xref: /freebsd/sys/vm/vm_map.c (revision 62cfcf62f627e5093fb37026a6d8c98e4d2ef04c)
1 /*-
2  * SPDX-License-Identifier: (BSD-3-Clause AND MIT-CMU)
3  *
4  * Copyright (c) 1991, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * This code is derived from software contributed to Berkeley by
8  * The Mach Operating System project at Carnegie-Mellon University.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	from: @(#)vm_map.c	8.3 (Berkeley) 1/12/94
35  *
36  *
37  * Copyright (c) 1987, 1990 Carnegie-Mellon University.
38  * All rights reserved.
39  *
40  * Authors: Avadis Tevanian, Jr., Michael Wayne Young
41  *
42  * Permission to use, copy, modify and distribute this software and
43  * its documentation is hereby granted, provided that both the copyright
44  * notice and this permission notice appear in all copies of the
45  * software, derivative works or modified versions, and any portions
46  * thereof, and that both notices appear in supporting documentation.
47  *
48  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
49  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
50  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
51  *
52  * Carnegie Mellon requests users of this software to return to
53  *
54  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
55  *  School of Computer Science
56  *  Carnegie Mellon University
57  *  Pittsburgh PA 15213-3890
58  *
59  * any improvements or extensions that they make and grant Carnegie the
60  * rights to redistribute these changes.
61  */
62 
63 /*
64  *	Virtual memory mapping module.
65  */
66 
67 #include <sys/cdefs.h>
68 __FBSDID("$FreeBSD$");
69 
70 #include <sys/param.h>
71 #include <sys/systm.h>
72 #include <sys/elf.h>
73 #include <sys/kernel.h>
74 #include <sys/ktr.h>
75 #include <sys/lock.h>
76 #include <sys/mutex.h>
77 #include <sys/proc.h>
78 #include <sys/vmmeter.h>
79 #include <sys/mman.h>
80 #include <sys/vnode.h>
81 #include <sys/racct.h>
82 #include <sys/resourcevar.h>
83 #include <sys/rwlock.h>
84 #include <sys/file.h>
85 #include <sys/sysctl.h>
86 #include <sys/sysent.h>
87 #include <sys/shm.h>
88 
89 #include <vm/vm.h>
90 #include <vm/vm_param.h>
91 #include <vm/pmap.h>
92 #include <vm/vm_map.h>
93 #include <vm/vm_page.h>
94 #include <vm/vm_pageout.h>
95 #include <vm/vm_object.h>
96 #include <vm/vm_pager.h>
97 #include <vm/vm_kern.h>
98 #include <vm/vm_extern.h>
99 #include <vm/vnode_pager.h>
100 #include <vm/swap_pager.h>
101 #include <vm/uma.h>
102 
103 /*
104  *	Virtual memory maps provide for the mapping, protection,
105  *	and sharing of virtual memory objects.  In addition,
106  *	this module provides for an efficient virtual copy of
107  *	memory from one map to another.
108  *
109  *	Synchronization is required prior to most operations.
110  *
111  *	Maps consist of an ordered doubly-linked list of simple
112  *	entries; a self-adjusting binary search tree of these
113  *	entries is used to speed up lookups.
114  *
115  *	Since portions of maps are specified by start/end addresses,
116  *	which may not align with existing map entries, all
117  *	routines merely "clip" entries to these start/end values.
118  *	[That is, an entry is split into two, bordering at a
119  *	start or end value.]  Note that these clippings may not
120  *	always be necessary (as the two resulting entries are then
121  *	not changed); however, the clipping is done for convenience.
122  *
123  *	As mentioned above, virtual copy operations are performed
124  *	by copying VM object references from one map to
125  *	another, and then marking both regions as copy-on-write.
126  */
127 
128 static struct mtx map_sleep_mtx;
129 static uma_zone_t mapentzone;
130 static uma_zone_t kmapentzone;
131 static uma_zone_t mapzone;
132 static uma_zone_t vmspace_zone;
133 static int vmspace_zinit(void *mem, int size, int flags);
134 static int vm_map_zinit(void *mem, int ize, int flags);
135 static void _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min,
136     vm_offset_t max);
137 static void vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map);
138 static void vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry);
139 static void vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry);
140 static int vm_map_growstack(vm_map_t map, vm_offset_t addr,
141     vm_map_entry_t gap_entry);
142 static void vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
143     vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags);
144 #ifdef INVARIANTS
145 static void vm_map_zdtor(void *mem, int size, void *arg);
146 static void vmspace_zdtor(void *mem, int size, void *arg);
147 #endif
148 static int vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos,
149     vm_size_t max_ssize, vm_size_t growsize, vm_prot_t prot, vm_prot_t max,
150     int cow);
151 static void vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry,
152     vm_offset_t failed_addr);
153 
154 #define	ENTRY_CHARGED(e) ((e)->cred != NULL || \
155     ((e)->object.vm_object != NULL && (e)->object.vm_object->cred != NULL && \
156      !((e)->eflags & MAP_ENTRY_NEEDS_COPY)))
157 
158 /*
159  * PROC_VMSPACE_{UN,}LOCK() can be a noop as long as vmspaces are type
160  * stable.
161  */
162 #define PROC_VMSPACE_LOCK(p) do { } while (0)
163 #define PROC_VMSPACE_UNLOCK(p) do { } while (0)
164 
165 /*
166  *	VM_MAP_RANGE_CHECK:	[ internal use only ]
167  *
168  *	Asserts that the starting and ending region
169  *	addresses fall within the valid range of the map.
170  */
171 #define	VM_MAP_RANGE_CHECK(map, start, end)		\
172 		{					\
173 		if (start < vm_map_min(map))		\
174 			start = vm_map_min(map);	\
175 		if (end > vm_map_max(map))		\
176 			end = vm_map_max(map);		\
177 		if (start > end)			\
178 			start = end;			\
179 		}
180 
181 /*
182  *	vm_map_startup:
183  *
184  *	Initialize the vm_map module.  Must be called before
185  *	any other vm_map routines.
186  *
187  *	Map and entry structures are allocated from the general
188  *	purpose memory pool with some exceptions:
189  *
190  *	- The kernel map and kmem submap are allocated statically.
191  *	- Kernel map entries are allocated out of a static pool.
192  *
193  *	These restrictions are necessary since malloc() uses the
194  *	maps and requires map entries.
195  */
196 
197 void
198 vm_map_startup(void)
199 {
200 	mtx_init(&map_sleep_mtx, "vm map sleep mutex", NULL, MTX_DEF);
201 	mapzone = uma_zcreate("MAP", sizeof(struct vm_map), NULL,
202 #ifdef INVARIANTS
203 	    vm_map_zdtor,
204 #else
205 	    NULL,
206 #endif
207 	    vm_map_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
208 	uma_prealloc(mapzone, MAX_KMAP);
209 	kmapentzone = uma_zcreate("KMAP ENTRY", sizeof(struct vm_map_entry),
210 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
211 	    UMA_ZONE_MTXCLASS | UMA_ZONE_VM);
212 	mapentzone = uma_zcreate("MAP ENTRY", sizeof(struct vm_map_entry),
213 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
214 	vmspace_zone = uma_zcreate("VMSPACE", sizeof(struct vmspace), NULL,
215 #ifdef INVARIANTS
216 	    vmspace_zdtor,
217 #else
218 	    NULL,
219 #endif
220 	    vmspace_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
221 }
222 
223 static int
224 vmspace_zinit(void *mem, int size, int flags)
225 {
226 	struct vmspace *vm;
227 
228 	vm = (struct vmspace *)mem;
229 
230 	vm->vm_map.pmap = NULL;
231 	(void)vm_map_zinit(&vm->vm_map, sizeof(vm->vm_map), flags);
232 	PMAP_LOCK_INIT(vmspace_pmap(vm));
233 	return (0);
234 }
235 
236 static int
237 vm_map_zinit(void *mem, int size, int flags)
238 {
239 	vm_map_t map;
240 
241 	map = (vm_map_t)mem;
242 	memset(map, 0, sizeof(*map));
243 	mtx_init(&map->system_mtx, "vm map (system)", NULL, MTX_DEF | MTX_DUPOK);
244 	sx_init(&map->lock, "vm map (user)");
245 	return (0);
246 }
247 
248 #ifdef INVARIANTS
249 static void
250 vmspace_zdtor(void *mem, int size, void *arg)
251 {
252 	struct vmspace *vm;
253 
254 	vm = (struct vmspace *)mem;
255 
256 	vm_map_zdtor(&vm->vm_map, sizeof(vm->vm_map), arg);
257 }
258 static void
259 vm_map_zdtor(void *mem, int size, void *arg)
260 {
261 	vm_map_t map;
262 
263 	map = (vm_map_t)mem;
264 	KASSERT(map->nentries == 0,
265 	    ("map %p nentries == %d on free.",
266 	    map, map->nentries));
267 	KASSERT(map->size == 0,
268 	    ("map %p size == %lu on free.",
269 	    map, (unsigned long)map->size));
270 }
271 #endif	/* INVARIANTS */
272 
273 /*
274  * Allocate a vmspace structure, including a vm_map and pmap,
275  * and initialize those structures.  The refcnt is set to 1.
276  *
277  * If 'pinit' is NULL then the embedded pmap is initialized via pmap_pinit().
278  */
279 struct vmspace *
280 vmspace_alloc(vm_offset_t min, vm_offset_t max, pmap_pinit_t pinit)
281 {
282 	struct vmspace *vm;
283 
284 	vm = uma_zalloc(vmspace_zone, M_WAITOK);
285 	KASSERT(vm->vm_map.pmap == NULL, ("vm_map.pmap must be NULL"));
286 	if (!pinit(vmspace_pmap(vm))) {
287 		uma_zfree(vmspace_zone, vm);
288 		return (NULL);
289 	}
290 	CTR1(KTR_VM, "vmspace_alloc: %p", vm);
291 	_vm_map_init(&vm->vm_map, vmspace_pmap(vm), min, max);
292 	vm->vm_refcnt = 1;
293 	vm->vm_shm = NULL;
294 	vm->vm_swrss = 0;
295 	vm->vm_tsize = 0;
296 	vm->vm_dsize = 0;
297 	vm->vm_ssize = 0;
298 	vm->vm_taddr = 0;
299 	vm->vm_daddr = 0;
300 	vm->vm_maxsaddr = 0;
301 	return (vm);
302 }
303 
304 #ifdef RACCT
305 static void
306 vmspace_container_reset(struct proc *p)
307 {
308 
309 	PROC_LOCK(p);
310 	racct_set(p, RACCT_DATA, 0);
311 	racct_set(p, RACCT_STACK, 0);
312 	racct_set(p, RACCT_RSS, 0);
313 	racct_set(p, RACCT_MEMLOCK, 0);
314 	racct_set(p, RACCT_VMEM, 0);
315 	PROC_UNLOCK(p);
316 }
317 #endif
318 
319 static inline void
320 vmspace_dofree(struct vmspace *vm)
321 {
322 
323 	CTR1(KTR_VM, "vmspace_free: %p", vm);
324 
325 	/*
326 	 * Make sure any SysV shm is freed, it might not have been in
327 	 * exit1().
328 	 */
329 	shmexit(vm);
330 
331 	/*
332 	 * Lock the map, to wait out all other references to it.
333 	 * Delete all of the mappings and pages they hold, then call
334 	 * the pmap module to reclaim anything left.
335 	 */
336 	(void)vm_map_remove(&vm->vm_map, vm_map_min(&vm->vm_map),
337 	    vm_map_max(&vm->vm_map));
338 
339 	pmap_release(vmspace_pmap(vm));
340 	vm->vm_map.pmap = NULL;
341 	uma_zfree(vmspace_zone, vm);
342 }
343 
344 void
345 vmspace_free(struct vmspace *vm)
346 {
347 
348 	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, NULL,
349 	    "vmspace_free() called");
350 
351 	if (vm->vm_refcnt == 0)
352 		panic("vmspace_free: attempt to free already freed vmspace");
353 
354 	if (atomic_fetchadd_int(&vm->vm_refcnt, -1) == 1)
355 		vmspace_dofree(vm);
356 }
357 
358 void
359 vmspace_exitfree(struct proc *p)
360 {
361 	struct vmspace *vm;
362 
363 	PROC_VMSPACE_LOCK(p);
364 	vm = p->p_vmspace;
365 	p->p_vmspace = NULL;
366 	PROC_VMSPACE_UNLOCK(p);
367 	KASSERT(vm == &vmspace0, ("vmspace_exitfree: wrong vmspace"));
368 	vmspace_free(vm);
369 }
370 
371 void
372 vmspace_exit(struct thread *td)
373 {
374 	int refcnt;
375 	struct vmspace *vm;
376 	struct proc *p;
377 
378 	/*
379 	 * Release user portion of address space.
380 	 * This releases references to vnodes,
381 	 * which could cause I/O if the file has been unlinked.
382 	 * Need to do this early enough that we can still sleep.
383 	 *
384 	 * The last exiting process to reach this point releases as
385 	 * much of the environment as it can. vmspace_dofree() is the
386 	 * slower fallback in case another process had a temporary
387 	 * reference to the vmspace.
388 	 */
389 
390 	p = td->td_proc;
391 	vm = p->p_vmspace;
392 	atomic_add_int(&vmspace0.vm_refcnt, 1);
393 	refcnt = vm->vm_refcnt;
394 	do {
395 		if (refcnt > 1 && p->p_vmspace != &vmspace0) {
396 			/* Switch now since other proc might free vmspace */
397 			PROC_VMSPACE_LOCK(p);
398 			p->p_vmspace = &vmspace0;
399 			PROC_VMSPACE_UNLOCK(p);
400 			pmap_activate(td);
401 		}
402 	} while (!atomic_fcmpset_int(&vm->vm_refcnt, &refcnt, refcnt - 1));
403 	if (refcnt == 1) {
404 		if (p->p_vmspace != vm) {
405 			/* vmspace not yet freed, switch back */
406 			PROC_VMSPACE_LOCK(p);
407 			p->p_vmspace = vm;
408 			PROC_VMSPACE_UNLOCK(p);
409 			pmap_activate(td);
410 		}
411 		pmap_remove_pages(vmspace_pmap(vm));
412 		/* Switch now since this proc will free vmspace */
413 		PROC_VMSPACE_LOCK(p);
414 		p->p_vmspace = &vmspace0;
415 		PROC_VMSPACE_UNLOCK(p);
416 		pmap_activate(td);
417 		vmspace_dofree(vm);
418 	}
419 #ifdef RACCT
420 	if (racct_enable)
421 		vmspace_container_reset(p);
422 #endif
423 }
424 
425 /* Acquire reference to vmspace owned by another process. */
426 
427 struct vmspace *
428 vmspace_acquire_ref(struct proc *p)
429 {
430 	struct vmspace *vm;
431 	int refcnt;
432 
433 	PROC_VMSPACE_LOCK(p);
434 	vm = p->p_vmspace;
435 	if (vm == NULL) {
436 		PROC_VMSPACE_UNLOCK(p);
437 		return (NULL);
438 	}
439 	refcnt = vm->vm_refcnt;
440 	do {
441 		if (refcnt <= 0) { 	/* Avoid 0->1 transition */
442 			PROC_VMSPACE_UNLOCK(p);
443 			return (NULL);
444 		}
445 	} while (!atomic_fcmpset_int(&vm->vm_refcnt, &refcnt, refcnt + 1));
446 	if (vm != p->p_vmspace) {
447 		PROC_VMSPACE_UNLOCK(p);
448 		vmspace_free(vm);
449 		return (NULL);
450 	}
451 	PROC_VMSPACE_UNLOCK(p);
452 	return (vm);
453 }
454 
455 /*
456  * Switch between vmspaces in an AIO kernel process.
457  *
458  * The new vmspace is either the vmspace of a user process obtained
459  * from an active AIO request or the initial vmspace of the AIO kernel
460  * process (when it is idling).  Because user processes will block to
461  * drain any active AIO requests before proceeding in exit() or
462  * execve(), the reference count for vmspaces from AIO requests can
463  * never be 0.  Similarly, AIO kernel processes hold an extra
464  * reference on their initial vmspace for the life of the process.  As
465  * a result, the 'newvm' vmspace always has a non-zero reference
466  * count.  This permits an additional reference on 'newvm' to be
467  * acquired via a simple atomic increment rather than the loop in
468  * vmspace_acquire_ref() above.
469  */
470 void
471 vmspace_switch_aio(struct vmspace *newvm)
472 {
473 	struct vmspace *oldvm;
474 
475 	/* XXX: Need some way to assert that this is an aio daemon. */
476 
477 	KASSERT(newvm->vm_refcnt > 0,
478 	    ("vmspace_switch_aio: newvm unreferenced"));
479 
480 	oldvm = curproc->p_vmspace;
481 	if (oldvm == newvm)
482 		return;
483 
484 	/*
485 	 * Point to the new address space and refer to it.
486 	 */
487 	curproc->p_vmspace = newvm;
488 	atomic_add_int(&newvm->vm_refcnt, 1);
489 
490 	/* Activate the new mapping. */
491 	pmap_activate(curthread);
492 
493 	vmspace_free(oldvm);
494 }
495 
496 void
497 _vm_map_lock(vm_map_t map, const char *file, int line)
498 {
499 
500 	if (map->system_map)
501 		mtx_lock_flags_(&map->system_mtx, 0, file, line);
502 	else
503 		sx_xlock_(&map->lock, file, line);
504 	map->timestamp++;
505 }
506 
507 void
508 vm_map_entry_set_vnode_text(vm_map_entry_t entry, bool add)
509 {
510 	vm_object_t object;
511 	struct vnode *vp;
512 	bool vp_held;
513 
514 	if ((entry->eflags & MAP_ENTRY_VN_EXEC) == 0)
515 		return;
516 	KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
517 	    ("Submap with execs"));
518 	object = entry->object.vm_object;
519 	KASSERT(object != NULL, ("No object for text, entry %p", entry));
520 	if ((object->flags & OBJ_ANON) != 0)
521 		object = object->handle;
522 	else
523 		KASSERT(object->backing_object == NULL,
524 		    ("non-anon object %p shadows", object));
525 	KASSERT(object != NULL, ("No content object for text, entry %p obj %p",
526 	    entry, entry->object.vm_object));
527 
528 	/*
529 	 * Mostly, we do not lock the backing object.  It is
530 	 * referenced by the entry we are processing, so it cannot go
531 	 * away.
532 	 */
533 	vp = NULL;
534 	vp_held = false;
535 	if (object->type == OBJT_DEAD) {
536 		/*
537 		 * For OBJT_DEAD objects, v_writecount was handled in
538 		 * vnode_pager_dealloc().
539 		 */
540 	} else if (object->type == OBJT_VNODE) {
541 		vp = object->handle;
542 	} else if (object->type == OBJT_SWAP) {
543 		KASSERT((object->flags & OBJ_TMPFS_NODE) != 0,
544 		    ("vm_map_entry_set_vnode_text: swap and !TMPFS "
545 		    "entry %p, object %p, add %d", entry, object, add));
546 		/*
547 		 * Tmpfs VREG node, which was reclaimed, has
548 		 * OBJ_TMPFS_NODE flag set, but not OBJ_TMPFS.  In
549 		 * this case there is no v_writecount to adjust.
550 		 */
551 		VM_OBJECT_RLOCK(object);
552 		if ((object->flags & OBJ_TMPFS) != 0) {
553 			vp = object->un_pager.swp.swp_tmpfs;
554 			if (vp != NULL) {
555 				vhold(vp);
556 				vp_held = true;
557 			}
558 		}
559 		VM_OBJECT_RUNLOCK(object);
560 	} else {
561 		KASSERT(0,
562 		    ("vm_map_entry_set_vnode_text: wrong object type, "
563 		    "entry %p, object %p, add %d", entry, object, add));
564 	}
565 	if (vp != NULL) {
566 		if (add) {
567 			VOP_SET_TEXT_CHECKED(vp);
568 		} else {
569 			vn_lock(vp, LK_SHARED | LK_RETRY);
570 			VOP_UNSET_TEXT_CHECKED(vp);
571 			VOP_UNLOCK(vp);
572 		}
573 		if (vp_held)
574 			vdrop(vp);
575 	}
576 }
577 
578 /*
579  * Use a different name for this vm_map_entry field when it's use
580  * is not consistent with its use as part of an ordered search tree.
581  */
582 #define defer_next right
583 
584 static void
585 vm_map_process_deferred(void)
586 {
587 	struct thread *td;
588 	vm_map_entry_t entry, next;
589 	vm_object_t object;
590 
591 	td = curthread;
592 	entry = td->td_map_def_user;
593 	td->td_map_def_user = NULL;
594 	while (entry != NULL) {
595 		next = entry->defer_next;
596 		MPASS((entry->eflags & (MAP_ENTRY_WRITECNT |
597 		    MAP_ENTRY_VN_EXEC)) != (MAP_ENTRY_WRITECNT |
598 		    MAP_ENTRY_VN_EXEC));
599 		if ((entry->eflags & MAP_ENTRY_WRITECNT) != 0) {
600 			/*
601 			 * Decrement the object's writemappings and
602 			 * possibly the vnode's v_writecount.
