xref: /freebsd/sys/kern/kern_exec.c (revision b197d4b893974c9eb4d7b38704c6d5c486235d6f)
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/asan.h>
41 #include <sys/capsicum.h>
42 #include <sys/compressor.h>
43 #include <sys/eventhandler.h>
44 #include <sys/exec.h>
45 #include <sys/fcntl.h>
46 #include <sys/filedesc.h>
47 #include <sys/imgact.h>
48 #include <sys/imgact_elf.h>
49 #include <sys/kernel.h>
50 #include <sys/lock.h>
51 #include <sys/malloc.h>
52 #include <sys/mman.h>
53 #include <sys/mount.h>
54 #include <sys/mutex.h>
55 #include <sys/namei.h>
56 #include <sys/priv.h>
57 #include <sys/proc.h>
58 #include <sys/ptrace.h>
59 #include <sys/reg.h>
60 #include <sys/resourcevar.h>
61 #include <sys/rwlock.h>
62 #include <sys/sched.h>
63 #include <sys/sdt.h>
64 #include <sys/sf_buf.h>
65 #include <sys/shm.h>
66 #include <sys/signalvar.h>
67 #include <sys/smp.h>
68 #include <sys/stat.h>
69 #include <sys/syscallsubr.h>
70 #include <sys/sysctl.h>
71 #include <sys/sysent.h>
72 #include <sys/sysproto.h>
73 #include <sys/timers.h>
74 #include <sys/umtxvar.h>
75 #include <sys/vnode.h>
76 #include <sys/wait.h>
77 #ifdef KTRACE
78 #include <sys/ktrace.h>
79 #endif
80 
81 #include <vm/vm.h>
82 #include <vm/vm_param.h>
83 #include <vm/pmap.h>
84 #include <vm/vm_page.h>
85 #include <vm/vm_map.h>
86 #include <vm/vm_kern.h>
87 #include <vm/vm_extern.h>
88 #include <vm/vm_object.h>
89 #include <vm/vm_pager.h>
90 
91 #ifdef	HWPMC_HOOKS
92 #include <sys/pmckern.h>
93 #endif
94 
95 #include <security/audit/audit.h>
96 #include <security/mac/mac_framework.h>
97 
98 #ifdef KDTRACE_HOOKS
99 #include <sys/dtrace_bsd.h>
100 dtrace_execexit_func_t	dtrace_fasttrap_exec;
101 #endif
102 
103 SDT_PROVIDER_DECLARE(proc);
104 SDT_PROBE_DEFINE1(proc, , , exec, "char *");
105 SDT_PROBE_DEFINE1(proc, , , exec__failure, "int");
106 SDT_PROBE_DEFINE1(proc, , , exec__success, "char *");
107 
108 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments");
109 
110 int coredump_pack_fileinfo = 1;
111 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_fileinfo, CTLFLAG_RWTUN,
112     &coredump_pack_fileinfo, 0,
113     "Enable file path packing in 'procstat -f' coredump notes");
114 
115 int coredump_pack_vmmapinfo = 1;
116 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_vmmapinfo, CTLFLAG_RWTUN,
117     &coredump_pack_vmmapinfo, 0,
118     "Enable file path packing in 'procstat -v' coredump notes");
119 
120 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS);
121 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS);
122 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS);
123 static int do_execve(struct thread *td, struct image_args *args,
124     struct mac *mac_p, struct vmspace *oldvmspace);
125 
126 /* XXX This should be vm_size_t. */
127 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD|
128     CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_ps_strings, "LU",
129     "Location of process' ps_strings structure");
130 
131 /* XXX This should be vm_size_t. */
132 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD|
133     CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_usrstack, "LU",
134     "Top of process stack");
135 
136 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD|CTLFLAG_MPSAFE,
137     NULL, 0, sysctl_kern_stackprot, "I",
138     "Stack memory permissions");
139 
140 u_long ps_arg_cache_limit = PAGE_SIZE / 16;
141 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW,
142     &ps_arg_cache_limit, 0,
143     "Process' command line characters cache limit");
144 
145 static int disallow_high_osrel;
146 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW,
147     &disallow_high_osrel, 0,
148     "Disallow execution of binaries built for higher version of the world");
149 
150 static int map_at_zero = 0;
151 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RWTUN, &map_at_zero, 0,
152     "Permit processes to map an object at virtual address 0.");
153 
154 static int core_dump_can_intr = 1;
155 SYSCTL_INT(_kern, OID_AUTO, core_dump_can_intr, CTLFLAG_RWTUN,
156     &core_dump_can_intr, 0,
157     "Core dumping interruptible with SIGKILL");
158 
159 static int
160 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS)
161 {
162 	struct proc *p;
163 	vm_offset_t ps_strings;
164 
165 	p = curproc;
166 #ifdef SCTL_MASK32
167 	if (req->flags & SCTL_MASK32) {
168 		unsigned int val;
169 		val = (unsigned int)PROC_PS_STRINGS(p);
170 		return (SYSCTL_OUT(req, &val, sizeof(val)));
171 	}
172 #endif
173 	ps_strings = PROC_PS_STRINGS(p);
174 	return (SYSCTL_OUT(req, &ps_strings, sizeof(ps_strings)));
175 }
176 
177 static int
178 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS)
179 {
180 	struct proc *p;
181 	vm_offset_t val;
182 
183 	p = curproc;
184 #ifdef SCTL_MASK32
185 	if (req->flags & SCTL_MASK32) {
186 		unsigned int val32;
187 
188 		val32 = round_page((unsigned int)p->p_vmspace->vm_stacktop);
189 		return (SYSCTL_OUT(req, &val32, sizeof(val32)));
190 	}
191 #endif
192 	val = round_page(p->p_vmspace->vm_stacktop);
193 	return (SYSCTL_OUT(req, &val, sizeof(val)));
194 }
195 
196 static int
197 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS)
198 {
199 	struct proc *p;
200 
201 	p = curproc;
202 	return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot,
203 	    sizeof(p->p_sysent->sv_stackprot)));
204 }
205 
206 /*
207  * Each of the items is a pointer to a `const struct execsw', hence the
208  * double pointer here.
209  */
210 static const struct execsw **execsw;
211 
212 #ifndef _SYS_SYSPROTO_H_
213 struct execve_args {
214 	char    *fname;
215 	char    **argv;
216 	char    **envv;
217 };
218 #endif
219 
220 int
221 sys_execve(struct thread *td, struct execve_args *uap)
222 {
223 	struct image_args args;
224 	struct vmspace *oldvmspace;
225 	int error;
226 
227 	error = pre_execve(td, &oldvmspace);
228 	if (error != 0)
229 		return (error);
230 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
231 	    uap->argv, uap->envv);
232 	if (error == 0)
233 		error = kern_execve(td, &args, NULL, oldvmspace);
234 	post_execve(td, error, oldvmspace);
235 	AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
236 	return (error);
237 }
238 
239 #ifndef _SYS_SYSPROTO_H_
240 struct fexecve_args {
241 	int	fd;
242 	char	**argv;
243 	char	**envv;
244 };
245 #endif
246 int
247 sys_fexecve(struct thread *td, struct fexecve_args *uap)
248 {
249 	struct image_args args;
250 	struct vmspace *oldvmspace;
251 	int error;
252 
253 	error = pre_execve(td, &oldvmspace);
254 	if (error != 0)
255 		return (error);
256 	error = exec_copyin_args(&args, NULL, UIO_SYSSPACE,
257 	    uap->argv, uap->envv);
258 	if (error == 0) {
259 		args.fd = uap->fd;
260 		error = kern_execve(td, &args, NULL, oldvmspace);
261 	}
262 	post_execve(td, error, oldvmspace);
263 	AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
264 	return (error);
265 }
266 
267 #ifndef _SYS_SYSPROTO_H_
268 struct __mac_execve_args {
269 	char	*fname;
270 	char	**argv;
271 	char	**envv;
272 	struct mac	*mac_p;
273 };
274 #endif
275 
276 int
277 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap)
278 {
279 #ifdef MAC
280 	struct image_args args;
281 	struct vmspace *oldvmspace;
282 	int error;
283 
284 	error = pre_execve(td, &oldvmspace);
285 	if (error != 0)
286 		return (error);
287 	error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE,
288 	    uap->argv, uap->envv);
289 	if (error == 0)
290 		error = kern_execve(td, &args, uap->mac_p, oldvmspace);
291 	post_execve(td, error, oldvmspace);
292 	AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td);
293 	return (error);
294 #else
295 	return (ENOSYS);
296 #endif
297 }
298 
299 int
300 pre_execve(struct thread *td, struct vmspace **oldvmspace)
301 {
302 	struct proc *p;
303 	int error;
304 
305 	KASSERT(td == curthread, ("non-current thread %p", td));
306 	error = 0;
307 	p = td->td_proc;
308 	if ((p->p_flag & P_HADTHREADS) != 0) {
309 		PROC_LOCK(p);
310 		if (thread_single(p, SINGLE_BOUNDARY) != 0)
311 			error = ERESTART;
312 		PROC_UNLOCK(p);
313 	}
314 	KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0,
315 	    ("nested execve"));
316 	*oldvmspace = p->p_vmspace;
317 	return (error);
318 }
319 
320 void
321 post_execve(struct thread *td, int error, struct vmspace *oldvmspace)
322 {
323 	struct proc *p;
324 
325 	KASSERT(td == curthread, ("non-current thread %p", td));
326 	p = td->td_proc;
327 	if ((p->p_flag & P_HADTHREADS) != 0) {
328 		PROC_LOCK(p);
329 		/*
330 		 * If success, we upgrade to SINGLE_EXIT state to
331 		 * force other threads to suicide.
