xref: /freebsd/sys/kern/kern_exec.c (revision e12ff891366cf94db4bfe4c2c810b26a5531053d)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 1993, David Greenman
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include "opt_capsicum.h"
33 #include "opt_hwpmc_hooks.h"
34 #include "opt_ktrace.h"
35 #include "opt_vm.h"
36 
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/acct.h>
40 #include <sys/capsicum.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/pioctl.h>
55 #include <sys/priv.h>
56 #include <sys/proc.h>
57 #include <sys/ptrace.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/vnode.h>
72 #include <sys/wait.h>
73 #ifdef KTRACE
74 #include <sys/ktrace.h>
75 #endif
76 
77 #include <vm/vm.h>
78 #include <vm/vm_param.h>
79 #include <vm/pmap.h>
80 #include <vm/vm_page.h>
81 #include <vm/vm_map.h>
82 #include <vm/vm_kern.h>
83 #include <vm/vm_extern.h>
84 #include <vm/vm_object.h>
85 #include <vm/vm_pager.h>
86 
87 #ifdef	HWPMC_HOOKS
88 #include <sys/pmckern.h>
89 #endif
90 
91 #include <machine/reg.h>
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);
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 
128 /* XXX This should be vm_size_t. */
129 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD|
130     CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_usrstack, "LU", "");
131 
132 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD|CTLFLAG_MPSAFE,
133     NULL, 0, sysctl_kern_stackprot, "I", "");
134 
135 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
136 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
137     &ps_arg_cache_limit, 0, "");
138 
139 static int disallow_high_osrel;
140 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW,
141     &disallow_high_osrel, 0,
142     "Disallow execution of binaries built for higher version of the world");
143 
144 static int map_at_zero = 0;
145 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RWTUN, &map_at_zero, 0,
146     "Permit processes to map an object at virtual address 0.");
147 
148 static int
149 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
150 {
151 	struct proc *p;
152 	int error;
153 
154 	p = curproc;
155 #ifdef SCTL_MASK32
156 	if (req->flags & SCTL_MASK32) {
157 		unsigned int val;
158 		val = (unsigned int)p->p_sysent->sv_psstrings;
159 		error = SYSCTL_OUT(req, &val, sizeof(val));
160 	} else
161 #endif
162 		error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
163 		   sizeof(p->p_sysent->sv_psstrings));
164 	return error;
165 }
166 
167 static int
168 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
169 {
170 	struct proc *p;
171 	int error;
172 
173 	p = curproc;
174 #ifdef SCTL_MASK32
175 	if (req->flags & SCTL_MASK32) {
176 		unsigned int val;
177 		val = (unsigned int)p->p_sysent->sv_usrstack;
178 		error = SYSCTL_OUT(req, &val, sizeof(val));
179 	} else
180 #endif
181 		error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
182 		    sizeof(p->p_sysent->sv_usrstack));
183 	return error;
184 }
185 
186 static int
187 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
188 {
189 	struct proc *p;
190 
191 	p = curproc;
192 	return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
193 	    sizeof(p->p_sysent->sv_stackprot)));
194 }
195 
196 /*
197  * Each of the items is a pointer to a `const struct execsw', hence the
198  * double pointer here.
199  */
200 static const struct execsw **execsw;
201 
202 #ifndef _SYS_SYSPROTO_H_
203 struct execve_args {
204 	char    *fname;
205 	char    **argv;
206 	char    **envv;
207 };
208 #endif
209 
210 int
211 sys_execve(struct thread *td, struct execve_args *uap)
212 {
213 	struct image_args args;
214 	struct vmspace *oldvmspace;
215 	int error;
216 
217 	error = pre_execve(td, &oldvmspace);
218 	if (error != 0)
219 		return (error);
220 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
221 	    uap->argv, uap->envv);
222 	if (error == 0)
223 		error = kern_execve(td, &args, NULL);
224 	post_execve(td, error, oldvmspace);
225 	return (error);
226 }
227 
228 #ifndef _SYS_SYSPROTO_H_
229 struct fexecve_args {
230 	int	fd;
231 	char	**argv;
232 	char	**envv;
233 }
234 #endif
235 int
236 sys_fexecve(struct thread *td, struct fexecve_args *uap)
237 {
238 	struct image_args args;
239 	struct vmspace *oldvmspace;
240 	int error;
241 
242 	error = pre_execve(td, &oldvmspace);
243 	if (error != 0)
244 		return (error);
245 	error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
246 	    uap->argv, uap->envv);
247 	if (error == 0) {
248 		args.fd = uap->fd;
249 		error = kern_execve(td, &args, NULL);
250 	}
251 	post_execve(td, error, oldvmspace);
252 	return (error);
253 }
254 
255 #ifndef _SYS_SYSPROTO_H_
256 struct __mac_execve_args {
257 	char	*fname;
258 	char	**argv;
259 	char	**envv;
260 	struct mac	*mac_p;
261 };
262 #endif
263 
264 int
265 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap)
266 {
267 #ifdef MAC
268 	struct image_args args;
269 	struct vmspace *oldvmspace;
270 	int error;
271 
272 	error = pre_execve(td, &oldvmspace);
273 	if (error != 0)
274 		return (error);
275 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
276 	    uap->argv, uap->envv);
277 	if (error == 0)
278 		error = kern_execve(td, &args, uap->mac_p);
279 	post_execve(td, error, oldvmspace);
280 	return (error);
281 #else
282 	return (ENOSYS);
283 #endif
284 }
285 
286 int
287 pre_execve(struct thread *td, struct vmspace **oldvmspace)
288 {
289 	struct proc *p;
290 	int error;
291 
292 	KASSERT(td == curthread, ("non-current thread %p", td));
293 	error = 0;
294 	p = td->td_proc;
295 	if ((p->p_flag & P_HADTHREADS) != 0) {
296 		PROC_LOCK(p);
297 		if (thread_single(p, SINGLE_BOUNDARY) != 0)
298 			error = ERESTART;
299 		PROC_UNLOCK(p);
300 	}
301 	KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0,
302 	    ("nested execve"));
303 	*oldvmspace = p->p_vmspace;
304 	return (error);
305 }
306 
307 void
308 post_execve(struct thread *td, int error, struct vmspace *oldvmspace)
309 {
310 	struct proc *p;
311 
312 	KASSERT(td == curthread, ("non-current thread %p", td));
313 	p = td->td_proc;
314 	if ((p->p_flag & P_HADTHREADS) != 0) {
315 		PROC_LOCK(p);
316 		/*
317 		 * If success, we upgrade to SINGLE_EXIT state to
318 		 * force other threads to suicide.
319 		 */
320 		if (error == EJUSTRETURN)
321 			thread_single(p, SINGLE_EXIT);
322 		else
323 			thread_single_end(p, SINGLE_BOUNDARY);
324 		PROC_UNLOCK(p);
325 	}
326 	if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
327 		KASSERT(p->p_vmspace != oldvmspace,
328 		    ("oldvmspace still used"));
329 		vmspace_free(oldvmspace);
330 		td->td_pflags &= ~TDP_EXECVMSPC;
331 	}
332 }
333 
334 /*
335  * XXX: kern_execve has the astonishing property of not always returning to
336  * the caller.  If sufficiently bad things happen during the call to
337  * do_execve(), it can end up calling exit1(); as a result, callers must
338  * avoid doing anything which they might need to undo (e.g., allocating
339  * memory).
340  */
341 int
342 kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p)
343 {
344 
345 	AUDIT_ARG_ARGV(args->begin_argv, args->argc,
346 	    exec_args_get_begin_envv(args) - args->begin_argv);
347 	AUDIT_ARG_ENVV(exec_args_get_begin_envv(args), args->envc,
348 	    args->endp - exec_args_get_begin_envv(args));
349 	return (do_execve(td, args, mac_p));
350 }
351 
352 /*
353  * In-kernel implementation of execve().  All arguments are assumed to be
354  * userspace pointers from the passed thread.