603 			 */
604 			KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
605 			    ("Submap with writecount"));
606 			object = entry->object.vm_object;
607 			KASSERT(object != NULL, ("No object for writecount"));
608 			vm_pager_release_writecount(object, entry->start,
609 			    entry->end);
610 		}
611 		vm_map_entry_set_vnode_text(entry, false);
612 		vm_map_entry_deallocate(entry, FALSE);
613 		entry = next;
614 	}
615 }
616 
617 #ifdef INVARIANTS
618 static void
619 _vm_map_assert_locked(vm_map_t map, const char *file, int line)
620 {
621 
622 	if (map->system_map)
623 		mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
624 	else
625 		sx_assert_(&map->lock, SA_XLOCKED, file, line);
626 }
627 
628 #define	VM_MAP_ASSERT_LOCKED(map) \
629     _vm_map_assert_locked(map, LOCK_FILE, LOCK_LINE)
630 
631 enum { VMMAP_CHECK_NONE, VMMAP_CHECK_UNLOCK, VMMAP_CHECK_ALL };
632 #ifdef DIAGNOSTIC
633 static int enable_vmmap_check = VMMAP_CHECK_UNLOCK;
634 #else
635 static int enable_vmmap_check = VMMAP_CHECK_NONE;
636 #endif
637 SYSCTL_INT(_debug, OID_AUTO, vmmap_check, CTLFLAG_RWTUN,
638     &enable_vmmap_check, 0, "Enable vm map consistency checking");
639 
640 static void _vm_map_assert_consistent(vm_map_t map, int check);
641 
642 #define VM_MAP_ASSERT_CONSISTENT(map) \
643     _vm_map_assert_consistent(map, VMMAP_CHECK_ALL)
644 #ifdef DIAGNOSTIC
645 #define VM_MAP_UNLOCK_CONSISTENT(map) do {				\
646 	if (map->nupdates > map->nentries) {				\
647 		_vm_map_assert_consistent(map, VMMAP_CHECK_UNLOCK);	\
648 		map->nupdates = 0;					\
649 	}								\
650 } while (0)
651 #else
652 #define VM_MAP_UNLOCK_CONSISTENT(map)
653 #endif
654 #else
655 #define	VM_MAP_ASSERT_LOCKED(map)
656 #define VM_MAP_ASSERT_CONSISTENT(map)
657 #define VM_MAP_UNLOCK_CONSISTENT(map)
658 #endif /* INVARIANTS */
659 
660 void
661 _vm_map_unlock(vm_map_t map, const char *file, int line)
662 {
663 
664 	VM_MAP_UNLOCK_CONSISTENT(map);
665 	if (map->system_map)
666 		mtx_unlock_flags_(&map->system_mtx, 0, file, line);
667 	else {
668 		sx_xunlock_(&map->lock, file, line);
669 		vm_map_process_deferred();
670 	}
671 }
672 
673 void
674 _vm_map_lock_read(vm_map_t map, const char *file, int line)
675 {
676 
677 	if (map->system_map)
678 		mtx_lock_flags_(&map->system_mtx, 0, file, line);
679 	else
680 		sx_slock_(&map->lock, file, line);
681 }
682 
683 void
684 _vm_map_unlock_read(vm_map_t map, const char *file, int line)
685 {
686 
687 	if (map->system_map)
688 		mtx_unlock_flags_(&map->system_mtx, 0, file, line);
689 	else {
690 		sx_sunlock_(&map->lock, file, line);
691 		vm_map_process_deferred();
692 	}
693 }
694 
695 int
696 _vm_map_trylock(vm_map_t map, const char *file, int line)
697 {
698 	int error;
699 
700 	error = map->system_map ?
701 	    !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
702 	    !sx_try_xlock_(&map->lock, file, line);
703 	if (error == 0)
704 		map->timestamp++;
705 	return (error == 0);
706 }
707 
708 int
709 _vm_map_trylock_read(vm_map_t map, const char *file, int line)
710 {
711 	int error;
712 
713 	error = map->system_map ?
714 	    !mtx_trylock_flags_(&map->system_mtx, 0, file, line) :
715 	    !sx_try_slock_(&map->lock, file, line);
716 	return (error == 0);
717 }
718 
719 /*
720  *	_vm_map_lock_upgrade:	[ internal use only ]
721  *
722  *	Tries to upgrade a read (shared) lock on the specified map to a write
723  *	(exclusive) lock.  Returns the value "0" if the upgrade succeeds and a
724  *	non-zero value if the upgrade fails.  If the upgrade fails, the map is
725  *	returned without a read or write lock held.
726  *
727  *	Requires that the map be read locked.
728  */
729 int
730 _vm_map_lock_upgrade(vm_map_t map, const char *file, int line)
731 {
732 	unsigned int last_timestamp;
733 
734 	if (map->system_map) {
735 		mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
736 	} else {
737 		if (!sx_try_upgrade_(&map->lock, file, line)) {
738 			last_timestamp = map->timestamp;
739 			sx_sunlock_(&map->lock, file, line);
740 			vm_map_process_deferred();
741 			/*
742 			 * If the map's timestamp does not change while the
743 			 * map is unlocked, then the upgrade succeeds.
744 			 */
745 			sx_xlock_(&map->lock, file, line);
746 			if (last_timestamp != map->timestamp) {
747 				sx_xunlock_(&map->lock, file, line);
748 				return (1);
749 			}
750 		}
751 	}
752 	map->timestamp++;
753 	return (0);
754 }
755 
756 void
757 _vm_map_lock_downgrade(vm_map_t map, const char *file, int line)
758 {
759 
760 	if (map->system_map) {
761 		mtx_assert_(&map->system_mtx, MA_OWNED, file, line);
762 	} else {
763 		VM_MAP_UNLOCK_CONSISTENT(map);
764 		sx_downgrade_(&map->lock, file, line);
765 	}
766 }
767 
768 /*
769  *	vm_map_locked:
770  *
771  *	Returns a non-zero value if the caller holds a write (exclusive) lock
772  *	on the specified map and the value "0" otherwise.
773  */
774 int
775 vm_map_locked(vm_map_t map)
776 {
777 
778 	if (map->system_map)
779 		return (mtx_owned(&map->system_mtx));
780 	else
781 		return (sx_xlocked(&map->lock));
782 }
783 
784 /*
785  *	_vm_map_unlock_and_wait:
786  *
787  *	Atomically releases the lock on the specified map and puts the calling
788  *	thread to sleep.  The calling thread will remain asleep until either
789  *	vm_map_wakeup() is performed on the map or the specified timeout is
790  *	exceeded.
791  *
792  *	WARNING!  This function does not perform deferred deallocations of
793  *	objects and map	entries.  Therefore, the calling thread is expected to
794  *	reacquire the map lock after reawakening and later perform an ordinary
795  *	unlock operation, such as vm_map_unlock(), before completing its
796  *	operation on the map.
797  */
798 int
799 _vm_map_unlock_and_wait(vm_map_t map, int timo, const char *file, int line)
800 {
801 
802 	VM_MAP_UNLOCK_CONSISTENT(map);
803 	mtx_lock(&map_sleep_mtx);
804 	if (map->system_map)
805 		mtx_unlock_flags_(&map->system_mtx, 0, file, line);
806 	else
807 		sx_xunlock_(&map->lock, file, line);
808 	return (msleep(&map->root, &map_sleep_mtx, PDROP | PVM, "vmmaps",
809 	    timo));
810 }
811 
812 /*
813  *	vm_map_wakeup:
814  *
815  *	Awaken any threads that have slept on the map using
816  *	vm_map_unlock_and_wait().
817  */
818 void
819 vm_map_wakeup(vm_map_t map)
820 {
821 
822 	/*
823 	 * Acquire and release map_sleep_mtx to prevent a wakeup()
824 	 * from being performed (and lost) between the map unlock
825 	 * and the msleep() in _vm_map_unlock_and_wait().
826 	 */
827 	mtx_lock(&map_sleep_mtx);
828 	mtx_unlock(&map_sleep_mtx);
829 	wakeup(&map->root);
830 }
831 
832 void
833 vm_map_busy(vm_map_t map)
834 {
835 
836 	VM_MAP_ASSERT_LOCKED(map);
837 	map->busy++;
838 }
839 
840 void
841 vm_map_unbusy(vm_map_t map)
842 {
843 
844 	VM_MAP_ASSERT_LOCKED(map);
845 	KASSERT(map->busy, ("vm_map_unbusy: not busy"));
846 	if (--map->busy == 0 && (map->flags & MAP_BUSY_WAKEUP)) {
847 		vm_map_modflags(map, 0, MAP_BUSY_WAKEUP);
848 		wakeup(&map->busy);
849 	}
850 }
851 
852 void
853 vm_map_wait_busy(vm_map_t map)
854 {
855 
856 	VM_MAP_ASSERT_LOCKED(map);
857 	while (map->busy) {
858 		vm_map_modflags(map, MAP_BUSY_WAKEUP, 0);
859 		if (map->system_map)
860 			msleep(&map->busy, &map->system_mtx, 0, "mbusy", 0);
861 		else
862 			sx_sleep(&map->busy, &map->lock, 0, "mbusy", 0);
863 	}
864 	map->timestamp++;
865 }
866 
867 long
868 vmspace_resident_count(struct vmspace *vmspace)
869 {
870 	return pmap_resident_count(vmspace_pmap(vmspace));
871 }
872 
873 /*
874  *	vm_map_create:
875  *
876  *	Creates and returns a new empty VM map with
877  *	the given physical map structure, and having
878  *	the given lower and upper address bounds.
879  */
880 vm_map_t
881 vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max)
882 {
883 	vm_map_t result;
884 
885 	result = uma_zalloc(mapzone, M_WAITOK);
886 	CTR1(KTR_VM, "vm_map_create: %p", result);
887 	_vm_map_init(result, pmap, min, max);
888 	return (result);
889 }
890 
891 /*
892  * Initialize an existing vm_map structure
893  * such as that in the vmspace structure.
894  */
895 static void
896 _vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
897 {
898 
899 	map->header.eflags = MAP_ENTRY_HEADER;
900 	map->needs_wakeup = FALSE;
901 	map->system_map = 0;
902 	map->pmap = pmap;
903 	map->header.end = min;
904 	map->header.start = max;
905 	map->flags = 0;
906 	map->header.left = map->header.right = &map->header;
907 	map->root = NULL;
908 	map->timestamp = 0;
909 	map->busy = 0;
910 	map->anon_loc = 0;
911 #ifdef DIAGNOSTIC
912 	map->nupdates = 0;
913 #endif
914 }
915 
916 void
917 vm_map_init(vm_map_t map, pmap_t pmap, vm_offset_t min, vm_offset_t max)
918 {
919 
920 	_vm_map_init(map, pmap, min, max);
921 	mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK);
922 	sx_init(&map->lock, "user map");
923 }
924 
925 /*
926  *	vm_map_entry_dispose:	[ internal use only ]
927  *
928  *	Inverse of vm_map_entry_create.
929  */
930 static void
931 vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry)
932 {
933 	uma_zfree(map->system_map ? kmapentzone : mapentzone, entry);
934 }
935 
936 /*
937  *	vm_map_entry_create:	[ internal use only ]
938  *
939  *	Allocates a VM map entry for insertion.
940  *	No entry fields are filled in.
941  */
942 static vm_map_entry_t
943 vm_map_entry_create(vm_map_t map)
944 {
945 	vm_map_entry_t new_entry;
946 
947 	if (map->system_map)
948 		new_entry = uma_zalloc(kmapentzone, M_NOWAIT);
949 	else
950 		new_entry = uma_zalloc(mapentzone, M_WAITOK);
951 	if (new_entry == NULL)
952 		panic("vm_map_entry_create: kernel resources exhausted");
953 	return (new_entry);
954 }
955 
956 /*
957  *	vm_map_entry_set_behavior:
958  *
959  *	Set the expected access behavior, either normal, random, or
960  *	sequential.
961  */
962 static inline void
963 vm_map_entry_set_behavior(vm_map_entry_t entry, u_char behavior)
964 {
965 	entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) |
966 	    (behavior & MAP_ENTRY_BEHAV_MASK);
967 }
968 
969 /*
970  *	vm_map_entry_max_free_{left,right}:
971  *
972  *	Compute the size of the largest free gap between two entries,
973  *	one the root of a tree and the other the ancestor of that root
974  *	that is the least or greatest ancestor found on the search path.
975  */
976 static inline vm_size_t
977 vm_map_entry_max_free_left(vm_map_entry_t root, vm_map_entry_t left_ancestor)
978 {
979 
980 	return (root->left != left_ancestor ?
981 	    root->left->max_free : root->start - left_ancestor->end);
982 }
983 
984 static inline vm_size_t
985 vm_map_entry_max_free_right(vm_map_entry_t root, vm_map_entry_t right_ancestor)
986 {
987 
988 	return (root->right != right_ancestor ?
989 	    root->right->max_free : right_ancestor->start - root->end);
990 }
991 
992 /*
993  *	vm_map_entry_{pred,succ}:
994  *
995  *	Find the {predecessor, successor} of the entry by taking one step
996  *	in the appropriate direction and backtracking as much as necessary.
997  *	vm_map_entry_succ is defined in vm_map.h.
998  */
999 static inline vm_map_entry_t
1000 vm_map_entry_pred(vm_map_entry_t entry)
1001 {
1002 	vm_map_entry_t prior;
1003 
1004 	prior = entry->left;
1005 	if (prior->right->start < entry->start) {
1006 		do
1007 			prior = prior->right;
1008 		while (prior->right != entry);
1009 	}
1010 	return (prior);
1011 }
1012 
1013 static inline vm_size_t
1014 vm_size_max(vm_size_t a, vm_size_t b)
1015 {
1016 
1017 	return (a > b ? a : b);
1018 }
1019 
1020 #define SPLAY_LEFT_STEP(root, y, llist, rlist, test) do {		\
1021 	vm_map_entry_t z;						\
1022 	vm_size_t max_free;						\
1023 									\
1024 	/*								\
1025 	 * Infer root->right->max_free == root->max_free when		\
1026 	 * y->max_free < root->max_free || root->max_free == 0.		\
1027 	 * Otherwise, look right to find it.				\
1028 	 */								\
1029 	y = root->left;							\
1030 	max_free = root->max_free;					\
1031 	KASSERT(max_free == vm_size_max(				\
1032 	    vm_map_entry_max_free_left(root, llist),			\
1033 	    vm_map_entry_max_free_right(root, rlist)),			\
1034 	    ("%s: max_free invariant fails", __func__));		\
1035 	if (max_free - 1 < vm_map_entry_max_free_left(root, llist))	\
1036 		max_free = vm_map_entry_max_free_right(root, rlist);	\
1037 	if (y != llist && (test)) {					\
1038 		/* Rotate right and make y root. */			\
1039 		z = y->right;						\
1040 		if (z != root) {					\
1041 			root->left = z;					\
1042 			y->right = root;				\
1043 			if (max_free < y->max_free)			\
1044 			    root->max_free = max_free =			\
1045 			    vm_size_max(max_free, z->max_free);		\
1046 		} else if (max_free < y->max_free)			\
1047 			root->max_free = max_free =			\
1048 			    vm_size_max(max_free, root->start - y->end);\
1049 		root = y;						\
1050 		y = root->left;						\
1051 	}								\
1052 	/* Copy right->max_free.  Put root on rlist. */			\
1053 	root->max_free = max_free;					\
1054 	KASSERT(max_free == vm_map_entry_max_free_right(root, rlist),	\
1055 	    ("%s: max_free not copied from right", __func__));		\
1056 	root->left = rlist;						\
1057 	rlist = root;							\
1058 	root = y != llist ? y : NULL;					\
1059 } while (0)
1060 
1061 #define SPLAY_RIGHT_STEP(root, y, llist, rlist, test) do {		\
1062 	vm_map_entry_t z;						\
1063 	vm_size_t max_free;						\
1064 									\
1065 	/*								\
1066 	 * Infer root->left->max_free == root->max_free when		\
1067 	 * y->max_free < root->max_free || root->max_free == 0.		\
1068 	 * Otherwise, look left to find it.				\
1069 	 */								\
1070 	y = root->right;						\
1071 	max_free = root->max_free;					\
1072 	KASSERT(max_free == vm_size_max(				\
1073 	    vm_map_entry_max_free_left(root, llist),			\
1074 	    vm_map_entry_max_free_right(root, rlist)),			\
1075 	    ("%s: max_free invariant fails", __func__));		\
1076 	if (max_free - 1 < vm_map_entry_max_free_right(root, rlist))	\
1077 		max_free = vm_map_entry_max_free_left(root, llist);	\
1078 	if (y != rlist && (test)) {					\
1079 		/* Rotate left and make y root. */			\
1080 		z = y->left;						\
1081 		if (z != root) {					\
1082 			root->right = z;				\
1083 			y->left = root;					\
1084 			if (max_free < y->max_free)			\
1085 			    root->max_free = max_free =			\
1086 			    vm_size_max(max_free, z->max_free);		\
1087 		} else if (max_free < y->max_free)			\
1088 			root->max_free = max_free =			\
1089 			    vm_size_max(max_free, y->start - root->end);\
1090 		root = y;						\
1091 		y = root->right;					\
1092 	}								\
1093 	/* Copy left->max_free.  Put root on llist. */			\
1094 	root->max_free = max_free;					\
1095 	KASSERT(max_free == vm_map_entry_max_free_left(root, llist),	\
1096 	    ("%s: max_free not copied from left", __func__));		\
1097 	root->right = llist;						\
1098 	llist = root;							\
1099 	root = y != rlist ? y : NULL;					\
1100 } while (0)
1101 
1102 /*
1103  * Walk down the tree until we find addr or a gap where addr would go, breaking
1104  * off left and right subtrees of nodes less than, or greater than addr.  Treat
1105  * subtrees with root->max_free < length as empty trees.  llist and rlist are
1106  * the two sides in reverse order (bottom-up), with llist linked by the right
1107  * pointer and rlist linked by the left pointer in the vm_map_entry, and both
1108  * lists terminated by &map->header.  This function, and the subsequent call to
1109  * vm_map_splay_merge_{left,right,pred,succ}, rely on the start and end address
1110  * values in &map->header.
1111  */
1112 static __always_inline vm_map_entry_t
1113 vm_map_splay_split(vm_map_t map, vm_offset_t addr, vm_size_t length,
1114     vm_map_entry_t *llist, vm_map_entry_t *rlist)
1115 {
1116 	vm_map_entry_t left, right, root, y;
1117 
1118 	left = right = &map->header;
1119 	root = map->root;
1120 	while (root != NULL && root->max_free >= length) {
1121 		KASSERT(left->end <= root->start &&
1122 		    root->end <= right->start,
1123 		    ("%s: root not within tree bounds", __func__));
1124 		if (addr < root->start) {
1125 			SPLAY_LEFT_STEP(root, y, left, right,
1126 			    y->max_free >= length && addr < y->start);
1127 		} else if (addr >= root->end) {
1128 			SPLAY_RIGHT_STEP(root, y, left, right,
1129 			    y->max_free >= length && addr >= y->end);
1130 		} else
1131 			break;
1132 	}
1133 	*llist = left;
1134 	*rlist = right;
1135 	return (root);
1136 }
1137 
1138 static __always_inline void
1139 vm_map_splay_findnext(vm_map_entry_t root, vm_map_entry_t *rlist)
1140 {
1141 	vm_map_entry_t hi, right, y;
1142 
1143 	right = *rlist;
1144 	hi = root->right == right ? NULL : root->right;
1145 	if (hi == NULL)
1146 		return;
1147 	do
1148 		SPLAY_LEFT_STEP(hi, y, root, right, true);
1149 	while (hi != NULL);
1150 	*rlist = right;
1151 }
1152 
1153 static __always_inline void
1154 vm_map_splay_findprev(vm_map_entry_t root, vm_map_entry_t *llist)
1155 {
1156 	vm_map_entry_t left, lo, y;
1157 
1158 	left = *llist;
1159 	lo = root->left == left ? NULL : root->left;
1160 	if (lo == NULL)
1161 		return;
1162 	do
1163 		SPLAY_RIGHT_STEP(lo, y, left, root, true);
1164 	while (lo != NULL);
1165 	*llist = left;
1166 }
1167 
1168 static inline void
1169 vm_map_entry_swap(vm_map_entry_t *a, vm_map_entry_t *b)
1170 {
1171 	vm_map_entry_t tmp;
1172 
1173 	tmp = *b;
1174 	*b = *a;
1175 	*a = tmp;
1176 }
1177 
1178 /*
1179  * Walk back up the two spines, flip the pointers and set max_free.  The
1180  * subtrees of the root go at the bottom of llist and rlist.
1181  */
1182 static vm_size_t
1183 vm_map_splay_merge_left_walk(vm_map_entry_t header, vm_map_entry_t root,
1184     vm_map_entry_t tail, vm_size_t max_free, vm_map_entry_t llist)
1185 {
1186 	do {
1187 		/*
1188 		 * The max_free values of the children of llist are in
1189 		 * llist->max_free and max_free.  Update with the
1190 		 * max value.
1191 		 */
1192 		llist->max_free = max_free =
1193 		    vm_size_max(llist->max_free, max_free);
1194 		vm_map_entry_swap(&llist->right, &tail);
1195 		vm_map_entry_swap(&tail, &llist);
1196 	} while (llist != header);
1197 	root->left = tail;
1198 	return (max_free);
1199 }
1200 
1201 /*
1202  * When llist is known to be the predecessor of root.
1203  */
1204 static inline vm_size_t
1205 vm_map_splay_merge_pred(vm_map_entry_t header, vm_map_entry_t root,
1206     vm_map_entry_t llist)
1207 {
1208 	vm_size_t max_free;
1209 
1210 	max_free = root->start - llist->end;
1211 	if (llist != header) {
1212 		max_free = vm_map_splay_merge_left_walk(header, root,
1213 		    root, max_free, llist);
1214 	} else {
1215 		root->left = header;
1216 		header->right = root;
1217 	}
1218 	return (max_free);
1219 }
1220 
1221 /*
1222  * When llist may or may not be the predecessor of root.
