xref: /freebsd/sys/kern/kern_exec.c (revision e17f5b1d307b7b8910d67883e57a9604305906d5)
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/priv.h>
55 #include <sys/proc.h>
56 #include <sys/ptrace.h>
57 #include <sys/resourcevar.h>
58 #include <sys/rwlock.h>
59 #include <sys/sched.h>
60 #include <sys/sdt.h>
61 #include <sys/sf_buf.h>
62 #include <sys/shm.h>
63 #include <sys/signalvar.h>
64 #include <sys/smp.h>
65 #include <sys/stat.h>
66 #include <sys/syscallsubr.h>
67 #include <sys/sysctl.h>
68 #include <sys/sysent.h>
69 #include <sys/sysproto.h>
70 #include <sys/vnode.h>
71 #include <sys/wait.h>
72 #ifdef KTRACE
73 #include <sys/ktrace.h>
74 #endif
75 
76 #include <vm/vm.h>
77 #include <vm/vm_param.h>
78 #include <vm/pmap.h>
79 #include <vm/vm_page.h>
80 #include <vm/vm_map.h>
81 #include <vm/vm_kern.h>
82 #include <vm/vm_extern.h>
83 #include <vm/vm_object.h>
84 #include <vm/vm_pager.h>
85 
86 #ifdef	HWPMC_HOOKS
87 #include <sys/pmckern.h>
88 #endif
89 
90 #include <machine/reg.h>
91 
92 #include <security/audit/audit.h>
93 #include <security/mac/mac_framework.h>
94 
95 #ifdef KDTRACE_HOOKS
96 #include <sys/dtrace_bsd.h>
97 dtrace_execexit_func_t	dtrace_fasttrap_exec;
98 #endif
99 
100 SDT_PROVIDER_DECLARE(proc);
101 SDT_PROBE_DEFINE1(proc, , , exec, "char *");
102 SDT_PROBE_DEFINE1(proc, , , exec__failure, "int");
103 SDT_PROBE_DEFINE1(proc, , , exec__success, "char *");
104 
105 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
106 
107 int coredump_pack_fileinfo = 1;
108 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_fileinfo, CTLFLAG_RWTUN,
109     &coredump_pack_fileinfo, 0,
110     "Enable file path packing in 'procstat -f' coredump notes");
111 
112 int coredump_pack_vmmapinfo = 1;
113 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_vmmapinfo, CTLFLAG_RWTUN,
114     &coredump_pack_vmmapinfo, 0,
115     "Enable file path packing in 'procstat -v' coredump notes");
116 
117 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
118 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
119 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
120 static int do_execve(struct thread *td, struct image_args *args,
121     struct mac *mac_p);
122 
123 /* XXX This should be vm_size_t. */
124 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD|
125     CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_ps_strings, "LU",
126     "Location of process' ps_strings structure");
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     "Top of process stack");
132 
133 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD|CTLFLAG_MPSAFE,
134     NULL, 0, sysctl_kern_stackprot, "I",
135     "Stack memory permissions");
136 
137 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
138 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
139     &ps_arg_cache_limit, 0,
140     "Process' command line characters cache limit");
141 
142 static int disallow_high_osrel;
143 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW,
144     &disallow_high_osrel, 0,
145     "Disallow execution of binaries built for higher version of the world");
146 
147 static int map_at_zero = 0;
148 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RWTUN, &map_at_zero, 0,
149     "Permit processes to map an object at virtual address 0.");
150 
151 static int
152 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
153 {
154 	struct proc *p;
155 	int error;
156 
157 	p = curproc;
158 #ifdef SCTL_MASK32
159 	if (req->flags & SCTL_MASK32) {
160 		unsigned int val;
161 		val = (unsigned int)p->p_sysent->sv_psstrings;
162 		error = SYSCTL_OUT(req, &val, sizeof(val));
163 	} else
164 #endif
165 		error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings,
166 		   sizeof(p->p_sysent->sv_psstrings));
167 	return error;
168 }
169 
170 static int
171 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
172 {
173 	struct proc *p;
174 	int error;
175 
176 	p = curproc;
177 #ifdef SCTL_MASK32
178 	if (req->flags & SCTL_MASK32) {
179 		unsigned int val;
180 		val = (unsigned int)p->p_sysent->sv_usrstack;
181 		error = SYSCTL_OUT(req, &val, sizeof(val));
182 	} else
183 #endif
184 		error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack,
185 		    sizeof(p->p_sysent->sv_usrstack));
186 	return error;
187 }
188 
189 static int
190 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
191 {
192 	struct proc *p;
193 
194 	p = curproc;
195 	return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
196 	    sizeof(p->p_sysent->sv_stackprot)));
197 }
198 
199 /*
200  * Each of the items is a pointer to a `const struct execsw', hence the
201  * double pointer here.
202  */
203 static const struct execsw **execsw;
204 
205 #ifndef _SYS_SYSPROTO_H_
206 struct execve_args {
207 	char    *fname;
208 	char    **argv;
209 	char    **envv;
210 };
211 #endif
212 
213 int
214 sys_execve(struct thread *td, struct execve_args *uap)
215 {
216 	struct image_args args;
217 	struct vmspace *oldvmspace;
218 	int error;
219 
220 	error = pre_execve(td, &oldvmspace);
221 	if (error != 0)
222 		return (error);
223 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
224 	    uap->argv, uap->envv);
225 	if (error == 0)
226 		error = kern_execve(td, &args, NULL);
227 	post_execve(td, error, oldvmspace);
228 	return (error);
229 }
230 
231 #ifndef _SYS_SYSPROTO_H_
232 struct fexecve_args {
233 	int	fd;
234 	char	**argv;
235 	char	**envv;
236 };
237 #endif
238 int
239 sys_fexecve(struct thread *td, struct fexecve_args *uap)
240 {
241 	struct image_args args;
242 	struct vmspace *oldvmspace;
243 	int error;
244 
245 	error = pre_execve(td, &oldvmspace);
246 	if (error != 0)
247 		return (error);
248 	error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
249 	    uap->argv, uap->envv);
250 	if (error == 0) {
251 		args.fd = uap->fd;
252 		error = kern_execve(td, &args, NULL);
253 	}
254 	post_execve(td, error, oldvmspace);
255 	return (error);
256 }
257 
258 #ifndef _SYS_SYSPROTO_H_
259 struct __mac_execve_args {
260 	char	*fname;
261 	char	**argv;
262 	char	**envv;
263 	struct mac	*mac_p;
264 };
265 #endif
266 
267 int
268 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap)
269 {
270 #ifdef MAC
271 	struct image_args args;
272 	struct vmspace *oldvmspace;
273 	int error;
274 
275 	error = pre_execve(td, &oldvmspace);
276 	if (error != 0)
277 		return (error);
278 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
279 	    uap->argv, uap->envv);
280 	if (error == 0)
281 		error = kern_execve(td, &args, uap->mac_p);
282 	post_execve(td, error, oldvmspace);
283 	return (error);
284 #else
285 	return (ENOSYS);
286 #endif
287 }
288 
289 int
290 pre_execve(struct thread *td, struct vmspace **oldvmspace)
291 {
292 	struct proc *p;
293 	int error;
294 
295 	KASSERT(td == curthread, ("non-current thread %p", td));
296 	error = 0;
297 	p = td->td_proc;
298 	if ((p->p_flag & P_HADTHREADS) != 0) {
299 		PROC_LOCK(p);
300 		if (thread_single(p, SINGLE_BOUNDARY) != 0)
301 			error = ERESTART;
302 		PROC_UNLOCK(p);
303 	}
304 	KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0,
305 	    ("nested execve"));
306 	*oldvmspace = p->p_vmspace;
307 	return (error);
308 }
309 
310 void
311 post_execve(struct thread *td, int error, struct vmspace *oldvmspace)
312 {
313 	struct proc *p;
314 
315 	KASSERT(td == curthread, ("non-current thread %p", td));
316 	p = td->td_proc;
317 	if ((p->p_flag & P_HADTHREADS) != 0) {
318 		PROC_LOCK(p);
319 		/*
320 		 * If success, we upgrade to SINGLE_EXIT state to
321 		 * force other threads to suicide.