355  */
356 static int
357 do_execve(struct thread *td, struct image_args *args, struct mac *mac_p)
358 {
359 	struct proc *p = td->td_proc;
360 	struct nameidata nd;
361 	struct ucred *oldcred;
362 	struct uidinfo *euip = NULL;
363 	register_t *stack_base;
364 	int error, i;
365 	struct image_params image_params, *imgp;
366 	struct vattr attr;
367 	int (*img_first)(struct image_params *);
368 	struct pargs *oldargs = NULL, *newargs = NULL;
369 	struct sigacts *oldsigacts = NULL, *newsigacts = NULL;
370 #ifdef KTRACE
371 	struct vnode *tracevp = NULL;
372 	struct ucred *tracecred = NULL;
373 #endif
374 	struct vnode *oldtextvp = NULL, *newtextvp;
375 	int credential_changing;
376 #ifdef MAC
377 	struct label *interpvplabel = NULL;
378 	int will_transition;
379 #endif
380 #ifdef HWPMC_HOOKS
381 	struct pmckern_procexec pe;
382 #endif
383 	static const char fexecv_proc_title[] = "(fexecv)";
384 
385 	imgp = &image_params;
386 
387 	/*
388 	 * Lock the process and set the P_INEXEC flag to indicate that
389 	 * it should be left alone until we're done here.  This is
390 	 * necessary to avoid race conditions - e.g. in ptrace() -
391 	 * that might allow a local user to illicitly obtain elevated
392 	 * privileges.
393 	 */
394 	PROC_LOCK(p);
395 	KASSERT((p->p_flag & P_INEXEC) == 0,
396 	    ("%s(): process already has P_INEXEC flag", __func__));
397 	p->p_flag |= P_INEXEC;
398 	PROC_UNLOCK(p);
399 
400 	/*
401 	 * Initialize part of the common data
402 	 */
403 	bzero(imgp, sizeof(*imgp));
404 	imgp->proc = p;
405 	imgp->attr = &attr;
406 	imgp->args = args;
407 	oldcred = p->p_ucred;
408 
409 #ifdef MAC
410 	error = mac_execve_enter(imgp, mac_p);
411 	if (error)
412 		goto exec_fail;
413 #endif
414 
415 	/*
416 	 * Translate the file name. namei() returns a vnode pointer
417 	 *	in ni_vp among other things.
418 	 *
419 	 * XXXAUDIT: It would be desirable to also audit the name of the
420 	 * interpreter if this is an interpreted binary.
421 	 */
422 	if (args->fname != NULL) {
423 		NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW |
424 		    SAVENAME | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
425 	}
426 
427 	SDT_PROBE1(proc, , , exec, args->fname);
428 
429 interpret:
430 	if (args->fname != NULL) {
431 #ifdef CAPABILITY_MODE
432 		/*
433 		 * While capability mode can't reach this point via direct
434 		 * path arguments to execve(), we also don't allow
435 		 * interpreters to be used in capability mode (for now).
436 		 * Catch indirect lookups and return a permissions error.
437 		 */
438 		if (IN_CAPABILITY_MODE(td)) {
439 			error = ECAPMODE;
440 			goto exec_fail;
441 		}
442 #endif
443 		error = namei(&nd);
444 		if (error)
445 			goto exec_fail;
446 
447 		newtextvp = nd.ni_vp;
448 		imgp->vp = newtextvp;
449 	} else {
450 		AUDIT_ARG_FD(args->fd);
451 		/*
452 		 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
453 		 */
454 		error = fgetvp_exec(td, args->fd, &cap_fexecve_rights, &newtextvp);
455 		if (error)
456 			goto exec_fail;
457 		vn_lock(newtextvp, LK_SHARED | LK_RETRY);
458 		AUDIT_ARG_VNODE1(newtextvp);
459 		imgp->vp = newtextvp;
460 	}
461 
462 	/*
463 	 * Check file permissions.  Also 'opens' file and sets its vnode to
464 	 * text mode.
465 	 */
466 	error = exec_check_permissions(imgp);
467 	if (error)
468 		goto exec_fail_dealloc;
469 
470 	imgp->object = imgp->vp->v_object;
471 	if (imgp->object != NULL)
472 		vm_object_reference(imgp->object);
473 
474 	error = exec_map_first_page(imgp);
475 	if (error)
476 		goto exec_fail_dealloc;
477 
478 	imgp->proc->p_osrel = 0;
479 	imgp->proc->p_fctl0 = 0;
480 
481 	/*
482 	 * Implement image setuid/setgid.
483 	 *
484 	 * Determine new credentials before attempting image activators
485 	 * so that it can be used by process_exec handlers to determine
486 	 * credential/setid changes.
487 	 *
488 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
489 	 * the process is being traced.
490 	 *
491 	 * We disable setuid/setgid/etc in capability mode on the basis
492 	 * that most setugid applications are not written with that
493 	 * environment in mind, and will therefore almost certainly operate
494 	 * incorrectly. In principle there's no reason that setugid
495 	 * applications might not be useful in capability mode, so we may want
496 	 * to reconsider this conservative design choice in the future.
497 	 *
498 	 * XXXMAC: For the time being, use NOSUID to also prohibit
499 	 * transitions on the file system.
500 	 */
501 	credential_changing = 0;
502 	credential_changing |= (attr.va_mode & S_ISUID) &&
503 	    oldcred->cr_uid != attr.va_uid;
504 	credential_changing |= (attr.va_mode & S_ISGID) &&
505 	    oldcred->cr_gid != attr.va_gid;
506 #ifdef MAC
507 	will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
508 	    interpvplabel, imgp);
509 	credential_changing |= will_transition;
510 #endif
511 
512 	/* Don't inherit PROC_PDEATHSIG_CTL value if setuid/setgid. */
513 	if (credential_changing)
514 		imgp->proc->p_pdeathsig = 0;
515 
516 	if (credential_changing &&
517 #ifdef CAPABILITY_MODE
518 	    ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
519 #endif
520 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
521 	    (p->p_flag & P_TRACED) == 0) {
522 		imgp->credential_setid = true;
523 		VOP_UNLOCK(imgp->vp, 0);
524 		imgp->newcred = crdup(oldcred);
525 		if (attr.va_mode & S_ISUID) {
526 			euip = uifind(attr.va_uid);
527 			change_euid(imgp->newcred, euip);
528 		}
529 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
530 		if (attr.va_mode & S_ISGID)
531 			change_egid(imgp->newcred, attr.va_gid);
532 		/*
533 		 * Implement correct POSIX saved-id behavior.
534 		 *
535 		 * XXXMAC: Note that the current logic will save the
536 		 * uid and gid if a MAC domain transition occurs, even
537 		 * though maybe it shouldn't.
538 		 */
539 		change_svuid(imgp->newcred, imgp->newcred->cr_uid);
540 		change_svgid(imgp->newcred, imgp->newcred->cr_gid);
541 	} else {
542 		/*
543 		 * Implement correct POSIX saved-id behavior.
544 		 *
545 		 * XXX: It's not clear that the existing behavior is
546 		 * POSIX-compliant.  A number of sources indicate that the
547 		 * saved uid/gid should only be updated if the new ruid is
548 		 * not equal to the old ruid, or the new euid is not equal
549 		 * to the old euid and the new euid is not equal to the old
550 		 * ruid.  The FreeBSD code always updates the saved uid/gid.
551 		 * Also, this code uses the new (replaced) euid and egid as
552 		 * the source, which may or may not be the right ones to use.