332 		 */
333 		if (error == EJUSTRETURN)
334 			thread_single(p, SINGLE_EXIT);
335 		else
336 			thread_single_end(p, SINGLE_BOUNDARY);
337 		PROC_UNLOCK(p);
338 	}
339 	exec_cleanup(td, oldvmspace);
340 }
341 
342 /*
343  * kern_execve() has the astonishing property of not always returning to
344  * the caller.  If sufficiently bad things happen during the call to
345  * do_execve(), it can end up calling exit1(); as a result, callers must
346  * avoid doing anything which they might need to undo (e.g., allocating
347  * memory).
348  */
349 int
350 kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p,
351     struct vmspace *oldvmspace)
352 {
353 
354 	TSEXEC(td->td_proc->p_pid, args->begin_argv);
355 	AUDIT_ARG_ARGV(args->begin_argv, args->argc,
356 	    exec_args_get_begin_envv(args) - args->begin_argv);
357 	AUDIT_ARG_ENVV(exec_args_get_begin_envv(args), args->envc,
358 	    args->endp - exec_args_get_begin_envv(args));
359 
360 	/* Must have at least one argument. */
361 	if (args->argc == 0) {
362 		exec_free_args(args);
363 		return (EINVAL);
364 	}
365 	return (do_execve(td, args, mac_p, oldvmspace));
366 }
367 
368 static void
369 execve_nosetid(struct image_params *imgp)
370 {
371 	imgp->credential_setid = false;
372 	if (imgp->newcred != NULL) {
373 		crfree(imgp->newcred);
374 		imgp->newcred = NULL;
375 	}
376 }
377 
378 /*
379  * In-kernel implementation of execve().  All arguments are assumed to be
380  * userspace pointers from the passed thread.
381  */
382 static int
383 do_execve(struct thread *td, struct image_args *args, struct mac *mac_p,
384     struct vmspace *oldvmspace)
385 {
386 	struct proc *p = td->td_proc;
387 	struct nameidata nd;
388 	struct ucred *oldcred;
389 	struct uidinfo *euip = NULL;
390 	uintptr_t stack_base;
391 	struct image_params image_params, *imgp;
392 	struct vattr attr;
393 	int (*img_first)(struct image_params *);
394 	struct pargs *oldargs = NULL, *newargs = NULL;
395 	struct sigacts *oldsigacts = NULL, *newsigacts = NULL;
396 #ifdef KTRACE
397 	struct ktr_io_params *kiop;
398 #endif
399 	struct vnode *oldtextvp, *newtextvp;
400 	struct vnode *oldtextdvp, *newtextdvp;
401 	char *oldbinname, *newbinname;
402 	bool credential_changing;
403 #ifdef MAC
404 	struct label *interpvplabel = NULL;
405 	bool will_transition;
406 #endif
407 #ifdef HWPMC_HOOKS
408 	struct pmckern_procexec pe;
409 #endif
410 	int error, i, orig_osrel;
411 	uint32_t orig_fctl0;
412 	Elf_Brandinfo *orig_brandinfo;
413 	size_t freepath_size;
414 	static const char fexecv_proc_title[] = "(fexecv)";
415 
416 	imgp = &image_params;
417 	oldtextvp = oldtextdvp = NULL;
418 	newtextvp = newtextdvp = NULL;
419 	newbinname = oldbinname = NULL;
420 #ifdef KTRACE
421 	kiop = NULL;
422 #endif
423 
424 	/*
425 	 * Lock the process and set the P_INEXEC flag to indicate that
426 	 * it should be left alone until we're done here.  This is
427 	 * necessary to avoid race conditions - e.g. in ptrace() -
428 	 * that might allow a local user to illicitly obtain elevated
429 	 * privileges.
430 	 */
431 	PROC_LOCK(p);
432 	KASSERT((p->p_flag & P_INEXEC) == 0,
433 	    ("%s(): process already has P_INEXEC flag", __func__));
434 	p->p_flag |= P_INEXEC;
435 	PROC_UNLOCK(p);
436 
437 	/*
438 	 * Initialize part of the common data
439 	 */
440 	bzero(imgp, sizeof(*imgp));
441 	imgp->proc = p;
442 	imgp->attr = &attr;
443 	imgp->args = args;
444 	oldcred = p->p_ucred;
445 	orig_osrel = p->p_osrel;
446 	orig_fctl0 = p->p_fctl0;
447 	orig_brandinfo = p->p_elf_brandinfo;
448 
449 #ifdef MAC
450 	error = mac_execve_enter(imgp, mac_p);
451 	if (error)
452 		goto exec_fail;
453 #endif
454 
455 	SDT_PROBE1(proc, , , exec, args->fname);
456 
457 interpret:
458 	if (args->fname != NULL) {
459 #ifdef CAPABILITY_MODE
460 		/*
461 		 * While capability mode can't reach this point via direct
462 		 * path arguments to execve(), we also don't allow
463 		 * interpreters to be used in capability mode (for now).
464 		 * Catch indirect lookups and return a permissions error.
465 		 */
466 		if (IN_CAPABILITY_MODE(td)) {
467 			error = ECAPMODE;
468 			goto exec_fail;
469 		}
470 #endif
471 
472 		/*
473 		 * Translate the file name. namei() returns a vnode
474 		 * pointer in ni_vp among other things.
475 		 */
476 		NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW |
477 		    SAVENAME | AUDITVNODE1 | WANTPARENT, UIO_SYSSPACE,
478 		    args->fname);
479 
480 		error = namei(&nd);
481 		if (error)
482 			goto exec_fail;
483 
484 		newtextvp = nd.ni_vp;
485 		newtextdvp = nd.ni_dvp;
486 		nd.ni_dvp = NULL;
487 		newbinname = malloc(nd.ni_cnd.cn_namelen + 1, M_PARGS,
488 		    M_WAITOK);
489 		memcpy(newbinname, nd.ni_cnd.cn_nameptr, nd.ni_cnd.cn_namelen);
490 		newbinname[nd.ni_cnd.cn_namelen] = '\0';
491 		imgp->vp = newtextvp;
492 
493 		/*
494 		 * Do the best to calculate the full path to the image file.
495 		 */
496 		if (args->fname[0] == '/') {
497 			imgp->execpath = args->fname;
498 		} else {
499 			VOP_UNLOCK(imgp->vp);
500 			freepath_size = MAXPATHLEN;
501 			if (vn_fullpath_hardlink(newtextvp, newtextdvp,
502 			    newbinname, nd.ni_cnd.cn_namelen, &imgp->execpath,
503 			    &imgp->freepath, &freepath_size) != 0)
504 				imgp->execpath = args->fname;
505 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
506 		}
507 	} else {
508 		AUDIT_ARG_FD(args->fd);
509 
510 		/*
511 		 * If the descriptors was not opened with O_PATH, then
512 		 * we require that it was opened with O_EXEC or
513 		 * O_RDONLY.  In either case, exec_check_permissions()
514 		 * below checks _current_ file access mode regardless
515 		 * of the permissions additionally checked at the
516 		 * open(2).
517 		 */
518 		error = fgetvp_exec(td, args->fd, &cap_fexecve_rights,
519 		    &newtextvp);
520 		if (error != 0)
521 			goto exec_fail;
522 
523 		if (vn_fullpath(newtextvp, &imgp->execpath,
524 		    &imgp->freepath) != 0)
525 			imgp->execpath = args->fname;
526 		vn_lock(newtextvp, LK_SHARED | LK_RETRY);
527 		AUDIT_ARG_VNODE1(newtextvp);
528 		imgp->vp = newtextvp;
529 	}
530 
531 	/*
532 	 * Check file permissions.  Also 'opens' file and sets its vnode to
533 	 * text mode.
534 	 */
535 	error = exec_check_permissions(imgp);
536 	if (error)
537 		goto exec_fail_dealloc;
538 
539 	imgp->object = imgp->vp->v_object;
540 	if (imgp->object != NULL)
541 		vm_object_reference(imgp->object);
542 
543 	error = exec_map_first_page(imgp);
544 	if (error)
545 		goto exec_fail_dealloc;
546 
547 	imgp->proc->p_osrel = 0;
548 	imgp->proc->p_fctl0 = 0;
549 	imgp->proc->p_elf_brandinfo = NULL;
550 
551 	/*
552 	 * Implement image setuid/setgid.
553 	 *
554 	 * Determine new credentials before attempting image activators
555 	 * so that it can be used by process_exec handlers to determine
556 	 * credential/setid changes.
557 	 *
558 	 * Don't honor setuid/setgid if the filesystem prohibits it or if
559 	 * the process is being traced.
560 	 *
561 	 * We disable setuid/setgid/etc in capability mode on the basis
562 	 * that most setugid applications are not written with that
563 	 * environment in mind, and will therefore almost certainly operate
564 	 * incorrectly. In principle there's no reason that setugid
565 	 * applications might not be useful in capability mode, so we may want
566 	 * to reconsider this conservative design choice in the future.
567 	 *
568 	 * XXXMAC: For the time being, use NOSUID to also prohibit
569 	 * transitions on the file system.
