAULT);
1780 		while (*stringp++ != 0)
1781 			ustringp++;
1782 		ustringp++;
1783 	}
1784 
1785 	/* a null vector table pointer separates the argp's from the envp's */
1786 	if (suword(vectp++, 0) != 0)
1787 		return (EFAULT);
1788 
1789 	imgp->envv = vectp;
1790 	if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 ||
1791 	    suword32(&arginfo->ps_nenvstr, envc) != 0)
1792 		return (EFAULT);
1793 
1794 	/*
1795 	 * Fill in environment portion of vector table.
1796 	 */
1797 	for (; envc > 0; --envc) {
1798 		if (suword(vectp++, ustringp) != 0)
1799 			return (EFAULT);
1800 		while (*stringp++ != 0)
1801 			ustringp++;
1802 		ustringp++;
1803 	}
1804 
1805 	/* end of vector table is a null pointer */
1806 	if (suword(vectp, 0) != 0)
1807 		return (EFAULT);
1808 
1809 	if (imgp->auxargs) {
1810 		vectp++;
1811 		error = imgp->sysent->sv_copyout_auxargs(imgp,
1812 		    (uintptr_t)vectp);
1813 		if (error != 0)
1814 			return (error);
1815 	}
1816 
1817 	return (0);
1818 }
1819 
1820 /*
1821  * Check permissions of file to execute.
1822  *	Called with imgp->vp locked.
1823  *	Return 0 for success or error code on failure.
1824  */
1825 int
1826 exec_check_permissions(struct image_params *imgp)
1827 {
1828 	struct vnode *vp = imgp->vp;
1829 	struct vattr *attr = imgp->attr;
1830 	struct thread *td;
1831 	int error;
1832 
1833 	td = curthread;
1834 
1835 	/* Get file attributes */
1836 	error = VOP_GETATTR(vp, attr, td->td_ucred);
1837 	if (error)
1838 		return (error);
1839 
1840 #ifdef MAC
1841 	error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1842 	if (error)
1843 		return (error);
1844 #endif
1845 
1846 	/*
1847 	 * 1) Check if file execution is disabled for the filesystem that
1848 	 *    this file resides on.
1849 	 * 2) Ensure that at least one execute bit is on. Otherwise, a
1850 	 *    privileged user will always succeed, and we don't want this
1851 	 *    to happen unless the file really is executable.
1852 	 * 3) Ensure that the file is a regular file.
1853 	 */
1854 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1855 	    (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1856 	    (attr->va_type != VREG))
1857 		return (EACCES);
1858 
1859 	/*
1860 	 * Zero length files can't be exec'd
1861 	 */
1862 	if (attr->va_size == 0)
1863 		return (ENOEXEC);
1864 
1865 	/*
1866 	 *  Check for execute permission to file based on current credentials.
1867 	 */
1868 	error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1869 	if (error)
1870 		return (error);
1871 
1872 	/*
1873 	 * Check number of open-for-writes on the file and deny execution
1874 	 * if there are any.
1875 	 *
1876 	 * Add a text reference now so no one can write to the
1877 	 * executable while we're activating it.
1878 	 *
1879 	 * Remember if this was set before and unset it in case this is not
1880 	 * actually an executable image.
1881 	 */
1882 	error = VOP_SET_TEXT(vp);
1883 	if (error != 0)
1884 		return (error);
1885 	imgp->textset = true;
1886 
1887 	/*
1888 	 * Call filesystem specific open routine (which does nothing in the
1889 	 * general case).
1890 	 */
1891 	error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1892 	if (error == 0)
1893 		imgp->opened = true;
1894 	return (error);
1895 }
1896 
1897 /*
1898  * Exec handler registration
1899  */
1900 int
1901 exec_register(const struct execsw *execsw_arg)
1902 {
1903 	const struct execsw **es, **xs, **newexecsw;
1904 	u_int count = 2;	/* New slot and trailing NULL */
1905 
1906 	if (execsw)
1907 		for (es = execsw; *es; es++)
1908 			count++;
1909 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1910 	xs = newexecsw;
1911 	if (execsw)
1912 		for (es = execsw; *es; es++)
1913 			*xs++ = *es;
1914 	*xs++ = execsw_arg;
1915 	*xs = NULL;
1916 	if (execsw)
1917 		free(execsw, M_TEMP);
1918 	execsw = newexecsw;
1919 	return (0);
1920 }
1921 
1922 int
1923 exec_unregister(const struct execsw *execsw_arg)
1924 {
1925 	const struct execsw **es, **xs, **newexecsw;
1926 	int count = 1;
1927 
1928 	if (execsw == NULL)
1929 		panic("unregister with no handlers left?\n");
1930 
1931 	for (es = execsw; *es; es++) {
1932 		if (*es == execsw_arg)
1933 			break;
1934 	}
1935 	if (*es == NULL)
1936 		return (ENOENT);
1937 	for (es = execsw; *es; es++)
1938 		if (*es != execsw_arg)
1939 			count++;
1940 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1941 	xs = newexecsw;
1942 	for (es = execsw; *es; es++)
1943 		if (*es != execsw_arg)
1944 			*xs++ = *es;
1945 	*xs = NULL;
1946 	if (execsw)
1947 		free(execsw, M_TEMP);
1948 	execsw = newexecsw;
1949 	return (0);
1950 }
1951 
1952 /*
1953  * Write out a core segment to the compression stream.