553 		 */
554 		if (oldcred->cr_svuid != oldcred->cr_uid ||
555 		    oldcred->cr_svgid != oldcred->cr_gid) {
556 			VOP_UNLOCK(imgp->vp, 0);
557 			imgp->newcred = crdup(oldcred);
558 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
559 			change_svuid(imgp->newcred, imgp->newcred->cr_uid);
560 			change_svgid(imgp->newcred, imgp->newcred->cr_gid);
561 		}
562 	}
563 	/* The new credentials are installed into the process later. */
564 
565 	/*
566 	 * Do the best to calculate the full path to the image file.
567 	 */
568 	if (args->fname != NULL && args->fname[0] == '/')
569 		imgp->execpath = args->fname;
570 	else {
571 		VOP_UNLOCK(imgp->vp, 0);
572 		if (vn_fullpath(td, imgp->vp, &imgp->execpath,
573 		    &imgp->freepath) != 0)
574 			imgp->execpath = args->fname;
575 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
576 	}
577 
578 	/*
579 	 *	If the current process has a special image activator it
580 	 *	wants to try first, call it.   For example, emulating shell
581 	 *	scripts differently.
582 	 */
583 	error = -1;
584 	if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
585 		error = img_first(imgp);
586 
587 	/*
588 	 *	Loop through the list of image activators, calling each one.
589 	 *	An activator returns -1 if there is no match, 0 on success,
590 	 *	and an error otherwise.
591 	 */
592 	for (i = 0; error == -1 && execsw[i]; ++i) {
593 		if (execsw[i]->ex_imgact == NULL ||
594 		    execsw[i]->ex_imgact == img_first) {
595 			continue;
596 		}
597 		error = (*execsw[i]->ex_imgact)(imgp);
598 	}
599 
600 	if (error) {
601 		if (error == -1)
602 			error = ENOEXEC;
603 		goto exec_fail_dealloc;
604 	}
605 
606 	/*
607 	 * Special interpreter operation, cleanup and loop up to try to
608 	 * activate the interpreter.
609 	 */
610 	if (imgp->interpreted) {
611 		exec_unmap_first_page(imgp);
612 		/*
613 		 * The text reference needs to be removed for scripts.
614 		 * There is a short period before we determine that
615 		 * something is a script where text reference is active.
616 		 * The vnode lock is held over this entire period
617 		 * so nothing should illegitimately be blocked.
618 		 */
619 		MPASS(imgp->textset);
620 		VOP_UNSET_TEXT_CHECKED(newtextvp);
621 		imgp->textset = false;
622 		/* free name buffer and old vnode */
623 		if (args->fname != NULL)
624 			NDFREE(&nd, NDF_ONLY_PNBUF);
625 #ifdef MAC
626 		mac_execve_interpreter_enter(newtextvp, &interpvplabel);
627 #endif
628 		if (imgp->opened) {
629 			VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td);
630 			imgp->opened = 0;
631 		}
632 		vput(newtextvp);
633 		vm_object_deallocate(imgp->object);
634 		imgp->object = NULL;
635 		imgp->credential_setid = false;
636 		if (imgp->newcred != NULL) {
637 			crfree(imgp->newcred);
638 			imgp->newcred = NULL;
639 		}
640 		imgp->execpath = NULL;
641 		free(imgp->freepath, M_TEMP);
642 		imgp->freepath = NULL;
643 		/* set new name to that of the interpreter */
644 		NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME,
645 		    UIO_SYSSPACE, imgp->interpreter_name, td);
646 		args->fname = imgp->interpreter_name;
647 		goto interpret;
648 	}
649 
650 	/*
651 	 * NB: We unlock the vnode here because it is believed that none
652 	 * of the sv_copyout_strings/sv_fixup operations require the vnode.
653 	 */
654 	VOP_UNLOCK(imgp->vp, 0);
655 
656 	if (disallow_high_osrel &&
657 	    P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
658 		error = ENOEXEC;
659 		uprintf("Osrel %d for image %s too high\n", p->p_osrel,
660 		    imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
661 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
662 		goto exec_fail_dealloc;
663 	}
664 
665 	/* ABI enforces the use of Capsicum. Switch into capabilities mode. */
666 	if (SV_PROC_FLAG(p, SV_CAPSICUM))
667 		sys_cap_enter(td, NULL);
668 
669 	/*
670 	 * Copy out strings (args and env) and initialize stack base.
671 	 */
672 	stack_base = (*p->p_sysent->sv_copyout_strings)(imgp);
673 
674 	/*
675 	 * Stack setup.
676 	 */
677 	error = (*p->p_sysent->sv_fixup)(&stack_base, imgp);
678 	if (error != 0) {
679 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
680 		goto exec_fail_dealloc;
681 	}
682 
683 	if (args->fdp != NULL) {
684 		/* Install a brand new file descriptor table. */
685 		fdinstall_remapped(td, args->fdp);
686 		args->fdp = NULL;
687 	} else {
688 		/*
689 		 * Keep on using the existing file descriptor table. For
690 		 * security and other reasons, the file descriptor table
691 		 * cannot be shared after an exec.
692 		 */
693 		fdunshare(td);
694 		/* close files on exec */
695 		fdcloseexec(td);
696 	}
697 
698 	/*
699 	 * Malloc things before we need locks.
700 	 */
701 	i = exec_args_get_begin_envv(imgp->args) - imgp->args->begin_argv;
702 	/* Cache arguments if they fit inside our allowance */
703 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
704 		newargs = pargs_alloc(i);
705 		bcopy(imgp->args->begin_argv, newargs->ar_args, i);
706 	}
707 
708 	/*
709 	 * For security and other reasons, signal handlers cannot
710 	 * be shared after an exec. The new process gets a copy of the old
711 	 * handlers. In execsigs(), the new process will have its signals
712 	 * reset.
713 	 */
714 	if (sigacts_shared(p->p_sigacts)) {
715 		oldsigacts = p->p_sigacts;
716 		newsigacts = sigacts_alloc();
717 		sigacts_copy(newsigacts, oldsigacts);
718 	}
719 
720 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
721 
722 	PROC_LOCK(p);
723 	if (oldsigacts)
724 		p->p_sigacts = newsigacts;
725 	/* Stop profiling */
726 	stopprofclock(p);
727 
728 	/* reset caught signals */
729 	execsigs(p);
730 
731 	/* name this process - nameiexec(p, ndp) */
732 	bzero(p->p_comm, sizeof(p->p_comm));
733 	if (args->fname)
734 		bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
735 		    min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
736 	else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0)
737 		bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
738 	bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
739 #ifdef KTR
740 	sched_clear_tdname(td);
741 #endif
742 
743 	/*
744 	 * mark as execed, wakeup the process that vforked (if any) and tell
745 	 * it that it now has its own resources back
746 	 */
747 	p->p_flag |= P_EXEC;
748 	if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0)
749 		p->p_flag2 &= ~P2_NOTRACE;
750 	if ((p->p_flag2 & P2_STKGAP_DISABLE_EXEC) == 0)
751 		p->p_flag2 &= ~P2_STKGAP_DISABLE;
752 	if (p->p_flag & P_PPWAIT) {
753 		p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
754 		cv_broadcast(&p->p_pwait);
755 		/* STOPs are no longer ignored, arrange for AST */
756 		signotify(td);
757 	}
758 
759 	/*
760 	 * Implement image setuid/setgid installation.
761 	 */
762 	if (imgp->credential_setid) {
763 		/*
764 		 * Turn off syscall tracing for set-id programs, except for
765 		 * root.  Record any set-id flags first to make sure that
766 		 * we do not regain any tracing during a possible block.