322 		 */
323 		if (error == EJUSTRETURN)
324 			thread_single(p, SINGLE_EXIT);
325 		else
326 			thread_single_end(p, SINGLE_BOUNDARY);
327 		PROC_UNLOCK(p);
328 	}
329 	if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
330 		KASSERT(p->p_vmspace != oldvmspace,
331 		    ("oldvmspace still used"));
332 		vmspace_free(oldvmspace);
333 		td->td_pflags &= ~TDP_EXECVMSPC;
334 	}
335 }
336 
337 /*
338  * XXX: 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 {
347 
348 	AUDIT_ARG_ARGV(args->begin_argv, args->argc,
349 	    exec_args_get_begin_envv(args) - args->begin_argv);
350 	AUDIT_ARG_ENVV(exec_args_get_begin_envv(args), args->envc,
351 	    args->endp - exec_args_get_begin_envv(args));
352 	return (do_execve(td, args, mac_p));
353 }
354 
355 /*
356  * In-kernel implementation of execve().  All arguments are assumed to be
357  * userspace pointers from the passed thread.
358  */
359 static int
360 do_execve(struct thread *td, struct image_args *args, struct mac *mac_p)
361 {
362 	struct proc *p = td->td_proc;
363 	struct nameidata nd;
364 	struct ucred *oldcred;
365 	struct uidinfo *euip = NULL;
366 	uintptr_t stack_base;
367 	struct image_params image_params, *imgp;
368 	struct vattr attr;
369 	int (*img_first)(struct image_params *);
370 	struct pargs *oldargs = NULL, *newargs = NULL;
371 	struct sigacts *oldsigacts = NULL, *newsigacts = NULL;
372 #ifdef KTRACE
373 	struct vnode *tracevp = NULL;
374 	struct ucred *tracecred = NULL;
375 #endif
376 	struct vnode *oldtextvp = NULL, *newtextvp;
377 	int credential_changing;
378 #ifdef MAC
379 	struct label *interpvplabel = NULL;
380 	int will_transition;
381 #endif
382 #ifdef HWPMC_HOOKS
383 	struct pmckern_procexec pe;
384 #endif
385 	int error, i, orig_osrel;
386 	uint32_t orig_fctl0;
387 	static const char fexecv_proc_title[] = "(fexecv)";
388 
389 	imgp = &image_params;
390 
391 	/*
392 	 * Lock the process and set the P_INEXEC flag to indicate that
393 	 * it should be left alone until we're done here.  This is
394 	 * necessary to avoid race conditions - e.g. in ptrace() -
395 	 * that might allow a local user to illicitly obtain elevated
396 	 * privileges.
397 	 */
398 	PROC_LOCK(p);
399 	KASSERT((p->p_flag & P_INEXEC) == 0,
400 	    ("%s(): process already has P_INEXEC flag", __func__));
401 	p->p_flag |= P_INEXEC;
402 	PROC_UNLOCK(p);
403 
404 	/*
405 	 * Initialize part of the common data
406 	 */
407 	bzero(imgp, sizeof(*imgp));
408 	imgp->proc = p;
409 	imgp->attr = &attr;
410 	imgp->args = args;
411 	oldcred = p->p_ucred;
412 	orig_osrel = p->p_osrel;
413 	orig_fctl0 = p->p_fctl0;
414 
415 #ifdef MAC
416 	error = mac_execve_enter(imgp, mac_p);
417 	if (error)
418 		goto exec_fail;
419 #endif
420 
421 	/*
422 	 * Translate the file name. namei() returns a vnode pointer
423 	 *	in ni_vp among other things.
424 	 *
425 	 * XXXAUDIT: It would be desirable to also audit the name of the
426 	 * interpreter if this is an interpreted binary.
427 	 */
428 	if (args->fname != NULL) {
429 		NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW |
430 		    SAVENAME | AUDITVNODE1, UIO_SYSSPACE, args->fname, td);
431 	}
432 
433 	SDT_PROBE1(proc, , , exec, args->fname);
434 
435 interpret:
436 	if (args->fname != NULL) {
437 #ifdef CAPABILITY_MODE
438 		/*
439 		 * While capability mode can't reach this point via direct
440 		 * path arguments to execve(), we also don't allow
441 		 * interpreters to be used in capability mode (for now).
442 		 * Catch indirect lookups and return a permissions error.
443 		 */
444 		if (IN_CAPABILITY_MODE(td)) {
445 			error = ECAPMODE;
446 			goto exec_fail;
447 		}
448 #endif
449 		error = namei(&nd);
450 		if (error)
451 			goto exec_fail;
452 
453 		newtextvp = nd.ni_vp;
454 		imgp->vp = newtextvp;
455 	} else {
456 		AUDIT_ARG_FD(args->fd);
457 		/*
458 		 * Descriptors opened only with O_EXEC or O_RDONLY are allowed.
459 		 */
460 		error = fgetvp_exec(td, args->fd, &cap_fexecve_rights, &newtextvp);
461 		if (error)
462 			goto exec_fail;
463 		vn_lock(newtextvp, LK_SHARED | LK_RETRY);
464 		AUDIT_ARG_VNODE1(newtextvp);
465 		imgp->vp = newtextvp;
466 	}
467 
468 	/*
469 	 * Check file permissions.  Also 'opens' file and sets its vnode to
470 	 * text mode.
471 	 */
472 	error = exec_check_permissions(imgp);
473 	if (error)
474 		goto exec_fail_dealloc;
475 
476 	imgp->object = imgp->vp->v_object;
477 	if (imgp->object != NULL)
478 		vm_object_reference(imgp->object);
479 
480 	error = exec_map_first_page(imgp);
481 	if (error)
482 		goto exec_fail_dealloc;
483 
484 	imgp->proc->p_osrel = 0;
485 	imgp->proc->p_fctl0 = 0;
486 
487 	/*
488 	 * Implement image setuid/setgid.
489 	 *
490 	 * Determine new credentials before attempting image activators
491 	 * so that it can be used by process_exec handlers to determine
492 	 * credential/setid changes.
493 	 *
494 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
495 	 * the process is being traced.
496 	 *
497 	 * We disable setuid/setgid/etc in capability mode on the basis
498 	 * that most setugid applications are not written with that
499 	 * environment in mind, and will therefore almost certainly operate
500 	 * incorrectly. In principle there's no reason that setugid
501 	 * applications might not be useful in capability mode, so we may want
502 	 * to reconsider this conservative design choice in the future.
503 	 *
504 	 * XXXMAC: For the time being, use NOSUID to also prohibit
505 	 * transitions on the file system.
506 	 */
507 	credential_changing = 0;
508 	credential_changing |= (attr.va_mode & S_ISUID) &&
509 	    oldcred->cr_uid != attr.va_uid;
510 	credential_changing |= (attr.va_mode & S_ISGID) &&
511 	    oldcred->cr_gid != attr.va_gid;
512 #ifdef MAC
513 	will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
514 	    interpvplabel, imgp);
515 	credential_changing |= will_transition;
516 #endif
517 
518 	/* Don't inherit PROC_PDEATHSIG_CTL value if setuid/setgid. */
519 	if (credential_changing)
520 		imgp->proc->p_pdeathsig = 0;
521 
522 	if (credential_changing &&
523 #ifdef CAPABILITY_MODE
524 	    ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
525 #endif
526 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
527 	    (p->p_flag & P_TRACED) == 0) {
528 		imgp->credential_setid = true;
529 		VOP_UNLOCK(imgp->vp);
530 		imgp->newcred = crdup(oldcred);
531 		if (attr.va_mode & S_ISUID) {
532 			euip = uifind(attr.va_uid);
533 			change_euid(imgp->newcred, euip);
534 		}
535 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
536 		if (attr.va_mode & S_ISGID)
537 			change_egid(imgp->newcred, attr.va_gid);
538 		/*
539 		 * Implement correct POSIX saved-id behavior.