570 	 */
571 	credential_changing = false;
572 	credential_changing |= (attr.va_mode & S_ISUID) &&
573 	    oldcred->cr_uid != attr.va_uid;
574 	credential_changing |= (attr.va_mode & S_ISGID) &&
575 	    oldcred->cr_gid != attr.va_gid;
576 #ifdef MAC
577 	will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp,
578 	    interpvplabel, imgp) != 0;
579 	credential_changing |= will_transition;
580 #endif
581 
582 	/* Don't inherit PROC_PDEATHSIG_CTL value if setuid/setgid. */
583 	if (credential_changing)
584 		imgp->proc->p_pdeathsig = 0;
585 
586 	if (credential_changing &&
587 #ifdef CAPABILITY_MODE
588 	    ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) &&
589 #endif
590 	    (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 &&
591 	    (p->p_flag & P_TRACED) == 0) {
592 		imgp->credential_setid = true;
593 		VOP_UNLOCK(imgp->vp);
594 		imgp->newcred = crdup(oldcred);
595 		if (attr.va_mode & S_ISUID) {
596 			euip = uifind(attr.va_uid);
597 			change_euid(imgp->newcred, euip);
598 		}
599 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
600 		if (attr.va_mode & S_ISGID)
601 			change_egid(imgp->newcred, attr.va_gid);
602 		/*
603 		 * Implement correct POSIX saved-id behavior.
604 		 *
605 		 * XXXMAC: Note that the current logic will save the
606 		 * uid and gid if a MAC domain transition occurs, even
607 		 * though maybe it shouldn't.
608 		 */
609 		change_svuid(imgp->newcred, imgp->newcred->cr_uid);
610 		change_svgid(imgp->newcred, imgp->newcred->cr_gid);
611 	} else {
612 		/*
613 		 * Implement correct POSIX saved-id behavior.
614 		 *
615 		 * XXX: It's not clear that the existing behavior is
616 		 * POSIX-compliant.  A number of sources indicate that the
617 		 * saved uid/gid should only be updated if the new ruid is
618 		 * not equal to the old ruid, or the new euid is not equal
619 		 * to the old euid and the new euid is not equal to the old
620 		 * ruid.  The FreeBSD code always updates the saved uid/gid.
621 		 * Also, this code uses the new (replaced) euid and egid as
622 		 * the source, which may or may not be the right ones to use.
623 		 */
624 		if (oldcred->cr_svuid != oldcred->cr_uid ||
625 		    oldcred->cr_svgid != oldcred->cr_gid) {
626 			VOP_UNLOCK(imgp->vp);
627 			imgp->newcred = crdup(oldcred);
628 			vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
629 			change_svuid(imgp->newcred, imgp->newcred->cr_uid);
630 			change_svgid(imgp->newcred, imgp->newcred->cr_gid);
631 		}
632 	}
633 	/* The new credentials are installed into the process later. */
634 
635 	/*
636 	 *	If the current process has a special image activator it
637 	 *	wants to try first, call it.   For example, emulating shell
638 	 *	scripts differently.
639 	 */
640 	error = -1;
641 	if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL)
642 		error = img_first(imgp);
643 
644 	/*
645 	 *	Loop through the list of image activators, calling each one.
646 	 *	An activator returns -1 if there is no match, 0 on success,
647 	 *	and an error otherwise.
648 	 */
649 	for (i = 0; error == -1 && execsw[i]; ++i) {
650 		if (execsw[i]->ex_imgact == NULL ||
651 		    execsw[i]->ex_imgact == img_first) {
652 			continue;
653 		}
654 		error = (*execsw[i]->ex_imgact)(imgp);
655 	}
656 
657 	if (error) {
658 		if (error == -1)
659 			error = ENOEXEC;
660 		goto exec_fail_dealloc;
661 	}
662 
663 	/*
664 	 * Special interpreter operation, cleanup and loop up to try to
665 	 * activate the interpreter.
666 	 */
667 	if (imgp->interpreted) {
668 		exec_unmap_first_page(imgp);
669 		/*
670 		 * The text reference needs to be removed for scripts.
671 		 * There is a short period before we determine that
672 		 * something is a script where text reference is active.
673 		 * The vnode lock is held over this entire period
674 		 * so nothing should illegitimately be blocked.
675 		 */
676 		MPASS(imgp->textset);
677 		VOP_UNSET_TEXT_CHECKED(newtextvp);
678 		imgp->textset = false;
679 		/* free name buffer and old vnode */
680 #ifdef MAC
681 		mac_execve_interpreter_enter(newtextvp, &interpvplabel);
682 #endif
683 		if (imgp->opened) {
684 			VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td);
685 			imgp->opened = false;
686 		}
687 		vput(newtextvp);
688 		imgp->vp = newtextvp = NULL;
689 		if (args->fname != NULL) {
690 			if (newtextdvp != NULL) {
691 				vrele(newtextdvp);
692 				newtextdvp = NULL;
693 			}
694 			NDFREE_PNBUF(&nd);
695 			free(newbinname, M_PARGS);
696 			newbinname = NULL;
697 		}
698 		vm_object_deallocate(imgp->object);
699 		imgp->object = NULL;
700 		execve_nosetid(imgp);
701 		imgp->execpath = NULL;
702 		free(imgp->freepath, M_TEMP);
703 		imgp->freepath = NULL;
704 		/* set new name to that of the interpreter */
705 		args->fname = imgp->interpreter_name;
706 		goto interpret;
707 	}
708 
709 	/*
710 	 * NB: We unlock the vnode here because it is believed that none
711 	 * of the sv_copyout_strings/sv_fixup operations require the vnode.
712 	 */
713 	VOP_UNLOCK(imgp->vp);
714 
715 	if (disallow_high_osrel &&
716 	    P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) {
717 		error = ENOEXEC;
718 		uprintf("Osrel %d for image %s too high\n", p->p_osrel,
719 		    imgp->execpath != NULL ? imgp->execpath : "<unresolved>");
720 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
721 		goto exec_fail_dealloc;
722 	}
723 
724 	/*
725 	 * Copy out strings (args and env) and initialize stack base.
726 	 */
727 	error = (*p->p_sysent->sv_copyout_strings)(imgp, &stack_base);
728 	if (error != 0) {
729 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
730 		goto exec_fail_dealloc;
731 	}
732 
733 	/*
734 	 * Stack setup.
735 	 */
736 	error = (*p->p_sysent->sv_fixup)(&stack_base, imgp);
737 	if (error != 0) {
738 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
739 		goto exec_fail_dealloc;
740 	}
741 
742 	/*
743 	 * For security and other reasons, the file descriptor table cannot be
744 	 * shared after an exec.
745 	 */
746 	fdunshare(td);
747 	pdunshare(td);
748 	/* close files on exec */
749 	fdcloseexec(td);
750 
751 	/*
752 	 * Malloc things before we need locks.
753 	 */
754 	i = exec_args_get_begin_envv(imgp->args) - imgp->args->begin_argv;
755 	/* Cache arguments if they fit inside our allowance */
756 	if (ps_arg_cache_limit >= i + sizeof(struct pargs)) {
757 		newargs = pargs_alloc(i);
758 		bcopy(imgp->args->begin_argv, newargs->ar_args, i);
759 	}
760 
761 	/*
762 	 * For security and other reasons, signal handlers cannot
763 	 * be shared after an exec. The new process gets a copy of the old
764 	 * handlers. In execsigs(), the new process will have its signals
765 	 * reset.
766 	 */
767 	if (sigacts_shared(p->p_sigacts)) {
768 		oldsigacts = p->p_sigacts;
769 		newsigacts = sigacts_alloc();
770 		sigacts_copy(newsigacts, oldsigacts);
771 	}
772 
773 	vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
774 
775 	PROC_LOCK(p);
776 	if (oldsigacts)
777 		p->p_sigacts = newsigacts;
778 	/* Stop profiling */
779 	stopprofclock(p);
780 
781 	/* reset caught signals */
782 	execsigs(p);
783 
784 	/* name this process - nameiexec(p, ndp) */
785 	bzero(p->p_comm, sizeof(p->p_comm));
786 	if (args->fname)
787 		bcopy(nd.ni_cnd.cn_nameptr, p->p_comm,
788 		    min(nd.ni_cnd.cn_namelen, MAXCOMLEN));
789 	else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0)
790 		bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title));
791 	bcopy(p->p_comm, td->td_name, sizeof(td->td_name));
792 #ifdef KTR
793 	sched_clear_tdname(td);
794 #endif
795 
796 	/*
797 	 * mark as execed, wakeup the process that vforked (if any) and tell
798 	 * it that it now has its own resources back
799 	 */
800 	p->p_flag |= P_EXEC;
801 	if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0)
802 		p->p_flag2 &= ~P2_NOTRACE;
803 	if ((p->p_flag2 & P2_STKGAP_DISABLE_EXEC) == 0)
804 		p->p_flag2 &= ~P2_STKGAP_DISABLE;
805 	if (p->p_flag & P_PPWAIT) {
806 		p->p_flag &= ~(P_PPWAIT | P_PPTRACE);
807 		cv_broadcast(&p->p_pwait);
808 		/* STOPs are no longer ignored, arrange for AST */
809 		signotify(td);
810 	}
811 
812 	if ((imgp->sysent->sv_setid_allowed != NULL &&
813 	    !(*imgp->sysent->sv_setid_allowed)(td, imgp)) ||
814 	    (p->p_flag2 & P2_NO_NEW_PRIVS) != 0)
815 		execve_nosetid(imgp);
816 
817 	/*
818 	 * Implement image setuid/setgid installation.
819 	 */
820 	if (imgp->credential_setid) {
821 		/*
822 		 * Turn off syscall tracing for set-id programs, except for
823 		 * root.  Record any set-id flags first to make sure that
824 		 * we do not regain any tracing during a possible block.