1223  */
1224 static inline vm_size_t
1225 vm_map_splay_merge_left(vm_map_entry_t header, vm_map_entry_t root,
1226     vm_map_entry_t llist)
1227 {
1228 	vm_size_t max_free;
1229 
1230 	max_free = vm_map_entry_max_free_left(root, llist);
1231 	if (llist != header) {
1232 		max_free = vm_map_splay_merge_left_walk(header, root,
1233 		    root->left == llist ? root : root->left,
1234 		    max_free, llist);
1235 	}
1236 	return (max_free);
1237 }
1238 
1239 static vm_size_t
1240 vm_map_splay_merge_right_walk(vm_map_entry_t header, vm_map_entry_t root,
1241     vm_map_entry_t tail, vm_size_t max_free, vm_map_entry_t rlist)
1242 {
1243 	do {
1244 		/*
1245 		 * The max_free values of the children of rlist are in
1246 		 * rlist->max_free and max_free.  Update with the
1247 		 * max value.
1248 		 */
1249 		rlist->max_free = max_free =
1250 		    vm_size_max(rlist->max_free, max_free);
1251 		vm_map_entry_swap(&rlist->left, &tail);
1252 		vm_map_entry_swap(&tail, &rlist);
1253 	} while (rlist != header);
1254 	root->right = tail;
1255 	return (max_free);
1256 }
1257 
1258 /*
1259  * When rlist is known to be the succecessor of root.
1260  */
1261 static inline vm_size_t
1262 vm_map_splay_merge_succ(vm_map_entry_t header, vm_map_entry_t root,
1263     vm_map_entry_t rlist)
1264 {
1265 	vm_size_t max_free;
1266 
1267 	max_free = rlist->start - root->end;
1268 	if (rlist != header) {
1269 		max_free = vm_map_splay_merge_right_walk(header, root,
1270 		    root, max_free, rlist);
1271 	} else {
1272 		root->right = header;
1273 		header->left = root;
1274 	}
1275 	return (max_free);
1276 }
1277 
1278 /*
1279  * When rlist may or may not be the succecessor of root.
1280  */
1281 static inline vm_size_t
1282 vm_map_splay_merge_right(vm_map_entry_t header, vm_map_entry_t root,
1283     vm_map_entry_t rlist)
1284 {
1285 	vm_size_t max_free;
1286 
1287 	max_free = vm_map_entry_max_free_right(root, rlist);
1288 	if (rlist != header) {
1289 		max_free = vm_map_splay_merge_right_walk(header, root,
1290 		    root->right == rlist ? root : root->right,
1291 		    max_free, rlist);
1292 	}
1293 	return (max_free);
1294 }
1295 
1296 /*
1297  *	vm_map_splay:
1298  *
1299  *	The Sleator and Tarjan top-down splay algorithm with the
1300  *	following variation.  Max_free must be computed bottom-up, so
1301  *	on the downward pass, maintain the left and right spines in
1302  *	reverse order.  Then, make a second pass up each side to fix
1303  *	the pointers and compute max_free.  The time bound is O(log n)
1304  *	amortized.
1305  *
1306  *	The tree is threaded, which means that there are no null pointers.
1307  *	When a node has no left child, its left pointer points to its
1308  *	predecessor, which the last ancestor on the search path from the root
1309  *	where the search branched right.  Likewise, when a node has no right
1310  *	child, its right pointer points to its successor.  The map header node
1311  *	is the predecessor of the first map entry, and the successor of the
1312  *	last.
1313  *
1314  *	The new root is the vm_map_entry containing "addr", or else an
1315  *	adjacent entry (lower if possible) if addr is not in the tree.
1316  *
1317  *	The map must be locked, and leaves it so.
1318  *
1319  *	Returns: the new root.
1320  */
1321 static vm_map_entry_t
1322 vm_map_splay(vm_map_t map, vm_offset_t addr)
1323 {
1324 	vm_map_entry_t header, llist, rlist, root;
1325 	vm_size_t max_free_left, max_free_right;
1326 
1327 	header = &map->header;
1328 	root = vm_map_splay_split(map, addr, 0, &llist, &rlist);
1329 	if (root != NULL) {
1330 		max_free_left = vm_map_splay_merge_left(header, root, llist);
1331 		max_free_right = vm_map_splay_merge_right(header, root, rlist);
1332 	} else if (llist != header) {
1333 		/*
1334 		 * Recover the greatest node in the left
1335 		 * subtree and make it the root.
1336 		 */
1337 		root = llist;
1338 		llist = root->right;
1339 		max_free_left = vm_map_splay_merge_left(header, root, llist);
1340 		max_free_right = vm_map_splay_merge_succ(header, root, rlist);
1341 	} else if (rlist != header) {
1342 		/*
1343 		 * Recover the least node in the right
1344 		 * subtree and make it the root.
1345 		 */
1346 		root = rlist;
1347 		rlist = root->left;
1348 		max_free_left = vm_map_splay_merge_pred(header, root, llist);
1349 		max_free_right = vm_map_splay_merge_right(header, root, rlist);
1350 	} else {
1351 		/* There is no root. */
1352 		return (NULL);
1353 	}
1354 	root->max_free = vm_size_max(max_free_left, max_free_right);
1355 	map->root = root;
1356 	VM_MAP_ASSERT_CONSISTENT(map);
1357 	return (root);
1358 }
1359 
1360 /*
1361  *	vm_map_entry_{un,}link:
1362  *
1363  *	Insert/remove entries from maps.  On linking, if new entry clips
1364  *	existing entry, trim existing entry to avoid overlap, and manage
1365  *	offsets.  On unlinking, merge disappearing entry with neighbor, if
1366  *	called for, and manage offsets.  Callers should not modify fields in
1367  *	entries already mapped.
1368  */
1369 static void
1370 vm_map_entry_link(vm_map_t map, vm_map_entry_t entry)
1371 {
1372 	vm_map_entry_t header, llist, rlist, root;
1373 	vm_size_t max_free_left, max_free_right;
1374 
1375 	CTR3(KTR_VM,
1376 	    "vm_map_entry_link: map %p, nentries %d, entry %p", map,
1377 	    map->nentries, entry);
1378 	VM_MAP_ASSERT_LOCKED(map);
1379 	map->nentries++;
1380 	header = &map->header;
1381 	root = vm_map_splay_split(map, entry->start, 0, &llist, &rlist);
1382 	if (root == NULL) {
1383 		/*
1384 		 * The new entry does not overlap any existing entry in the
1385 		 * map, so it becomes the new root of the map tree.
1386 		 */
1387 		max_free_left = vm_map_splay_merge_pred(header, entry, llist);
1388 		max_free_right = vm_map_splay_merge_succ(header, entry, rlist);
1389 	} else if (entry->start == root->start) {
1390 		/*
1391 		 * The new entry is a clone of root, with only the end field
1392 		 * changed.  The root entry will be shrunk to abut the new
1393 		 * entry, and will be the right child of the new root entry in
1394 		 * the modified map.
1395 		 */
1396 		KASSERT(entry->end < root->end,
1397 		    ("%s: clip_start not within entry", __func__));
1398 		vm_map_splay_findprev(root, &llist);
1399 		root->offset += entry->end - root->start;
1400 		root->start = entry->end;
1401 		max_free_left = vm_map_splay_merge_pred(header, entry, llist);
1402 		max_free_right = root->max_free = vm_size_max(
1403 		    vm_map_splay_merge_pred(entry, root, entry),
1404 		    vm_map_splay_merge_right(header, root, rlist));
1405 	} else {
1406 		/*
1407 		 * The new entry is a clone of root, with only the start field
1408 		 * changed.  The root entry will be shrunk to abut the new
1409 		 * entry, and will be the left child of the new root entry in
1410 		 * the modified map.
1411 		 */
1412 		KASSERT(entry->end == root->end,
1413 		    ("%s: clip_start not within entry", __func__));
1414 		vm_map_splay_findnext(root, &rlist);
1415 		entry->offset += entry->start - root->start;
1416 		root->end = entry->start;
1417 		max_free_left = root->max_free = vm_size_max(
1418 		    vm_map_splay_merge_left(header, root, llist),
1419 		    vm_map_splay_merge_succ(entry, root, entry));
1420 		max_free_right = vm_map_splay_merge_succ(header, entry, rlist);
1421 	}
1422 	entry->max_free = vm_size_max(max_free_left, max_free_right);
1423 	map->root = entry;
1424 	VM_MAP_ASSERT_CONSISTENT(map);
1425 }
1426 
1427 enum unlink_merge_type {
1428 	UNLINK_MERGE_NONE,
1429 	UNLINK_MERGE_NEXT
1430 };
1431 
1432 static void
1433 vm_map_entry_unlink(vm_map_t map, vm_map_entry_t entry,
1434     enum unlink_merge_type op)
1435 {
1436 	vm_map_entry_t header, llist, rlist, root;
1437 	vm_size_t max_free_left, max_free_right;
1438 
1439 	VM_MAP_ASSERT_LOCKED(map);
1440 	header = &map->header;
1441 	root = vm_map_splay_split(map, entry->start, 0, &llist, &rlist);
1442 	KASSERT(root != NULL,
1443 	    ("vm_map_entry_unlink: unlink object not mapped"));
1444 
1445 	vm_map_splay_findprev(root, &llist);
1446 	vm_map_splay_findnext(root, &rlist);
1447 	if (op == UNLINK_MERGE_NEXT) {
1448 		rlist->start = root->start;
1449 		rlist->offset = root->offset;
1450 	}
1451 	if (llist != header) {
1452 		root = llist;
1453 		llist = root->right;
1454 		max_free_left = vm_map_splay_merge_left(header, root, llist);
1455 		max_free_right = vm_map_splay_merge_succ(header, root, rlist);
1456 	} else if (rlist != header) {
1457 		root = rlist;
1458 		rlist = root->left;
1459 		max_free_left = vm_map_splay_merge_pred(header, root, llist);
1460 		max_free_right = vm_map_splay_merge_right(header, root, rlist);
1461 	} else {
1462 		header->left = header->right = header;
1463 		root = NULL;
1464 	}
1465 	if (root != NULL)
1466 		root->max_free = vm_size_max(max_free_left, max_free_right);
1467 	map->root = root;
1468 	VM_MAP_ASSERT_CONSISTENT(map);
1469 	map->nentries--;
1470 	CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map,
1471 	    map->nentries, entry);
1472 }
1473 
1474 /*
1475  *	vm_map_entry_resize:
1476  *
1477  *	Resize a vm_map_entry, recompute the amount of free space that
1478  *	follows it and propagate that value up the tree.
1479  *
1480  *	The map must be locked, and leaves it so.
1481  */
1482 static void
1483 vm_map_entry_resize(vm_map_t map, vm_map_entry_t entry, vm_size_t grow_amount)
1484 {
1485 	vm_map_entry_t header, llist, rlist, root;
1486 
1487 	VM_MAP_ASSERT_LOCKED(map);
1488 	header = &map->header;
1489 	root = vm_map_splay_split(map, entry->start, 0, &llist, &rlist);
1490 	KASSERT(root != NULL, ("%s: resize object not mapped", __func__));
1491 	vm_map_splay_findnext(root, &rlist);
1492 	entry->end += grow_amount;
1493 	root->max_free = vm_size_max(
1494 	    vm_map_splay_merge_left(header, root, llist),
1495 	    vm_map_splay_merge_succ(header, root, rlist));
1496 	map->root = root;
1497 	VM_MAP_ASSERT_CONSISTENT(map);
1498 	CTR4(KTR_VM, "%s: map %p, nentries %d, entry %p",
1499 	    __func__, map, map->nentries, entry);
1500 }
1501 
1502 /*
1503  *	vm_map_lookup_entry:	[ internal use only ]
1504  *
1505  *	Finds the map entry containing (or
1506  *	immediately preceding) the specified address
1507  *	in the given map; the entry is returned
1508  *	in the "entry" parameter.  The boolean
1509  *	result indicates whether the address is
1510  *	actually contained in the map.
1511  */
1512 boolean_t
1513 vm_map_lookup_entry(
1514 	vm_map_t map,
1515 	vm_offset_t address,
1516 	vm_map_entry_t *entry)	/* OUT */
1517 {
1518 	vm_map_entry_t cur, header, lbound, ubound;
1519 	boolean_t locked;
1520 
1521 	/*
1522 	 * If the map is empty, then the map entry immediately preceding
1523 	 * "address" is the map's header.
1524 	 */
1525 	header = &map->header;
1526 	cur = map->root;
1527 	if (cur == NULL) {
1528 		*entry = header;
1529 		return (FALSE);
1530 	}
1531 	if (address >= cur->start && cur->end > address) {
1532 		*entry = cur;
1533 		return (TRUE);
1534 	}
1535 	if ((locked = vm_map_locked(map)) ||
1536 	    sx_try_upgrade(&map->lock)) {
1537 		/*
1538 		 * Splay requires a write lock on the map.  However, it only
1539 		 * restructures the binary search tree; it does not otherwise
1540 		 * change the map.  Thus, the map's timestamp need not change
1541 		 * on a temporary upgrade.
1542 		 */
1543 		cur = vm_map_splay(map, address);
1544 		if (!locked) {
1545 			VM_MAP_UNLOCK_CONSISTENT(map);
1546 			sx_downgrade(&map->lock);
1547 		}
1548 
1549 		/*
1550 		 * If "address" is contained within a map entry, the new root
1551 		 * is that map entry.  Otherwise, the new root is a map entry
1552 		 * immediately before or after "address".
1553 		 */
1554 		if (address < cur->start) {
1555 			*entry = header;
1556 			return (FALSE);
1557 		}
1558 		*entry = cur;
1559 		return (address < cur->end);
1560 	}
1561 	/*
1562 	 * Since the map is only locked for read access, perform a
1563 	 * standard binary search tree lookup for "address".
1564 	 */
1565 	lbound = ubound = header;
1566 	for (;;) {
1567 		if (address < cur->start) {
1568 			ubound = cur;
1569 			cur = cur->left;
1570 			if (cur == lbound)
1571 				break;
1572 		} else if (cur->end <= address) {
1573 			lbound = cur;
1574 			cur = cur->right;
1575 			if (cur == ubound)
1576 				break;
1577 		} else {
1578 			*entry = cur;
1579 			return (TRUE);
1580 		}
1581 	}
1582 	*entry = lbound;
1583 	return (FALSE);
1584 }
1585 
1586 /*
1587  *	vm_map_insert:
1588  *
1589  *	Inserts the given whole VM object into the target
1590  *	map at the specified address range.  The object's
1591  *	size should match that of the address range.
1592  *
1593  *	Requires that the map be locked, and leaves it so.
1594  *
1595  *	If object is non-NULL, ref count must be bumped by caller
1596  *	prior to making call to account for the new entry.
1597  */
1598 int
1599 vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1600     vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max, int cow)
1601 {
1602 	vm_map_entry_t new_entry, next_entry, prev_entry;
1603 	struct ucred *cred;
1604 	vm_eflags_t protoeflags;
1605 	vm_inherit_t inheritance;
1606 
1607 	VM_MAP_ASSERT_LOCKED(map);
1608 	KASSERT(object != kernel_object ||
1609 	    (cow & MAP_COPY_ON_WRITE) == 0,
1610 	    ("vm_map_insert: kernel object and COW"));
1611 	KASSERT(object == NULL || (cow & MAP_NOFAULT) == 0,
1612 	    ("vm_map_insert: paradoxical MAP_NOFAULT request"));
1613 	KASSERT((prot & ~max) == 0,
1614 	    ("prot %#x is not subset of max_prot %#x", prot, max));
1615 
1616 	/*
1617 	 * Check that the start and end points are not bogus.
1618 	 */
1619 	if (start == end || !vm_map_range_valid(map, start, end))
1620 		return (KERN_INVALID_ADDRESS);
1621 
1622 	/*
1623 	 * Find the entry prior to the proposed starting address; if it's part
1624 	 * of an existing entry, this range is bogus.
1625 	 */
1626 	if (vm_map_lookup_entry(map, start, &prev_entry))
1627 		return (KERN_NO_SPACE);
1628 
1629 	/*
1630 	 * Assert that the next entry doesn't overlap the end point.
1631 	 */
1632 	next_entry = vm_map_entry_succ(prev_entry);
1633 	if (next_entry->start < end)
1634 		return (KERN_NO_SPACE);
1635 
1636 	if ((cow & MAP_CREATE_GUARD) != 0 && (object != NULL ||
1637 	    max != VM_PROT_NONE))
1638 		return (KERN_INVALID_ARGUMENT);
1639 
1640 	protoeflags = 0;
1641 	if (cow & MAP_COPY_ON_WRITE)
1642 		protoeflags |= MAP_ENTRY_COW | MAP_ENTRY_NEEDS_COPY;
1643 	if (cow & MAP_NOFAULT)
1644 		protoeflags |= MAP_ENTRY_NOFAULT;
1645 	if (cow & MAP_DISABLE_SYNCER)
1646 		protoeflags |= MAP_ENTRY_NOSYNC;
1647 	if (cow & MAP_DISABLE_COREDUMP)
1648 		protoeflags |= MAP_ENTRY_NOCOREDUMP;
1649 	if (cow & MAP_STACK_GROWS_DOWN)
1650 		protoeflags |= MAP_ENTRY_GROWS_DOWN;
1651 	if (cow & MAP_STACK_GROWS_UP)
1652 		protoeflags |= MAP_ENTRY_GROWS_UP;
1653 	if (cow & MAP_WRITECOUNT)
1654 		protoeflags |= MAP_ENTRY_WRITECNT;
1655 	if (cow & MAP_VN_EXEC)
1656 		protoeflags |= MAP_ENTRY_VN_EXEC;
1657 	if ((cow & MAP_CREATE_GUARD) != 0)
1658 		protoeflags |= MAP_ENTRY_GUARD;
1659 	if ((cow & MAP_CREATE_STACK_GAP_DN) != 0)
1660 		protoeflags |= MAP_ENTRY_STACK_GAP_DN;
1661 	if ((cow & MAP_CREATE_STACK_GAP_UP) != 0)
1662 		protoeflags |= MAP_ENTRY_STACK_GAP_UP;
1663 	if (cow & MAP_INHERIT_SHARE)
1664 		inheritance = VM_INHERIT_SHARE;
1665 	else
1666 		inheritance = VM_INHERIT_DEFAULT;
1667 
1668 	cred = NULL;
1669 	if ((cow & (MAP_ACC_NO_CHARGE | MAP_NOFAULT | MAP_CREATE_GUARD)) != 0)
1670 		goto charged;
1671 	if ((cow & MAP_ACC_CHARGED) || ((prot & VM_PROT_WRITE) &&
1672 	    ((protoeflags & MAP_ENTRY_NEEDS_COPY) || object == NULL))) {
1673 		if (!(cow & MAP_ACC_CHARGED) && !swap_reserve(end - start))
1674 			return (KERN_RESOURCE_SHORTAGE);
1675 		KASSERT(object == NULL ||
1676 		    (protoeflags & MAP_ENTRY_NEEDS_COPY) != 0 ||
1677 		    object->cred == NULL,
1678 		    ("overcommit: vm_map_insert o %p", object));
1679 		cred = curthread->td_ucred;
1680 	}
1681 
1682 charged:
1683 	/* Expand the kernel pmap, if necessary. */
1684 	if (map == kernel_map && end > kernel_vm_end)
1685 		pmap_growkernel(end);
1686 	if (object != NULL) {
1687 		/*
1688 		 * OBJ_ONEMAPPING must be cleared unless this mapping
1689 		 * is trivially proven to be the only mapping for any
1690 		 * of the object's pages.  (Object granularity
1691 		 * reference counting is insufficient to recognize
1692 		 * aliases with precision.)
1693 		 */
1694 		if ((object->flags & OBJ_ANON) != 0) {
1695 			VM_OBJECT_WLOCK(object);
1696 			if (object->ref_count > 1 || object->shadow_count != 0)
1697 				vm_object_clear_flag(object, OBJ_ONEMAPPING);
1698 			VM_OBJECT_WUNLOCK(object);
1699 		}
1700 	} else if ((prev_entry->eflags & ~MAP_ENTRY_USER_WIRED) ==
1701 	    protoeflags &&
1702 	    (cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP |
1703 	    MAP_VN_EXEC)) == 0 &&
1704 	    prev_entry->end == start && (prev_entry->cred == cred ||
1705 	    (prev_entry->object.vm_object != NULL &&
1706 	    prev_entry->object.vm_object->cred == cred)) &&
1707 	    vm_object_coalesce(prev_entry->object.vm_object,
1708 	    prev_entry->offset,
1709 	    (vm_size_t)(prev_entry->end - prev_entry->start),
1710 	    (vm_size_t)(end - prev_entry->end), cred != NULL &&
1711 	    (protoeflags & MAP_ENTRY_NEEDS_COPY) == 0)) {
1712 		/*
1713 		 * We were able to extend the object.  Determine if we
1714 		 * can extend the previous map entry to include the
1715 		 * new range as well.
1716 		 */
1717 		if (prev_entry->inheritance == inheritance &&
1718 		    prev_entry->protection == prot &&
1719 		    prev_entry->max_protection == max &&
1720 		    prev_entry->wired_count == 0) {
1721 			KASSERT((prev_entry->eflags & MAP_ENTRY_USER_WIRED) ==
1722 			    0, ("prev_entry %p has incoherent wiring",
1723 			    prev_entry));
1724 			if ((prev_entry->eflags & MAP_ENTRY_GUARD) == 0)
1725 				map->size += end - prev_entry->end;
1726 			vm_map_entry_resize(map, prev_entry,
1727 			    end - prev_entry->end);
1728 			vm_map_try_merge_entries(map, prev_entry, next_entry);
1729 			return (KERN_SUCCESS);
1730 		}
1731 
1732 		/*
1733 		 * If we can extend the object but cannot extend the
1734 		 * map entry, we have to create a new map entry.  We
1735 		 * must bump the ref count on the extended object to
1736 		 * account for it.  object may be NULL.