767 		 */
768 		setsugid(p);
769 
770 #ifdef KTRACE
771 		if (p->p_tracecred != NULL &&
772 		    priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED))
773 			ktrprocexec(p, &tracecred, &tracevp);
774 #endif
775 		/*
776 		 * Close any file descriptors 0..2 that reference procfs,
777 		 * then make sure file descriptors 0..2 are in use.
778 		 *
779 		 * Both fdsetugidsafety() and fdcheckstd() may call functions
780 		 * taking sleepable locks, so temporarily drop our locks.
781 		 */
782 		PROC_UNLOCK(p);
783 		VOP_UNLOCK(imgp->vp, 0);
784 		fdsetugidsafety(td);
785 		error = fdcheckstd(td);
786 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
787 		if (error != 0)
788 			goto exec_fail_dealloc;
789 		PROC_LOCK(p);
790 #ifdef MAC
791 		if (will_transition) {
792 			mac_vnode_execve_transition(oldcred, imgp->newcred,
793 			    imgp->vp, interpvplabel, imgp);
794 		}
795 #endif
796 	} else {
797 		if (oldcred->cr_uid == oldcred->cr_ruid &&
798 		    oldcred->cr_gid == oldcred->cr_rgid)
799 			p->p_flag &= ~P_SUGID;
800 	}
801 	/*
802 	 * Set the new credentials.
803 	 */
804 	if (imgp->newcred != NULL) {
805 		proc_set_cred(p, imgp->newcred);
806 		crfree(oldcred);
807 		oldcred = NULL;
808 	}
809 
810 	/*
811 	 * Store the vp for use in procfs.  This vnode was referenced by namei
812 	 * or fgetvp_exec.
813 	 */
814 	oldtextvp = p->p_textvp;
815 	p->p_textvp = newtextvp;
816 
817 #ifdef KDTRACE_HOOKS
818 	/*
819 	 * Tell the DTrace fasttrap provider about the exec if it
820 	 * has declared an interest.
821 	 */
822 	if (dtrace_fasttrap_exec)
823 		dtrace_fasttrap_exec(p);
824 #endif
825 
826 	/*
827 	 * Notify others that we exec'd, and clear the P_INEXEC flag
828 	 * as we're now a bona fide freshly-execed process.
829 	 */
830 	KNOTE_LOCKED(p->p_klist, NOTE_EXEC);
831 	p->p_flag &= ~P_INEXEC;
832 
833 	/* clear "fork but no exec" flag, as we _are_ execing */
834 	p->p_acflag &= ~AFORK;
835 
836 	/*
837 	 * Free any previous argument cache and replace it with
838 	 * the new argument cache, if any.
839 	 */
840 	oldargs = p->p_args;
841 	p->p_args = newargs;
842 	newargs = NULL;
843 
844 	PROC_UNLOCK(p);
845 
846 #ifdef	HWPMC_HOOKS
847 	/*
848 	 * Check if system-wide sampling is in effect or if the
849 	 * current process is using PMCs.  If so, do exec() time
850 	 * processing.  This processing needs to happen AFTER the
851 	 * P_INEXEC flag is cleared.
852 	 */
853 	if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
854 		VOP_UNLOCK(imgp->vp, 0);
855 		pe.pm_credentialschanged = credential_changing;
856 		pe.pm_entryaddr = imgp->entry_addr;
857 
858 		PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
859 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
860 	}
861 #endif
862 
863 	/* Set values passed into the program in registers. */
864 	(*p->p_sysent->sv_setregs)(td, imgp, (u_long)(uintptr_t)stack_base);
865 
866 	vfs_mark_atime(imgp->vp, td->td_ucred);
867 
868 	SDT_PROBE1(proc, , , exec__success, args->fname);
869 
870 exec_fail_dealloc:
871 	if (imgp->firstpage != NULL)
872 		exec_unmap_first_page(imgp);
873 
874 	if (imgp->vp != NULL) {
875 		if (args->fname)
876 			NDFREE(&nd, NDF_ONLY_PNBUF);
877 		if (imgp->opened)
878 			VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
879 		if (imgp->textset)
880 			VOP_UNSET_TEXT_CHECKED(imgp->vp);
881 		if (error != 0)
882 			vput(imgp->vp);
883 		else
884 			VOP_UNLOCK(imgp->vp, 0);
885 	}
886 
887 	if (imgp->object != NULL)
888 		vm_object_deallocate(imgp->object);
889 
890 	free(imgp->freepath, M_TEMP);
891 
892 	if (error == 0) {
893 		if (p->p_ptevents & PTRACE_EXEC) {
894 			PROC_LOCK(p);
895 			if (p->p_ptevents & PTRACE_EXEC)
896 				td->td_dbgflags |= TDB_EXEC;
897 			PROC_UNLOCK(p);
898 		}
899 
900 		/*
901 		 * Stop the process here if its stop event mask has
902 		 * the S_EXEC bit set.
903 		 */
904 		STOPEVENT(p, S_EXEC, 0);
905 	} else {
906 exec_fail:
907 		/* we're done here, clear P_INEXEC */
908 		PROC_LOCK(p);
909 		p->p_flag &= ~P_INEXEC;
910 		PROC_UNLOCK(p);
911 
912 		SDT_PROBE1(proc, , , exec__failure, error);
913 	}
914 
915 	if (imgp->newcred != NULL && oldcred != NULL)
916 		crfree(imgp->newcred);
917 
918 #ifdef MAC
919 	mac_execve_exit(imgp);
920 	mac_execve_interpreter_exit(interpvplabel);
921 #endif
922 	exec_free_args(args);
923 
924 	/*
925 	 * Handle deferred decrement of ref counts.
926 	 */
927 	if (oldtextvp != NULL)
928 		vrele(oldtextvp);
929 #ifdef KTRACE
930 	if (tracevp != NULL)
931 		vrele(tracevp);
932 	if (tracecred != NULL)
933 		crfree(tracecred);
934 #endif
935 	pargs_drop(oldargs);
936 	pargs_drop(newargs);
937 	if (oldsigacts != NULL)
938 		sigacts_free(oldsigacts);
939 	if (euip != NULL)
940 		uifree(euip);
941 
942 	if (error && imgp->vmspace_destroyed) {
943 		/* sorry, no more process anymore. exit gracefully */
944 		exit1(td, 0, SIGABRT);
945 		/* NOT REACHED */
946 	}
947 
948 #ifdef KTRACE
949 	if (error == 0)
950 		ktrprocctor(p);
951 #endif
952 
953 	/*
954 	 * We don't want cpu_set_syscall_retval() to overwrite any of
955 	 * the register values put in place by exec_setregs().
956 	 * Implementations of cpu_set_syscall_retval() will leave
957 	 * registers unmodified when returning EJUSTRETURN.