825 		 */
826 		setsugid(p);
827 #ifdef KTRACE
828 		kiop = ktrprocexec(p);
829 #endif
830 		/*
831 		 * Close any file descriptors 0..2 that reference procfs,
832 		 * then make sure file descriptors 0..2 are in use.
833 		 *
834 		 * Both fdsetugidsafety() and fdcheckstd() may call functions
835 		 * taking sleepable locks, so temporarily drop our locks.
836 		 */
837 		PROC_UNLOCK(p);
838 		VOP_UNLOCK(imgp->vp);
839 		fdsetugidsafety(td);
840 		error = fdcheckstd(td);
841 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
842 		if (error != 0)
843 			goto exec_fail_dealloc;
844 		PROC_LOCK(p);
845 #ifdef MAC
846 		if (will_transition) {
847 			mac_vnode_execve_transition(oldcred, imgp->newcred,
848 			    imgp->vp, interpvplabel, imgp);
849 		}
850 #endif
851 	} else {
852 		if (oldcred->cr_uid == oldcred->cr_ruid &&
853 		    oldcred->cr_gid == oldcred->cr_rgid)
854 			p->p_flag &= ~P_SUGID;
855 	}
856 	/*
857 	 * Set the new credentials.
858 	 */
859 	if (imgp->newcred != NULL) {
860 		proc_set_cred(p, imgp->newcred);
861 		crfree(oldcred);
862 		oldcred = NULL;
863 	}
864 
865 	/*
866 	 * Store the vp for use in kern.proc.pathname.  This vnode was
867 	 * referenced by namei() or by fexecve variant of fname handling.
868 	 */
869 	oldtextvp = p->p_textvp;
870 	p->p_textvp = newtextvp;
871 	oldtextdvp = p->p_textdvp;
872 	p->p_textdvp = newtextdvp;
873 	newtextdvp = NULL;
874 	oldbinname = p->p_binname;
875 	p->p_binname = newbinname;
876 	newbinname = NULL;
877 
878 #ifdef KDTRACE_HOOKS
879 	/*
880 	 * Tell the DTrace fasttrap provider about the exec if it
881 	 * has declared an interest.
882 	 */
883 	if (dtrace_fasttrap_exec)
884 		dtrace_fasttrap_exec(p);
885 #endif
886 
887 	/*
888 	 * Notify others that we exec'd, and clear the P_INEXEC flag
889 	 * as we're now a bona fide freshly-execed process.
890 	 */
891 	KNOTE_LOCKED(p->p_klist, NOTE_EXEC);
892 	p->p_flag &= ~P_INEXEC;
893 
894 	/* clear "fork but no exec" flag, as we _are_ execing */
895 	p->p_acflag &= ~AFORK;
896 
897 	/*
898 	 * Free any previous argument cache and replace it with
899 	 * the new argument cache, if any.
900 	 */
901 	oldargs = p->p_args;
902 	p->p_args = newargs;
903 	newargs = NULL;
904 
905 	PROC_UNLOCK(p);
906 
907 #ifdef	HWPMC_HOOKS
908 	/*
909 	 * Check if system-wide sampling is in effect or if the
910 	 * current process is using PMCs.  If so, do exec() time
911 	 * processing.  This processing needs to happen AFTER the
912 	 * P_INEXEC flag is cleared.
913 	 */
914 	if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) {
915 		VOP_UNLOCK(imgp->vp);
916 		pe.pm_credentialschanged = credential_changing;
917 		pe.pm_entryaddr = imgp->entry_addr;
918 
919 		PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe);
920 		vn_lock(imgp->vp, LK_SHARED | LK_RETRY);
921 	}
922 #endif
923 
924 	/* Set values passed into the program in registers. */
925 	(*p->p_sysent->sv_setregs)(td, imgp, stack_base);
926 
927 	VOP_MMAPPED(imgp->vp);
928 
929 	SDT_PROBE1(proc, , , exec__success, args->fname);
930 
931 exec_fail_dealloc:
932 	if (error != 0) {
933 		p->p_osrel = orig_osrel;
934 		p->p_fctl0 = orig_fctl0;
935 		p->p_elf_brandinfo = orig_brandinfo;
936 	}
937 
938 	if (imgp->firstpage != NULL)
939 		exec_unmap_first_page(imgp);
940 
941 	if (imgp->vp != NULL) {
942 		if (imgp->opened)
943 			VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td);
944 		if (imgp->textset)
945 			VOP_UNSET_TEXT_CHECKED(imgp->vp);
946 		if (error != 0)
947 			vput(imgp->vp);
948 		else
949 			VOP_UNLOCK(imgp->vp);
950 		if (args->fname != NULL)
951 			NDFREE_PNBUF(&nd);
952 		if (newtextdvp != NULL)
953 			vrele(newtextdvp);
954 		free(newbinname, M_PARGS);
955 	}
956 
957 	if (imgp->object != NULL)
958 		vm_object_deallocate(imgp->object);
959 
960 	free(imgp->freepath, M_TEMP);
961 
962 	if (error == 0) {
963 		if (p->p_ptevents & PTRACE_EXEC) {
964 			PROC_LOCK(p);
965 			if (p->p_ptevents & PTRACE_EXEC)
966 				td->td_dbgflags |= TDB_EXEC;
967 			PROC_UNLOCK(p);
968 		}
969 	} else {
970 exec_fail:
971 		/* we're done here, clear P_INEXEC */
972 		PROC_LOCK(p);
973 		p->p_flag &= ~P_INEXEC;
974 		PROC_UNLOCK(p);
975 
976 		SDT_PROBE1(proc, , , exec__failure, error);
977 	}
978 
979 	if (imgp->newcred != NULL && oldcred != NULL)
980 		crfree(imgp->newcred);
981 
982 #ifdef MAC
983 	mac_execve_exit(imgp);
984 	mac_execve_interpreter_exit(interpvplabel);
985 #endif
986 	exec_free_args(args);
987 
988 	/*
989 	 * Handle deferred decrement of ref counts.
990 	 */
991 	if (oldtextvp != NULL)
992 		vrele(oldtextvp);
993 	if (oldtextdvp != NULL)
994 		vrele(oldtextdvp);
995 	free(oldbinname, M_PARGS);
996 #ifdef KTRACE
997 	ktr_io_params_free(kiop);
998 #endif
999 	pargs_drop(oldargs);
1000 	pargs_drop(newargs);
1001 	if (oldsigacts != NULL)
1002 		sigacts_free(oldsigacts);
1003 	if (euip != NULL)
1004 		uifree(euip);
1005 
1006 	if (error && imgp->vmspace_destroyed) {
1007 		/* sorry, no more process anymore. exit gracefully */
1008 		exec_cleanup(td, oldvmspace);
1009 		exit1(td, 0, SIGABRT);
1010 		/* NOT REACHED */
1011 	}
1012 
1013 #ifdef KTRACE
1014 	if (error == 0)
1015 		ktrprocctor(p);
1016 #endif
1017 
1018 	/*
1019 	 * We don't want cpu_set_syscall_retval() to overwrite any of
1020 	 * the register values put in place by exec_setregs().
1021 	 * Implementations of cpu_set_syscall_retval() will leave
1022 	 * registers unmodified when returning EJUSTRETURN.
1023 	 */
1024 	return (error == 0 ? EJUSTRETURN : error);
1025 }
1026 
1027 void
1028 exec_cleanup(struct thread *td, struct vmspace *oldvmspace)
1029 {
1030 	if ((td->td_pflags & TDP_EXECVMSPC) != 0) {
1031 		KASSERT(td->td_proc->p_vmspace != oldvmspace,
1032 		    ("oldvmspace still used"));
1033 		vmspace_free(oldvmspace);
1034 		td->td_pflags &= ~TDP_EXECVMSPC;
1035 	}
1036 }
1037 
1038 int
1039 exec_map_first_page(struct image_params *imgp)
1040 {
1041 	vm_object_t object;
1042 	vm_page_t m;
1043 	int error;
1044 
1045 	if (imgp->firstpage != NULL)
1046 		exec_unmap_first_page(imgp);
1047 
1048 	object = imgp->vp->v_object;
1049 	if (object == NULL)
1050 		return (EACCES);
1051 #if VM_NRESERVLEVEL > 0
1052 	if ((object->flags & OBJ_COLORED) == 0) {
1053 		VM_OBJECT_WLOCK(object);
1054 		vm_object_color(object, 0);
1055 		VM_OBJECT_WUNLOCK(object);
1056 	}
1057 #endif
1058 	error = vm_page_grab_valid_unlocked(&m, object, 0,
1059 	    VM_ALLOC_COUNT(VM_INITIAL_PAGEIN) |
1060 	    VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED);
1061 
1062 	if (error != VM_PAGER_OK)
1063 		return (EIO);
1064 	imgp->firstpage = sf_buf_alloc(m, 0);
1065 	imgp->image_header = (char *)sf_buf_kva(imgp->firstpage);
1066 
1067 	return (0);
1068 }
1069 
1070 void
1071 exec_unmap_first_page(struct image_params *imgp)
1072 {
1073 	vm_page_t m;
1074 
1075 	if (imgp->firstpage != NULL) {
1076 		m = sf_buf_page(imgp->firstpage);
1077 		sf_buf_free(imgp->firstpage);
1078 		imgp->firstpage = NULL;
1079 		vm_page_unwire(m, PQ_ACTIVE);
1080 	}
1081 }
1082 
1083 void
1084 exec_onexec_old(struct thread *td)
1085 {
1086 	sigfastblock_clear(td);
1087 	umtx_exec(td->td_proc);
1088 }
1089 
1090 /*
1091  * This is an optimization which removes the unmanaged shared page
1092  * mapping. In combination with pmap_remove_pages(), which cleans all
1093  * managed mappings in the process' vmspace pmap, no work will be left
1094  * for pmap_remove(min, max).