540 		 *
541 		 * XXXMAC: Note that the current logic will save the
542 		 * uid and gid if a MAC domain transition occurs, even
543 		 * though maybe it shouldn't.
544 		 */
545 		change_svuid(imgp->newcred, imgp->newcred->cr_uid);
546 		change_svgid(imgp->newcred, imgp->newcred->cr_gid);
547 	} else {
548 		/*
549 		 * Implement correct POSIX saved-id behavior.
550 		 *
551 		 * XXX: It's not clear that the existing behavior is
552 		 * POSIX-compliant.  A number of sources indicate that the
553 		 * saved uid/gid should only be updated if the new ruid is
554 		 * not equal to the old ruid, or the new euid is not equal
555 		 * to the old euid and the new euid is not equal to the old
556 		 * ruid.  The FreeBSD code always updates the saved uid/gid.
557 		 * Also, this code uses the new (replaced) euid and egid as
558 		 * the source, which may or may not be the right ones to use.
559 		 */
560 		if (oldcred->cr_svuid != oldcred->cr_uid ||
561 		    oldcred->cr_svgid != oldcred->cr_gid) {
562 			VOP_UNLOCK(imgp->vp);
563 			imgp->newcred = crdup(oldcred);
564 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
565 			change_svuid(imgp->newcred, imgp->newcred->cr_uid);
566 			change_svgid(imgp->newcred, imgp->newcred->cr_gid);
567 		}
568 	}
569 	/* The new credentials are installed into the process later. */
570 
571 	/*
572 	 * Do the best to calculate the full path to the image file.
573 	 */
574 	if (args->fname != NULL && args->fname[0] == '/')
575 		imgp->execpath = args->fname;
576 	else {
577 		VOP_UNLOCK(imgp->vp);
578 		if (vn_fullpath(td, imgp->vp, &imgp->execpath,
579 		    &imgp->freepath) != 0)
580 			imgp->execpath = args->fname;
581 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
582 	}
583 
584 	/*
585 	 *	If the current process has a special image activator it
586 	 *	wants to try first, call it.   For example, emulating shell
587 	 *	scripts differently.
588 	 */
589 	error = -1;
590 	if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
591 		error = img_first(imgp);
592 
593 	/*
594 	 *	Loop through the list of image activators, calling each one.
595 	 *	An activator returns -1 if there is no match, 0 on success,
596 	 *	and an error otherwise.
597 	 */
598 	for (i = 0; error == -1 && execsw[i]; ++i) {
599 		if (execsw[i]->ex_imgact == NULL ||
600 		    execsw[i]->ex_imgact == img_first) {
601 			continue;
602 		}
603 		error = (*execsw[i]->ex_imgact)(imgp);
604 	}
605 
606 	if (error) {
607 		if (error == -1)
608 			error = ENOEXEC;
609 		goto exec_fail_dealloc;
610 	}
611 
612 	/*
613 	 * Special interpreter operation, cleanup and loop up to try to
614 	 * activate the interpreter.
615 	 */
616 	if (imgp->interpreted) {
617 		exec_unmap_first_page(imgp);
618 		/*
619 		 * The text reference needs to be removed for scripts.
620 		 * There is a short period before we determine that
621 		 * something is a script where text reference is active.
622 		 * The vnode lock is held over this entire period
623 		 * so nothing should illegitimately be blocked.
624 		 */
625 		MPASS(imgp->textset);
626 		VOP_UNSET_TEXT_CHECKED(newtextvp);
627 		imgp->textset = false;
628 		/* free name buffer and old vnode */
629 		if (args->fname != NULL)
630 			NDFREE(&nd, NDF_ONLY_PNBUF);
631 #ifdef MAC
632 		mac_execve_interpreter_enter(newtextvp, &interpvplabel);
633 #endif
634 		if (imgp->opened) {
635 			VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td);
636 			imgp->opened = 0;
637 		}
638 		vput(newtextvp);
639 		vm_object_deallocate(imgp->object);
640 		imgp->object = NULL;
641 		imgp->credential_setid = false;
642 		if (imgp->newcred != NULL) {
643 			crfree(imgp->newcred);
644 			imgp->newcred = NULL;
645 		}
646 		imgp->execpath = NULL;
647 		free(imgp->freepath, M_TEMP);
648 		imgp->freepath = NULL;
649 		/* set new name to that of the interpreter */
650 		NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW |
651 		    SAVENAME, UIO_SYSSPACE, imgp->interpreter_name, td);
652 		args->fname = imgp->interpreter_name;
653 		goto interpret;
654 	}
655 
656 	/*
657 	 * NB: We unlock the vnode here because it is believed that none
658 	 * of the sv_copyout_strings/sv_fixup operations require the vnode.
659 	 */
660 	VOP_UNLOCK(imgp->vp);
661 
662 	if (disallow_high_osrel &&
663 	    P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
664 		error = ENOEXEC;
665 		uprintf("Osrel %d for image %s too high\n", p->p_osrel,
666 		    imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
667 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
668 		goto exec_fail_dealloc;
669 	}
670 
671 	/* ABI enforces the use of Capsicum. Switch into capabilities mode. */
672 	if (SV_PROC_FLAG(p, SV_CAPSICUM))
673 		sys_cap_enter(td, NULL);
674 
675 	/*
676 	 * Copy out strings (args and env) and initialize stack base.
677 	 */
678 	error = (*p->p_sysent->sv_copyout_strings)(imgp, &stack_base);
679 	if (error != 0) {
680 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
681 		goto exec_fail_dealloc;
682 	}
683 
684 	/*
685 	 * Stack setup.
686 	 */
687 	error = (*p->p_sysent->sv_fixup)(&stack_base, imgp);
688 	if (error != 0) {
689 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
690 		goto exec_fail_dealloc;
691 	}
692 
693 	if (args->fdp != NULL) {
694 		/* Install a brand new file descriptor table. */
695 		fdinstall_remapped(td, args->fdp);
696 		args->fdp = NULL;
697 	} else {
698 		/*
699 		 * Keep on using the existing file descriptor table. For
700 		 * security and other reasons, the file descriptor table
701 		 * cannot be shared after an exec.
702 		 */
703 		fdunshare(td);
704 		/* close files on exec */
705 		fdcloseexec(td);
706 	}
707 
708 	/*
709 	 * Malloc things before we need locks.
710 	 */
711 	i = exec_args_get_begin_envv(imgp->args) - imgp->args->begin_argv;
712 	/* Cache arguments if they fit inside our allowance */
713 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
714 		newargs = pargs_alloc(i);
715 		bcopy(imgp->args->begin_argv, newargs->ar_args, i);
716 	}
717 
718 	/*
719 	 * For security and other reasons, signal handlers cannot
720 	 * be shared after an exec. The new process gets a copy of the old
721 	 * handlers. In execsigs(), the new process will have its signals
722 	 * reset.