958 	 */
959 	return (error == 0 ? EJUSTRETURN : error);
960 }
961 
962 int
963 exec_map_first_page(struct image_params *imgp)
964 {
965 	int rv, i, after, initial_pagein;
966 	vm_page_t ma[VM_INITIAL_PAGEIN];
967 	vm_object_t object;
968 
969 	if (imgp->firstpage != NULL)
970 		exec_unmap_first_page(imgp);
971 
972 	object = imgp->vp->v_object;
973 	if (object == NULL)
974 		return (EACCES);
975 	VM_OBJECT_WLOCK(object);
976 #if VM_NRESERVLEVEL > 0
977 	vm_object_color(object, 0);
978 #endif
979 	ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY |
980 	    VM_ALLOC_WIRED);
981 	if (ma[0]->valid != VM_PAGE_BITS_ALL) {
982 		vm_page_xbusy(ma[0]);
983 		if (!vm_pager_has_page(object, 0, NULL, &after)) {
984 			if (vm_page_unwire_noq(ma[0]))
985 				vm_page_free(ma[0]);
986 			else
987 				vm_page_xunbusy(ma[0]);
988 			VM_OBJECT_WUNLOCK(object);
989 			return (EIO);
990 		}
991 		initial_pagein = min(after, VM_INITIAL_PAGEIN);
992 		KASSERT(initial_pagein <= object->size,
993 		    ("%s: initial_pagein %d object->size %ju",
994 		    __func__, initial_pagein, (uintmax_t )object->size));
995 		for (i = 1; i < initial_pagein; i++) {
996 			if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) {
997 				if (ma[i]->valid)
998 					break;
999 				if (!vm_page_tryxbusy(ma[i]))
1000 					break;
1001 			} else {
1002 				ma[i] = vm_page_alloc(object, i,
1003 				    VM_ALLOC_NORMAL);
1004 				if (ma[i] == NULL)
1005 					break;
1006 			}
1007 		}
1008 		initial_pagein = i;
1009 		rv = vm_pager_get_pages(object, ma, initial_pagein, NULL, NULL);
1010 		if (rv != VM_PAGER_OK) {
1011 			if (vm_page_unwire_noq(ma[0]))
1012 				vm_page_free(ma[0]);
1013 			else
1014 				vm_page_xunbusy(ma[0]);
1015 			for (i = 1; i < initial_pagein; i++) {
1016 				if (!vm_page_wired(ma[i]))
1017 					vm_page_free(ma[i]);
1018 				else
1019 					vm_page_xunbusy(ma[i]);
1020 			}
1021 			VM_OBJECT_WUNLOCK(object);
1022 			return (EIO);
1023 		}
1024 		vm_page_xunbusy(ma[0]);
1025 		for (i = 1; i < initial_pagein; i++)
1026 			vm_page_readahead_finish(ma[i]);
1027 	}
1028 	VM_OBJECT_WUNLOCK(object);
1029 
1030 	imgp->firstpage = sf_buf_alloc(ma[0], 0);
1031 	imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
1032 
1033 	return (0);
1034 }
1035 
1036 void
1037 exec_unmap_first_page(struct image_params *imgp)
1038 {
1039 	vm_page_t m;
1040 
1041 	if (imgp->firstpage != NULL) {
1042 		m = sf_buf_page(imgp->firstpage);
1043 		sf_buf_free(imgp->firstpage);
1044 		imgp->firstpage = NULL;
1045 		vm_page_unwire(m, PQ_ACTIVE);
1046 	}
1047 }
1048 
1049 /*
1050  * Destroy old address space, and allocate a new stack.
1051  *	The new stack is only sgrowsiz large because it is grown
1052  *	automatically on a page fault.
1053  */
1054 int
1055 exec_new_vmspace(struct image_params *imgp, struct sysentvec *sv)
1056 {
1057 	int error;
1058 	struct proc *p = imgp->proc;
1059 	struct vmspace *vmspace = p->p_vmspace;
1060 	vm_object_t obj;
1061 	struct rlimit rlim_stack;
1062 	vm_offset_t sv_minuser, stack_addr;
1063 	vm_map_t map;
1064 	u_long ssiz;
1065 
1066 	imgp->vmspace_destroyed = 1;
1067 	imgp->sysent = sv;
1068 
1069 	/* May be called with Giant held */
1070 	EVENTHANDLER_DIRECT_INVOKE(process_exec, p, imgp);
1071 
1072 	/*
1073 	 * Blow away entire process VM, if address space not shared,
1074 	 * otherwise, create a new VM space so that other threads are
1075 	 * not disrupted
1076 	 */
1077 	map = &vmspace->vm_map;
1078 	if (map_at_zero)
1079 		sv_minuser = sv->sv_minuser;
1080 	else
1081 		sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1082 	if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser &&
1083 	    vm_map_max(map) == sv->sv_maxuser &&
1084 	    cpu_exec_vmspace_reuse(p, map)) {
1085 		shmexit(vmspace);
1086 		pmap_remove_pages(vmspace_pmap(vmspace));
1087 		vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1088 		/*
1089 		 * An exec terminates mlockall(MCL_FUTURE), ASLR state
1090 		 * must be re-evaluated.
1091 		 */
1092 		vm_map_lock(map);
1093 		vm_map_modflags(map, 0, MAP_WIREFUTURE | MAP_ASLR |
1094 		    MAP_ASLR_IGNSTART);
1095 		vm_map_unlock(map);
1096 	} else {
1097 		error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1098 		if (error)
1099 			return (error);
1100 		vmspace = p->p_vmspace;
1101 		map = &vmspace->vm_map;
1102 	}
1103 	map->flags |= imgp->map_flags;
1104 
1105 	/* Map a shared page */
1106 	obj = sv->sv_shared_page_obj;
1107 	if (obj != NULL) {
1108 		vm_object_reference(obj);
1109 		error = vm_map_fixed(map, obj, 0,
1110 		    sv->sv_shared_page_base, sv->sv_shared_page_len,
1111 		    VM_PROT_READ | VM_PROT_EXECUTE,
1112 		    VM_PROT_READ | VM_PROT_EXECUTE,
1113 		    MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1114 		if (error != KERN_SUCCESS) {
1115 			vm_object_deallocate(obj);
1116 			return (vm_mmap_to_errno(error));
1117 		}
1118 	}
1119 
1120 	/* Allocate a new stack */
1121 	if (imgp->stack_sz != 0) {
1122 		ssiz = trunc_page(imgp->stack_sz);
1123 		PROC_LOCK(p);
1124 		lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack);
1125 		PROC_UNLOCK(p);
1126 		if (ssiz > rlim_stack.rlim_max)
1127 			ssiz = rlim_stack.rlim_max;
1128 		if (ssiz > rlim_stack.rlim_cur) {
1129 			rlim_stack.rlim_cur = ssiz;
1130 			kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack);
1131 		}
1132 	} else if (sv->sv_maxssiz != NULL) {
1133 		ssiz = *sv->sv_maxssiz;
1134 	} else {
1135 		ssiz = maxssiz;
1136 	}
1137 	imgp->eff_stack_sz = lim_cur(curthread, RLIMIT_STACK);
1138 	if (ssiz < imgp->eff_stack_sz)
1139 		imgp->eff_stack_sz = ssiz;
1140 	stack_addr = sv->sv_usrstack - ssiz;
1141 	error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1142 	    obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1143 	    sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1144 	if (error != KERN_SUCCESS)
1145 		return (vm_mmap_to_errno(error));
1146 
1147 	/*
1148 	 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they
1149 	 * are still used to enforce the stack rlimit on the process stack.
1150 	 */
1151 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1152 	vmspace->vm_maxsaddr = (char *)stack_addr;
1153 
1154 	return (0);
1155 }
1156 
1157 /*
1158  * Copy out argument and environment strings from the old process address
1159  * space into the temporary string buffer.
1160  */
1161 int
1162 exec_copyin_args(struct image_args *args, const char *fname,
1163     enum uio_seg segflg, char **argv, char **envv)
1164 {
1165 	u_long arg, env;
1166 	int error;
1167 
1168 	bzero(args, sizeof(*args));
1169 	if (argv == NULL)
1170 		return (EFAULT);
1171 
1172 	/*
1173 	 * Allocate demand-paged memory for the file name, argument, and
1174 	 * environment strings.
1175 	 */
1176 	error = exec_alloc_args(args);
1177 	if (error != 0)
1178 		return (error);
1179 
1180 	/*
1181 	 * Copy the file name.