1737 		 */
1738 		object = prev_entry->object.vm_object;
1739 		offset = prev_entry->offset +
1740 		    (prev_entry->end - prev_entry->start);
1741 		vm_object_reference(object);
1742 		if (cred != NULL && object != NULL && object->cred != NULL &&
1743 		    !(prev_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
1744 			/* Object already accounts for this uid. */
1745 			cred = NULL;
1746 		}
1747 	}
1748 	if (cred != NULL)
1749 		crhold(cred);
1750 
1751 	/*
1752 	 * Create a new entry
1753 	 */
1754 	new_entry = vm_map_entry_create(map);
1755 	new_entry->start = start;
1756 	new_entry->end = end;
1757 	new_entry->cred = NULL;
1758 
1759 	new_entry->eflags = protoeflags;
1760 	new_entry->object.vm_object = object;
1761 	new_entry->offset = offset;
1762 
1763 	new_entry->inheritance = inheritance;
1764 	new_entry->protection = prot;
1765 	new_entry->max_protection = max;
1766 	new_entry->wired_count = 0;
1767 	new_entry->wiring_thread = NULL;
1768 	new_entry->read_ahead = VM_FAULT_READ_AHEAD_INIT;
1769 	new_entry->next_read = start;
1770 
1771 	KASSERT(cred == NULL || !ENTRY_CHARGED(new_entry),
1772 	    ("overcommit: vm_map_insert leaks vm_map %p", new_entry));
1773 	new_entry->cred = cred;
1774 
1775 	/*
1776 	 * Insert the new entry into the list
1777 	 */
1778 	vm_map_entry_link(map, new_entry);
1779 	if ((new_entry->eflags & MAP_ENTRY_GUARD) == 0)
1780 		map->size += new_entry->end - new_entry->start;
1781 
1782 	/*
1783 	 * Try to coalesce the new entry with both the previous and next
1784 	 * entries in the list.  Previously, we only attempted to coalesce
1785 	 * with the previous entry when object is NULL.  Here, we handle the
1786 	 * other cases, which are less common.
1787 	 */
1788 	vm_map_try_merge_entries(map, prev_entry, new_entry);
1789 	vm_map_try_merge_entries(map, new_entry, next_entry);
1790 
1791 	if ((cow & (MAP_PREFAULT | MAP_PREFAULT_PARTIAL)) != 0) {
1792 		vm_map_pmap_enter(map, start, prot, object, OFF_TO_IDX(offset),
1793 		    end - start, cow & MAP_PREFAULT_PARTIAL);
1794 	}
1795 
1796 	return (KERN_SUCCESS);
1797 }
1798 
1799 /*
1800  *	vm_map_findspace:
1801  *
1802  *	Find the first fit (lowest VM address) for "length" free bytes
1803  *	beginning at address >= start in the given map.
1804  *
1805  *	In a vm_map_entry, "max_free" is the maximum amount of
1806  *	contiguous free space between an entry in its subtree and a
1807  *	neighbor of that entry.  This allows finding a free region in
1808  *	one path down the tree, so O(log n) amortized with splay
1809  *	trees.
1810  *
1811  *	The map must be locked, and leaves it so.
1812  *
1813  *	Returns: starting address if sufficient space,
1814  *		 vm_map_max(map)-length+1 if insufficient space.
1815  */
1816 vm_offset_t
1817 vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length)
1818 {
1819 	vm_map_entry_t header, llist, rlist, root, y;
1820 	vm_size_t left_length, max_free_left, max_free_right;
1821 	vm_offset_t gap_end;
1822 
1823 	/*
1824 	 * Request must fit within min/max VM address and must avoid
1825 	 * address wrap.
1826 	 */
1827 	start = MAX(start, vm_map_min(map));
1828 	if (start >= vm_map_max(map) || length > vm_map_max(map) - start)
1829 		return (vm_map_max(map) - length + 1);
1830 
1831 	/* Empty tree means wide open address space. */
1832 	if (map->root == NULL)
1833 		return (start);
1834 
1835 	/*
1836 	 * After splay_split, if start is within an entry, push it to the start
1837 	 * of the following gap.  If rlist is at the end of the gap containing
1838 	 * start, save the end of that gap in gap_end to see if the gap is big
1839 	 * enough; otherwise set gap_end to start skip gap-checking and move
1840 	 * directly to a search of the right subtree.
1841 	 */
1842 	header = &map->header;
1843 	root = vm_map_splay_split(map, start, length, &llist, &rlist);
1844 	gap_end = rlist->start;
1845 	if (root != NULL) {
1846 		start = root->end;
1847 		if (root->right != rlist)
1848 			gap_end = start;
1849 		max_free_left = vm_map_splay_merge_left(header, root, llist);
1850 		max_free_right = vm_map_splay_merge_right(header, root, rlist);
1851 	} else if (rlist != header) {
1852 		root = rlist;
1853 		rlist = root->left;
1854 		max_free_left = vm_map_splay_merge_pred(header, root, llist);
1855 		max_free_right = vm_map_splay_merge_right(header, root, rlist);
1856 	} else {
1857 		root = llist;
1858 		llist = root->right;
1859 		max_free_left = vm_map_splay_merge_left(header, root, llist);
1860 		max_free_right = vm_map_splay_merge_succ(header, root, rlist);
1861 	}
1862 	root->max_free = vm_size_max(max_free_left, max_free_right);
1863 	map->root = root;
1864 	VM_MAP_ASSERT_CONSISTENT(map);
1865 	if (length <= gap_end - start)
1866 		return (start);
1867 
1868 	/* With max_free, can immediately tell if no solution. */
1869 	if (root->right == header || length > root->right->max_free)
1870 		return (vm_map_max(map) - length + 1);
1871 
1872 	/*
1873 	 * Splay for the least large-enough gap in the right subtree.
1874 	 */
1875 	llist = rlist = header;
1876 	for (left_length = 0;;
1877 	    left_length = vm_map_entry_max_free_left(root, llist)) {
1878 		if (length <= left_length)
1879 			SPLAY_LEFT_STEP(root, y, llist, rlist,
1880 			    length <= vm_map_entry_max_free_left(y, llist));
1881 		else
1882 			SPLAY_RIGHT_STEP(root, y, llist, rlist,
1883 			    length > vm_map_entry_max_free_left(y, root));
1884 		if (root == NULL)
1885 			break;
1886 	}
1887 	root = llist;
1888 	llist = root->right;
1889 	max_free_left = vm_map_splay_merge_left(header, root, llist);
1890 	if (rlist == header) {
1891 		root->max_free = vm_size_max(max_free_left,
1892 		    vm_map_splay_merge_succ(header, root, rlist));
1893 	} else {
1894 		y = rlist;
1895 		rlist = y->left;
1896 		y->max_free = vm_size_max(
1897 		    vm_map_splay_merge_pred(root, y, root),
1898 		    vm_map_splay_merge_right(header, y, rlist));
1899 		root->max_free = vm_size_max(max_free_left, y->max_free);
1900 	}
1901 	map->root = root;
1902 	VM_MAP_ASSERT_CONSISTENT(map);
1903 	return (root->end);
1904 }
1905 
1906 int
1907 vm_map_fixed(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1908     vm_offset_t start, vm_size_t length, vm_prot_t prot,
1909     vm_prot_t max, int cow)
1910 {
1911 	vm_offset_t end;
1912 	int result;
1913 
1914 	end = start + length;
1915 	KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 ||
1916 	    object == NULL,
1917 	    ("vm_map_fixed: non-NULL backing object for stack"));
1918 	vm_map_lock(map);
1919 	VM_MAP_RANGE_CHECK(map, start, end);
1920 	if ((cow & MAP_CHECK_EXCL) == 0)
1921 		vm_map_delete(map, start, end);
1922 	if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) {
1923 		result = vm_map_stack_locked(map, start, length, sgrowsiz,
1924 		    prot, max, cow);
1925 	} else {
1926 		result = vm_map_insert(map, object, offset, start, end,
1927 		    prot, max, cow);
1928 	}
1929 	vm_map_unlock(map);
1930 	return (result);
1931 }
1932 
1933 static const int aslr_pages_rnd_64[2] = {0x1000, 0x10};
1934 static const int aslr_pages_rnd_32[2] = {0x100, 0x4};
1935 
1936 static int cluster_anon = 1;
1937 SYSCTL_INT(_vm, OID_AUTO, cluster_anon, CTLFLAG_RW,
1938     &cluster_anon, 0,
1939     "Cluster anonymous mappings: 0 = no, 1 = yes if no hint, 2 = always");
1940 
1941 static bool
1942 clustering_anon_allowed(vm_offset_t addr)
1943 {
1944 
1945 	switch (cluster_anon) {
1946 	case 0:
1947 		return (false);
1948 	case 1:
1949 		return (addr == 0);
1950 	case 2:
1951 	default:
1952 		return (true);
1953 	}
1954 }
1955 
1956 static long aslr_restarts;
1957 SYSCTL_LONG(_vm, OID_AUTO, aslr_restarts, CTLFLAG_RD,
1958     &aslr_restarts, 0,
1959     "Number of aslr failures");
1960 
1961 #define	MAP_32BIT_MAX_ADDR	((vm_offset_t)1 << 31)
1962 
1963 /*
1964  * Searches for the specified amount of free space in the given map with the
1965  * specified alignment.  Performs an address-ordered, first-fit search from
1966  * the given address "*addr", with an optional upper bound "max_addr".  If the
1967  * parameter "alignment" is zero, then the alignment is computed from the
1968  * given (object, offset) pair so as to enable the greatest possible use of
1969  * superpage mappings.  Returns KERN_SUCCESS and the address of the free space
1970  * in "*addr" if successful.  Otherwise, returns KERN_NO_SPACE.
1971  *
1972  * The map must be locked.  Initially, there must be at least "length" bytes
1973  * of free space at the given address.
1974  */
1975 static int
1976 vm_map_alignspace(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1977     vm_offset_t *addr, vm_size_t length, vm_offset_t max_addr,
1978     vm_offset_t alignment)
1979 {
1980 	vm_offset_t aligned_addr, free_addr;
1981 
1982 	VM_MAP_ASSERT_LOCKED(map);
1983 	free_addr = *addr;
1984 	KASSERT(free_addr == vm_map_findspace(map, free_addr, length),
1985 	    ("caller failed to provide space %#jx at address %p",
1986 	     (uintmax_t)length, (void *)free_addr));
1987 	for (;;) {
1988 		/*
1989 		 * At the start of every iteration, the free space at address
1990 		 * "*addr" is at least "length" bytes.
1991 		 */
1992 		if (alignment == 0)
1993 			pmap_align_superpage(object, offset, addr, length);
1994 		else if ((*addr & (alignment - 1)) != 0) {
1995 			*addr &= ~(alignment - 1);
1996 			*addr += alignment;
1997 		}
1998 		aligned_addr = *addr;
1999 		if (aligned_addr == free_addr) {
2000 			/*
2001 			 * Alignment did not change "*addr", so "*addr" must
2002 			 * still provide sufficient free space.
2003 			 */
2004 			return (KERN_SUCCESS);
2005 		}
2006 
2007 		/*
2008 		 * Test for address wrap on "*addr".  A wrapped "*addr" could
2009 		 * be a valid address, in which case vm_map_findspace() cannot
2010 		 * be relied upon to fail.
2011 		 */
2012 		if (aligned_addr < free_addr)
2013 			return (KERN_NO_SPACE);
2014 		*addr = vm_map_findspace(map, aligned_addr, length);
2015 		if (*addr + length > vm_map_max(map) ||
2016 		    (max_addr != 0 && *addr + length > max_addr))
2017 			return (KERN_NO_SPACE);
2018 		free_addr = *addr;
2019 		if (free_addr == aligned_addr) {
2020 			/*
2021 			 * If a successful call to vm_map_findspace() did not
2022 			 * change "*addr", then "*addr" must still be aligned
2023 			 * and provide sufficient free space.
2024 			 */
2025 			return (KERN_SUCCESS);
2026 		}
2027 	}
2028 }
2029 
2030 /*
2031  *	vm_map_find finds an unallocated region in the target address
2032  *	map with the given length.  The search is defined to be
2033  *	first-fit from the specified address; the region found is
2034  *	returned in the same parameter.
2035  *
2036  *	If object is non-NULL, ref count must be bumped by caller
2037  *	prior to making call to account for the new entry.
2038  */
2039 int
2040 vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
2041 	    vm_offset_t *addr,	/* IN/OUT */
2042 	    vm_size_t length, vm_offset_t max_addr, int find_space,
2043 	    vm_prot_t prot, vm_prot_t max, int cow)
2044 {
2045 	vm_offset_t alignment, curr_min_addr, min_addr;
2046 	int gap, pidx, rv, try;
2047 	bool cluster, en_aslr, update_anon;
2048 
2049 	KASSERT((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0 ||
2050 	    object == NULL,
2051 	    ("vm_map_find: non-NULL backing object for stack"));
2052 	MPASS((cow & MAP_REMAP) == 0 || (find_space == VMFS_NO_SPACE &&
2053 	    (cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) == 0));
2054 	if (find_space == VMFS_OPTIMAL_SPACE && (object == NULL ||
2055 	    (object->flags & OBJ_COLORED) == 0))
2056 		find_space = VMFS_ANY_SPACE;
2057 	if (find_space >> 8 != 0) {
2058 		KASSERT((find_space & 0xff) == 0, ("bad VMFS flags"));
2059 		alignment = (vm_offset_t)1 << (find_space >> 8);
2060 	} else
2061 		alignment = 0;
2062 	en_aslr = (map->flags & MAP_ASLR) != 0;
2063 	update_anon = cluster = clustering_anon_allowed(*addr) &&
2064 	    (map->flags & MAP_IS_SUB_MAP) == 0 && max_addr == 0 &&
2065 	    find_space != VMFS_NO_SPACE && object == NULL &&
2066 	    (cow & (MAP_INHERIT_SHARE | MAP_STACK_GROWS_UP |
2067 	    MAP_STACK_GROWS_DOWN)) == 0 && prot != PROT_NONE;
2068 	curr_min_addr = min_addr = *addr;
2069 	if (en_aslr && min_addr == 0 && !cluster &&
2070 	    find_space != VMFS_NO_SPACE &&
2071 	    (map->flags & MAP_ASLR_IGNSTART) != 0)
2072 		curr_min_addr = min_addr = vm_map_min(map);
2073 	try = 0;
2074 	vm_map_lock(map);
2075 	if (cluster) {
2076 		curr_min_addr = map->anon_loc;
2077 		if (curr_min_addr == 0)
2078 			cluster = false;
2079 	}
2080 	if (find_space != VMFS_NO_SPACE) {
2081 		KASSERT(find_space == VMFS_ANY_SPACE ||
2082 		    find_space == VMFS_OPTIMAL_SPACE ||
2083 		    find_space == VMFS_SUPER_SPACE ||
2084 		    alignment != 0, ("unexpected VMFS flag"));
2085 again:
2086 		/*
2087 		 * When creating an anonymous mapping, try clustering
2088 		 * with an existing anonymous mapping first.
2089 		 *
2090 		 * We make up to two attempts to find address space
2091 		 * for a given find_space value. The first attempt may
2092 		 * apply randomization or may cluster with an existing
2093 		 * anonymous mapping. If this first attempt fails,
2094 		 * perform a first-fit search of the available address
2095 		 * space.
2096 		 *
2097 		 * If all tries failed, and find_space is
2098 		 * VMFS_OPTIMAL_SPACE, fallback to VMFS_ANY_SPACE.
2099 		 * Again enable clustering and randomization.
2100 		 */
2101 		try++;
2102 		MPASS(try <= 2);
2103 
2104 		if (try == 2) {
2105 			/*
2106 			 * Second try: we failed either to find a
2107 			 * suitable region for randomizing the
2108 			 * allocation, or to cluster with an existing
2109 			 * mapping.  Retry with free run.
2110 			 */
2111 			curr_min_addr = (map->flags & MAP_ASLR_IGNSTART) != 0 ?
2112 			    vm_map_min(map) : min_addr;
2113 			atomic_add_long(&aslr_restarts, 1);
2114 		}
2115 
2116 		if (try == 1 && en_aslr && !cluster) {
2117 			/*
2118 			 * Find space for allocation, including
2119 			 * gap needed for later randomization.
2120 			 */
2121 			pidx = MAXPAGESIZES > 1 && pagesizes[1] != 0 &&
2122 			    (find_space == VMFS_SUPER_SPACE || find_space ==
2123 			    VMFS_OPTIMAL_SPACE) ? 1 : 0;
2124 			gap = vm_map_max(map) > MAP_32BIT_MAX_ADDR &&
2125 			    (max_addr == 0 || max_addr > MAP_32BIT_MAX_ADDR) ?
2126 			    aslr_pages_rnd_64[pidx] : aslr_pages_rnd_32[pidx];
2127 			*addr = vm_map_findspace(map, curr_min_addr,
2128 			    length + gap * pagesizes[pidx]);
2129 			if (*addr + length + gap * pagesizes[pidx] >
2130 			    vm_map_max(map))
2131 				goto again;
2132 			/* And randomize the start address. */
2133 			*addr += (arc4random() % gap) * pagesizes[pidx];
2134 			if (max_addr != 0 && *addr + length > max_addr)
2135 				goto again;
2136 		} else {
2137 			*addr = vm_map_findspace(map, curr_min_addr, length);
2138 			if (*addr + length > vm_map_max(map) ||
2139 			    (max_addr != 0 && *addr + length > max_addr)) {
2140 				if (cluster) {
2141 					cluster = false;
2142 					MPASS(try == 1);
2143 					goto again;
2144 				}
2145 				rv = KERN_NO_SPACE;
2146 				goto done;
2147 			}
2148 		}
2149 
2150 		if (find_space != VMFS_ANY_SPACE &&
2151 		    (rv = vm_map_alignspace(map, object, offset, addr, length,
2152 		    max_addr, alignment)) != KERN_SUCCESS) {
2153 			if (find_space == VMFS_OPTIMAL_SPACE) {
2154 				find_space = VMFS_ANY_SPACE;
2155 				curr_min_addr = min_addr;
2156 				cluster = update_anon;
2157 				try = 0;
2158 				goto again;
2159 			}
2160 			goto done;
2161 		}
2162 	} else if ((cow & MAP_REMAP) != 0) {
2163 		if (!vm_map_range_valid(map, *addr, *addr + length)) {
2164 			rv = KERN_INVALID_ADDRESS;
2165 			goto done;
2166 		}
2167 		vm_map_delete(map, *addr, *addr + length);
2168 	}
2169 	if ((cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP)) != 0) {
2170 		rv = vm_map_stack_locked(map, *addr, length, sgrowsiz, prot,
2171 		    max, cow);
2172 	} else {
2173 		rv = vm_map_insert(map, object, offset, *addr, *addr + length,
2174 		    prot, max, cow);
2175 	}
2176 	if (rv == KERN_SUCCESS && update_anon)
2177 		map->anon_loc = *addr + length;
2178 done:
2179 	vm_map_unlock(map);
2180 	return (rv);
2181 }
2182 
2183 /*
2184  *	vm_map_find_min() is a variant of vm_map_find() that takes an
2185  *	additional parameter (min_addr) and treats the given address
2186  *	(*addr) differently.  Specifically, it treats *addr as a hint
2187  *	and not as the minimum address where the mapping is created.
2188  *
2189  *	This function works in two phases.  First, it tries to
2190  *	allocate above the hint.  If that fails and the hint is
2191  *	greater than min_addr, it performs a second pass, replacing
2192  *	the hint with min_addr as the minimum address for the
2193  *	allocation.
2194  */
2195 int
2196 vm_map_find_min(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
2197     vm_offset_t *addr, vm_size_t length, vm_offset_t min_addr,
2198     vm_offset_t max_addr, int find_space, vm_prot_t prot, vm_prot_t max,
2199     int cow)
2200 {
2201 	vm_offset_t hint;
2202 	int rv;
2203 
2204 	hint = *addr;
2205 	for (;;) {
2206 		rv = vm_map_find(map, object, offset, addr, length, max_addr,
2207 		    find_space, prot, max, cow);
2208 		if (rv == KERN_SUCCESS || min_addr >= hint)
2209 			return (rv);
2210 		*addr = hint = min_addr;
2211 	}
2212 }
2213 
2214 /*
2215  * A map entry with any of the following flags set must not be merged with
2216  * another entry.
2217  */
2218 #define	MAP_ENTRY_NOMERGE_MASK	(MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP | \
2219 	    MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP | MAP_ENTRY_VN_EXEC)
2220 
2221 static bool
2222 vm_map_mergeable_neighbors(vm_map_entry_t prev, vm_map_entry_t entry)
2223 {
2224 
2225 	KASSERT((prev->eflags & MAP_ENTRY_NOMERGE_MASK) == 0 ||
2226 	    (entry->eflags & MAP_ENTRY_NOMERGE_MASK) == 0,
2227 	    ("vm_map_mergeable_neighbors: neither %p nor %p are mergeable",
2228 	    prev, entry));
2229 	return (prev->end == entry->start &&
2230 	    prev->object.vm_object == entry->object.vm_object &&
2231 	    (prev->object.vm_object == NULL ||
2232 	    prev->offset + (prev->end - prev->start) == entry->offset) &&
2233 	    prev->eflags == entry->eflags &&
2234 	    prev->protection == entry->protection &&
2235 	    prev->max_protection == entry->max_protection &&
2236 	    prev->inheritance == entry->inheritance &&
2237 	    prev->wired_count == entry->wired_count &&
2238 	    prev->cred == entry->cred);
2239 }
2240 
2241 static void
2242 vm_map_merged_neighbor_dispose(vm_map_t map, vm_map_entry_t entry)
2243 {
2244 
2245 	/*
2246 	 * If the backing object is a vnode object, vm_object_deallocate()
2247 	 * calls vrele().  However, vrele() does not lock the vnode because
2248 	 * the vnode has additional references.  Thus, the map lock can be
2249 	 * kept without causing a lock-order reversal with the vnode lock.