723 	 */
724 	if (sigacts_shared(p->p_sigacts)) {
725 		oldsigacts = p->p_sigacts;
726 		newsigacts = sigacts_alloc();
727 		sigacts_copy(newsigacts, oldsigacts);
728 	}
729 
730 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
731 
732 	PROC_LOCK(p);
733 	if (oldsigacts)
734 		p->p_sigacts = newsigacts;
735 	/* Stop profiling */
736 	stopprofclock(p);
737 
738 	/* reset caught signals */
739 	execsigs(p);
740 
741 	/* name this process - nameiexec(p, ndp) */
742 	bzero(p->p_comm, sizeof(p->p_comm));
743 	if (args->fname)
744 		bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
745 		    min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
746 	else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0)
747 		bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
748 	bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
749 #ifdef KTR
750 	sched_clear_tdname(td);
751 #endif
752 
753 	/*
754 	 * mark as execed, wakeup the process that vforked (if any) and tell
755 	 * it that it now has its own resources back
756 	 */
757 	p->p_flag |= P_EXEC;
758 	if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0)
759 		p->p_flag2 &= ~P2_NOTRACE;
760 	if ((p->p_flag2 & P2_STKGAP_DISABLE_EXEC) == 0)
761 		p->p_flag2 &= ~P2_STKGAP_DISABLE;
762 	if (p->p_flag & P_PPWAIT) {
763 		p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
764 		cv_broadcast(&p->p_pwait);
765 		/* STOPs are no longer ignored, arrange for AST */
766 		signotify(td);
767 	}
768 
769 	/*
770 	 * Implement image setuid/setgid installation.
771 	 */
772 	if (imgp->credential_setid) {
773 		/*
774 		 * Turn off syscall tracing for set-id programs, except for
775 		 * root.  Record any set-id flags first to make sure that
776 		 * we do not regain any tracing during a possible block.
777 		 */
778 		setsugid(p);
779 
780 #ifdef KTRACE
781 		if (p->p_tracecred != NULL &&
782 		    priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED))
783 			ktrprocexec(p, &tracecred, &tracevp);
784 #endif
785 		/*
786 		 * Close any file descriptors 0..2 that reference procfs,
787 		 * then make sure file descriptors 0..2 are in use.
788 		 *
789 		 * Both fdsetugidsafety() and fdcheckstd() may call functions
790 		 * taking sleepable locks, so temporarily drop our locks.
791 		 */
792 		PROC_UNLOCK(p);
793 		VOP_UNLOCK(imgp->vp);
794 		fdsetugidsafety(td);
795 		error = fdcheckstd(td);
796 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
797 		if (error != 0)
798 			goto exec_fail_dealloc;
799 		PROC_LOCK(p);
800 #ifdef MAC
801 		if (will_transition) {
802 			mac_vnode_execve_transition(oldcred, imgp->newcred,
803 			    imgp->vp, interpvplabel, imgp);
804 		}
805 #endif
806 	} else {
807 		if (oldcred->cr_uid == oldcred->cr_ruid &&
808 		    oldcred->cr_gid == oldcred->cr_rgid)
809 			p->p_flag &= ~P_SUGID;
810 	}
811 	/*
812 	 * Set the new credentials.
813 	 */
814 	if (imgp->newcred != NULL) {
815 		proc_set_cred(p, imgp->newcred);
816 		crfree(oldcred);
817 		oldcred = NULL;
818 	}
819 
820 	/*
821 	 * Store the vp for use in procfs.  This vnode was referenced by namei
822 	 * or fgetvp_exec.
823 	 */
824 	oldtextvp = p->p_textvp;
825 	p->p_textvp = newtextvp;
826 
827 #ifdef KDTRACE_HOOKS
828 	/*
829 	 * Tell the DTrace fasttrap provider about the exec if it
830 	 * has declared an interest.
831 	 */
832 	if (dtrace_fasttrap_exec)
833 		dtrace_fasttrap_exec(p);
834 #endif
835 
836 	/*
837 	 * Notify others that we exec'd, and clear the P_INEXEC flag
838 	 * as we're now a bona fide freshly-execed process.
839 	 */
840 	KNOTE_LOCKED(p->p_klist, NOTE_EXEC);
841 	p->p_flag &= ~P_INEXEC;
842 
843 	/* clear "fork but no exec" flag, as we _are_ execing */
844 	p->p_acflag &= ~AFORK;
845 
846 	/*
847 	 * Free any previous argument cache and replace it with
848 	 * the new argument cache, if any.
849 	 */
850 	oldargs = p->p_args;
851 	p->p_args = newargs;
852 	newargs = NULL;
853 
854 	PROC_UNLOCK(p);
855 
856 #ifdef	HWPMC_HOOKS
857 	/*
858 	 * Check if system-wide sampling is in effect or if the
859 	 * current process is using PMCs.  If so, do exec() time
860 	 * processing.  This processing needs to happen AFTER the
861 	 * P_INEXEC flag is cleared.
862 	 */
863 	if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
864 		VOP_UNLOCK(imgp->vp);
865 		pe.pm_credentialschanged = credential_changing;
866 		pe.pm_entryaddr = imgp->entry_addr;
867 
868 		PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
869 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
870 	}
871 #endif
872 
873 	/* Set values passed into the program in registers. */
874 	(*p->p_sysent->sv_setregs)(td, imgp, stack_base);
875 
876 	VOP_MMAPPED(imgp->vp);
877 
878 	SDT_PROBE1(proc, , , exec__success, args->fname);
879 
880 exec_fail_dealloc:
881 	if (error != 0) {
882 		p->p_osrel = orig_osrel;
883 		p->p_fctl0 = orig_fctl0;
884 	}
885 
886 	if (imgp->firstpage != NULL)
887 		exec_unmap_first_page(imgp);
888 
889 	if (imgp->vp != NULL) {
890 		if (args->fname)
891 			NDFREE(&nd, NDF_ONLY_PNBUF);
892 		if (imgp->opened)
893 			VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
894 		if (imgp->textset)
895 			VOP_UNSET_TEXT_CHECKED(imgp->vp);
896 		if (error != 0)
897 			vput(imgp->vp);
898 		else
899 			VOP_UNLOCK(imgp->vp);
900 	}
901 
902 	if (imgp->object != NULL)
903 		vm_object_deallocate(imgp->object);
904 
905 	free(imgp->freepath, M_TEMP);
906 
907 	if (error == 0) {
908 		if (p->p_ptevents & PTRACE_EXEC) {
909 			PROC_LOCK(p);
910 			if (p->p_ptevents & PTRACE_EXEC)
911 				td->td_dbgflags |= TDB_EXEC;
912 			PROC_UNLOCK(p);
913 		}
914 	} else {
915 exec_fail:
916 		/* we're done here, clear P_INEXEC */
917 		PROC_LOCK(p);
918 		p->p_flag &= ~P_INEXEC;
919 		PROC_UNLOCK(p);
920 
921 		SDT_PROBE1(proc, , , exec__failure, error);
922 	}
923 
924 	if (imgp->newcred != NULL && oldcred != NULL)
925 		crfree(imgp->newcred);
926 
927 #ifdef MAC
928 	mac_execve_exit(imgp);
929 	mac_execve_interpreter_exit(interpvplabel);
930 #endif
931 	exec_free_args(args);
932 
933 	/*
934 	 * Handle deferred decrement of ref counts.
935 	 */
936 	if (oldtextvp != NULL)
937 		vrele(oldtextvp);
938 #ifdef KTRACE
939 	if (tracevp != NULL)
940 		vrele(tracevp);
941 	if (tracecred != NULL)
942 		crfree(tracecred);
943 #endif
944 	pargs_drop(oldargs);
945 	pargs_drop(newargs);
946 	if (oldsigacts != NULL)
947 		sigacts_free(oldsigacts);
948 	if (euip != NULL)
949 		uifree(euip);
950 
951 	if (error && imgp->vmspace_destroyed) {
952 		/* sorry, no more process anymore. exit gracefully */
953 		exit1(td, 0, SIGABRT);
954 		/* NOT REACHED */
955 	}
956 
957 #ifdef KTRACE
958 	if (error == 0)
959 		ktrprocctor(p);
960 #endif
961 
962 	/*
963 	 * We don't want cpu_set_syscall_retval() to overwrite any of
964 	 * the register values put in place by exec_setregs().