1182 	 */
1183 	error = exec_args_add_fname(args, fname, segflg);
1184 	if (error != 0)
1185 		goto err_exit;
1186 
1187 	/*
1188 	 * extract arguments first
1189 	 */
1190 	for (;;) {
1191 		error = fueword(argv++, &arg);
1192 		if (error == -1) {
1193 			error = EFAULT;
1194 			goto err_exit;
1195 		}
1196 		if (arg == 0)
1197 			break;
1198 		error = exec_args_add_arg(args, (char *)(uintptr_t)arg,
1199 		    UIO_USERSPACE);
1200 		if (error != 0)
1201 			goto err_exit;
1202 	}
1203 
1204 	/*
1205 	 * extract environment strings
1206 	 */
1207 	if (envv) {
1208 		for (;;) {
1209 			error = fueword(envv++, &env);
1210 			if (error == -1) {
1211 				error = EFAULT;
1212 				goto err_exit;
1213 			}
1214 			if (env == 0)
1215 				break;
1216 			error = exec_args_add_env(args,
1217 			    (char *)(uintptr_t)env, UIO_USERSPACE);
1218 			if (error != 0)
1219 				goto err_exit;
1220 		}
1221 	}
1222 
1223 	return (0);
1224 
1225 err_exit:
1226 	exec_free_args(args);
1227 	return (error);
1228 }
1229 
1230 int
1231 exec_copyin_data_fds(struct thread *td, struct image_args *args,
1232     const void *data, size_t datalen, const int *fds, size_t fdslen)
1233 {
1234 	struct filedesc *ofdp;
1235 	const char *p;
1236 	int *kfds;
1237 	int error;
1238 
1239 	memset(args, '\0', sizeof(*args));
1240 	ofdp = td->td_proc->p_fd;
1241 	if (datalen >= ARG_MAX || fdslen > ofdp->fd_lastfile + 1)
1242 		return (E2BIG);
1243 	error = exec_alloc_args(args);
1244 	if (error != 0)
1245 		return (error);
1246 
1247 	args->begin_argv = args->buf;
1248 	args->stringspace = ARG_MAX;
1249 
1250 	if (datalen > 0) {
1251 		/*
1252 		 * Argument buffer has been provided. Copy it into the
1253 		 * kernel as a single string and add a terminating null
1254 		 * byte.
1255 		 */
1256 		error = copyin(data, args->begin_argv, datalen);
1257 		if (error != 0)
1258 			goto err_exit;
1259 		args->begin_argv[datalen] = '\0';
1260 		args->endp = args->begin_argv + datalen + 1;
1261 		args->stringspace -= datalen + 1;
1262 
1263 		/*
1264 		 * Traditional argument counting. Count the number of
1265 		 * null bytes.
1266 		 */
1267 		for (p = args->begin_argv; p < args->endp; ++p)
1268 			if (*p == '\0')
1269 				++args->argc;
1270 	} else {
1271 		/* No argument buffer provided. */
1272 		args->endp = args->begin_argv;
1273 	}
1274 
1275 	/* Create new file descriptor table. */
1276 	kfds = malloc(fdslen * sizeof(int), M_TEMP, M_WAITOK);
1277 	error = copyin(fds, kfds, fdslen * sizeof(int));
1278 	if (error != 0) {
1279 		free(kfds, M_TEMP);
1280 		goto err_exit;
1281 	}
1282 	error = fdcopy_remapped(ofdp, kfds, fdslen, &args->fdp);
1283 	free(kfds, M_TEMP);
1284 	if (error != 0)
1285 		goto err_exit;
1286 
1287 	return (0);
1288 err_exit:
1289 	exec_free_args(args);
1290 	return (error);
1291 }
1292 
1293 struct exec_args_kva {
1294 	vm_offset_t addr;
1295 	u_int gen;
1296 	SLIST_ENTRY(exec_args_kva) next;
1297 };
1298 
1299 DPCPU_DEFINE_STATIC(struct exec_args_kva *, exec_args_kva);
1300 
1301 static SLIST_HEAD(, exec_args_kva) exec_args_kva_freelist;
1302 static struct mtx exec_args_kva_mtx;
1303 static u_int exec_args_gen;
1304 
1305 static void
1306 exec_prealloc_args_kva(void *arg __unused)
1307 {
1308 	struct exec_args_kva *argkva;
1309 	u_int i;
1310 
1311 	SLIST_INIT(&exec_args_kva_freelist);
1312 	mtx_init(&exec_args_kva_mtx, "exec args kva", NULL, MTX_DEF);
1313 	for (i = 0; i < exec_map_entries; i++) {
1314 		argkva = malloc(sizeof(*argkva), M_PARGS, M_WAITOK);
1315 		argkva->addr = kmap_alloc_wait(exec_map, exec_map_entry_size);
1316 		argkva->gen = exec_args_gen;
1317 		SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1318 	}
1319 }
1320 SYSINIT(exec_args_kva, SI_SUB_EXEC, SI_ORDER_ANY, exec_prealloc_args_kva, NULL);
1321 
1322 static vm_offset_t
1323 exec_alloc_args_kva(void **cookie)
1324 {
1325 	struct exec_args_kva *argkva;
1326 
1327 	argkva = (void *)atomic_readandclear_ptr(
1328 	    (uintptr_t *)DPCPU_PTR(exec_args_kva));
1329 	if (argkva == NULL) {
1330 		mtx_lock(&exec_args_kva_mtx);
1331 		while ((argkva = SLIST_FIRST(&exec_args_kva_freelist)) == NULL)
1332 			(void)mtx_sleep(&exec_args_kva_freelist,
1333 			    &exec_args_kva_mtx, 0, "execkva", 0);
1334 		SLIST_REMOVE_HEAD(&exec_args_kva_freelist, next);
1335 		mtx_unlock(&exec_args_kva_mtx);
1336 	}
1337 	*(struct exec_args_kva **)cookie = argkva;
1338 	return (argkva->addr);
1339 }
1340 
1341 static void
1342 exec_release_args_kva(struct exec_args_kva *argkva, u_int gen)
1343 {
1344 	vm_offset_t base;
1345 
1346 	base = argkva->addr;
1347 	if (argkva->gen != gen) {
1348 		(void)vm_map_madvise(exec_map, base, base + exec_map_entry_size,
1349 		    MADV_FREE);
1350 		argkva->gen = gen;
1351 	}
1352 	if (!atomic_cmpset_ptr((uintptr_t *)DPCPU_PTR(exec_args_kva),
1353 	    (uintptr_t)NULL, (uintptr_t)argkva)) {
1354 		mtx_lock(&exec_args_kva_mtx);
1355 		SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1356 		wakeup_one(&exec_args_kva_freelist);
1357 		mtx_unlock(&exec_args_kva_mtx);
1358 	}
1359 }
1360 
1361 static void
1362 exec_free_args_kva(void *cookie)
1363 {
1364 
1365 	exec_release_args_kva(cookie, exec_args_gen);
1366 }
1367 
1368 static void
1369 exec_args_kva_lowmem(void *arg __unused)
1370 {
1371 	SLIST_HEAD(, exec_args_kva) head;
1372 	struct exec_args_kva *argkva;
1373 	u_int gen;
1374 	int i;
1375 
1376 	gen = atomic_fetchadd_int(&exec_args_gen, 1) + 1;
1377 
1378 	/*
1379 	 * Force an madvise of each KVA range. Any currently allocated ranges
1380 	 * will have MADV_FREE applied once they are freed.