2250 	 *
2251 	 * Since we count the number of virtual page mappings in
2252 	 * object->un_pager.vnp.writemappings, the writemappings value
2253 	 * should not be adjusted when the entry is disposed of.
2254 	 */
2255 	if (entry->object.vm_object != NULL)
2256 		vm_object_deallocate(entry->object.vm_object);
2257 	if (entry->cred != NULL)
2258 		crfree(entry->cred);
2259 	vm_map_entry_dispose(map, entry);
2260 }
2261 
2262 /*
2263  *	vm_map_try_merge_entries:
2264  *
2265  *	Compare the given map entry to its predecessor, and merge its precessor
2266  *	into it if possible.  The entry remains valid, and may be extended.
2267  *	The predecessor may be deleted.
2268  *
2269  *	The map must be locked.
2270  */
2271 void
2272 vm_map_try_merge_entries(vm_map_t map, vm_map_entry_t prev_entry,
2273     vm_map_entry_t entry)
2274 {
2275 
2276 	VM_MAP_ASSERT_LOCKED(map);
2277 	if ((entry->eflags & MAP_ENTRY_NOMERGE_MASK) == 0 &&
2278 	    vm_map_mergeable_neighbors(prev_entry, entry)) {
2279 		vm_map_entry_unlink(map, prev_entry, UNLINK_MERGE_NEXT);
2280 		vm_map_merged_neighbor_dispose(map, prev_entry);
2281 	}
2282 }
2283 
2284 /*
2285  *	vm_map_entry_back:
2286  *
2287  *	Allocate an object to back a map entry.
2288  */
2289 static inline void
2290 vm_map_entry_back(vm_map_entry_t entry)
2291 {
2292 	vm_object_t object;
2293 
2294 	KASSERT(entry->object.vm_object == NULL,
2295 	    ("map entry %p has backing object", entry));
2296 	KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
2297 	    ("map entry %p is a submap", entry));
2298 	object = vm_object_allocate_anon(atop(entry->end - entry->start), NULL,
2299 	    entry->cred, entry->end - entry->start);
2300 	entry->object.vm_object = object;
2301 	entry->offset = 0;
2302 	entry->cred = NULL;
2303 }
2304 
2305 /*
2306  *	vm_map_entry_charge_object
2307  *
2308  *	If there is no object backing this entry, create one.  Otherwise, if
2309  *	the entry has cred, give it to the backing object.
2310  */
2311 static inline void
2312 vm_map_entry_charge_object(vm_map_t map, vm_map_entry_t entry)
2313 {
2314 
2315 	VM_MAP_ASSERT_LOCKED(map);
2316 	KASSERT((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0,
2317 	    ("map entry %p is a submap", entry));
2318 	if (entry->object.vm_object == NULL && !map->system_map &&
2319 	    (entry->eflags & MAP_ENTRY_GUARD) == 0)
2320 		vm_map_entry_back(entry);
2321 	else if (entry->object.vm_object != NULL &&
2322 	    ((entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) &&
2323 	    entry->cred != NULL) {
2324 		VM_OBJECT_WLOCK(entry->object.vm_object);
2325 		KASSERT(entry->object.vm_object->cred == NULL,
2326 		    ("OVERCOMMIT: %s: both cred e %p", __func__, entry));
2327 		entry->object.vm_object->cred = entry->cred;
2328 		entry->object.vm_object->charge = entry->end - entry->start;
2329 		VM_OBJECT_WUNLOCK(entry->object.vm_object);
2330 		entry->cred = NULL;
2331 	}
2332 }
2333 
2334 /*
2335  *	vm_map_entry_clone
2336  *
2337  *	Create a duplicate map entry for clipping.
2338  */
2339 static vm_map_entry_t
2340 vm_map_entry_clone(vm_map_t map, vm_map_entry_t entry)
2341 {
2342 	vm_map_entry_t new_entry;
2343 
2344 	VM_MAP_ASSERT_LOCKED(map);
2345 
2346 	/*
2347 	 * Create a backing object now, if none exists, so that more individual
2348 	 * objects won't be created after the map entry is split.
2349 	 */
2350 	vm_map_entry_charge_object(map, entry);
2351 
2352 	/* Clone the entry. */
2353 	new_entry = vm_map_entry_create(map);
2354 	*new_entry = *entry;
2355 	if (new_entry->cred != NULL)
2356 		crhold(entry->cred);
2357 	if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
2358 		vm_object_reference(new_entry->object.vm_object);
2359 		vm_map_entry_set_vnode_text(new_entry, true);
2360 		/*
2361 		 * The object->un_pager.vnp.writemappings for the object of
2362 		 * MAP_ENTRY_WRITECNT type entry shall be kept as is here.  The
2363 		 * virtual pages are re-distributed among the clipped entries,
2364 		 * so the sum is left the same.
2365 		 */
2366 	}
2367 	return (new_entry);
2368 }
2369 
2370 /*
2371  *	vm_map_clip_start:	[ internal use only ]
2372  *
2373  *	Asserts that the given entry begins at or after
2374  *	the specified address; if necessary,
2375  *	it splits the entry into two.
2376  */
2377 static inline void
2378 vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start)
2379 {
2380 	vm_map_entry_t new_entry;
2381 
2382 	if (start <= entry->start)
2383 		return;
2384 
2385 	VM_MAP_ASSERT_LOCKED(map);
2386 	KASSERT(entry->end > start && entry->start < start,
2387 	    ("%s: invalid clip of entry %p", __func__, entry));
2388 
2389 	new_entry = vm_map_entry_clone(map, entry);
2390 
2391 	/*
2392 	 * Split off the front portion.  Insert the new entry BEFORE this one,
2393 	 * so that this entry has the specified starting address.
2394 	 */
2395 	new_entry->end = start;
2396 	vm_map_entry_link(map, new_entry);
2397 }
2398 
2399 /*
2400  *	vm_map_lookup_clip_start:
2401  *
2402  *	Find the entry at or just after 'start', and clip it if 'start' is in
2403  *	the interior of the entry.  Return entry after 'start', and in
2404  *	prev_entry set the entry before 'start'.
2405  */
2406 static inline vm_map_entry_t
2407 vm_map_lookup_clip_start(vm_map_t map, vm_offset_t start,
2408     vm_map_entry_t *prev_entry)
2409 {
2410 	vm_map_entry_t entry;
2411 
2412 	if (vm_map_lookup_entry(map, start, prev_entry)) {
2413 		entry = *prev_entry;
2414 		vm_map_clip_start(map, entry, start);
2415 		*prev_entry = vm_map_entry_pred(entry);
2416 	} else
2417 		entry = vm_map_entry_succ(*prev_entry);
2418 	return (entry);
2419 }
2420 
2421 /*
2422  *	vm_map_clip_end:	[ internal use only ]
2423  *
2424  *	Asserts that the given entry ends at or before
2425  *	the specified address; if necessary,
2426  *	it splits the entry into two.
2427  */
2428 static inline void
2429 vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end)
2430 {
2431 	vm_map_entry_t new_entry;
2432 
2433 	if (end >= entry->end)
2434 		return;
2435 
2436 	VM_MAP_ASSERT_LOCKED(map);
2437 	KASSERT(entry->start < end && entry->end > end,
2438 	    ("%s: invalid clip of entry %p", __func__, entry));
2439 
2440 	new_entry = vm_map_entry_clone(map, entry);
2441 
2442 	/*
2443 	 * Split off the back portion.  Insert the new entry AFTER this one,
2444 	 * so that this entry has the specified ending address.
2445 	 */
2446 	new_entry->start = end;
2447 	vm_map_entry_link(map, new_entry);
2448 }
2449 
2450 /*
2451  *	vm_map_submap:		[ kernel use only ]
2452  *
2453  *	Mark the given range as handled by a subordinate map.
2454  *
2455  *	This range must have been created with vm_map_find,
2456  *	and no other operations may have been performed on this
2457  *	range prior to calling vm_map_submap.
2458  *
2459  *	Only a limited number of operations can be performed
2460  *	within this rage after calling vm_map_submap:
2461  *		vm_fault
2462  *	[Don't try vm_map_copy!]
2463  *
2464  *	To remove a submapping, one must first remove the
2465  *	range from the superior map, and then destroy the
2466  *	submap (if desired).  [Better yet, don't try it.]
2467  */
2468 int
2469 vm_map_submap(
2470 	vm_map_t map,
2471 	vm_offset_t start,
2472 	vm_offset_t end,
2473 	vm_map_t submap)
2474 {
2475 	vm_map_entry_t entry;
2476 	int result;
2477 
2478 	result = KERN_INVALID_ARGUMENT;
2479 
2480 	vm_map_lock(submap);
2481 	submap->flags |= MAP_IS_SUB_MAP;
2482 	vm_map_unlock(submap);
2483 
2484 	vm_map_lock(map);
2485 	VM_MAP_RANGE_CHECK(map, start, end);
2486 	if (vm_map_lookup_entry(map, start, &entry) && entry->end >= end &&
2487 	    (entry->eflags & MAP_ENTRY_COW) == 0 &&
2488 	    entry->object.vm_object == NULL) {
2489 		vm_map_clip_start(map, entry, start);
2490 		vm_map_clip_end(map, entry, end);
2491 		entry->object.sub_map = submap;
2492 		entry->eflags |= MAP_ENTRY_IS_SUB_MAP;
2493 		result = KERN_SUCCESS;
2494 	}
2495 	vm_map_unlock(map);
2496 
2497 	if (result != KERN_SUCCESS) {
2498 		vm_map_lock(submap);
2499 		submap->flags &= ~MAP_IS_SUB_MAP;
2500 		vm_map_unlock(submap);
2501 	}
2502 	return (result);
2503 }
2504 
2505 /*
2506  * The maximum number of pages to map if MAP_PREFAULT_PARTIAL is specified
2507  */
2508 #define	MAX_INIT_PT	96
2509 
2510 /*
2511  *	vm_map_pmap_enter:
2512  *
2513  *	Preload the specified map's pmap with mappings to the specified
2514  *	object's memory-resident pages.  No further physical pages are
2515  *	allocated, and no further virtual pages are retrieved from secondary
2516  *	storage.  If the specified flags include MAP_PREFAULT_PARTIAL, then a
2517  *	limited number of page mappings are created at the low-end of the
2518  *	specified address range.  (For this purpose, a superpage mapping
2519  *	counts as one page mapping.)  Otherwise, all resident pages within
2520  *	the specified address range are mapped.
2521  */
2522 static void
2523 vm_map_pmap_enter(vm_map_t map, vm_offset_t addr, vm_prot_t prot,
2524     vm_object_t object, vm_pindex_t pindex, vm_size_t size, int flags)
2525 {
2526 	vm_offset_t start;
2527 	vm_page_t p, p_start;
2528 	vm_pindex_t mask, psize, threshold, tmpidx;
2529 
2530 	if ((prot & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0 || object == NULL)
2531 		return;
2532 	if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
2533 		VM_OBJECT_WLOCK(object);
2534 		if (object->type == OBJT_DEVICE || object->type == OBJT_SG) {
2535 			pmap_object_init_pt(map->pmap, addr, object, pindex,
2536 			    size);
2537 			VM_OBJECT_WUNLOCK(object);
2538 			return;
2539 		}
2540 		VM_OBJECT_LOCK_DOWNGRADE(object);
2541 	} else
2542 		VM_OBJECT_RLOCK(object);
2543 
2544 	psize = atop(size);
2545 	if (psize + pindex > object->size) {
2546 		if (pindex >= object->size) {
2547 			VM_OBJECT_RUNLOCK(object);
2548 			return;
2549 		}
2550 		psize = object->size - pindex;
2551 	}
2552 
2553 	start = 0;
2554 	p_start = NULL;
2555 	threshold = MAX_INIT_PT;
2556 
2557 	p = vm_page_find_least(object, pindex);
2558 	/*
2559 	 * Assert: the variable p is either (1) the page with the
2560 	 * least pindex greater than or equal to the parameter pindex
2561 	 * or (2) NULL.
2562 	 */
2563 	for (;
2564 	     p != NULL && (tmpidx = p->pindex - pindex) < psize;
2565 	     p = TAILQ_NEXT(p, listq)) {
2566 		/*
2567 		 * don't allow an madvise to blow away our really
2568 		 * free pages allocating pv entries.
2569 		 */
2570 		if (((flags & MAP_PREFAULT_MADVISE) != 0 &&
2571 		    vm_page_count_severe()) ||
2572 		    ((flags & MAP_PREFAULT_PARTIAL) != 0 &&
2573 		    tmpidx >= threshold)) {
2574 			psize = tmpidx;
2575 			break;
2576 		}
2577 		if (vm_page_all_valid(p)) {
2578 			if (p_start == NULL) {
2579 				start = addr + ptoa(tmpidx);
2580 				p_start = p;
2581 			}
2582 			/* Jump ahead if a superpage mapping is possible. */
2583 			if (p->psind > 0 && ((addr + ptoa(tmpidx)) &
2584 			    (pagesizes[p->psind] - 1)) == 0) {
2585 				mask = atop(pagesizes[p->psind]) - 1;
2586 				if (tmpidx + mask < psize &&
2587 				    vm_page_ps_test(p, PS_ALL_VALID, NULL)) {
2588 					p += mask;
2589 					threshold += mask;
2590 				}
2591 			}
2592 		} else if (p_start != NULL) {
2593 			pmap_enter_object(map->pmap, start, addr +
2594 			    ptoa(tmpidx), p_start, prot);
2595 			p_start = NULL;
2596 		}
2597 	}
2598 	if (p_start != NULL)
2599 		pmap_enter_object(map->pmap, start, addr + ptoa(psize),
2600 		    p_start, prot);
2601 	VM_OBJECT_RUNLOCK(object);
2602 }
2603 
2604 /*
2605  *	vm_map_protect:
2606  *
2607  *	Sets the protection of the specified address
2608  *	region in the target map.  If "set_max" is
2609  *	specified, the maximum protection is to be set;
2610  *	otherwise, only the current protection is affected.
2611  */
2612 int
2613 vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end,
2614 	       vm_prot_t new_prot, boolean_t set_max)
2615 {
2616 	vm_map_entry_t entry, first_entry, in_tran, prev_entry;
2617 	vm_object_t obj;
2618 	struct ucred *cred;
2619 	vm_prot_t old_prot;
2620 	int rv;
2621 
2622 	if (start == end)
2623 		return (KERN_SUCCESS);
2624 
2625 again:
2626 	in_tran = NULL;
2627 	vm_map_lock(map);
2628 
2629 	/*
2630 	 * Ensure that we are not concurrently wiring pages.  vm_map_wire() may
2631 	 * need to fault pages into the map and will drop the map lock while
2632 	 * doing so, and the VM object may end up in an inconsistent state if we
2633 	 * update the protection on the map entry in between faults.
2634 	 */
2635 	vm_map_wait_busy(map);
2636 
2637 	VM_MAP_RANGE_CHECK(map, start, end);
2638 
2639 	if (!vm_map_lookup_entry(map, start, &first_entry))
2640 		first_entry = vm_map_entry_succ(first_entry);
2641 
2642 	/*
2643 	 * Make a first pass to check for protection violations.
2644 	 */
2645 	for (entry = first_entry; entry->start < end;
2646 	    entry = vm_map_entry_succ(entry)) {
2647 		if ((entry->eflags & MAP_ENTRY_GUARD) != 0)
2648 			continue;
2649 		if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0) {
2650 			vm_map_unlock(map);
2651 			return (KERN_INVALID_ARGUMENT);
2652 		}
2653 		if ((new_prot & entry->max_protection) != new_prot) {
2654 			vm_map_unlock(map);
2655 			return (KERN_PROTECTION_FAILURE);
2656 		}
2657 		if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0)
2658 			in_tran = entry;
2659 	}
2660 
2661 	/*
2662 	 * Postpone the operation until all in-transition map entries have
2663 	 * stabilized.  An in-transition entry might already have its pages
2664 	 * wired and wired_count incremented, but not yet have its
2665 	 * MAP_ENTRY_USER_WIRED flag set.  In which case, we would fail to call
2666 	 * vm_fault_copy_entry() in the final loop below.
2667 	 */
2668 	if (in_tran != NULL) {
2669 		in_tran->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2670 		vm_map_unlock_and_wait(map, 0);
2671 		goto again;
2672 	}
2673 
2674 	/*
2675 	 * Before changing the protections, try to reserve swap space for any
2676 	 * private (i.e., copy-on-write) mappings that are transitioning from
2677 	 * read-only to read/write access.  If a reservation fails, break out
2678 	 * of this loop early and let the next loop simplify the entries, since
2679 	 * some may now be mergeable.
2680 	 */
2681 	rv = KERN_SUCCESS;
2682 	vm_map_clip_start(map, first_entry, start);
2683 	for (entry = first_entry; entry->start < end;
2684 	    entry = vm_map_entry_succ(entry)) {
2685 		vm_map_clip_end(map, entry, end);
2686 
2687 		if (set_max ||
2688 		    ((new_prot & ~entry->protection) & VM_PROT_WRITE) == 0 ||
2689 		    ENTRY_CHARGED(entry) ||
2690 		    (entry->eflags & MAP_ENTRY_GUARD) != 0) {
2691 			continue;
2692 		}
2693 
2694 		cred = curthread->td_ucred;
2695 		obj = entry->object.vm_object;
2696 
2697 		if (obj == NULL ||
2698 		    (entry->eflags & MAP_ENTRY_NEEDS_COPY) != 0) {
2699 			if (!swap_reserve(entry->end - entry->start)) {
2700 				rv = KERN_RESOURCE_SHORTAGE;
2701 				end = entry->end;
2702 				break;
2703 			}
2704 			crhold(cred);
2705 			entry->cred = cred;
2706 			continue;
2707 		}
2708 
2709 		if (obj->type != OBJT_DEFAULT && obj->type != OBJT_SWAP)
2710 			continue;
2711 		VM_OBJECT_WLOCK(obj);
2712 		if (obj->type != OBJT_DEFAULT && obj->type != OBJT_SWAP) {
2713 			VM_OBJECT_WUNLOCK(obj);
2714 			continue;
2715 		}
2716 
2717 		/*
2718 		 * Charge for the whole object allocation now, since
2719 		 * we cannot distinguish between non-charged and
2720 		 * charged clipped mapping of the same object later.
2721 		 */
2722 		KASSERT(obj->charge == 0,
2723 		    ("vm_map_protect: object %p overcharged (entry %p)",
2724 		    obj, entry));
2725 		if (!swap_reserve(ptoa(obj->size))) {
2726 			VM_OBJECT_WUNLOCK(obj);
2727 			rv = KERN_RESOURCE_SHORTAGE;
2728 			end = entry->end;
2729 			break;
2730 		}
2731 
2732 		crhold(cred);
2733 		obj->cred = cred;
2734 		obj->charge = ptoa(obj->size);
2735 		VM_OBJECT_WUNLOCK(obj);
2736 	}
2737 
2738 	/*
2739 	 * If enough swap space was available, go back and fix up protections.
2740 	 * Otherwise, just simplify entries, since some may have been modified.
2741 	 * [Note that clipping is not necessary the second time.]
2742 	 */
2743 	for (prev_entry = vm_map_entry_pred(first_entry), entry = first_entry;
2744 	    entry->start < end;
2745 	    vm_map_try_merge_entries(map, prev_entry, entry),
2746 	    prev_entry = entry, entry = vm_map_entry_succ(entry)) {
2747 		if (rv != KERN_SUCCESS ||
2748 		    (entry->eflags & MAP_ENTRY_GUARD) != 0)
2749 			continue;
2750 
2751 		old_prot = entry->protection;
2752 
2753 		if (set_max)
2754 			entry->protection =
2755 			    (entry->max_protection = new_prot) &
2756 			    old_prot;
2757 		else
2758 			entry->protection = new_prot;
2759 
2760 		/*
2761 		 * For user wired map entries, the normal lazy evaluation of
2762 		 * write access upgrades through soft page faults is
2763 		 * undesirable.  Instead, immediately copy any pages that are
2764 		 * copy-on-write and enable write access in the physical map.
2765 		 */
2766 		if ((entry->eflags & MAP_ENTRY_USER_WIRED) != 0 &&
2767 		    (entry->protection & VM_PROT_WRITE) != 0 &&
2768 		    (old_prot & VM_PROT_WRITE) == 0)
2769 			vm_fault_copy_entry(map, map, entry, entry, NULL);
2770 
2771 		/*
2772 		 * When restricting access, update the physical map.  Worry
2773 		 * about copy-on-write here.
2774 		 */
2775 		if ((old_prot & ~entry->protection) != 0) {
2776 #define MASK(entry)	(((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \
2777 							VM_PROT_ALL)
2778 			pmap_protect(map->pmap, entry->start,
2779 			    entry->end,
2780 			    entry->protection & MASK(entry));
2781 #undef	MASK
2782 		}
2783 	}
2784 	vm_map_try_merge_entries(map, prev_entry, entry);
2785 	vm_map_unlock(map);
2786 	return (rv);
2787 }
2788 
2789 /*
2790  *	vm_map_madvise:
2791  *
2792  *	This routine traverses a processes map handling the madvise
2793  *	system call.  Advisories are classified as either those effecting
2794  *	the vm_map_entry structure, or those effecting the underlying
2795  *	objects.