1095  */
1096 void
1097 exec_free_abi_mappings(struct proc *p)
1098 {
1099 	struct vmspace *vmspace;
1100 
1101 	vmspace = p->p_vmspace;
1102 	if (refcount_load(&vmspace->vm_refcnt) != 1)
1103 		return;
1104 
1105 	if (!PROC_HAS_SHP(p))
1106 		return;
1107 
1108 	pmap_remove(vmspace_pmap(vmspace), vmspace->vm_shp_base,
1109 	    vmspace->vm_shp_base + p->p_sysent->sv_shared_page_len);
1110 }
1111 
1112 /*
1113  * Run down the current address space and install a new one.
1114  */
1115 int
1116 exec_new_vmspace(struct image_params *imgp, struct sysentvec *sv)
1117 {
1118 	int error;
1119 	struct proc *p = imgp->proc;
1120 	struct vmspace *vmspace = p->p_vmspace;
1121 	struct thread *td = curthread;
1122 	vm_offset_t sv_minuser;
1123 	vm_map_t map;
1124 
1125 	imgp->vmspace_destroyed = true;
1126 	imgp->sysent = sv;
1127 
1128 	if (p->p_sysent->sv_onexec_old != NULL)
1129 		p->p_sysent->sv_onexec_old(td);
1130 	itimers_exec(p);
1131 
1132 	EVENTHANDLER_DIRECT_INVOKE(process_exec, p, imgp);
1133 
1134 	/*
1135 	 * Blow away entire process VM, if address space not shared,
1136 	 * otherwise, create a new VM space so that other threads are
1137 	 * not disrupted
1138 	 */
1139 	map = &vmspace->vm_map;
1140 	if (map_at_zero)
1141 		sv_minuser = sv->sv_minuser;
1142 	else
1143 		sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE);
1144 	if (refcount_load(&vmspace->vm_refcnt) == 1 &&
1145 	    vm_map_min(map) == sv_minuser &&
1146 	    vm_map_max(map) == sv->sv_maxuser &&
1147 	    cpu_exec_vmspace_reuse(p, map)) {
1148 		exec_free_abi_mappings(p);
1149 		shmexit(vmspace);
1150 		pmap_remove_pages(vmspace_pmap(vmspace));
1151 		vm_map_remove(map, vm_map_min(map), vm_map_max(map));
1152 		/*
1153 		 * An exec terminates mlockall(MCL_FUTURE).
1154 		 * ASLR and W^X states must be re-evaluated.
1155 		 */
1156 		vm_map_lock(map);
1157 		vm_map_modflags(map, 0, MAP_WIREFUTURE | MAP_ASLR |
1158 		    MAP_ASLR_IGNSTART | MAP_ASLR_STACK | MAP_WXORX);
1159 		vm_map_unlock(map);
1160 	} else {
1161 		error = vmspace_exec(p, sv_minuser, sv->sv_maxuser);
1162 		if (error)
1163 			return (error);
1164 		vmspace = p->p_vmspace;
1165 		map = &vmspace->vm_map;
1166 	}
1167 	map->flags |= imgp->map_flags;
1168 
1169 	return (sv->sv_onexec != NULL ? sv->sv_onexec(p, imgp) : 0);
1170 }
1171 
1172 /*
1173  * Compute the stack size limit and map the main process stack.
1174  * Map the shared page.
1175  */
1176 int
1177 exec_map_stack(struct image_params *imgp)
1178 {
1179 	struct rlimit rlim_stack;
1180 	struct sysentvec *sv;
1181 	struct proc *p;
1182 	vm_map_t map;
1183 	struct vmspace *vmspace;
1184 	vm_offset_t stack_addr, stack_top;
1185 	vm_offset_t sharedpage_addr;
1186 	u_long ssiz;
1187 	int error, find_space, stack_off;
1188 	vm_prot_t stack_prot;
1189 	vm_object_t obj;
1190 
1191 	p = imgp->proc;
1192 	sv = p->p_sysent;
1193 
1194 	if (imgp->stack_sz != 0) {
1195 		ssiz = trunc_page(imgp->stack_sz);
1196 		PROC_LOCK(p);
1197 		lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack);
1198 		PROC_UNLOCK(p);
1199 		if (ssiz > rlim_stack.rlim_max)
1200 			ssiz = rlim_stack.rlim_max;
1201 		if (ssiz > rlim_stack.rlim_cur) {
1202 			rlim_stack.rlim_cur = ssiz;
1203 			kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack);
1204 		}
1205 	} else if (sv->sv_maxssiz != NULL) {
1206 		ssiz = *sv->sv_maxssiz;
1207 	} else {
1208 		ssiz = maxssiz;
1209 	}
1210 
1211 	vmspace = p->p_vmspace;
1212 	map = &vmspace->vm_map;
1213 
1214 	stack_prot = sv->sv_shared_page_obj != NULL && imgp->stack_prot != 0 ?
1215 	    imgp->stack_prot : sv->sv_stackprot;
1216 	if ((map->flags & MAP_ASLR_STACK) != 0) {
1217 		stack_addr = round_page((vm_offset_t)p->p_vmspace->vm_daddr +
1218 		    lim_max(curthread, RLIMIT_DATA));
1219 		find_space = VMFS_ANY_SPACE;
1220 	} else {
1221 		stack_addr = sv->sv_usrstack - ssiz;
1222 		find_space = VMFS_NO_SPACE;
1223 	}
1224 	error = vm_map_find(map, NULL, 0, &stack_addr, (vm_size_t)ssiz,
1225 	    sv->sv_usrstack, find_space, stack_prot, VM_PROT_ALL,
1226 	    MAP_STACK_GROWS_DOWN);
1227 	if (error != KERN_SUCCESS) {
1228 		uprintf("exec_new_vmspace: mapping stack size %#jx prot %#x "
1229 		    "failed, mach error %d errno %d\n", (uintmax_t)ssiz,
1230 		    stack_prot, error, vm_mmap_to_errno(error));
1231 		return (vm_mmap_to_errno(error));
1232 	}
1233 
1234 	stack_top = stack_addr + ssiz;
1235 	if ((map->flags & MAP_ASLR_STACK) != 0) {
1236 		/* Randomize within the first page of the stack. */
1237 		arc4rand(&stack_off, sizeof(stack_off), 0);
1238 		stack_top -= rounddown2(stack_off & PAGE_MASK, sizeof(void *));
1239 	}
1240 
1241 	/* Map a shared page */
1242 	obj = sv->sv_shared_page_obj;
1243 	if (obj == NULL) {
1244 		sharedpage_addr = 0;
1245 		goto out;
1246 	}
1247 
1248 	/*
1249 	 * If randomization is disabled then the shared page will
1250 	 * be mapped at address specified in sysentvec.
1251 	 * Otherwise any address above .data section can be selected.
1252 	 * Same logic is used for stack address randomization.
1253 	 * If the address randomization is applied map a guard page
1254 	 * at the top of UVA.
1255 	 */
1256 	vm_object_reference(obj);
1257 	if ((imgp->imgp_flags & IMGP_ASLR_SHARED_PAGE) != 0) {
1258 		sharedpage_addr = round_page((vm_offset_t)p->p_vmspace->vm_daddr +
1259 		    lim_max(curthread, RLIMIT_DATA));
1260 
1261 		error = vm_map_fixed(map, NULL, 0,
1262 		    sv->sv_maxuser - PAGE_SIZE, PAGE_SIZE,
1263 		    VM_PROT_NONE, VM_PROT_NONE, MAP_CREATE_GUARD);
1264 		if (error != KERN_SUCCESS) {
1265 			/*
1266 			 * This is not fatal, so let's just print a warning
1267 			 * and continue.
1268 			 */
1269 			uprintf("%s: Mapping guard page at the top of UVA failed"
1270 			    " mach error %d errno %d",
1271 			    __func__, error, vm_mmap_to_errno(error));
1272 		}
1273 
1274 		error = vm_map_find(map, obj, 0,
1275 		    &sharedpage_addr, sv->sv_shared_page_len,
1276 		    sv->sv_maxuser, VMFS_ANY_SPACE,
1277 		    VM_PROT_READ | VM_PROT_EXECUTE,
1278 		    VM_PROT_READ | VM_PROT_EXECUTE,
1279 		    MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1280 	} else {
1281 		sharedpage_addr = sv->sv_shared_page_base;
1282 		vm_map_fixed(map, obj, 0,
1283 		    sharedpage_addr, sv->sv_shared_page_len,
1284 		    VM_PROT_READ | VM_PROT_EXECUTE,
1285 		    VM_PROT_READ | VM_PROT_EXECUTE,
1286 		    MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE);
1287 	}
1288 	if (error != KERN_SUCCESS) {
1289 		uprintf("%s: mapping shared page at addr: %p"
1290 		    "failed, mach error %d errno %d\n", __func__,
1291 		    (void *)sharedpage_addr, error, vm_mmap_to_errno(error));
1292 		vm_object_deallocate(obj);
1293 		return (vm_mmap_to_errno(error));
1294 	}
1295 out:
1296 	/*
1297 	 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they
1298 	 * are still used to enforce the stack rlimit on the process stack.