1381 	 */
1382 	SLIST_INIT(&head);
1383 	mtx_lock(&exec_args_kva_mtx);
1384 	SLIST_SWAP(&head, &exec_args_kva_freelist, exec_args_kva);
1385 	mtx_unlock(&exec_args_kva_mtx);
1386 	while ((argkva = SLIST_FIRST(&head)) != NULL) {
1387 		SLIST_REMOVE_HEAD(&head, next);
1388 		exec_release_args_kva(argkva, gen);
1389 	}
1390 
1391 	CPU_FOREACH(i) {
1392 		argkva = (void *)atomic_readandclear_ptr(
1393 		    (uintptr_t *)DPCPU_ID_PTR(i, exec_args_kva));
1394 		if (argkva != NULL)
1395 			exec_release_args_kva(argkva, gen);
1396 	}
1397 }
1398 EVENTHANDLER_DEFINE(vm_lowmem, exec_args_kva_lowmem, NULL,
1399     EVENTHANDLER_PRI_ANY);
1400 
1401 /*
1402  * Allocate temporary demand-paged, zero-filled memory for the file name,
1403  * argument, and environment strings.
1404  */
1405 int
1406 exec_alloc_args(struct image_args *args)
1407 {
1408 
1409 	args->buf = (char *)exec_alloc_args_kva(&args->bufkva);
1410 	return (0);
1411 }
1412 
1413 void
1414 exec_free_args(struct image_args *args)
1415 {
1416 
1417 	if (args->buf != NULL) {
1418 		exec_free_args_kva(args->bufkva);
1419 		args->buf = NULL;
1420 	}
1421 	if (args->fname_buf != NULL) {
1422 		free(args->fname_buf, M_TEMP);
1423 		args->fname_buf = NULL;
1424 	}
1425 	if (args->fdp != NULL)
1426 		fdescfree_remapped(args->fdp);
1427 }
1428 
1429 /*
1430  * A set to functions to fill struct image args.
1431  *
1432  * NOTE: exec_args_add_fname() must be called (possibly with a NULL
1433  * fname) before the other functions.  All exec_args_add_arg() calls must
1434  * be made before any exec_args_add_env() calls.  exec_args_adjust_args()
1435  * may be called any time after exec_args_add_fname().
1436  *
1437  * exec_args_add_fname() - install path to be executed
1438  * exec_args_add_arg() - append an argument string
1439  * exec_args_add_env() - append an env string
1440  * exec_args_adjust_args() - adjust location of the argument list to
1441  *                           allow new arguments to be prepended
1442  */
1443 int
1444 exec_args_add_fname(struct image_args *args, const char *fname,
1445     enum uio_seg segflg)
1446 {
1447 	int error;
1448 	size_t length;
1449 
1450 	KASSERT(args->fname == NULL, ("fname already appended"));
1451 	KASSERT(args->endp == NULL, ("already appending to args"));
1452 
1453 	if (fname != NULL) {
1454 		args->fname = args->buf;
1455 		error = segflg == UIO_SYSSPACE ?
1456 		    copystr(fname, args->fname, PATH_MAX, &length) :
1457 		    copyinstr(fname, args->fname, PATH_MAX, &length);
1458 		if (error != 0)
1459 			return (error == ENAMETOOLONG ? E2BIG : error);
1460 	} else
1461 		length = 0;
1462 
1463 	/* Set up for _arg_*()/_env_*() */
1464 	args->endp = args->buf + length;
1465 	/* begin_argv must be set and kept updated */
1466 	args->begin_argv = args->endp;
1467 	KASSERT(exec_map_entry_size - length >= ARG_MAX,
1468 	    ("too little space remaining for arguments %zu < %zu",
1469 	    exec_map_entry_size - length, (size_t)ARG_MAX));
1470 	args->stringspace = ARG_MAX;
1471 
1472 	return (0);
1473 }
1474 
1475 static int
1476 exec_args_add_str(struct image_args *args, const char *str,
1477     enum uio_seg segflg, int *countp)
1478 {
1479 	int error;
1480 	size_t length;
1481 
1482 	KASSERT(args->endp != NULL, ("endp not initialized"));
1483 	KASSERT(args->begin_argv != NULL, ("begin_argp not initialized"));
1484 
1485 	error = (segflg == UIO_SYSSPACE) ?
1486 	    copystr(str, args->endp, args->stringspace, &length) :
1487 	    copyinstr(str, args->endp, args->stringspace, &length);
1488 	if (error != 0)
1489 		return (error == ENAMETOOLONG ? E2BIG : error);
1490 	args->stringspace -= length;
1491 	args->endp += length;
1492 	(*countp)++;
1493 
1494 	return (0);
1495 }
1496 
1497 int
1498 exec_args_add_arg(struct image_args *args, const char *argp,
1499     enum uio_seg segflg)
1500 {
1501 
1502 	KASSERT(args->envc == 0, ("appending args after env"));
1503 
1504 	return (exec_args_add_str(args, argp, segflg, &args->argc));
1505 }
1506 
1507 int
1508 exec_args_add_env(struct image_args *args, const char *envp,
1509     enum uio_seg segflg)
1510 {
1511 
1512 	if (args->envc == 0)
1513 		args->begin_envv = args->endp;
1514 
1515 	return (exec_args_add_str(args, envp, segflg, &args->envc));
1516 }
1517 
1518 int
1519 exec_args_adjust_args(struct image_args *args, size_t consume, ssize_t extend)
1520 {
1521 	ssize_t offset;
1522 
1523 	KASSERT(args->endp != NULL, ("endp not initialized"));
1524 	KASSERT(args->begin_argv != NULL, ("begin_argp not initialized"));
1525 
1526 	offset = extend - consume;
1527 	if (args->stringspace < offset)
1528 		return (E2BIG);
1529 	memmove(args->begin_argv + extend, args->begin_argv + consume,
1530 	    args->endp - args->begin_argv + consume);
1531 	if (args->envc > 0)
1532 		args->begin_envv += offset;
1533 	args->endp += offset;
1534 	args->stringspace -= offset;
1535 	return (0);
1536 }
1537 
1538 char *
1539 exec_args_get_begin_envv(struct image_args *args)
1540 {
1541 
1542 	KASSERT(args->endp != NULL, ("endp not initialized"));
1543 
1544 	if (args->envc > 0)
1545 		return (args->begin_envv);
1546 	return (args->endp);
1547 }
1548 
1549 /*
1550  * Copy strings out to the new process address space, constructing new arg
1551  * and env vector tables. Return a pointer to the base so that it can be used
1552  * as the initial stack pointer.
1553  */
1554 register_t *
1555 exec_copyout_strings(struct image_params *imgp)
1556 {
1557 	int argc, envc;
1558 	char **vectp;
1559 	char *stringp;
1560 	uintptr_t destp;
1561 	register_t *stack_base;
1562 	struct ps_strings *arginfo;
1563 	struct proc *p;
1564 	size_t execpath_len;
1565 	int szsigcode, szps;
1566 	char canary[sizeof(long) * 8];
1567 
1568 	szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1569 	/*
1570 	 * Calculate string base and vector table pointers.
1571 	 * Also deal with signal trampoline code for this exec type.
1572 	 */
1573 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
1574 		execpath_len = strlen(imgp->execpath) + 1;
1575 	else
1576 		execpath_len = 0;
1577 	p = imgp->proc;
1578 	szsigcode = 0;
1579 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1580 	if (p->p_sysent->sv_sigcode_base == 0) {
1581 		if (p->p_sysent->sv_szsigcode != NULL)
1582 			szsigcode = *(p->p_sysent->sv_szsigcode);
1583 	}
1584 	destp =	(uintptr_t)arginfo;
1585 
1586 	/*
1587 	 * install sigcode
1588 	 */
1589 	if (szsigcode != 0) {
1590 		destp -= szsigcode;
1591 		destp = rounddown2(destp, sizeof(void *));
1592 		copyout(p->p_sysent->sv_sigcode, (void *)destp, szsigcode);
1593 	}
1594 
1595 	/*
1596 	 * Copy the image path for the rtld.