965 	 * Implementations of cpu_set_syscall_retval() will leave
966 	 * registers unmodified when returning EJUSTRETURN.
967 	 */
968 	return (error == 0 ? EJUSTRETURN : error);
969 }
970 
971 int
972 exec_map_first_page(struct image_params *imgp)
973 {
974 	vm_object_t object;
975 	vm_page_t m;
976 	int error;
977 
978 	if (imgp->firstpage != NULL)
979 		exec_unmap_first_page(imgp);
980 
981 	object = imgp->vp->v_object;
982 	if (object == NULL)
983 		return (EACCES);
984 #if VM_NRESERVLEVEL > 0
985 	if ((object->flags & OBJ_COLORED) == 0) {
986 		VM_OBJECT_WLOCK(object);
987 		vm_object_color(object, 0);
988 		VM_OBJECT_WUNLOCK(object);
989 	}
990 #endif
991 	error = vm_page_grab_valid_unlocked(&m, object, 0,
992 	    VM_ALLOC_COUNT(VM_INITIAL_PAGEIN) |
993             VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED);
994 
995 	if (error != VM_PAGER_OK)
996 		return (EIO);
997 	imgp->firstpage = sf_buf_alloc(m, 0);
998 	imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
999 
1000 	return (0);
1001 }
1002 
1003 void
1004 exec_unmap_first_page(struct image_params *imgp)
1005 {
1006 	vm_page_t m;
1007 
1008 	if (imgp->firstpage != NULL) {
1009 		m = sf_buf_page(imgp->firstpage);
1010 		sf_buf_free(imgp->firstpage);
1011 		imgp->firstpage = NULL;
1012 		vm_page_unwire(m, PQ_ACTIVE);
1013 	}
1014 }
1015 
1016 /*
1017  * Destroy old address space, and allocate a new stack.
1018  *	The new stack is only sgrowsiz large because it is grown
1019  *	automatically on a page fault.
1020  */
1021 int
1022 exec_new_vmspace(struct image_params *imgp, struct sysentvec *sv)
1023 {
1024 	int error;
1025 	struct proc *p = imgp->proc;
1026 	struct vmspace *vmspace = p->p_vmspace;
1027 	struct thread *td = curthread;
1028 	vm_object_t obj;
1029 	struct rlimit rlim_stack;
1030 	vm_offset_t sv_minuser, stack_addr;
1031 	vm_map_t map;
1032 	u_long ssiz;
1033 
1034 	imgp->vmspace_destroyed = 1;
1035 	imgp->sysent = sv;
1036 
1037 	sigfastblock_clear(td);
1038 
1039 	/* May be called with Giant held */
1040 	EVENTHANDLER_DIRECT_INVOKE(process_exec, p, imgp);
1041 
1042 	/*
1043 	 * Blow away entire process VM, if address space not shared,
1044 	 * otherwise, create a new VM space so that other threads are
1045 	 * not disrupted
1046 	 */
1047 	map = &vmspace->vm_map;
1048 	if (map_at_zero)
1049 		sv_minuser = sv->sv_minuser;
1050 	else
1051 		sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1052 	if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser &&
1053 	    vm_map_max(map) == sv->sv_maxuser &&
1054 	    cpu_exec_vmspace_reuse(p, map)) {
1055 		shmexit(vmspace);
1056 		pmap_remove_pages(vmspace_pmap(vmspace));
1057 		vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1058 		/*
1059 		 * An exec terminates mlockall(MCL_FUTURE), ASLR state
1060 		 * must be re-evaluated.
1061 		 */
1062 		vm_map_lock(map);
1063 		vm_map_modflags(map, 0, MAP_WIREFUTURE | MAP_ASLR |
1064 		    MAP_ASLR_IGNSTART);
1065 		vm_map_unlock(map);
1066 	} else {
1067 		error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1068 		if (error)
1069 			return (error);
1070 		vmspace = p->p_vmspace;
1071 		map = &vmspace->vm_map;
1072 	}
1073 	map->flags |= imgp->map_flags;
1074 
1075 	/* Map a shared page */
1076 	obj = sv->sv_shared_page_obj;
1077 	if (obj != NULL) {
1078 		vm_object_reference(obj);
1079 		error = vm_map_fixed(map, obj, 0,
1080 		    sv->sv_shared_page_base, sv->sv_shared_page_len,
1081 		    VM_PROT_READ | VM_PROT_EXECUTE,
1082 		    VM_PROT_READ | VM_PROT_EXECUTE,
1083 		    MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1084 		if (error != KERN_SUCCESS) {
1085 			vm_object_deallocate(obj);
1086 			return (vm_mmap_to_errno(error));
1087 		}
1088 	}
1089 
1090 	/* Allocate a new stack */
1091 	if (imgp->stack_sz != 0) {
1092 		ssiz = trunc_page(imgp->stack_sz);
1093 		PROC_LOCK(p);
1094 		lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack);
1095 		PROC_UNLOCK(p);
1096 		if (ssiz > rlim_stack.rlim_max)
1097 			ssiz = rlim_stack.rlim_max;
1098 		if (ssiz > rlim_stack.rlim_cur) {
1099 			rlim_stack.rlim_cur = ssiz;
1100 			kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack);
1101 		}
1102 	} else if (sv->sv_maxssiz != NULL) {
1103 		ssiz = *sv->sv_maxssiz;
1104 	} else {
1105 		ssiz = maxssiz;
1106 	}
1107 	imgp->eff_stack_sz = lim_cur(curthread, RLIMIT_STACK);
1108 	if (ssiz < imgp->eff_stack_sz)
1109 		imgp->eff_stack_sz = ssiz;
1110 	stack_addr = sv->sv_usrstack - ssiz;
1111 	error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz,
1112 	    obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot :
1113 	    sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN);
1114 	if (error != KERN_SUCCESS)
1115 		return (vm_mmap_to_errno(error));
1116 
1117 	/*
1118 	 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they
1119 	 * are still used to enforce the stack rlimit on the process stack.
1120 	 */
1121 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1122 	vmspace->vm_maxsaddr = (char *)stack_addr;
1123 
1124 	return (0);
1125 }
1126 
1127 /*
1128  * Copy out argument and environment strings from the old process address
1129  * space into the temporary string buffer.
1130  */
1131 int
1132 exec_copyin_args(struct image_args *args, const char *fname,
1133     enum uio_seg segflg, char **argv, char **envv)
1134 {
1135 	u_long arg, env;
1136 	int error;
1137 
1138 	bzero(args, sizeof(*args));
1139 	if (argv == NULL)
1140 		return (EFAULT);
1141 
1142 	/*
1143 	 * Allocate demand-paged memory for the file name, argument, and
1144 	 * environment strings.
1145 	 */
1146 	error = exec_alloc_args(args);
1147 	if (error != 0)
1148 		return (error);
1149 
1150 	/*
1151 	 * Copy the file name.