1299 	 */
1300 	vmspace->vm_maxsaddr = (char *)stack_addr;
1301 	vmspace->vm_stacktop = stack_top;
1302 	vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT;
1303 	vmspace->vm_shp_base = sharedpage_addr;
1304 
1305 	return (0);
1306 }
1307 
1308 /*
1309  * Copy out argument and environment strings from the old process address
1310  * space into the temporary string buffer.
1311  */
1312 int
1313 exec_copyin_args(struct image_args *args, const char *fname,
1314     enum uio_seg segflg, char **argv, char **envv)
1315 {
1316 	u_long arg, env;
1317 	int error;
1318 
1319 	bzero(args, sizeof(*args));
1320 	if (argv == NULL)
1321 		return (EFAULT);
1322 
1323 	/*
1324 	 * Allocate demand-paged memory for the file name, argument, and
1325 	 * environment strings.
1326 	 */
1327 	error = exec_alloc_args(args);
1328 	if (error != 0)
1329 		return (error);
1330 
1331 	/*
1332 	 * Copy the file name.
1333 	 */
1334 	error = exec_args_add_fname(args, fname, segflg);
1335 	if (error != 0)
1336 		goto err_exit;
1337 
1338 	/*
1339 	 * extract arguments first
1340 	 */
1341 	for (;;) {
1342 		error = fueword(argv++, &arg);
1343 		if (error == -1) {
1344 			error = EFAULT;
1345 			goto err_exit;
1346 		}
1347 		if (arg == 0)
1348 			break;
1349 		error = exec_args_add_arg(args, (char *)(uintptr_t)arg,
1350 		    UIO_USERSPACE);
1351 		if (error != 0)
1352 			goto err_exit;
1353 	}
1354 
1355 	/*
1356 	 * extract environment strings
1357 	 */
1358 	if (envv) {
1359 		for (;;) {
1360 			error = fueword(envv++, &env);
1361 			if (error == -1) {
1362 				error = EFAULT;
1363 				goto err_exit;
1364 			}
1365 			if (env == 0)
1366 				break;
1367 			error = exec_args_add_env(args,
1368 			    (char *)(uintptr_t)env, UIO_USERSPACE);
1369 			if (error != 0)
1370 				goto err_exit;
1371 		}
1372 	}
1373 
1374 	return (0);
1375 
1376 err_exit:
1377 	exec_free_args(args);
1378 	return (error);
1379 }
1380 
1381 struct exec_args_kva {
1382 	vm_offset_t addr;
1383 	u_int gen;
1384 	SLIST_ENTRY(exec_args_kva) next;
1385 };
1386 
1387 DPCPU_DEFINE_STATIC(struct exec_args_kva *, exec_args_kva);
1388 
1389 static SLIST_HEAD(, exec_args_kva) exec_args_kva_freelist;
1390 static struct mtx exec_args_kva_mtx;
1391 static u_int exec_args_gen;
1392 
1393 static void
1394 exec_prealloc_args_kva(void *arg __unused)
1395 {
1396 	struct exec_args_kva *argkva;
1397 	u_int i;
1398 
1399 	SLIST_INIT(&exec_args_kva_freelist);
1400 	mtx_init(&exec_args_kva_mtx, "exec args kva", NULL, MTX_DEF);
1401 	for (i = 0; i < exec_map_entries; i++) {
1402 		argkva = malloc(sizeof(*argkva), M_PARGS, M_WAITOK);
1403 		argkva->addr = kmap_alloc_wait(exec_map, exec_map_entry_size);
1404 		argkva->gen = exec_args_gen;
1405 		SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1406 	}
1407 }
1408 SYSINIT(exec_args_kva, SI_SUB_EXEC, SI_ORDER_ANY, exec_prealloc_args_kva, NULL);
1409 
1410 static vm_offset_t
1411 exec_alloc_args_kva(void **cookie)
1412 {
1413 	struct exec_args_kva *argkva;
1414 
1415 	argkva = (void *)atomic_readandclear_ptr(
1416 	    (uintptr_t *)DPCPU_PTR(exec_args_kva));
1417 	if (argkva == NULL) {
1418 		mtx_lock(&exec_args_kva_mtx);
1419 		while ((argkva = SLIST_FIRST(&exec_args_kva_freelist)) == NULL)
1420 			(void)mtx_sleep(&exec_args_kva_freelist,
1421 			    &exec_args_kva_mtx, 0, "execkva", 0);
1422 		SLIST_REMOVE_HEAD(&exec_args_kva_freelist, next);
1423 		mtx_unlock(&exec_args_kva_mtx);
1424 	}
1425 	kasan_mark((void *)argkva->addr, exec_map_entry_size,
1426 	    exec_map_entry_size, 0);
1427 	*(struct exec_args_kva **)cookie = argkva;
1428 	return (argkva->addr);
1429 }
1430 
1431 static void
1432 exec_release_args_kva(struct exec_args_kva *argkva, u_int gen)
1433 {
1434 	vm_offset_t base;
1435 
1436 	base = argkva->addr;
1437 	kasan_mark((void *)argkva->addr, 0, exec_map_entry_size,
1438 	    KASAN_EXEC_ARGS_FREED);
1439 	if (argkva->gen != gen) {
1440 		(void)vm_map_madvise(exec_map, base, base + exec_map_entry_size,
1441 		    MADV_FREE);
1442 		argkva->gen = gen;
1443 	}
1444 	if (!atomic_cmpset_ptr((uintptr_t *)DPCPU_PTR(exec_args_kva),
1445 	    (uintptr_t)NULL, (uintptr_t)argkva)) {
1446 		mtx_lock(&exec_args_kva_mtx);
1447 		SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next);
1448 		wakeup_one(&exec_args_kva_freelist);
1449 		mtx_unlock(&exec_args_kva_mtx);
1450 	}
1451 }
1452 
1453 static void
1454 exec_free_args_kva(void *cookie)
1455 {
1456 
1457 	exec_release_args_kva(cookie, exec_args_gen);
1458 }
1459 
1460 static void
1461 exec_args_kva_lowmem(void *arg __unused)
1462 {
1463 	SLIST_HEAD(, exec_args_kva) head;
1464 	struct exec_args_kva *argkva;
1465 	u_int gen;
1466 	int i;
1467 
1468 	gen = atomic_fetchadd_int(&exec_args_gen, 1) + 1;
1469 
1470 	/*
1471 	 * Force an madvise of each KVA range. Any currently allocated ranges
1472 	 * will have MADV_FREE applied once they are freed.
1473 	 */
1474 	SLIST_INIT(&head);
1475 	mtx_lock(&exec_args_kva_mtx);
1476 	SLIST_SWAP(&head, &exec_args_kva_freelist, exec_args_kva);
1477 	mtx_unlock(&exec_args_kva_mtx);
1478 	while ((argkva = SLIST_FIRST(&head)) != NULL) {
1479 		SLIST_REMOVE_HEAD(&head, next);
1480 		exec_release_args_kva(argkva, gen);
1481 	}
1482 
1483 	CPU_FOREACH(i) {
1484 		argkva = (void *)atomic_readandclear_ptr(
1485 		    (uintptr_t *)DPCPU_ID_PTR(i, exec_args_kva));
1486 		if (argkva != NULL)
1487 			exec_release_args_kva(argkva, gen);
1488 	}
1489 }
1490 EVENTHANDLER_DEFINE(vm_lowmem, exec_args_kva_lowmem, NULL,
1491     EVENTHANDLER_PRI_ANY);
1492 
1493 /*
1494  * Allocate temporary demand-paged, zero-filled memory for the file name,
1495  * argument, and environment strings.
1496  */
1497 int
1498 exec_alloc_args(struct image_args *args)
1499 {
1500 
1501 	args->buf = (char *)exec_alloc_args_kva(&args->bufkva);
1502 	return (0);
1503 }
1504 
1505 void
1506 exec_free_args(struct image_args *args)
1507 {
1508 
1509 	if (args->buf != NULL) {
1510 		exec_free_args_kva(args->bufkva);
1511 		args->buf = NULL;
1512 	}
1513 	if (args->fname_buf != NULL) {
1514 		free(args->fname_buf, M_TEMP);
1515 		args->fname_buf = NULL;
1516 	}
1517 }
1518 
1519 /*
1520  * A set to functions to fill struct image args.
1521  *
1522  * NOTE: exec_args_add_fname() must be called (possibly with a NULL
1523  * fname) before the other functions.  All exec_args_add_arg() calls must
1524  * be made before any exec_args_add_env() calls.  exec_args_adjust_args()
1525  * may be called any time after exec_args_add_fname().
1526  *
1527  * exec_args_add_fname() - install path to be executed
1528  * exec_args_add_arg() - append an argument string
1529  * exec_args_add_env() - append an env string
1530  * exec_args_adjust_args() - adjust location of the argument list to
1531  *                           allow new arguments to be prepended
1532  */
1533 int
1534 exec_args_add_fname(struct image_args *args, const char *fname,
1535     enum uio_seg segflg)
1536 {
1537 	int error;
1538 	size_t length;
1539 
1540 	KASSERT(args->fname == NULL, ("fname already appended"));
1541 	KASSERT(args->endp == NULL, ("already appending to args"));
1542 
1543 	if (fname != NULL) {
1544 		args->fname = args->buf;
1545 		error = segflg == UIO_SYSSPACE ?