2796  */
2797 int
2798 vm_map_madvise(
2799 	vm_map_t map,
2800 	vm_offset_t start,
2801 	vm_offset_t end,
2802 	int behav)
2803 {
2804 	vm_map_entry_t entry, prev_entry;
2805 	bool modify_map;
2806 
2807 	/*
2808 	 * Some madvise calls directly modify the vm_map_entry, in which case
2809 	 * we need to use an exclusive lock on the map and we need to perform
2810 	 * various clipping operations.  Otherwise we only need a read-lock
2811 	 * on the map.
2812 	 */
2813 	switch(behav) {
2814 	case MADV_NORMAL:
2815 	case MADV_SEQUENTIAL:
2816 	case MADV_RANDOM:
2817 	case MADV_NOSYNC:
2818 	case MADV_AUTOSYNC:
2819 	case MADV_NOCORE:
2820 	case MADV_CORE:
2821 		if (start == end)
2822 			return (0);
2823 		modify_map = true;
2824 		vm_map_lock(map);
2825 		break;
2826 	case MADV_WILLNEED:
2827 	case MADV_DONTNEED:
2828 	case MADV_FREE:
2829 		if (start == end)
2830 			return (0);
2831 		modify_map = false;
2832 		vm_map_lock_read(map);
2833 		break;
2834 	default:
2835 		return (EINVAL);
2836 	}
2837 
2838 	/*
2839 	 * Locate starting entry and clip if necessary.
2840 	 */
2841 	VM_MAP_RANGE_CHECK(map, start, end);
2842 
2843 	if (modify_map) {
2844 		/*
2845 		 * madvise behaviors that are implemented in the vm_map_entry.
2846 		 *
2847 		 * We clip the vm_map_entry so that behavioral changes are
2848 		 * limited to the specified address range.
2849 		 */
2850 		for (entry = vm_map_lookup_clip_start(map, start, &prev_entry);
2851 		    entry->start < end;
2852 		    prev_entry = entry, entry = vm_map_entry_succ(entry)) {
2853 			if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0)
2854 				continue;
2855 
2856 			vm_map_clip_end(map, entry, end);
2857 
2858 			switch (behav) {
2859 			case MADV_NORMAL:
2860 				vm_map_entry_set_behavior(entry,
2861 				    MAP_ENTRY_BEHAV_NORMAL);
2862 				break;
2863 			case MADV_SEQUENTIAL:
2864 				vm_map_entry_set_behavior(entry,
2865 				    MAP_ENTRY_BEHAV_SEQUENTIAL);
2866 				break;
2867 			case MADV_RANDOM:
2868 				vm_map_entry_set_behavior(entry,
2869 				    MAP_ENTRY_BEHAV_RANDOM);
2870 				break;
2871 			case MADV_NOSYNC:
2872 				entry->eflags |= MAP_ENTRY_NOSYNC;
2873 				break;
2874 			case MADV_AUTOSYNC:
2875 				entry->eflags &= ~MAP_ENTRY_NOSYNC;
2876 				break;
2877 			case MADV_NOCORE:
2878 				entry->eflags |= MAP_ENTRY_NOCOREDUMP;
2879 				break;
2880 			case MADV_CORE:
2881 				entry->eflags &= ~MAP_ENTRY_NOCOREDUMP;
2882 				break;
2883 			default:
2884 				break;
2885 			}
2886 			vm_map_try_merge_entries(map, prev_entry, entry);
2887 		}
2888 		vm_map_try_merge_entries(map, prev_entry, entry);
2889 		vm_map_unlock(map);
2890 	} else {
2891 		vm_pindex_t pstart, pend;
2892 
2893 		/*
2894 		 * madvise behaviors that are implemented in the underlying
2895 		 * vm_object.
2896 		 *
2897 		 * Since we don't clip the vm_map_entry, we have to clip
2898 		 * the vm_object pindex and count.
2899 		 */
2900 		if (!vm_map_lookup_entry(map, start, &entry))
2901 			entry = vm_map_entry_succ(entry);
2902 		for (; entry->start < end;
2903 		    entry = vm_map_entry_succ(entry)) {
2904 			vm_offset_t useEnd, useStart;
2905 
2906 			if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0)
2907 				continue;
2908 
2909 			/*
2910 			 * MADV_FREE would otherwise rewind time to
2911 			 * the creation of the shadow object.  Because
2912 			 * we hold the VM map read-locked, neither the
2913 			 * entry's object nor the presence of a
2914 			 * backing object can change.
2915 			 */
2916 			if (behav == MADV_FREE &&
2917 			    entry->object.vm_object != NULL &&
2918 			    entry->object.vm_object->backing_object != NULL)
2919 				continue;
2920 
2921 			pstart = OFF_TO_IDX(entry->offset);
2922 			pend = pstart + atop(entry->end - entry->start);
2923 			useStart = entry->start;
2924 			useEnd = entry->end;
2925 
2926 			if (entry->start < start) {
2927 				pstart += atop(start - entry->start);
2928 				useStart = start;
2929 			}
2930 			if (entry->end > end) {
2931 				pend -= atop(entry->end - end);
2932 				useEnd = end;
2933 			}
2934 
2935 			if (pstart >= pend)
2936 				continue;
2937 
2938 			/*
2939 			 * Perform the pmap_advise() before clearing
2940 			 * PGA_REFERENCED in vm_page_advise().  Otherwise, a
2941 			 * concurrent pmap operation, such as pmap_remove(),
2942 			 * could clear a reference in the pmap and set
2943 			 * PGA_REFERENCED on the page before the pmap_advise()
2944 			 * had completed.  Consequently, the page would appear
2945 			 * referenced based upon an old reference that
2946 			 * occurred before this pmap_advise() ran.
2947 			 */
2948 			if (behav == MADV_DONTNEED || behav == MADV_FREE)
2949 				pmap_advise(map->pmap, useStart, useEnd,
2950 				    behav);
2951 
2952 			vm_object_madvise(entry->object.vm_object, pstart,
2953 			    pend, behav);
2954 
2955 			/*
2956 			 * Pre-populate paging structures in the
2957 			 * WILLNEED case.  For wired entries, the
2958 			 * paging structures are already populated.
2959 			 */
2960 			if (behav == MADV_WILLNEED &&
2961 			    entry->wired_count == 0) {
2962 				vm_map_pmap_enter(map,
2963 				    useStart,
2964 				    entry->protection,
2965 				    entry->object.vm_object,
2966 				    pstart,
2967 				    ptoa(pend - pstart),
2968 				    MAP_PREFAULT_MADVISE
2969 				);
2970 			}
2971 		}
2972 		vm_map_unlock_read(map);
2973 	}
2974 	return (0);
2975 }
2976 
2977 
2978 /*
2979  *	vm_map_inherit:
2980  *
2981  *	Sets the inheritance of the specified address
2982  *	range in the target map.  Inheritance
2983  *	affects how the map will be shared with
2984  *	child maps at the time of vmspace_fork.
2985  */
2986 int
2987 vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end,
2988 	       vm_inherit_t new_inheritance)
2989 {
2990 	vm_map_entry_t entry, prev_entry;
2991 
2992 	switch (new_inheritance) {
2993 	case VM_INHERIT_NONE:
2994 	case VM_INHERIT_COPY:
2995 	case VM_INHERIT_SHARE:
2996 	case VM_INHERIT_ZERO:
2997 		break;
2998 	default:
2999 		return (KERN_INVALID_ARGUMENT);
3000 	}
3001 	if (start == end)
3002 		return (KERN_SUCCESS);
3003 	vm_map_lock(map);
3004 	VM_MAP_RANGE_CHECK(map, start, end);
3005 	for (entry = vm_map_lookup_clip_start(map, start, &prev_entry);
3006 	    entry->start < end;
3007 	    prev_entry = entry, entry = vm_map_entry_succ(entry)) {
3008 		vm_map_clip_end(map, entry, end);
3009 		if ((entry->eflags & MAP_ENTRY_GUARD) == 0 ||
3010 		    new_inheritance != VM_INHERIT_ZERO)
3011 			entry->inheritance = new_inheritance;
3012 		vm_map_try_merge_entries(map, prev_entry, entry);
3013 	}
3014 	vm_map_try_merge_entries(map, prev_entry, entry);
3015 	vm_map_unlock(map);
3016 	return (KERN_SUCCESS);
3017 }
3018 
3019 /*
3020  *	vm_map_entry_in_transition:
3021  *
3022  *	Release the map lock, and sleep until the entry is no longer in
3023  *	transition.  Awake and acquire the map lock.  If the map changed while
3024  *	another held the lock, lookup a possibly-changed entry at or after the
3025  *	'start' position of the old entry.
3026  */
3027 static vm_map_entry_t
3028 vm_map_entry_in_transition(vm_map_t map, vm_offset_t in_start,
3029     vm_offset_t *io_end, bool holes_ok, vm_map_entry_t in_entry)
3030 {
3031 	vm_map_entry_t entry;
3032 	vm_offset_t start;
3033 	u_int last_timestamp;
3034 
3035 	VM_MAP_ASSERT_LOCKED(map);
3036 	KASSERT((in_entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
3037 	    ("not in-tranition map entry %p", in_entry));
3038 	/*
3039 	 * We have not yet clipped the entry.
3040 	 */
3041 	start = MAX(in_start, in_entry->start);
3042 	in_entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
3043 	last_timestamp = map->timestamp;
3044 	if (vm_map_unlock_and_wait(map, 0)) {
3045 		/*
3046 		 * Allow interruption of user wiring/unwiring?
3047 		 */
3048 	}
3049 	vm_map_lock(map);
3050 	if (last_timestamp + 1 == map->timestamp)
3051 		return (in_entry);
3052 
3053 	/*
3054 	 * Look again for the entry because the map was modified while it was
3055 	 * unlocked.  Specifically, the entry may have been clipped, merged, or
3056 	 * deleted.
3057 	 */
3058 	if (!vm_map_lookup_entry(map, start, &entry)) {
3059 		if (!holes_ok) {
3060 			*io_end = start;
3061 			return (NULL);
3062 		}
3063 		entry = vm_map_entry_succ(entry);
3064 	}
3065 	return (entry);
3066 }
3067 
3068 /*
3069  *	vm_map_unwire:
3070  *
3071  *	Implements both kernel and user unwiring.
3072  */
3073 int
3074 vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end,
3075     int flags)
3076 {
3077 	vm_map_entry_t entry, first_entry, next_entry, prev_entry;
3078 	int rv;
3079 	bool holes_ok, need_wakeup, user_unwire;
3080 
3081 	if (start == end)
3082 		return (KERN_SUCCESS);
3083 	holes_ok = (flags & VM_MAP_WIRE_HOLESOK) != 0;
3084 	user_unwire = (flags & VM_MAP_WIRE_USER) != 0;
3085 	vm_map_lock(map);
3086 	VM_MAP_RANGE_CHECK(map, start, end);
3087 	if (!vm_map_lookup_entry(map, start, &first_entry)) {
3088 		if (holes_ok)
3089 			first_entry = vm_map_entry_succ(first_entry);
3090 		else {
3091 			vm_map_unlock(map);
3092 			return (KERN_INVALID_ADDRESS);
3093 		}
3094 	}
3095 	rv = KERN_SUCCESS;
3096 	for (entry = first_entry; entry->start < end; entry = next_entry) {
3097 		if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
3098 			/*
3099 			 * We have not yet clipped the entry.
3100 			 */
3101 			next_entry = vm_map_entry_in_transition(map, start,
3102 			    &end, holes_ok, entry);
3103 			if (next_entry == NULL) {
3104 				if (entry == first_entry) {
3105 					vm_map_unlock(map);
3106 					return (KERN_INVALID_ADDRESS);
3107 				}
3108 				rv = KERN_INVALID_ADDRESS;
3109 				break;
3110 			}
3111 			first_entry = (entry == first_entry) ?
3112 			    next_entry : NULL;
3113 			continue;
3114 		}
3115 		vm_map_clip_start(map, entry, start);
3116 		vm_map_clip_end(map, entry, end);
3117 		/*
3118 		 * Mark the entry in case the map lock is released.  (See
3119 		 * above.)
3120 		 */
3121 		KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 &&
3122 		    entry->wiring_thread == NULL,
3123 		    ("owned map entry %p", entry));
3124 		entry->eflags |= MAP_ENTRY_IN_TRANSITION;
3125 		entry->wiring_thread = curthread;
3126 		next_entry = vm_map_entry_succ(entry);
3127 		/*
3128 		 * Check the map for holes in the specified region.
3129 		 * If holes_ok, skip this check.
3130 		 */
3131 		if (!holes_ok &&
3132 		    entry->end < end && next_entry->start > entry->end) {
3133 			end = entry->end;
3134 			rv = KERN_INVALID_ADDRESS;
3135 			break;
3136 		}
3137 		/*
3138 		 * If system unwiring, require that the entry is system wired.
3139 		 */
3140 		if (!user_unwire &&
3141 		    vm_map_entry_system_wired_count(entry) == 0) {
3142 			end = entry->end;
3143 			rv = KERN_INVALID_ARGUMENT;
3144 			break;
3145 		}
3146 	}
3147 	need_wakeup = false;
3148 	if (first_entry == NULL &&
3149 	    !vm_map_lookup_entry(map, start, &first_entry)) {
3150 		KASSERT(holes_ok, ("vm_map_unwire: lookup failed"));
3151 		prev_entry = first_entry;
3152 		entry = vm_map_entry_succ(first_entry);
3153 	} else {
3154 		prev_entry = vm_map_entry_pred(first_entry);
3155 		entry = first_entry;
3156 	}
3157 	for (; entry->start < end;
3158 	    prev_entry = entry, entry = vm_map_entry_succ(entry)) {
3159 		/*
3160 		 * If holes_ok was specified, an empty
3161 		 * space in the unwired region could have been mapped
3162 		 * while the map lock was dropped for draining
3163 		 * MAP_ENTRY_IN_TRANSITION.  Moreover, another thread
3164 		 * could be simultaneously wiring this new mapping
3165 		 * entry.  Detect these cases and skip any entries
3166 		 * marked as in transition by us.
3167 		 */
3168 		if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
3169 		    entry->wiring_thread != curthread) {
3170 			KASSERT(holes_ok,
3171 			    ("vm_map_unwire: !HOLESOK and new/changed entry"));
3172 			continue;
3173 		}
3174 
3175 		if (rv == KERN_SUCCESS && (!user_unwire ||
3176 		    (entry->eflags & MAP_ENTRY_USER_WIRED))) {
3177 			if (entry->wired_count == 1)
3178 				vm_map_entry_unwire(map, entry);
3179 			else
3180 				entry->wired_count--;
3181 			if (user_unwire)
3182 				entry->eflags &= ~MAP_ENTRY_USER_WIRED;
3183 		}
3184 		KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
3185 		    ("vm_map_unwire: in-transition flag missing %p", entry));
3186 		KASSERT(entry->wiring_thread == curthread,
3187 		    ("vm_map_unwire: alien wire %p", entry));
3188 		entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
3189 		entry->wiring_thread = NULL;
3190 		if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
3191 			entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
3192 			need_wakeup = true;
3193 		}
3194 		vm_map_try_merge_entries(map, prev_entry, entry);
3195 	}
3196 	vm_map_try_merge_entries(map, prev_entry, entry);
3197 	vm_map_unlock(map);
3198 	if (need_wakeup)
3199 		vm_map_wakeup(map);
3200 	return (rv);
3201 }
3202 
3203 static void
3204 vm_map_wire_user_count_sub(u_long npages)
3205 {
3206 
3207 	atomic_subtract_long(&vm_user_wire_count, npages);
3208 }
3209 
3210 static bool
3211 vm_map_wire_user_count_add(u_long npages)
3212 {
3213 	u_long wired;
3214 
3215 	wired = vm_user_wire_count;
3216 	do {
3217 		if (npages + wired > vm_page_max_user_wired)
3218 			return (false);
3219 	} while (!atomic_fcmpset_long(&vm_user_wire_count, &wired,
3220 	    npages + wired));
3221 
3222 	return (true);
3223 }
3224 
3225 /*
3226  *	vm_map_wire_entry_failure:
3227  *
3228  *	Handle a wiring failure on the given entry.
3229  *
3230  *	The map should be locked.
3231  */
3232 static void
3233 vm_map_wire_entry_failure(vm_map_t map, vm_map_entry_t entry,
3234     vm_offset_t failed_addr)
3235 {
3236 
3237 	VM_MAP_ASSERT_LOCKED(map);
3238 	KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 &&
3239 	    entry->wired_count == 1,
3240 	    ("vm_map_wire_entry_failure: entry %p isn't being wired", entry));
3241 	KASSERT(failed_addr < entry->end,
3242 	    ("vm_map_wire_entry_failure: entry %p was fully wired", entry));
3243 
3244 	/*
3245 	 * If any pages at the start of this entry were successfully wired,
3246 	 * then unwire them.
3247 	 */
3248 	if (failed_addr > entry->start) {
3249 		pmap_unwire(map->pmap, entry->start, failed_addr);
3250 		vm_object_unwire(entry->object.vm_object, entry->offset,
3251 		    failed_addr - entry->start, PQ_ACTIVE);
3252 	}
3253 
3254 	/*
3255 	 * Assign an out-of-range value to represent the failure to wire this
3256 	 * entry.
3257 	 */
3258 	entry->wired_count = -1;
3259 }
3260 
3261 int
3262 vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end, int flags)
3263 {
3264 	int rv;
3265 
3266 	vm_map_lock(map);
3267 	rv = vm_map_wire_locked(map, start, end, flags);
3268 	vm_map_unlock(map);
3269 	return (rv);
3270 }
3271 
3272 
3273 /*
3274  *	vm_map_wire_locked:
3275  *
3276  *	Implements both kernel and user wiring.  Returns with the map locked,
3277  *	the map lock may be dropped.
3278  */
3279 int
3280 vm_map_wire_locked(vm_map_t map, vm_offset_t start, vm_offset_t end, int flags)
3281 {
3282 	vm_map_entry_t entry, first_entry, next_entry, prev_entry;
3283 	vm_offset_t faddr, saved_end, saved_start;
3284 	u_long npages;
3285 	u_int last_timestamp;
3286 	int rv;
3287 	bool holes_ok, need_wakeup, user_wire;
3288 	vm_prot_t prot;
3289 
3290 	VM_MAP_ASSERT_LOCKED(map);
3291 
3292 	if (start == end)
3293 		return (KERN_SUCCESS);
3294 	prot = 0;
3295 	if (flags & VM_MAP_WIRE_WRITE)
3296 		prot |= VM_PROT_WRITE;
3297 	holes_ok = (flags & VM_MAP_WIRE_HOLESOK) != 0;
3298 	user_wire = (flags & VM_MAP_WIRE_USER) != 0;
3299 	VM_MAP_RANGE_CHECK(map, start, end);
3300 	if (!vm_map_lookup_entry(map, start, &first_entry)) {
3301 		if (holes_ok)
3302 			first_entry = vm_map_entry_succ(first_entry);
3303 		else
3304 			return (KERN_INVALID_ADDRESS);
3305 	}
3306 	for (entry = first_entry; entry->start < end; entry = next_entry) {
3307 		if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
3308 			/*
3309 			 * We have not yet clipped the entry.
3310 			 */
3311 			next_entry = vm_map_entry_in_transition(map, start,
3312 			    &end, holes_ok, entry);
3313 			if (next_entry == NULL) {
3314 				if (entry == first_entry)
3315 					return (KERN_INVALID_ADDRESS);
3316 				rv = KERN_INVALID_ADDRESS;
3317 				goto done;
3318 			}
3319 			first_entry = (entry == first_entry) ?
3320 			    next_entry : NULL;
3321 			continue;
3322 		}
3323 		vm_map_clip_start(map, entry, start);
3324 		vm_map_clip_end(map, entry, end);
3325 		/*
3326 		 * Mark the entry in case the map lock is released.  (See
3327 		 * above.)
3328 		 */
3329 		KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 &&
3330 		    entry->wiring_thread == NULL,
3331 		    ("owned map entry %p", entry));
3332 		entry->eflags |= MAP_ENTRY_IN_TRANSITION;
3333 		entry->wiring_thread = curthread;
3334 		if ((entry->protection & (VM_PROT_READ | VM_PROT_EXECUTE)) == 0
3335 		    || (entry->protection & prot) != prot) {
3336 			entry->eflags |= MAP_ENTRY_WIRE_SKIPPED;
3337 			if (!holes_ok) {
3338 				end = entry->end;
3339 				rv = KERN_INVALID_ADDRESS;
3340 				goto done;
3341 			}
3342 		} else if (entry->wired_count == 0) {
3343 			entry->wired_count++;
3344 
3345 			npages = atop(entry->end - entry->start);
3346 			if (user_wire && !vm_map_wire_user_count_add(npages)) {
3347 				vm_map_wire_entry_failure(map, entry,
3348 				    entry->start);
3349 				end = entry->end;
3350 				rv = KERN_RESOURCE_SHORTAGE;
3351 				goto done;
3352 			}
3353 
3354 			/*
3355 			 * Release the map lock, relying on the in-transition
3356 			 * mark.  Mark the map busy for fork.