1597 	 */
1598 	if (execpath_len != 0) {
1599 		destp -= execpath_len;
1600 		destp = rounddown2(destp, sizeof(void *));
1601 		imgp->execpathp = destp;
1602 		copyout(imgp->execpath, (void *)destp, execpath_len);
1603 	}
1604 
1605 	/*
1606 	 * Prepare the canary for SSP.
1607 	 */
1608 	arc4rand(canary, sizeof(canary), 0);
1609 	destp -= sizeof(canary);
1610 	imgp->canary = destp;
1611 	copyout(canary, (void *)destp, sizeof(canary));
1612 	imgp->canarylen = sizeof(canary);
1613 
1614 	/*
1615 	 * Prepare the pagesizes array.
1616 	 */
1617 	destp -= szps;
1618 	destp = rounddown2(destp, sizeof(void *));
1619 	imgp->pagesizes = destp;
1620 	copyout(pagesizes, (void *)destp, szps);
1621 	imgp->pagesizeslen = szps;
1622 
1623 	destp -= ARG_MAX - imgp->args->stringspace;
1624 	destp = rounddown2(destp, sizeof(void *));
1625 
1626 	vectp = (char **)destp;
1627 	if (imgp->sysent->sv_stackgap != NULL)
1628 		imgp->sysent->sv_stackgap(imgp, (u_long *)&vectp);
1629 
1630 	if (imgp->auxargs) {
1631 		/*
1632 		 * Allocate room on the stack for the ELF auxargs
1633 		 * array.  It has up to AT_COUNT entries.
1634 		 */
1635 		vectp -= howmany(AT_COUNT * sizeof(Elf_Auxinfo),
1636 		    sizeof(*vectp));
1637 	}
1638 
1639 	/*
1640 	 * Allocate room for the argv[] and env vectors including the
1641 	 * terminating NULL pointers.
1642 	 */
1643 	vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
1644 
1645 	/*
1646 	 * vectp also becomes our initial stack base
1647 	 */
1648 	stack_base = (register_t *)vectp;
1649 
1650 	stringp = imgp->args->begin_argv;
1651 	argc = imgp->args->argc;
1652 	envc = imgp->args->envc;
1653 
1654 	/*
1655 	 * Copy out strings - arguments and environment.
1656 	 */
1657 	copyout(stringp, (void *)destp, ARG_MAX - imgp->args->stringspace);
1658 
1659 	/*
1660 	 * Fill in "ps_strings" struct for ps, w, etc.
1661 	 */
1662 	suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp);
1663 	suword32(&arginfo->ps_nargvstr, argc);
1664 
1665 	/*
1666 	 * Fill in argument portion of vector table.
1667 	 */
1668 	for (; argc > 0; --argc) {
1669 		suword(vectp++, (long)(intptr_t)destp);
1670 		while (*stringp++ != 0)
1671 			destp++;
1672 		destp++;
1673 	}
1674 
1675 	/* a null vector table pointer separates the argp's from the envp's */
1676 	suword(vectp++, 0);
1677 
1678 	suword(&arginfo->ps_envstr, (long)(intptr_t)vectp);
1679 	suword32(&arginfo->ps_nenvstr, envc);
1680 
1681 	/*
1682 	 * Fill in environment portion of vector table.
1683 	 */
1684 	for (; envc > 0; --envc) {
1685 		suword(vectp++, (long)(intptr_t)destp);
1686 		while (*stringp++ != 0)
1687 			destp++;
1688 		destp++;
1689 	}
1690 
1691 	/* end of vector table is a null pointer */
1692 	suword(vectp, 0);
1693 
1694 	return (stack_base);
1695 }
1696 
1697 /*
1698  * Check permissions of file to execute.
1699  *	Called with imgp->vp locked.
1700  *	Return 0 for success or error code on failure.
1701  */
1702 int
1703 exec_check_permissions(struct image_params *imgp)
1704 {
1705 	struct vnode *vp = imgp->vp;
1706 	struct vattr *attr = imgp->attr;
1707 	struct thread *td;
1708 	int error;
1709 
1710 	td = curthread;
1711 
1712 	/* Get file attributes */
1713 	error = VOP_GETATTR(vp, attr, td->td_ucred);
1714 	if (error)
1715 		return (error);
1716 
1717 #ifdef MAC
1718 	error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1719 	if (error)
1720 		return (error);
1721 #endif
1722 
1723 	/*
1724 	 * 1) Check if file execution is disabled for the filesystem that
1725 	 *    this file resides on.
1726 	 * 2) Ensure that at least one execute bit is on. Otherwise, a
1727 	 *    privileged user will always succeed, and we don't want this
1728 	 *    to happen unless the file really is executable.
1729 	 * 3) Ensure that the file is a regular file.
1730 	 */
1731 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1732 	    (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1733 	    (attr->va_type != VREG))
1734 		return (EACCES);
1735 
1736 	/*
1737 	 * Zero length files can't be exec'd
1738 	 */
1739 	if (attr->va_size == 0)
1740 		return (ENOEXEC);
1741 
1742 	/*
1743 	 *  Check for execute permission to file based on current credentials.
1744 	 */
1745 	error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1746 	if (error)
1747 		return (error);
1748 
1749 	/*
1750 	 * Check number of open-for-writes on the file and deny execution
1751 	 * if there are any.
1752 	 *
1753 	 * Add a text reference now so no one can write to the
1754 	 * executable while we're activating it.
1755 	 *
1756 	 * Remember if this was set before and unset it in case this is not
1757 	 * actually an executable image.
1758 	 */
1759 	error = VOP_SET_TEXT(vp);
1760 	if (error != 0)
1761 		return (error);
1762 	imgp->textset = true;
1763 
1764 	/*
1765 	 * Call filesystem specific open routine (which does nothing in the
1766 	 * general case).
1767 	 */
1768 	error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1769 	if (error == 0)
1770 		imgp->opened = 1;
1771 	return (error);
1772 }
1773 
1774 /*
1775  * Exec handler registration
1776  */
1777 int
1778 exec_register(const struct execsw *execsw_arg)
1779 {
1780 	const struct execsw **es, **xs, **newexecsw;
1781 	u_int count = 2;	/* New slot and trailing NULL */
1782 
1783 	if (execsw)
1784 		for (es = execsw; *es; es++)
1785 			count++;
1786 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1787 	xs = newexecsw;
1788 	if (execsw)
1789 		for (es = execsw; *es; es++)
1790 			*xs++ = *es;
1791 	*xs++ = execsw_arg;
1792 	*xs = NULL;
1793 	if (execsw)
1794 		free(execsw, M_TEMP);
1795 	execsw = newexecsw;
1796 	return (0);
1797 }
1798 
1799 int
1800 exec_unregister(const struct execsw *execsw_arg)
1801 {
1802 	const struct execsw **es, **xs, **newexecsw;
1803 	int count = 1;
1804 
1805 	if (execsw == NULL)
1806 		panic("unregister with no handlers left?\n");
1807 
1808 	for (es = execsw; *es; es++) {
1809 		if (*es == execsw_arg)
1810 			break;
1811 	}
1812 	if (*es == NULL)
1813 		return (ENOENT);
1814 	for (es = execsw; *es; es++)
1815 		if (*es != execsw_arg)
1816 			count++;
1817 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1818 	xs = newexecsw;
1819 	for (es = execsw; *es; es++)
1820 		if (*es != execsw_arg)
1821 			*xs++ = *es;
1822 	*xs = NULL;
1823 	if (execsw)
1824 		free(execsw, M_TEMP);
1825 	execsw = newexecsw;
1826 	return (0);
1827 }
1828