1152 	 */
1153 	error = exec_args_add_fname(args, fname, segflg);
1154 	if (error != 0)
1155 		goto err_exit;
1156 
1157 	/*
1158 	 * extract arguments first
1159 	 */
1160 	for (;;) {
1161 		error = fueword(argv++, &arg);
1162 		if (error == -1) {
1163 			error = EFAULT;
1164 			goto err_exit;
1165 		}
1166 		if (arg == 0)
1167 			break;
1168 		error = exec_args_add_arg(args, (char *)(uintptr_t)arg,
1169 		    UIO_USERSPACE);
1170 		if (error != 0)
1171 			goto err_exit;
1172 	}
1173 
1174 	/*
1175 	 * extract environment strings
1176 	 */
1177 	if (envv) {
1178 		for (;;) {
1179 			error = fueword(envv++, &env);
1180 			if (error == -1) {
1181 				error = EFAULT;
1182 				goto err_exit;
1183 			}
1184 			if (env == 0)
1185 				break;
1186 			error = exec_args_add_env(args,
1187 			    (char *)(uintptr_t)env, UIO_USERSPACE);
1188 			if (error != 0)
1189 				goto err_exit;
1190 		}
1191 	}
1192 
1193 	return (0);
1194 
1195 err_exit:
1196 	exec_free_args(args);
1197 	return (error);
1198 }
1199 
1200 int
1201 exec_copyin_data_fds(struct thread *td, struct image_args *args,
1202     const void *data, size_t datalen, const int *fds, size_t fdslen)
1203 {
1204 	struct filedesc *ofdp;
1205 	const char *p;
1206 	int *kfds;
1207 	int error;
1208 
1209 	memset(args, '\0', sizeof(*args));
1210 	ofdp = td->td_proc->p_fd;
1211 	if (datalen >= ARG_MAX || fdslen >= ofdp->fd_nfiles)
1212 		return (E2BIG);
1213 	error = exec_alloc_args(args);
1214 	if (error != 0)
1215 		return (error);
1216 
1217 	args->begin_argv = args->buf;
1218 	args->stringspace = ARG_MAX;
1219 
1220 	if (datalen > 0) {
1221 		/*
1222 		 * Argument buffer has been provided. Copy it into the
1223 		 * kernel as a single string and add a terminating null
1224 		 * byte.
1225 		 */
1226 		error = copyin(data, args->begin_argv, datalen);
1227 		if (error != 0)
1228 			goto err_exit;
1229 		args->begin_argv[datalen] = '\0';
1230 		args->endp = args->begin_argv + datalen + 1;
1231 		args->stringspace -= datalen + 1;
1232 
1233 		/*
1234 		 * Traditional argument counting. Count the number of
1235 		 * null bytes.
1236 		 */
1237 		for (p = args->begin_argv; p < args->endp; ++p)
1238 			if (*p == '\0')
1239 				++args->argc;
1240 	} else {
1241 		/* No argument buffer provided. */
1242 		args->endp = args->begin_argv;
1243 	}
1244 
1245 	/* Create new file descriptor table. */
1246 	kfds = malloc(fdslen * sizeof(int), M_TEMP, M_WAITOK);
1247 	error = copyin(fds, kfds, fdslen * sizeof(int));
1248 	if (error != 0) {
1249 		free(kfds, M_TEMP);
1250 		goto err_exit;
1251 	}
1252 	error = fdcopy_remapped(ofdp, kfds, fdslen, &args->fdp);
1253 	free(kfds, M_TEMP);
1254 	if (error != 0)
1255 		goto err_exit;
1256 
1257 	return (0);
1258 err_exit:
1259 	exec_free_args(args);
1260 	return (error);
1261 }
1262 
1263 struct exec_args_kva {
1264 	vm_offset_t addr;
1265 	u_int gen;
1266 	SLIST_ENTRY(exec_args_kva) next;
1267 };
1268 
1269 DPCPU_DEFINE_STATIC(struct exec_args_kva *, exec_args_kva);
1270 
1271 static SLIST_HEAD(, exec_args_kva) exec_args_kva_freelist;
1272 static struct mtx exec_args_kva_mtx;
1273 static u_int exec_args_gen;
1274 
1275 static void
1276 exec_prealloc_args_kva(void *arg __unused)
1277 {
1278 	struct exec_args_kva *argkva;
1279 	u_int i;
1280 
1281 	SLIST_INIT(&exec_args_kva_freelist);
1282 	mtx_init(&exec_args_kva_mtx, "exec args kva", NULL, MTX_DEF);
1283 	for (i = 0; i < exec_map_entries; i++) {
1284 		argkva = malloc(sizeof(*argkva), M_PARGS, M_WAITOK);
1285 		argkva->addr = kmap_alloc_wait(exec_map, exec_map_entry_size);
1286 		argkva->gen = exec_args_gen;
1287 		SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1288 	}
1289 }
1290 SYSINIT(exec_args_kva, SI_SUB_EXEC, SI_ORDER_ANY, exec_prealloc_args_kva, NULL);
1291 
1292 static vm_offset_t
1293 exec_alloc_args_kva(void **cookie)
1294 {
1295 	struct exec_args_kva *argkva;
1296 
1297 	argkva = (void *)atomic_readandclear_ptr(
1298 	    (uintptr_t *)DPCPU_PTR(exec_args_kva));
1299 	if (argkva == NULL) {
1300 		mtx_lock(&exec_args_kva_mtx);
1301 		while ((argkva = SLIST_FIRST(&exec_args_kva_freelist)) == NULL)
1302 			(void)mtx_sleep(&exec_args_kva_freelist,
1303 			    &exec_args_kva_mtx, 0, "execkva", 0);
1304 		SLIST_REMOVE_HEAD(&exec_args_kva_freelist, next);
1305 		mtx_unlock(&exec_args_kva_mtx);
1306 	}
1307 	*(struct exec_args_kva **)cookie = argkva;
1308 	return (argkva->addr);
1309 }
1310 
1311 static void
1312 exec_release_args_kva(struct exec_args_kva *argkva, u_int gen)
1313 {
1314 	vm_offset_t base;
1315 
1316 	base = argkva->addr;
1317 	if (argkva->gen != gen) {
1318 		(void)vm_map_madvise(exec_map, base, base + exec_map_entry_size,
1319 		    MADV_FREE);
1320 		argkva->gen = gen;
1321 	}
1322 	if (!atomic_cmpset_ptr((uintptr_t *)DPCPU_PTR(exec_args_kva),
1323 	    (uintptr_t)NULL, (uintptr_t)argkva)) {
1324 		mtx_lock(&exec_args_kva_mtx);
1325 		SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1326 		wakeup_one(&exec_args_kva_freelist);
1327 		mtx_unlock(&exec_args_kva_mtx);
1328 	}
1329 }
1330 
1331 static void
1332 exec_free_args_kva(void *cookie)
1333 {
1334 
1335 	exec_release_args_kva(cookie, exec_args_gen);
1336 }
1337 
1338 static void
1339 exec_args_kva_lowmem(void *arg __unused)
1340 {
1341 	SLIST_HEAD(, exec_args_kva) head;
1342 	struct exec_args_kva *argkva;
1343 	u_int gen;
1344 	int i;
1345 
1346 	gen = atomic_fetchadd_int(&exec_args_gen, 1) + 1;
1347 
1348 	/*
1349 	 * Force an madvise of each KVA range. Any currently allocated ranges
1350 	 * will have MADV_FREE applied once they are freed.
1351 	 */
1352 	SLIST_INIT(&head);
1353 	mtx_lock(&exec_args_kva_mtx);
1354 	SLIST_SWAP(&head, &exec_args_kva_freelist, exec_args_kva);
1355 	mtx_unlock(&exec_args_kva_mtx);
1356 	while ((argkva = SLIST_FIRST(&head)) != NULL) {
1357 		SLIST_REMOVE_HEAD(&head, next);
1358 		exec_release_args_kva(argkva, gen);
1359 	}
1360 
1361 	CPU_FOREACH(i) {
1362 		argkva = (void *)atomic_readandclear_ptr(
1363 		    (uintptr_t *)DPCPU_ID_PTR(i, exec_args_kva));
1364 		if (argkva != NULL)
1365 			exec_release_args_kva(argkva, gen);
1366 	}
1367 }
1368 EVENTHANDLER_DEFINE(vm_lowmem, exec_args_kva_lowmem, NULL,
1369     EVENTHANDLER_PRI_ANY);
1370 
1371 /*
1372  * Allocate temporary demand-paged, zero-filled memory for the file name,
1373  * argument, and environment strings.