1954  */
1955 static int
1956 compress_chunk(struct coredump_params *cp, char *base, char *buf, size_t len)
1957 {
1958 	size_t chunk_len;
1959 	int error;
1960 
1961 	error = 0;
1962 	while (len > 0) {
1963 		chunk_len = MIN(len, CORE_BUF_SIZE);
1964 
1965 		/*
1966 		 * We can get EFAULT error here.
1967 		 * In that case zero out the current chunk of the segment.
1968 		 */
1969 		error = copyin(base, buf, chunk_len);
1970 		if (error != 0)
1971 			bzero(buf, chunk_len);
1972 		error = compressor_write(cp->comp, buf, chunk_len);
1973 		if (error != 0)
1974 			break;
1975 		base += chunk_len;
1976 		len -= chunk_len;
1977 	}
1978 	return (error);
1979 }
1980 
1981 int
1982 core_write(struct coredump_params *cp, const void *base, size_t len,
1983     off_t offset, enum uio_seg seg, size_t *resid)
1984 {
1985 
1986 	return (vn_rdwr_inchunks(UIO_WRITE, cp->vp, __DECONST(void *, base),
1987 	    len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1988 	    cp->active_cred, cp->file_cred, resid, cp->td));
1989 }
1990 
1991 int
1992 core_output(char *base, size_t len, off_t offset, struct coredump_params *cp,
1993     void *tmpbuf)
1994 {
1995 	vm_map_t map;
1996 	struct mount *mp;
1997 	size_t resid, runlen;
1998 	int error;
1999 	bool success;
2000 
2001 	KASSERT((uintptr_t)base % PAGE_SIZE == 0,
2002 	    ("%s: user address %p is not page-aligned", __func__, base));
2003 
2004 	if (cp->comp != NULL)
2005 		return (compress_chunk(cp, base, tmpbuf, len));
2006 
2007 	error = 0;
2008 	map = &cp->td->td_proc->p_vmspace->vm_map;
2009 	for (; len > 0; base += runlen, offset += runlen, len -= runlen) {
2010 		/*
2011 		 * Attempt to page in all virtual pages in the range.  If a
2012 		 * virtual page is not backed by the pager, it is represented as
2013 		 * a hole in the file.  This can occur with zero-filled
2014 		 * anonymous memory or truncated files, for example.
2015 		 */
2016 		for (runlen = 0; runlen < len; runlen += PAGE_SIZE) {
2017 			if (core_dump_can_intr && curproc_sigkilled())
2018 				return (EINTR);
2019 			error = vm_fault(map, (uintptr_t)base + runlen,
2020 			    VM_PROT_READ, VM_FAULT_NOFILL, NULL);
2021 			if (runlen == 0)
2022 				success = error == KERN_SUCCESS;
2023 			else if ((error == KERN_SUCCESS) != success)
2024 				break;
2025 		}
2026 
2027 		if (success) {
2028 			error = core_write(cp, base, runlen, offset,
2029 			    UIO_USERSPACE, &resid);
2030 			if (error != 0) {
2031 				if (error != EFAULT)
2032 					break;
2033 
2034 				/*
2035 				 * EFAULT may be returned if the user mapping
2036 				 * could not be accessed, e.g., because a mapped
2037 				 * file has been truncated.  Skip the page if no
2038 				 * progress was made, to protect against a
2039 				 * hypothetical scenario where vm_fault() was
2040 				 * successful but core_write() returns EFAULT
2041 				 * anyway.
2042 				 */
2043 				runlen -= resid;
2044 				if (runlen == 0) {
2045 					success = false;
2046 					runlen = PAGE_SIZE;
2047 				}
2048 			}
2049 		}
2050 		if (!success) {
2051 			error = vn_start_write(cp->vp, &mp, V_WAIT);
2052 			if (error != 0)
2053 				break;
2054 			vn_lock(cp->vp, LK_EXCLUSIVE | LK_RETRY);
2055 			error = vn_truncate_locked(cp->vp, offset + runlen,
2056 			    false, cp->td->td_ucred);
2057 			VOP_UNLOCK(cp->vp);
2058 			vn_finished_write(mp);
2059 			if (error != 0)
2060 				break;
2061 		}
2062 	}
2063 	return (error);
2064 }
2065 
2066 /*
2067  * Drain into a core file.
2068  */
2069 int
2070 sbuf_drain_core_output(void *arg, const char *data, int len)
2071 {
2072 	struct coredump_params *cp;
2073 	struct proc *p;
2074 	int error, locked;
2075 
2076 	cp = arg;
2077 	p = cp->td->td_proc;
2078 
2079 	/*
2080 	 * Some kern_proc out routines that print to this sbuf may
2081 	 * call us with the process lock held. Draining with the
2082 	 * non-sleepable lock held is unsafe. The lock is needed for
2083 	 * those routines when dumping a live process. In our case we
2084 	 * can safely release the lock before draining and acquire
2085 	 * again after.
2086 	 */
2087 	locked = PROC_LOCKED(p);
2088 	if (locked)
2089 		PROC_UNLOCK(p);
2090 	if (cp->comp != NULL)
2091 		error = compressor_write(cp->comp, __DECONST(char *, data),
2092 		    len);
2093 	else
2094 		error = core_write(cp, __DECONST(void *, data), len, cp->offset,
2095 		    UIO_SYSSPACE, NULL);
2096 	if (locked)
2097 		PROC_LOCK(p);
2098 	if (error != 0)
2099 		return (-error);
2100 	cp->offset += len;
2101 	return (len);
2102 }
2103