3357 			 */
3358 			saved_start = entry->start;
3359 			saved_end = entry->end;
3360 			last_timestamp = map->timestamp;
3361 			vm_map_busy(map);
3362 			vm_map_unlock(map);
3363 
3364 			faddr = saved_start;
3365 			do {
3366 				/*
3367 				 * Simulate a fault to get the page and enter
3368 				 * it into the physical map.
3369 				 */
3370 				if ((rv = vm_fault(map, faddr,
3371 				    VM_PROT_NONE, VM_FAULT_WIRE, NULL)) !=
3372 				    KERN_SUCCESS)
3373 					break;
3374 			} while ((faddr += PAGE_SIZE) < saved_end);
3375 			vm_map_lock(map);
3376 			vm_map_unbusy(map);
3377 			if (last_timestamp + 1 != map->timestamp) {
3378 				/*
3379 				 * Look again for the entry because the map was
3380 				 * modified while it was unlocked.  The entry
3381 				 * may have been clipped, but NOT merged or
3382 				 * deleted.
3383 				 */
3384 				if (!vm_map_lookup_entry(map, saved_start,
3385 				    &next_entry))
3386 					KASSERT(false,
3387 					    ("vm_map_wire: lookup failed"));
3388 				first_entry = (entry == first_entry) ?
3389 				    next_entry : NULL;
3390 				for (entry = next_entry; entry->end < saved_end;
3391 				    entry = vm_map_entry_succ(entry)) {
3392 					/*
3393 					 * In case of failure, handle entries
3394 					 * that were not fully wired here;
3395 					 * fully wired entries are handled
3396 					 * later.
3397 					 */
3398 					if (rv != KERN_SUCCESS &&
3399 					    faddr < entry->end)
3400 						vm_map_wire_entry_failure(map,
3401 						    entry, faddr);
3402 				}
3403 			}
3404 			if (rv != KERN_SUCCESS) {
3405 				vm_map_wire_entry_failure(map, entry, faddr);
3406 				if (user_wire)
3407 					vm_map_wire_user_count_sub(npages);
3408 				end = entry->end;
3409 				goto done;
3410 			}
3411 		} else if (!user_wire ||
3412 			   (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
3413 			entry->wired_count++;
3414 		}
3415 		/*
3416 		 * Check the map for holes in the specified region.
3417 		 * If holes_ok was specified, skip this check.
3418 		 */
3419 		next_entry = vm_map_entry_succ(entry);
3420 		if (!holes_ok &&
3421 		    entry->end < end && next_entry->start > entry->end) {
3422 			end = entry->end;
3423 			rv = KERN_INVALID_ADDRESS;
3424 			goto done;
3425 		}
3426 	}
3427 	rv = KERN_SUCCESS;
3428 done:
3429 	need_wakeup = false;
3430 	if (first_entry == NULL &&
3431 	    !vm_map_lookup_entry(map, start, &first_entry)) {
3432 		KASSERT(holes_ok, ("vm_map_wire: lookup failed"));
3433 		prev_entry = first_entry;
3434 		entry = vm_map_entry_succ(first_entry);
3435 	} else {
3436 		prev_entry = vm_map_entry_pred(first_entry);
3437 		entry = first_entry;
3438 	}
3439 	for (; entry->start < end;
3440 	    prev_entry = entry, entry = vm_map_entry_succ(entry)) {
3441 		/*
3442 		 * If holes_ok was specified, an empty
3443 		 * space in the unwired region could have been mapped
3444 		 * while the map lock was dropped for faulting in the
3445 		 * pages or draining MAP_ENTRY_IN_TRANSITION.
3446 		 * Moreover, another thread could be simultaneously
3447 		 * wiring this new mapping entry.  Detect these cases
3448 		 * and skip any entries marked as in transition not by us.
3449 		 */
3450 		if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) == 0 ||
3451 		    entry->wiring_thread != curthread) {
3452 			KASSERT(holes_ok,
3453 			    ("vm_map_wire: !HOLESOK and new/changed entry"));
3454 			continue;
3455 		}
3456 
3457 		if ((entry->eflags & MAP_ENTRY_WIRE_SKIPPED) != 0) {
3458 			/* do nothing */
3459 		} else if (rv == KERN_SUCCESS) {
3460 			if (user_wire)
3461 				entry->eflags |= MAP_ENTRY_USER_WIRED;
3462 		} else if (entry->wired_count == -1) {
3463 			/*
3464 			 * Wiring failed on this entry.  Thus, unwiring is
3465 			 * unnecessary.
3466 			 */
3467 			entry->wired_count = 0;
3468 		} else if (!user_wire ||
3469 		    (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
3470 			/*
3471 			 * Undo the wiring.  Wiring succeeded on this entry
3472 			 * but failed on a later entry.
3473 			 */
3474 			if (entry->wired_count == 1) {
3475 				vm_map_entry_unwire(map, entry);
3476 				if (user_wire)
3477 					vm_map_wire_user_count_sub(
3478 					    atop(entry->end - entry->start));
3479 			} else
3480 				entry->wired_count--;
3481 		}
3482 		KASSERT((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0,
3483 		    ("vm_map_wire: in-transition flag missing %p", entry));
3484 		KASSERT(entry->wiring_thread == curthread,
3485 		    ("vm_map_wire: alien wire %p", entry));
3486 		entry->eflags &= ~(MAP_ENTRY_IN_TRANSITION |
3487 		    MAP_ENTRY_WIRE_SKIPPED);
3488 		entry->wiring_thread = NULL;
3489 		if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
3490 			entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
3491 			need_wakeup = true;
3492 		}
3493 		vm_map_try_merge_entries(map, prev_entry, entry);
3494 	}
3495 	vm_map_try_merge_entries(map, prev_entry, entry);
3496 	if (need_wakeup)
3497 		vm_map_wakeup(map);
3498 	return (rv);
3499 }
3500 
3501 /*
3502  * vm_map_sync
3503  *
3504  * Push any dirty cached pages in the address range to their pager.
3505  * If syncio is TRUE, dirty pages are written synchronously.
3506  * If invalidate is TRUE, any cached pages are freed as well.
3507  *
3508  * If the size of the region from start to end is zero, we are
3509  * supposed to flush all modified pages within the region containing
3510  * start.  Unfortunately, a region can be split or coalesced with
3511  * neighboring regions, making it difficult to determine what the
3512  * original region was.  Therefore, we approximate this requirement by
3513  * flushing the current region containing start.
3514  *
3515  * Returns an error if any part of the specified range is not mapped.
3516  */
3517 int
3518 vm_map_sync(
3519 	vm_map_t map,
3520 	vm_offset_t start,
3521 	vm_offset_t end,
3522 	boolean_t syncio,
3523 	boolean_t invalidate)
3524 {
3525 	vm_map_entry_t entry, first_entry, next_entry;
3526 	vm_size_t size;
3527 	vm_object_t object;
3528 	vm_ooffset_t offset;
3529 	unsigned int last_timestamp;
3530 	boolean_t failed;
3531 
3532 	vm_map_lock_read(map);
3533 	VM_MAP_RANGE_CHECK(map, start, end);
3534 	if (!vm_map_lookup_entry(map, start, &first_entry)) {
3535 		vm_map_unlock_read(map);
3536 		return (KERN_INVALID_ADDRESS);
3537 	} else if (start == end) {
3538 		start = first_entry->start;
3539 		end = first_entry->end;
3540 	}
3541 	/*
3542 	 * Make a first pass to check for user-wired memory and holes.
3543 	 */
3544 	for (entry = first_entry; entry->start < end; entry = next_entry) {
3545 		if (invalidate &&
3546 		    (entry->eflags & MAP_ENTRY_USER_WIRED) != 0) {
3547 			vm_map_unlock_read(map);
3548 			return (KERN_INVALID_ARGUMENT);
3549 		}
3550 		next_entry = vm_map_entry_succ(entry);
3551 		if (end > entry->end &&
3552 		    entry->end != next_entry->start) {
3553 			vm_map_unlock_read(map);
3554 			return (KERN_INVALID_ADDRESS);
3555 		}
3556 	}
3557 
3558 	if (invalidate)
3559 		pmap_remove(map->pmap, start, end);
3560 	failed = FALSE;
3561 
3562 	/*
3563 	 * Make a second pass, cleaning/uncaching pages from the indicated
3564 	 * objects as we go.
3565 	 */
3566 	for (entry = first_entry; entry->start < end;) {
3567 		offset = entry->offset + (start - entry->start);
3568 		size = (end <= entry->end ? end : entry->end) - start;
3569 		if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0) {
3570 			vm_map_t smap;
3571 			vm_map_entry_t tentry;
3572 			vm_size_t tsize;
3573 
3574 			smap = entry->object.sub_map;
3575 			vm_map_lock_read(smap);
3576 			(void) vm_map_lookup_entry(smap, offset, &tentry);
3577 			tsize = tentry->end - offset;
3578 			if (tsize < size)
3579 				size = tsize;
3580 			object = tentry->object.vm_object;
3581 			offset = tentry->offset + (offset - tentry->start);
3582 			vm_map_unlock_read(smap);
3583 		} else {
3584 			object = entry->object.vm_object;
3585 		}
3586 		vm_object_reference(object);
3587 		last_timestamp = map->timestamp;
3588 		vm_map_unlock_read(map);
3589 		if (!vm_object_sync(object, offset, size, syncio, invalidate))
3590 			failed = TRUE;
3591 		start += size;
3592 		vm_object_deallocate(object);
3593 		vm_map_lock_read(map);
3594 		if (last_timestamp == map->timestamp ||
3595 		    !vm_map_lookup_entry(map, start, &entry))
3596 			entry = vm_map_entry_succ(entry);
3597 	}
3598 
3599 	vm_map_unlock_read(map);
3600 	return (failed ? KERN_FAILURE : KERN_SUCCESS);
3601 }
3602 
3603 /*
3604  *	vm_map_entry_unwire:	[ internal use only ]
3605  *
3606  *	Make the region specified by this entry pageable.
3607  *
3608  *	The map in question should be locked.
3609  *	[This is the reason for this routine's existence.]
3610  */
3611 static void
3612 vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry)
3613 {
3614 	vm_size_t size;
3615 
3616 	VM_MAP_ASSERT_LOCKED(map);
3617 	KASSERT(entry->wired_count > 0,
3618 	    ("vm_map_entry_unwire: entry %p isn't wired", entry));
3619 
3620 	size = entry->end - entry->start;
3621 	if ((entry->eflags & MAP_ENTRY_USER_WIRED) != 0)
3622 		vm_map_wire_user_count_sub(atop(size));
3623 	pmap_unwire(map->pmap, entry->start, entry->end);
3624 	vm_object_unwire(entry->object.vm_object, entry->offset, size,
3625 	    PQ_ACTIVE);
3626 	entry->wired_count = 0;
3627 }
3628 
3629 static void
3630 vm_map_entry_deallocate(vm_map_entry_t entry, boolean_t system_map)
3631 {
3632 
3633 	if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0)
3634 		vm_object_deallocate(entry->object.vm_object);
3635 	uma_zfree(system_map ? kmapentzone : mapentzone, entry);
3636 }
3637 
3638 /*
3639  *	vm_map_entry_delete:	[ internal use only ]
3640  *
3641  *	Deallocate the given entry from the target map.
3642  */
3643 static void
3644 vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry)
3645 {
3646 	vm_object_t object;
3647 	vm_pindex_t offidxstart, offidxend, size1;
3648 	vm_size_t size;
3649 
3650 	vm_map_entry_unlink(map, entry, UNLINK_MERGE_NONE);
3651 	object = entry->object.vm_object;
3652 
3653 	if ((entry->eflags & MAP_ENTRY_GUARD) != 0) {
3654 		MPASS(entry->cred == NULL);
3655 		MPASS((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0);
3656 		MPASS(object == NULL);
3657 		vm_map_entry_deallocate(entry, map->system_map);
3658 		return;
3659 	}
3660 
3661 	size = entry->end - entry->start;
3662 	map->size -= size;
3663 
3664 	if (entry->cred != NULL) {
3665 		swap_release_by_cred(size, entry->cred);
3666 		crfree(entry->cred);
3667 	}
3668 
3669 	if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0 || object == NULL) {
3670 		entry->object.vm_object = NULL;
3671 	} else if ((object->flags & OBJ_ANON) != 0 ||
3672 	    object == kernel_object) {
3673 		KASSERT(entry->cred == NULL || object->cred == NULL ||
3674 		    (entry->eflags & MAP_ENTRY_NEEDS_COPY),
3675 		    ("OVERCOMMIT vm_map_entry_delete: both cred %p", entry));
3676 		offidxstart = OFF_TO_IDX(entry->offset);
3677 		offidxend = offidxstart + atop(size);
3678 		VM_OBJECT_WLOCK(object);
3679 		if (object->ref_count != 1 &&
3680 		    ((object->flags & OBJ_ONEMAPPING) != 0 ||
3681 		    object == kernel_object)) {
3682 			vm_object_collapse(object);
3683 
3684 			/*
3685 			 * The option OBJPR_NOTMAPPED can be passed here
3686 			 * because vm_map_delete() already performed
3687 			 * pmap_remove() on the only mapping to this range
3688 			 * of pages.
3689 			 */
3690 			vm_object_page_remove(object, offidxstart, offidxend,
3691 			    OBJPR_NOTMAPPED);
3692 			if (offidxend >= object->size &&
3693 			    offidxstart < object->size) {
3694 				size1 = object->size;
3695 				object->size = offidxstart;
3696 				if (object->cred != NULL) {
3697 					size1 -= object->size;
3698 					KASSERT(object->charge >= ptoa(size1),
3699 					    ("object %p charge < 0", object));
3700 					swap_release_by_cred(ptoa(size1),
3701 					    object->cred);
3702 					object->charge -= ptoa(size1);
3703 				}
3704 			}
3705 		}
3706 		VM_OBJECT_WUNLOCK(object);
3707 	}
3708 	if (map->system_map)
3709 		vm_map_entry_deallocate(entry, TRUE);
3710 	else {
3711 		entry->defer_next = curthread->td_map_def_user;
3712 		curthread->td_map_def_user = entry;
3713 	}
3714 }
3715 
3716 /*
3717  *	vm_map_delete:	[ internal use only ]
3718  *
3719  *	Deallocates the given address range from the target
3720  *	map.
3721  */
3722 int
3723 vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end)
3724 {
3725 	vm_map_entry_t entry, next_entry;
3726 
3727 	VM_MAP_ASSERT_LOCKED(map);
3728 	if (start == end)
3729 		return (KERN_SUCCESS);
3730 
3731 	/*
3732 	 * Find the start of the region, and clip it.
3733 	 * Step through all entries in this region.
3734 	 */
3735 	for (entry = vm_map_lookup_clip_start(map, start, &entry);
3736 	    entry->start < end; entry = next_entry) {
3737 		/*
3738 		 * Wait for wiring or unwiring of an entry to complete.
3739 		 * Also wait for any system wirings to disappear on
3740 		 * user maps.
3741 		 */
3742 		if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0 ||
3743 		    (vm_map_pmap(map) != kernel_pmap &&
3744 		    vm_map_entry_system_wired_count(entry) != 0)) {
3745 			unsigned int last_timestamp;
3746 			vm_offset_t saved_start;
3747 
3748 			saved_start = entry->start;
3749 			entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
3750 			last_timestamp = map->timestamp;
3751 			(void) vm_map_unlock_and_wait(map, 0);
3752 			vm_map_lock(map);
3753 			if (last_timestamp + 1 != map->timestamp) {
3754 				/*
3755 				 * Look again for the entry because the map was
3756 				 * modified while it was unlocked.
3757 				 * Specifically, the entry may have been
3758 				 * clipped, merged, or deleted.
3759 				 */
3760 				next_entry = vm_map_lookup_clip_start(map,
3761 				    saved_start, &next_entry);
3762 			} else
3763 				next_entry = entry;
3764 			continue;
3765 		}
3766 		vm_map_clip_end(map, entry, end);
3767 		next_entry = vm_map_entry_succ(entry);
3768 
3769 		/*
3770 		 * Unwire before removing addresses from the pmap; otherwise,
3771 		 * unwiring will put the entries back in the pmap.
3772 		 */
3773 		if (entry->wired_count != 0)
3774 			vm_map_entry_unwire(map, entry);
3775 
3776 		/*
3777 		 * Remove mappings for the pages, but only if the
3778 		 * mappings could exist.  For instance, it does not
3779 		 * make sense to call pmap_remove() for guard entries.
3780 		 */
3781 		if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0 ||
3782 		    entry->object.vm_object != NULL)
3783 			pmap_remove(map->pmap, entry->start, entry->end);
3784 
3785 		if (entry->end == map->anon_loc)
3786 			map->anon_loc = entry->start;
3787 
3788 		/*
3789 		 * Delete the entry only after removing all pmap
3790 		 * entries pointing to its pages.  (Otherwise, its
3791 		 * page frames may be reallocated, and any modify bits
3792 		 * will be set in the wrong object!)
3793 		 */
3794 		vm_map_entry_delete(map, entry);
3795 	}
3796 	return (KERN_SUCCESS);
3797 }
3798 
3799 /*
3800  *	vm_map_remove:
3801  *
3802  *	Remove the given address range from the target map.
3803  *	This is the exported form of vm_map_delete.
3804  */
3805 int
3806 vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end)
3807 {
3808 	int result;
3809 
3810 	vm_map_lock(map);
3811 	VM_MAP_RANGE_CHECK(map, start, end);
3812 	result = vm_map_delete(map, start, end);
3813 	vm_map_unlock(map);
3814 	return (result);
3815 }
3816 
3817 /*
3818  *	vm_map_check_protection:
3819  *
3820  *	Assert that the target map allows the specified privilege on the
3821  *	entire address region given.  The entire region must be allocated.
3822  *
3823  *	WARNING!  This code does not and should not check whether the
3824  *	contents of the region is accessible.  For example a smaller file
3825  *	might be mapped into a larger address space.
3826  *
3827  *	NOTE!  This code is also called by munmap().
3828  *
3829  *	The map must be locked.  A read lock is sufficient.
3830  */
3831 boolean_t
3832 vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end,
3833 			vm_prot_t protection)
3834 {
3835 	vm_map_entry_t entry;
3836 	vm_map_entry_t tmp_entry;
3837 
3838 	if (!vm_map_lookup_entry(map, start, &tmp_entry))
3839 		return (FALSE);
3840 	entry = tmp_entry;
3841 
3842 	while (start < end) {
3843 		/*
3844 		 * No holes allowed!
3845 		 */
3846 		if (start < entry->start)
3847 			return (FALSE);
3848 		/*
3849 		 * Check protection associated with entry.
3850 		 */
3851 		if ((entry->protection & protection) != protection)
3852 			return (FALSE);
3853 		/* go to next entry */
3854 		start = entry->end;
3855 		entry = vm_map_entry_succ(entry);
3856 	}
3857 	return (TRUE);
3858 }
3859 
3860 
3861 /*
3862  *
3863  *	vm_map_copy_swap_object:
3864  *
3865  *	Copies a swap-backed object from an existing map entry to a
3866  *	new one.  Carries forward the swap charge.  May change the
3867  *	src object on return.
3868  */
3869 static void
3870 vm_map_copy_swap_object(vm_map_entry_t src_entry, vm_map_entry_t dst_entry,
3871     vm_offset_t size, vm_ooffset_t *fork_charge)
3872 {
3873 	vm_object_t src_object;
3874 	struct ucred *cred;
3875 	int charged;
3876 
3877 	src_object = src_entry->object.vm_object;
3878 	charged = ENTRY_CHARGED(src_entry);
3879 	if ((src_object->flags & OBJ_ANON) != 0) {
3880 		VM_OBJECT_WLOCK(src_object);
3881 		vm_object_collapse(src_object);
3882 		if ((src_object->flags & OBJ_ONEMAPPING) != 0) {
3883 			vm_object_split(src_entry);
3884 			src_object = src_entry->object.vm_object;
3885 		}
3886 		vm_object_reference_locked(src_object);
3887 		vm_object_clear_flag(src_object, OBJ_ONEMAPPING);
3888 		VM_OBJECT_WUNLOCK(src_object);
3889 	} else
3890 		vm_object_reference(src_object);
3891 	if (src_entry->cred != NULL &&
3892 	    !(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
3893 		KASSERT(src_object->cred == NULL,
3894 		    ("OVERCOMMIT: vm_map_copy_anon_entry: cred %p",
3895 		     src_object));
3896 		src_object->cred = src_entry->cred;
3897 		src_object->charge = size;
3898 	}
3899 	dst_entry->object.vm_object = src_object;
3900 	if (charged) {
3901 		cred = curthread->td_ucred;
3902 		crhold(cred);
3903 		dst_entry->cred = cred;
3904 		*fork_charge += size;
3905 		if (!(src_entry->eflags & MAP_ENTRY_NEEDS_COPY)) {
3906 			crhold(cred);
3907 			src_entry->cred = cred;
3908 			*fork_charge += size;
3909 		}
3910 	}
3911 }
3912 
3913 /*
3914  *	vm_map_copy_entry:
3915  *
3916  *	Copies the contents of the source entry to the destination
3917  *	entry.  The entries *must* be aligned properly.
3918  */
3919 static void
3920 vm_map_copy_entry(
3921 	vm_map_t src_map,
3922 	vm_map_t dst_map,
3923 	vm_map_entry_t src_entry,
3924 	vm_map_entry_t dst_entry,
3925 	vm_ooffset_t *fork_charge)
3926 {
3927 	vm_object_t src_object;
3928 	vm_map_entry_t fake_entry;
3929 	vm_offset_t size;
3930 
3931 	VM_MAP_ASSERT_LOCKED(dst_map);
3932 
3933 	if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP)
3934 		return;
3935 
3936 	if (src_entry->wired_count == 0 ||
3937 	    (src_entry->protection & VM_PROT_WRITE) == 0) {
3938 		/*
3939 		 * If the source entry is marked needs_copy, it is already
3940 		 * write-protected.