1374  */
1375 int
1376 exec_alloc_args(struct image_args *args)
1377 {
1378 
1379 	args->buf = (char *)exec_alloc_args_kva(&args->bufkva);
1380 	return (0);
1381 }
1382 
1383 void
1384 exec_free_args(struct image_args *args)
1385 {
1386 
1387 	if (args->buf != NULL) {
1388 		exec_free_args_kva(args->bufkva);
1389 		args->buf = NULL;
1390 	}
1391 	if (args->fname_buf != NULL) {
1392 		free(args->fname_buf, M_TEMP);
1393 		args->fname_buf = NULL;
1394 	}
1395 	if (args->fdp != NULL)
1396 		fdescfree_remapped(args->fdp);
1397 }
1398 
1399 /*
1400  * A set to functions to fill struct image args.
1401  *
1402  * NOTE: exec_args_add_fname() must be called (possibly with a NULL
1403  * fname) before the other functions.  All exec_args_add_arg() calls must
1404  * be made before any exec_args_add_env() calls.  exec_args_adjust_args()
1405  * may be called any time after exec_args_add_fname().
1406  *
1407  * exec_args_add_fname() - install path to be executed
1408  * exec_args_add_arg() - append an argument string
1409  * exec_args_add_env() - append an env string
1410  * exec_args_adjust_args() - adjust location of the argument list to
1411  *                           allow new arguments to be prepended
1412  */
1413 int
1414 exec_args_add_fname(struct image_args *args, const char *fname,
1415     enum uio_seg segflg)
1416 {
1417 	int error;
1418 	size_t length;
1419 
1420 	KASSERT(args->fname == NULL, ("fname already appended"));
1421 	KASSERT(args->endp == NULL, ("already appending to args"));
1422 
1423 	if (fname != NULL) {
1424 		args->fname = args->buf;
1425 		error = segflg == UIO_SYSSPACE ?
1426 		    copystr(fname, args->fname, PATH_MAX, &length) :
1427 		    copyinstr(fname, args->fname, PATH_MAX, &length);
1428 		if (error != 0)
1429 			return (error == ENAMETOOLONG ? E2BIG : error);
1430 	} else
1431 		length = 0;
1432 
1433 	/* Set up for _arg_*()/_env_*() */
1434 	args->endp = args->buf + length;
1435 	/* begin_argv must be set and kept updated */
1436 	args->begin_argv = args->endp;
1437 	KASSERT(exec_map_entry_size - length >= ARG_MAX,
1438 	    ("too little space remaining for arguments %zu < %zu",
1439 	    exec_map_entry_size - length, (size_t)ARG_MAX));
1440 	args->stringspace = ARG_MAX;
1441 
1442 	return (0);
1443 }
1444 
1445 static int
1446 exec_args_add_str(struct image_args *args, const char *str,
1447     enum uio_seg segflg, int *countp)
1448 {
1449 	int error;
1450 	size_t length;
1451 
1452 	KASSERT(args->endp != NULL, ("endp not initialized"));
1453 	KASSERT(args->begin_argv != NULL, ("begin_argp not initialized"));
1454 
1455 	error = (segflg == UIO_SYSSPACE) ?
1456 	    copystr(str, args->endp, args->stringspace, &length) :
1457 	    copyinstr(str, args->endp, args->stringspace, &length);
1458 	if (error != 0)
1459 		return (error == ENAMETOOLONG ? E2BIG : error);
1460 	args->stringspace -= length;
1461 	args->endp += length;
1462 	(*countp)++;
1463 
1464 	return (0);
1465 }
1466 
1467 int
1468 exec_args_add_arg(struct image_args *args, const char *argp,
1469     enum uio_seg segflg)
1470 {
1471 
1472 	KASSERT(args->envc == 0, ("appending args after env"));
1473 
1474 	return (exec_args_add_str(args, argp, segflg, &args->argc));
1475 }
1476 
1477 int
1478 exec_args_add_env(struct image_args *args, const char *envp,
1479     enum uio_seg segflg)
1480 {
1481 
1482 	if (args->envc == 0)
1483 		args->begin_envv = args->endp;
1484 
1485 	return (exec_args_add_str(args, envp, segflg, &args->envc));
1486 }
1487 
1488 int
1489 exec_args_adjust_args(struct image_args *args, size_t consume, ssize_t extend)
1490 {
1491 	ssize_t offset;
1492 
1493 	KASSERT(args->endp != NULL, ("endp not initialized"));
1494 	KASSERT(args->begin_argv != NULL, ("begin_argp not initialized"));
1495 
1496 	offset = extend - consume;
1497 	if (args->stringspace < offset)
1498 		return (E2BIG);
1499 	memmove(args->begin_argv + extend, args->begin_argv + consume,
1500 	    args->endp - args->begin_argv + consume);
1501 	if (args->envc > 0)
1502 		args->begin_envv += offset;
1503 	args->endp += offset;
1504 	args->stringspace -= offset;
1505 	return (0);
1506 }
1507 
1508 char *
1509 exec_args_get_begin_envv(struct image_args *args)
1510 {
1511 
1512 	KASSERT(args->endp != NULL, ("endp not initialized"));
1513 
1514 	if (args->envc > 0)
1515 		return (args->begin_envv);
1516 	return (args->endp);
1517 }
1518 
1519 /*
1520  * Copy strings out to the new process address space, constructing new arg
1521  * and env vector tables. Return a pointer to the base so that it can be used
1522  * as the initial stack pointer.
1523  */
1524 int
1525 exec_copyout_strings(struct image_params *imgp, uintptr_t *stack_base)
1526 {
1527 	int argc, envc;
1528 	char **vectp;
1529 	char *stringp;
1530 	uintptr_t destp, ustringp;
1531 	struct ps_strings *arginfo;
1532 	struct proc *p;
1533 	size_t execpath_len;
1534 	int error, szsigcode, szps;
1535 	char canary[sizeof(long) * 8];
1536 
1537 	szps = sizeof(pagesizes[0]) * MAXPAGESIZES;
1538 	/*
1539 	 * Calculate string base and vector table pointers.
1540 	 * Also deal with signal trampoline code for this exec type.
1541 	 */
1542 	if (imgp->execpath != NULL && imgp->auxargs != NULL)
1543 		execpath_len = strlen(imgp->execpath) + 1;
1544 	else
1545 		execpath_len = 0;
1546 	p = imgp->proc;
1547 	szsigcode = 0;
1548 	arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings;
1549 	imgp->ps_strings = arginfo;
1550 	if (p->p_sysent->sv_sigcode_base == 0) {
1551 		if (p->p_sysent->sv_szsigcode != NULL)
1552 			szsigcode = *(p->p_sysent->sv_szsigcode);
1553 	}
1554 	destp =	(uintptr_t)arginfo;
1555 
1556 	/*
1557 	 * install sigcode
1558 	 */
1559 	if (szsigcode != 0) {
1560 		destp -= szsigcode;
1561 		destp = rounddown2(destp, sizeof(void *));
1562 		error = copyout(p->p_sysent->sv_sigcode, (void *)destp,
1563 		    szsigcode);
1564 		if (error != 0)
1565 			return (error);
1566 	}
1567 
1568 	/*
1569 	 * Copy the image path for the rtld.
1570 	 */
1571 	if (execpath_len != 0) {
1572 		destp -= execpath_len;
1573 		destp = rounddown2(destp, sizeof(void *));
1574 		imgp->execpathp = (void *)destp;
1575 		error = copyout(imgp->execpath, imgp->execpathp, execpath_len);
1576 		if (error != 0)
1577 			return (error);
1578 	}
1579 
1580 	/*
1581 	 * Prepare the canary for SSP.
1582 	 */
1583 	arc4rand(canary, sizeof(canary), 0);
1584 	destp -= sizeof(canary);
1585 	imgp->canary = (void *)destp;
1586 	error = copyout(canary, imgp->canary, sizeof(canary));
1587 	if (error != 0)
1588 		return (error);
1589 	imgp->canarylen = sizeof(canary);
1590 
1591 	/*
1592 	 * Prepare the pagesizes array.