1546 		    copystr(fname, args->fname, PATH_MAX, &length) :
1547 		    copyinstr(fname, args->fname, PATH_MAX, &length);
1548 		if (error != 0)
1549 			return (error == ENAMETOOLONG ? E2BIG : error);
1550 	} else
1551 		length = 0;
1552 
1553 	/* Set up for _arg_*()/_env_*() */
1554 	args->endp = args->buf + length;
1555 	/* begin_argv must be set and kept updated */
1556 	args->begin_argv = args->endp;
1557 	KASSERT(exec_map_entry_size - length >= ARG_MAX,
1558 	    ("too little space remaining for arguments %zu < %zu",
1559 	    exec_map_entry_size - length, (size_t)ARG_MAX));
1560 	args->stringspace = ARG_MAX;
1561 
1562 	return (0);
1563 }
1564 
1565 static int
1566 exec_args_add_str(struct image_args *args, const char *str,
1567     enum uio_seg segflg, int *countp)
1568 {
1569 	int error;
1570 	size_t length;
1571 
1572 	KASSERT(args->endp != NULL, ("endp not initialized"));
1573 	KASSERT(args->begin_argv != NULL, ("begin_argp not initialized"));
1574 
1575 	error = (segflg == UIO_SYSSPACE) ?
1576 	    copystr(str, args->endp, args->stringspace, &length) :
1577 	    copyinstr(str, args->endp, args->stringspace, &length);
1578 	if (error != 0)
1579 		return (error == ENAMETOOLONG ? E2BIG : error);
1580 	args->stringspace -= length;
1581 	args->endp += length;
1582 	(*countp)++;
1583 
1584 	return (0);
1585 }
1586 
1587 int
1588 exec_args_add_arg(struct image_args *args, const char *argp,
1589     enum uio_seg segflg)
1590 {
1591 
1592 	KASSERT(args->envc == 0, ("appending args after env"));
1593 
1594 	return (exec_args_add_str(args, argp, segflg, &args->argc));
1595 }
1596 
1597 int
1598 exec_args_add_env(struct image_args *args, const char *envp,
1599     enum uio_seg segflg)
1600 {
1601 
1602 	if (args->envc == 0)
1603 		args->begin_envv = args->endp;
1604 
1605 	return (exec_args_add_str(args, envp, segflg, &args->envc));
1606 }
1607 
1608 int
1609 exec_args_adjust_args(struct image_args *args, size_t consume, ssize_t extend)
1610 {
1611 	ssize_t offset;
1612 
1613 	KASSERT(args->endp != NULL, ("endp not initialized"));
1614 	KASSERT(args->begin_argv != NULL, ("begin_argp not initialized"));
1615 
1616 	offset = extend - consume;
1617 	if (args->stringspace < offset)
1618 		return (E2BIG);
1619 	memmove(args->begin_argv + extend, args->begin_argv + consume,
1620 	    args->endp - args->begin_argv + consume);
1621 	if (args->envc > 0)
1622 		args->begin_envv += offset;
1623 	args->endp += offset;
1624 	args->stringspace -= offset;
1625 	return (0);
1626 }
1627 
1628 char *
1629 exec_args_get_begin_envv(struct image_args *args)
1630 {
1631 
1632 	KASSERT(args->endp != NULL, ("endp not initialized"));
1633 
1634 	if (args->envc > 0)
1635 		return (args->begin_envv);
1636 	return (args->endp);
1637 }
1638 
1639 /*
1640  * Copy strings out to the new process address space, constructing new arg
1641  * and env vector tables. Return a pointer to the base so that it can be used
1642  * as the initial stack pointer.
1643  */
1644 int
1645 exec_copyout_strings(struct image_params *imgp, uintptr_t *stack_base)
1646 {
1647 	int argc, envc;
1648 	char **vectp;
1649 	char *stringp;
1650 	uintptr_t destp, ustringp;
1651 	struct ps_strings *arginfo;
1652 	struct proc *p;
1653 	struct sysentvec *sysent;
1654 	size_t execpath_len;
1655 	int error, szsigcode;
1656 	char canary[sizeof(long) * 8];
1657 
1658 	p = imgp->proc;
1659 	sysent = p->p_sysent;
1660 
1661 	destp =	PROC_PS_STRINGS(p);
1662 	arginfo = imgp->ps_strings = (void *)destp;
1663 
1664 	/*
1665 	 * Install sigcode.
1666 	 */
1667 	if (sysent->sv_shared_page_base == 0 && sysent->sv_szsigcode != NULL) {
1668 		szsigcode = *(sysent->sv_szsigcode);
1669 		destp -= szsigcode;
1670 		destp = rounddown2(destp, sizeof(void *));
1671 		error = copyout(sysent->sv_sigcode, (void *)destp, szsigcode);
1672 		if (error != 0)
1673 			return (error);
1674 	}
1675 
1676 	/*
1677 	 * Copy the image path for the rtld.
1678 	 */
1679 	if (imgp->execpath != NULL && imgp->auxargs != NULL) {
1680 		execpath_len = strlen(imgp->execpath) + 1;
1681 		destp -= execpath_len;
1682 		destp = rounddown2(destp, sizeof(void *));
1683 		imgp->execpathp = (void *)destp;
1684 		error = copyout(imgp->execpath, imgp->execpathp, execpath_len);
1685 		if (error != 0)
1686 			return (error);
1687 	}
1688 
1689 	/*
1690 	 * Prepare the canary for SSP.
1691 	 */
1692 	arc4rand(canary, sizeof(canary), 0);
1693 	destp -= sizeof(canary);
1694 	imgp->canary = (void *)destp;
1695 	error = copyout(canary, imgp->canary, sizeof(canary));
1696 	if (error != 0)
1697 		return (error);
1698 	imgp->canarylen = sizeof(canary);
1699 
1700 	/*
1701 	 * Prepare the pagesizes array.
1702 	 */
1703 	imgp->pagesizeslen = sizeof(pagesizes[0]) * MAXPAGESIZES;
1704 	destp -= imgp->pagesizeslen;
1705 	destp = rounddown2(destp, sizeof(void *));
1706 	imgp->pagesizes = (void *)destp;
1707 	error = copyout(pagesizes, imgp->pagesizes, imgp->pagesizeslen);
1708 	if (error != 0)
1709 		return (error);
1710 
1711 	/*
1712 	 * Allocate room for the argument and environment strings.
1713 	 */
1714 	destp -= ARG_MAX - imgp->args->stringspace;
1715 	destp = rounddown2(destp, sizeof(void *));
1716 	ustringp = destp;
1717 
1718 	if (imgp->auxargs) {
1719 		/*
1720 		 * Allocate room on the stack for the ELF auxargs
1721 		 * array.  It has up to AT_COUNT entries.
1722 		 */
1723 		destp -= AT_COUNT * sizeof(Elf_Auxinfo);
1724 		destp = rounddown2(destp, sizeof(void *));
1725 	}
1726 
1727 	vectp = (char **)destp;
1728 
1729 	/*
1730 	 * Allocate room for the argv[] and env vectors including the
1731 	 * terminating NULL pointers.
1732 	 */
1733 	vectp -= imgp->args->argc + 1 + imgp->args->envc + 1;
1734 
1735 	/*
1736 	 * vectp also becomes our initial stack base
1737 	 */
1738 	*stack_base = (uintptr_t)vectp;
1739 
1740 	stringp = imgp->args->begin_argv;
1741 	argc = imgp->args->argc;
1742 	envc = imgp->args->envc;
1743 
1744 	/*
1745 	 * Copy out strings - arguments and environment.
1746 	 */
1747 	error = copyout(stringp, (void *)ustringp,
1748 	    ARG_MAX - imgp->args->stringspace);
1749 	if (error != 0)
1750 		return (error);
1751 
1752 	/*
1753 	 * Fill in "ps_strings" struct for ps, w, etc.
1754 	 */
1755 	imgp->argv = vectp;
1756 	if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 ||
1757 	    suword32(&arginfo->ps_nargvstr, argc) != 0)
1758 		return (EFAULT);
1759 
1760 	/*
1761 	 * Fill in argument portion of vector table.
1762 	 */
1763 	for (; argc > 0; --argc) {
1764 		if (suword(vectp++, ustringp) != 0)
1765 			return (EFAULT);
1766 		while (*stringp++ != 0)
1767 			ustringp++;
1768 		ustringp++;
1769 	}
1770 
1771 	/* a null vector table pointer separates the argp's from the envp's */
1772 	if (suword(vectp++, 0) != 0)
1773 		return (EFAULT);
1774 
1775 	imgp->envv = vectp;
1776 	if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 ||
1777 	    suword32(&arginfo->ps_nenvstr, envc) != 0)
1778 		return (EFAULT);
1779 
1780 	/*
1781 	 * Fill in environment portion of vector table.
1782 	 */
1783 	for (; envc > 0; --envc) {
1784 		if (suword(vectp++, ustringp) != 0)
1785 			return (EFAULT);
1786 		while (*stringp++ != 0)
1787 			ustringp++;
1788 		ustringp++;
1789 	}
1790 
1791 	/* end of vector table is a null pointer */
1792 	if (suword(vectp, 0) != 0)
1793 		return (EFAULT);
1794 
1795 	if (imgp->auxargs) {
1796 		vectp++;
1797 		error = imgp->sysent->sv_copyout_auxargs(imgp,
1798 		    (uintptr_t)vectp);
1799 		if (error != 0)
1800 			return (error);
1801 	}
1802 
1803 	return (0);
1804 }
1805 
1806 /*
1807  * Check permissions of file to execute.
1808  *	Called with imgp->vp locked.