1593 	 */
1594 	destp -= szps;
1595 	destp = rounddown2(destp, sizeof(void *));
1596 	imgp->pagesizes = (void *)destp;
1597 	error = copyout(pagesizes, imgp->pagesizes, szps);
1598 	if (error != 0)
1599 		return (error);
1600 	imgp->pagesizeslen = szps;
1601 
1602 	/*
1603 	 * Allocate room for the argument and environment strings.
1604 	 */
1605 	destp -= ARG_MAX - imgp->args->stringspace;
1606 	destp = rounddown2(destp, sizeof(void *));
1607 	ustringp = destp;
1608 
1609 	if (imgp->sysent->sv_stackgap != NULL)
1610 		imgp->sysent->sv_stackgap(imgp, &destp);
1611 
1612 	if (imgp->auxargs) {
1613 		/*
1614 		 * Allocate room on the stack for the ELF auxargs
1615 		 * array.  It has up to AT_COUNT entries.
1616 		 */
1617 		destp -= AT_COUNT * sizeof(Elf_Auxinfo);
1618 		destp = rounddown2(destp, sizeof(void *));
1619 	}
1620 
1621 	vectp = (char **)destp;
1622 
1623 	/*
1624 	 * Allocate room for the argv[] and env vectors including the
1625 	 * terminating NULL pointers.
1626 	 */
1627 	vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
1628 
1629 	/*
1630 	 * vectp also becomes our initial stack base
1631 	 */
1632 	*stack_base = (uintptr_t)vectp;
1633 
1634 	stringp = imgp->args->begin_argv;
1635 	argc = imgp->args->argc;
1636 	envc = imgp->args->envc;
1637 
1638 	/*
1639 	 * Copy out strings - arguments and environment.
1640 	 */
1641 	error = copyout(stringp, (void *)ustringp,
1642 	    ARG_MAX - imgp->args->stringspace);
1643 	if (error != 0)
1644 		return (error);
1645 
1646 	/*
1647 	 * Fill in "ps_strings" struct for ps, w, etc.
1648 	 */
1649 	imgp->argv = vectp;
1650 	if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 ||
1651 	    suword32(&arginfo->ps_nargvstr, argc) != 0)
1652 		return (EFAULT);
1653 
1654 	/*
1655 	 * Fill in argument portion of vector table.
1656 	 */
1657 	for (; argc > 0; --argc) {
1658 		if (suword(vectp++, ustringp) != 0)
1659 			return (EFAULT);
1660 		while (*stringp++ != 0)
1661 			ustringp++;
1662 		ustringp++;
1663 	}
1664 
1665 	/* a null vector table pointer separates the argp's from the envp's */
1666 	if (suword(vectp++, 0) != 0)
1667 		return (EFAULT);
1668 
1669 	imgp->envv = vectp;
1670 	if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 ||
1671 	    suword32(&arginfo->ps_nenvstr, envc) != 0)
1672 		return (EFAULT);
1673 
1674 	/*
1675 	 * Fill in environment portion of vector table.
1676 	 */
1677 	for (; envc > 0; --envc) {
1678 		if (suword(vectp++, ustringp) != 0)
1679 			return (EFAULT);
1680 		while (*stringp++ != 0)
1681 			ustringp++;
1682 		ustringp++;
1683 	}
1684 
1685 	/* end of vector table is a null pointer */
1686 	if (suword(vectp, 0) != 0)
1687 		return (EFAULT);
1688 
1689 	if (imgp->auxargs) {
1690 		vectp++;
1691 		error = imgp->sysent->sv_copyout_auxargs(imgp,
1692 		    (uintptr_t)vectp);
1693 		if (error != 0)
1694 			return (error);
1695 	}
1696 
1697 	return (0);
1698 }
1699 
1700 /*
1701  * Check permissions of file to execute.
1702  *	Called with imgp->vp locked.
1703  *	Return 0 for success or error code on failure.
1704  */
1705 int
1706 exec_check_permissions(struct image_params *imgp)
1707 {
1708 	struct vnode *vp = imgp->vp;
1709 	struct vattr *attr = imgp->attr;
1710 	struct thread *td;
1711 	int error;
1712 
1713 	td = curthread;
1714 
1715 	/* Get file attributes */
1716 	error = VOP_GETATTR(vp, attr, td->td_ucred);
1717 	if (error)
1718 		return (error);
1719 
1720 #ifdef MAC
1721 	error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1722 	if (error)
1723 		return (error);
1724 #endif
1725 
1726 	/*
1727 	 * 1) Check if file execution is disabled for the filesystem that
1728 	 *    this file resides on.
1729 	 * 2) Ensure that at least one execute bit is on. Otherwise, a
1730 	 *    privileged user will always succeed, and we don't want this
1731 	 *    to happen unless the file really is executable.
1732 	 * 3) Ensure that the file is a regular file.
1733 	 */
1734 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1735 	    (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1736 	    (attr->va_type != VREG))
1737 		return (EACCES);
1738 
1739 	/*
1740 	 * Zero length files can't be exec'd
1741 	 */
1742 	if (attr->va_size == 0)
1743 		return (ENOEXEC);
1744 
1745 	/*
1746 	 *  Check for execute permission to file based on current credentials.
1747 	 */
1748 	error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1749 	if (error)
1750 		return (error);
1751 
1752 	/*
1753 	 * Check number of open-for-writes on the file and deny execution
1754 	 * if there are any.
1755 	 *
1756 	 * Add a text reference now so no one can write to the
1757 	 * executable while we're activating it.
1758 	 *
1759 	 * Remember if this was set before and unset it in case this is not
1760 	 * actually an executable image.
1761 	 */
1762 	error = VOP_SET_TEXT(vp);
1763 	if (error != 0)
1764 		return (error);
1765 	imgp->textset = true;
1766 
1767 	/*
1768 	 * Call filesystem specific open routine (which does nothing in the
1769 	 * general case).
1770 	 */
1771 	error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1772 	if (error == 0)
1773 		imgp->opened = 1;
1774 	return (error);
1775 }
1776 
1777 /*
1778  * Exec handler registration
1779  */
1780 int
1781 exec_register(const struct execsw *execsw_arg)
1782 {
1783 	const struct execsw **es, **xs, **newexecsw;
1784 	u_int count = 2;	/* New slot and trailing NULL */
1785 
1786 	if (execsw)
1787 		for (es = execsw; *es; es++)
1788 			count++;
1789 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1790 	xs = newexecsw;
1791 	if (execsw)
1792 		for (es = execsw; *es; es++)
1793 			*xs++ = *es;
1794 	*xs++ = execsw_arg;
1795 	*xs = NULL;
1796 	if (execsw)
1797 		free(execsw, M_TEMP);
1798 	execsw = newexecsw;
1799 	return (0);
1800 }
1801 
1802 int
1803 exec_unregister(const struct execsw *execsw_arg)
1804 {
1805 	const struct execsw **es, **xs, **newexecsw;
1806 	int count = 1;
1807 
1808 	if (execsw == NULL)
1809 		panic("unregister with no handlers left?\n");
1810 
1811 	for (es = execsw; *es; es++) {
1812 		if (*es == execsw_arg)
1813 			break;
1814 	}
1815 	if (*es == NULL)
1816 		return (ENOENT);
1817 	for (es = execsw; *es; es++)
1818 		if (*es != execsw_arg)
1819 			count++;
1820 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1821 	xs = newexecsw;
1822 	for (es = execsw; *es; es++)
1823 		if (*es != execsw_arg)
1824 			*xs++ = *es;
1825 	*xs = NULL;
1826 	if (execsw)
1827 		free(execsw, M_TEMP);
1828 	execsw = newexecsw;
1829 	return (0);
1830 }
1831