1809  *	Return 0 for success or error code on failure.
1810  */
1811 int
1812 exec_check_permissions(struct image_params *imgp)
1813 {
1814 	struct vnode *vp = imgp->vp;
1815 	struct vattr *attr = imgp->attr;
1816 	struct thread *td;
1817 	int error;
1818 
1819 	td = curthread;
1820 
1821 	/* Get file attributes */
1822 	error = VOP_GETATTR(vp, attr, td->td_ucred);
1823 	if (error)
1824 		return (error);
1825 
1826 #ifdef MAC
1827 	error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp);
1828 	if (error)
1829 		return (error);
1830 #endif
1831 
1832 	/*
1833 	 * 1) Check if file execution is disabled for the filesystem that
1834 	 *    this file resides on.
1835 	 * 2) Ensure that at least one execute bit is on. Otherwise, a
1836 	 *    privileged user will always succeed, and we don't want this
1837 	 *    to happen unless the file really is executable.
1838 	 * 3) Ensure that the file is a regular file.
1839 	 */
1840 	if ((vp->v_mount->mnt_flag & MNT_NOEXEC) ||
1841 	    (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 ||
1842 	    (attr->va_type != VREG))
1843 		return (EACCES);
1844 
1845 	/*
1846 	 * Zero length files can't be exec'd
1847 	 */
1848 	if (attr->va_size == 0)
1849 		return (ENOEXEC);
1850 
1851 	/*
1852 	 *  Check for execute permission to file based on current credentials.
1853 	 */
1854 	error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td);
1855 	if (error)
1856 		return (error);
1857 
1858 	/*
1859 	 * Check number of open-for-writes on the file and deny execution
1860 	 * if there are any.
1861 	 *
1862 	 * Add a text reference now so no one can write to the
1863 	 * executable while we're activating it.
1864 	 *
1865 	 * Remember if this was set before and unset it in case this is not
1866 	 * actually an executable image.
1867 	 */
1868 	error = VOP_SET_TEXT(vp);
1869 	if (error != 0)
1870 		return (error);
1871 	imgp->textset = true;
1872 
1873 	/*
1874 	 * Call filesystem specific open routine (which does nothing in the
1875 	 * general case).
1876 	 */
1877 	error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL);
1878 	if (error == 0)
1879 		imgp->opened = true;
1880 	return (error);
1881 }
1882 
1883 /*
1884  * Exec handler registration
1885  */
1886 int
1887 exec_register(const struct execsw *execsw_arg)
1888 {
1889 	const struct execsw **es, **xs, **newexecsw;
1890 	u_int count = 2;	/* New slot and trailing NULL */
1891 
1892 	if (execsw)
1893 		for (es = execsw; *es; es++)
1894 			count++;
1895 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1896 	xs = newexecsw;
1897 	if (execsw)
1898 		for (es = execsw; *es; es++)
1899 			*xs++ = *es;
1900 	*xs++ = execsw_arg;
1901 	*xs = NULL;
1902 	if (execsw)
1903 		free(execsw, M_TEMP);
1904 	execsw = newexecsw;
1905 	return (0);
1906 }
1907 
1908 int
1909 exec_unregister(const struct execsw *execsw_arg)
1910 {
1911 	const struct execsw **es, **xs, **newexecsw;
1912 	int count = 1;
1913 
1914 	if (execsw == NULL)
1915 		panic("unregister with no handlers left?\n");
1916 
1917 	for (es = execsw; *es; es++) {
1918 		if (*es == execsw_arg)
1919 			break;
1920 	}
1921 	if (*es == NULL)
1922 		return (ENOENT);
1923 	for (es = execsw; *es; es++)
1924 		if (*es != execsw_arg)
1925 			count++;
1926 	newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK);
1927 	xs = newexecsw;
1928 	for (es = execsw; *es; es++)
1929 		if (*es != execsw_arg)
1930 			*xs++ = *es;
1931 	*xs = NULL;
1932 	if (execsw)
1933 		free(execsw, M_TEMP);
1934 	execsw = newexecsw;
1935 	return (0);
1936 }
1937 
1938 /*
1939  * Write out a core segment to the compression stream.
1940  */
1941 static int
1942 compress_chunk(struct coredump_params *cp, char *base, char *buf, size_t len)
1943 {
1944 	size_t chunk_len;
1945 	int error;
1946 
1947 	while (len > 0) {
1948 		chunk_len = MIN(len, CORE_BUF_SIZE);
1949 
1950 		/*
1951 		 * We can get EFAULT error here.
1952 		 * In that case zero out the current chunk of the segment.
1953 		 */
1954 		error = copyin(base, buf, chunk_len);
1955 		if (error != 0)
1956 			bzero(buf, chunk_len);
1957 		error = compressor_write(cp->comp, buf, chunk_len);
1958 		if (error != 0)
1959 			break;
1960 		base += chunk_len;
1961 		len -= chunk_len;
1962 	}
1963 	return (error);
1964 }
1965 
1966 int
1967 core_write(struct coredump_params *cp, const void *base, size_t len,
1968     off_t offset, enum uio_seg seg, size_t *resid)
1969 {
1970 
1971 	return (vn_rdwr_inchunks(UIO_WRITE, cp->vp, __DECONST(void *, base),
1972 	    len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED,
1973 	    cp->active_cred, cp->file_cred, resid, cp->td));
1974 }
1975 
1976 int
1977 core_output(char *base, size_t len, off_t offset, struct coredump_params *cp,
1978     void *tmpbuf)
1979 {
1980 	vm_map_t map;
1981 	struct mount *mp;
1982 	size_t resid, runlen;
1983 	int error;
1984 	bool success;
1985 
1986 	KASSERT((uintptr_t)base % PAGE_SIZE == 0,
1987 	    ("%s: user address %p is not page-aligned", __func__, base));
1988 
1989 	if (cp->comp != NULL)
1990 		return (compress_chunk(cp, base, tmpbuf, len));
1991 
1992 	map = &cp->td->td_proc->p_vmspace->vm_map;
1993 	for (; len > 0; base += runlen, offset += runlen, len -= runlen) {
1994 		/*
1995 		 * Attempt to page in all virtual pages in the range.  If a
1996 		 * virtual page is not backed by the pager, it is represented as
1997 		 * a hole in the file.  This can occur with zero-filled
1998 		 * anonymous memory or truncated files, for example.
1999 		 */
2000 		for (runlen = 0; runlen < len; runlen += PAGE_SIZE) {
2001 			if (core_dump_can_intr && curproc_sigkilled())
2002 				return (EINTR);
2003 			error = vm_fault(map, (uintptr_t)base + runlen,
2004 			    VM_PROT_READ, VM_FAULT_NOFILL, NULL);
2005 			if (runlen == 0)
2006 				success = error == KERN_SUCCESS;
2007 			else if ((error == KERN_SUCCESS) != success)
2008 				break;
2009 		}
2010 
2011 		if (success) {
2012 			error = core_write(cp, base, runlen, offset,
2013 			    UIO_USERSPACE, &resid);
2014 			if (error != 0) {
2015 				if (error != EFAULT)
2016 					break;
2017 
2018 				/*
2019 				 * EFAULT may be returned if the user mapping
2020 				 * could not be accessed, e.g., because a mapped
2021 				 * file has been truncated.  Skip the page if no
2022 				 * progress was made, to protect against a
2023 				 * hypothetical scenario where vm_fault() was
2024 				 * successful but core_write() returns EFAULT
2025 				 * anyway.
2026 				 */
2027 				runlen -= resid;
2028 				if (runlen == 0) {
2029 					success = false;
2030 					runlen = PAGE_SIZE;
2031 				}
2032 			}
2033 		}
2034 		if (!success) {
2035 			error = vn_start_write(cp->vp, &mp, V_WAIT);
2036 			if (error != 0)
2037 				break;
2038 			vn_lock(cp->vp, LK_EXCLUSIVE | LK_RETRY);
2039 			error = vn_truncate_locked(cp->vp, offset + runlen,
2040 			    false, cp->td->td_ucred);
2041 			VOP_UNLOCK(cp->vp);
2042 			vn_finished_write(mp);
2043 			if (error != 0)
2044 				break;
2045 		}
2046 	}
2047 	return (error);
2048 }
2049 
2050 /*
2051  * Drain into a core file.
2052  */
2053 int
2054 sbuf_drain_core_output(void *arg, const char *data, int len)
2055 {
2056 	struct coredump_params *cp;
2057 	struct proc *p;
2058 	int error, locked;
2059 
2060 	cp = arg;
2061 	p = cp->td->td_proc;
2062 
2063 	/*
2064 	 * Some kern_proc out routines that print to this sbuf may
2065 	 * call us with the process lock held. Draining with the
2066 	 * non-sleepable lock held is unsafe. The lock is needed for
2067 	 * those routines when dumping a live process. In our case we
2068 	 * can safely release the lock before draining and acquire
2069 	 * again after.
2070 	 */
2071 	locked = PROC_LOCKED(p);
2072 	if (locked)
2073 		PROC_UNLOCK(p);
2074 	if (cp->comp != NULL)
2075 		error = compressor_write(cp->comp, __DECONST(char *, data),
2076 		    len);
2077 	else
2078 		error = core_write(cp, __DECONST(void *, data), len, cp->offset,
2079 		    UIO_SYSSPACE, NULL);
2080 	if (locked)
2081 		PROC_LOCK(p);
2082 	if (error != 0)
2083 		return (-error);
2084 	cp->offset += len;
2085 	return (len);
2086 }
2087