3941 		 */
3942 		if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0 &&
3943 		    (src_entry->protection & VM_PROT_WRITE) != 0) {
3944 			pmap_protect(src_map->pmap,
3945 			    src_entry->start,
3946 			    src_entry->end,
3947 			    src_entry->protection & ~VM_PROT_WRITE);
3948 		}
3949 
3950 		/*
3951 		 * Make a copy of the object.
3952 		 */
3953 		size = src_entry->end - src_entry->start;
3954 		if ((src_object = src_entry->object.vm_object) != NULL) {
3955 			if (src_object->type == OBJT_DEFAULT ||
3956 			    src_object->type == OBJT_SWAP) {
3957 				vm_map_copy_swap_object(src_entry, dst_entry,
3958 				    size, fork_charge);
3959 				/* May have split/collapsed, reload obj. */
3960 				src_object = src_entry->object.vm_object;
3961 			} else {
3962 				vm_object_reference(src_object);
3963 				dst_entry->object.vm_object = src_object;
3964 			}
3965 			src_entry->eflags |= MAP_ENTRY_COW |
3966 			    MAP_ENTRY_NEEDS_COPY;
3967 			dst_entry->eflags |= MAP_ENTRY_COW |
3968 			    MAP_ENTRY_NEEDS_COPY;
3969 			dst_entry->offset = src_entry->offset;
3970 			if (src_entry->eflags & MAP_ENTRY_WRITECNT) {
3971 				/*
3972 				 * MAP_ENTRY_WRITECNT cannot
3973 				 * indicate write reference from
3974 				 * src_entry, since the entry is
3975 				 * marked as needs copy.  Allocate a
3976 				 * fake entry that is used to
3977 				 * decrement object->un_pager writecount
3978 				 * at the appropriate time.  Attach
3979 				 * fake_entry to the deferred list.
3980 				 */
3981 				fake_entry = vm_map_entry_create(dst_map);
3982 				fake_entry->eflags = MAP_ENTRY_WRITECNT;
3983 				src_entry->eflags &= ~MAP_ENTRY_WRITECNT;
3984 				vm_object_reference(src_object);
3985 				fake_entry->object.vm_object = src_object;
3986 				fake_entry->start = src_entry->start;
3987 				fake_entry->end = src_entry->end;
3988 				fake_entry->defer_next =
3989 				    curthread->td_map_def_user;
3990 				curthread->td_map_def_user = fake_entry;
3991 			}
3992 
3993 			pmap_copy(dst_map->pmap, src_map->pmap,
3994 			    dst_entry->start, dst_entry->end - dst_entry->start,
3995 			    src_entry->start);
3996 		} else {
3997 			dst_entry->object.vm_object = NULL;
3998 			dst_entry->offset = 0;
3999 			if (src_entry->cred != NULL) {
4000 				dst_entry->cred = curthread->td_ucred;
4001 				crhold(dst_entry->cred);
4002 				*fork_charge += size;
4003 			}
4004 		}
4005 	} else {
4006 		/*
4007 		 * We don't want to make writeable wired pages copy-on-write.
4008 		 * Immediately copy these pages into the new map by simulating
4009 		 * page faults.  The new pages are pageable.
4010 		 */
4011 		vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry,
4012 		    fork_charge);
4013 	}
4014 }
4015 
4016 /*
4017  * vmspace_map_entry_forked:
4018  * Update the newly-forked vmspace each time a map entry is inherited
4019  * or copied.  The values for vm_dsize and vm_tsize are approximate
4020  * (and mostly-obsolete ideas in the face of mmap(2) et al.)
4021  */
4022 static void
4023 vmspace_map_entry_forked(const struct vmspace *vm1, struct vmspace *vm2,
4024     vm_map_entry_t entry)
4025 {
4026 	vm_size_t entrysize;
4027 	vm_offset_t newend;
4028 
4029 	if ((entry->eflags & MAP_ENTRY_GUARD) != 0)
4030 		return;
4031 	entrysize = entry->end - entry->start;
4032 	vm2->vm_map.size += entrysize;
4033 	if (entry->eflags & (MAP_ENTRY_GROWS_DOWN | MAP_ENTRY_GROWS_UP)) {
4034 		vm2->vm_ssize += btoc(entrysize);
4035 	} else if (entry->start >= (vm_offset_t)vm1->vm_daddr &&
4036 	    entry->start < (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize)) {
4037 		newend = MIN(entry->end,
4038 		    (vm_offset_t)vm1->vm_daddr + ctob(vm1->vm_dsize));
4039 		vm2->vm_dsize += btoc(newend - entry->start);
4040 	} else if (entry->start >= (vm_offset_t)vm1->vm_taddr &&
4041 	    entry->start < (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize)) {
4042 		newend = MIN(entry->end,
4043 		    (vm_offset_t)vm1->vm_taddr + ctob(vm1->vm_tsize));
4044 		vm2->vm_tsize += btoc(newend - entry->start);
4045 	}
4046 }
4047 
4048 /*
4049  * vmspace_fork:
4050  * Create a new process vmspace structure and vm_map
4051  * based on those of an existing process.  The new map
4052  * is based on the old map, according to the inheritance
4053  * values on the regions in that map.
4054  *
4055  * XXX It might be worth coalescing the entries added to the new vmspace.
4056  *
4057  * The source map must not be locked.
4058  */
4059 struct vmspace *
4060 vmspace_fork(struct vmspace *vm1, vm_ooffset_t *fork_charge)
4061 {
4062 	struct vmspace *vm2;
4063 	vm_map_t new_map, old_map;
4064 	vm_map_entry_t new_entry, old_entry;
4065 	vm_object_t object;
4066 	int error, locked;
4067 	vm_inherit_t inh;
4068 
4069 	old_map = &vm1->vm_map;
4070 	/* Copy immutable fields of vm1 to vm2. */
4071 	vm2 = vmspace_alloc(vm_map_min(old_map), vm_map_max(old_map),
4072 	    pmap_pinit);
4073 	if (vm2 == NULL)
4074 		return (NULL);
4075 
4076 	vm2->vm_taddr = vm1->vm_taddr;
4077 	vm2->vm_daddr = vm1->vm_daddr;
4078 	vm2->vm_maxsaddr = vm1->vm_maxsaddr;
4079 	vm_map_lock(old_map);
4080 	if (old_map->busy)
4081 		vm_map_wait_busy(old_map);
4082 	new_map = &vm2->vm_map;
4083 	locked = vm_map_trylock(new_map); /* trylock to silence WITNESS */
4084 	KASSERT(locked, ("vmspace_fork: lock failed"));
4085 
4086 	error = pmap_vmspace_copy(new_map->pmap, old_map->pmap);
4087 	if (error != 0) {
4088 		sx_xunlock(&old_map->lock);
4089 		sx_xunlock(&new_map->lock);
4090 		vm_map_process_deferred();
4091 		vmspace_free(vm2);
4092 		return (NULL);
4093 	}
4094 
4095 	new_map->anon_loc = old_map->anon_loc;
4096 
4097 	VM_MAP_ENTRY_FOREACH(old_entry, old_map) {
4098 		if ((old_entry->eflags & MAP_ENTRY_IS_SUB_MAP) != 0)
4099 			panic("vm_map_fork: encountered a submap");
4100 
4101 		inh = old_entry->inheritance;
4102 		if ((old_entry->eflags & MAP_ENTRY_GUARD) != 0 &&
4103 		    inh != VM_INHERIT_NONE)
4104 			inh = VM_INHERIT_COPY;
4105 
4106 		switch (inh) {
4107 		case VM_INHERIT_NONE:
4108 			break;
4109 
4110 		case VM_INHERIT_SHARE:
4111 			/*
4112 			 * Clone the entry, creating the shared object if
4113 			 * necessary.
4114 			 */
4115 			object = old_entry->object.vm_object;
4116 			if (object == NULL) {
4117 				vm_map_entry_back(old_entry);
4118 				object = old_entry->object.vm_object;
4119 			}
4120 
4121 			/*
4122 			 * Add the reference before calling vm_object_shadow
4123 			 * to insure that a shadow object is created.
4124 			 */
4125 			vm_object_reference(object);
4126 			if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) {
4127 				vm_object_shadow(&old_entry->object.vm_object,
4128 				    &old_entry->offset,
4129 				    old_entry->end - old_entry->start,
4130 				    old_entry->cred,
4131 				    /* Transfer the second reference too. */
4132 				    true);
4133 				old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
4134 				old_entry->cred = NULL;
4135 
4136 				/*
4137 				 * As in vm_map_merged_neighbor_dispose(),
4138 				 * the vnode lock will not be acquired in
4139 				 * this call to vm_object_deallocate().
4140 				 */
4141 				vm_object_deallocate(object);
4142 				object = old_entry->object.vm_object;
4143 			} else {
4144 				VM_OBJECT_WLOCK(object);
4145 				vm_object_clear_flag(object, OBJ_ONEMAPPING);
4146 				if (old_entry->cred != NULL) {
4147 					KASSERT(object->cred == NULL,
4148 					    ("vmspace_fork both cred"));
4149 					object->cred = old_entry->cred;
4150 					object->charge = old_entry->end -
4151 					    old_entry->start;
4152 					old_entry->cred = NULL;
4153 				}
4154 
4155 				/*
4156 				 * Assert the correct state of the vnode
4157 				 * v_writecount while the object is locked, to
4158 				 * not relock it later for the assertion
4159 				 * correctness.
4160 				 */
4161 				if (old_entry->eflags & MAP_ENTRY_WRITECNT &&
4162 				    object->type == OBJT_VNODE) {
4163 					KASSERT(((struct vnode *)object->
4164 					    handle)->v_writecount > 0,
4165 					    ("vmspace_fork: v_writecount %p",
4166 					    object));
4167 					KASSERT(object->un_pager.vnp.
4168 					    writemappings > 0,
4169 					    ("vmspace_fork: vnp.writecount %p",
4170 					    object));
4171 				}
4172 				VM_OBJECT_WUNLOCK(object);
4173 			}
4174 
4175 			/*
4176 			 * Clone the entry, referencing the shared object.
4177 			 */
4178 			new_entry = vm_map_entry_create(new_map);
4179 			*new_entry = *old_entry;
4180 			new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
4181 			    MAP_ENTRY_IN_TRANSITION);
4182 			new_entry->wiring_thread = NULL;
4183 			new_entry->wired_count = 0;
4184 			if (new_entry->eflags & MAP_ENTRY_WRITECNT) {
4185 				vm_pager_update_writecount(object,
4186 				    new_entry->start, new_entry->end);
4187 			}
4188 			vm_map_entry_set_vnode_text(new_entry, true);
4189 
4190 			/*
4191 			 * Insert the entry into the new map -- we know we're
4192 			 * inserting at the end of the new map.
4193 			 */
4194 			vm_map_entry_link(new_map, new_entry);
4195 			vmspace_map_entry_forked(vm1, vm2, new_entry);
4196 
4197 			/*
4198 			 * Update the physical map
4199 			 */
4200 			pmap_copy(new_map->pmap, old_map->pmap,
4201 			    new_entry->start,
4202 			    (old_entry->end - old_entry->start),
4203 			    old_entry->start);
4204 			break;
4205 
4206 		case VM_INHERIT_COPY:
4207 			/*
4208 			 * Clone the entry and link into the map.
4209 			 */
4210 			new_entry = vm_map_entry_create(new_map);
4211 			*new_entry = *old_entry;
4212 			/*
4213 			 * Copied entry is COW over the old object.
4214 			 */
4215 			new_entry->eflags &= ~(MAP_ENTRY_USER_WIRED |
4216 			    MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_WRITECNT);
4217 			new_entry->wiring_thread = NULL;
4218 			new_entry->wired_count = 0;
4219 			new_entry->object.vm_object = NULL;
4220 			new_entry->cred = NULL;
4221 			vm_map_entry_link(new_map, new_entry);
4222 			vmspace_map_entry_forked(vm1, vm2, new_entry);
4223 			vm_map_copy_entry(old_map, new_map, old_entry,
4224 			    new_entry, fork_charge);
4225 			vm_map_entry_set_vnode_text(new_entry, true);
4226 			break;
4227 
4228 		case VM_INHERIT_ZERO:
4229 			/*
4230 			 * Create a new anonymous mapping entry modelled from
4231 			 * the old one.
4232 			 */
4233 			new_entry = vm_map_entry_create(new_map);
4234 			memset(new_entry, 0, sizeof(*new_entry));
4235 
4236 			new_entry->start = old_entry->start;
4237 			new_entry->end = old_entry->end;
4238 			new_entry->eflags = old_entry->eflags &
4239 			    ~(MAP_ENTRY_USER_WIRED | MAP_ENTRY_IN_TRANSITION |
4240 			    MAP_ENTRY_WRITECNT | MAP_ENTRY_VN_EXEC);
4241 			new_entry->protection = old_entry->protection;
4242 			new_entry->max_protection = old_entry->max_protection;
4243 			new_entry->inheritance = VM_INHERIT_ZERO;
4244 
4245 			vm_map_entry_link(new_map, new_entry);
4246 			vmspace_map_entry_forked(vm1, vm2, new_entry);
4247 
4248 			new_entry->cred = curthread->td_ucred;
4249 			crhold(new_entry->cred);
4250 			*fork_charge += (new_entry->end - new_entry->start);
4251 
4252 			break;
4253 		}
4254 	}
4255 	/*
4256 	 * Use inlined vm_map_unlock() to postpone handling the deferred
4257 	 * map entries, which cannot be done until both old_map and
4258 	 * new_map locks are released.
4259 	 */
4260 	sx_xunlock(&old_map->lock);
4261 	sx_xunlock(&new_map->lock);
4262 	vm_map_process_deferred();
4263 
4264 	return (vm2);
4265 }
4266 
4267 /*
4268  * Create a process's stack for exec_new_vmspace().  This function is never
4269  * asked to wire the newly created stack.
4270  */
4271 int
4272 vm_map_stack(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
4273     vm_prot_t prot, vm_prot_t max, int cow)
4274 {
4275 	vm_size_t growsize, init_ssize;
4276 	rlim_t vmemlim;
4277 	int rv;
4278 
4279 	MPASS((map->flags & MAP_WIREFUTURE) == 0);
4280 	growsize = sgrowsiz;
4281 	init_ssize = (max_ssize < growsize) ? max_ssize : growsize;
4282 	vm_map_lock(map);
4283 	vmemlim = lim_cur(curthread, RLIMIT_VMEM);
4284 	/* If we would blow our VMEM resource limit, no go */
4285 	if (map->size + init_ssize > vmemlim) {
4286 		rv = KERN_NO_SPACE;
4287 		goto out;
4288 	}
4289 	rv = vm_map_stack_locked(map, addrbos, max_ssize, growsize, prot,
4290 	    max, cow);
4291 out:
4292 	vm_map_unlock(map);
4293 	return (rv);
4294 }
4295 
4296 static int stack_guard_page = 1;
4297 SYSCTL_INT(_security_bsd, OID_AUTO, stack_guard_page, CTLFLAG_RWTUN,
4298     &stack_guard_page, 0,
4299     "Specifies the number of guard pages for a stack that grows");
4300 
4301 static int
4302 vm_map_stack_locked(vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
4303     vm_size_t growsize, vm_prot_t prot, vm_prot_t max, int cow)
4304 {
4305 	vm_map_entry_t new_entry, prev_entry;
4306 	vm_offset_t bot, gap_bot, gap_top, top;
4307 	vm_size_t init_ssize, sgp;
4308 	int orient, rv;
4309 
4310 	/*
4311 	 * The stack orientation is piggybacked with the cow argument.
4312 	 * Extract it into orient and mask the cow argument so that we
4313 	 * don't pass it around further.
4314 	 */
4315 	orient = cow & (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP);
4316 	KASSERT(orient != 0, ("No stack grow direction"));
4317 	KASSERT(orient != (MAP_STACK_GROWS_DOWN | MAP_STACK_GROWS_UP),
4318 	    ("bi-dir stack"));
4319 
4320 	if (max_ssize == 0 ||
4321 	    !vm_map_range_valid(map, addrbos, addrbos + max_ssize))
4322 		return (KERN_INVALID_ADDRESS);
4323 	sgp = ((curproc->p_flag2 & P2_STKGAP_DISABLE) != 0 ||
4324 	    (curproc->p_fctl0 & NT_FREEBSD_FCTL_STKGAP_DISABLE) != 0) ? 0 :
4325 	    (vm_size_t)stack_guard_page * PAGE_SIZE;
4326 	if (sgp >= max_ssize)
4327 		return (KERN_INVALID_ARGUMENT);
4328 
4329 	init_ssize = growsize;
4330 	if (max_ssize < init_ssize + sgp)
4331 		init_ssize = max_ssize - sgp;
4332 
4333 	/* If addr is already mapped, no go */
4334 	if (vm_map_lookup_entry(map, addrbos, &prev_entry))
4335 		return (KERN_NO_SPACE);
4336 
4337 	/*
4338 	 * If we can't accommodate max_ssize in the current mapping, no go.
4339 	 */
4340 	if (vm_map_entry_succ(prev_entry)->start < addrbos + max_ssize)
4341 		return (KERN_NO_SPACE);
4342 
4343 	/*
4344 	 * We initially map a stack of only init_ssize.  We will grow as
4345 	 * needed later.  Depending on the orientation of the stack (i.e.
4346 	 * the grow direction) we either map at the top of the range, the
4347 	 * bottom of the range or in the middle.
4348 	 *
4349 	 * Note: we would normally expect prot and max to be VM_PROT_ALL,
4350 	 * and cow to be 0.  Possibly we should eliminate these as input
4351 	 * parameters, and just pass these values here in the insert call.
4352 	 */
4353 	if (orient == MAP_STACK_GROWS_DOWN) {
4354 		bot = addrbos + max_ssize - init_ssize;
4355 		top = bot + init_ssize;
4356 		gap_bot = addrbos;
4357 		gap_top = bot;
4358 	} else /* if (orient == MAP_STACK_GROWS_UP) */ {
4359 		bot = addrbos;
4360 		top = bot + init_ssize;
4361 		gap_bot = top;
4362 		gap_top = addrbos + max_ssize;
4363 	}
4364 	rv = vm_map_insert(map, NULL, 0, bot, top, prot, max, cow);
4365 	if (rv != KERN_SUCCESS)
4366 		return (rv);
4367 	new_entry = vm_map_entry_succ(prev_entry);
4368 	KASSERT(new_entry->end == top || new_entry->start == bot,
4369 	    ("Bad entry start/end for new stack entry"));
4370 	KASSERT((orient & MAP_STACK_GROWS_DOWN) == 0 ||
4371 	    (new_entry->eflags & MAP_ENTRY_GROWS_DOWN) != 0,
4372 	    ("new entry lacks MAP_ENTRY_GROWS_DOWN"));
4373 	KASSERT((orient & MAP_STACK_GROWS_UP) == 0 ||
4374 	    (new_entry->eflags & MAP_ENTRY_GROWS_UP) != 0,
4375 	    ("new entry lacks MAP_ENTRY_GROWS_UP"));
4376 	if (gap_bot == gap_top)
4377 		return (KERN_SUCCESS);
4378 	rv = vm_map_insert(map, NULL, 0, gap_bot, gap_top, VM_PROT_NONE,
4379 	    VM_PROT_NONE, MAP_CREATE_GUARD | (orient == MAP_STACK_GROWS_DOWN ?
4380 	    MAP_CREATE_STACK_GAP_DN : MAP_CREATE_STACK_GAP_UP));
4381 	if (rv == KERN_SUCCESS) {
4382 		/*
4383 		 * Gap can never successfully handle a fault, so
4384 		 * read-ahead logic is never used for it.  Re-use
4385 		 * next_read of the gap entry to store
4386 		 * stack_guard_page for vm_map_growstack().
4387 		 */
4388 		if (orient == MAP_STACK_GROWS_DOWN)
4389 			vm_map_entry_pred(new_entry)->next_read = sgp;
4390 		else
4391 			vm_map_entry_succ(new_entry)->next_read = sgp;
4392 	} else {
4393 		(void)vm_map_delete(map, bot, top);
4394 	}
4395 	return (rv);
4396 }
4397 
4398 /*
4399  * Attempts to grow a vm stack entry.  Returns KERN_SUCCESS if we
4400  * successfully grow the stack.
4401  */
4402 static int
4403 vm_map_growstack(vm_map_t map, vm_offset_t addr, vm_map_entry_t gap_entry)
4404 {
4405 	vm_map_entry_t stack_entry;
4406 	struct proc *p;
4407 	struct vmspace *vm;
4408 	struct ucred *cred;
4409 	vm_offset_t gap_end, gap_start, grow_start;
4410 	vm_size_t grow_amount, guard, max_grow;
4411 	rlim_t lmemlim, stacklim, vmemlim;
4412 	int rv, rv1;
4413 	bool gap_deleted, grow_down, is_procstack;
4414 #ifdef notyet
4415 	uint64_t limit;
4416 #endif
4417 #ifdef RACCT
4418 	int error;
4419 #endif
4420 
4421 	p = curproc;
4422 	vm = p->p_vmspace;
4423 
4424 	/*
4425 	 * Disallow stack growth when the access is performed by a
4426 	 * debugger or AIO daemon.  The reason is that the wrong
4427 	 * resource limits are applied.
4428 	 */
4429 	if (p != initproc && (map != &p->p_vmspace->vm_map ||
4430 	    p->p_textvp == NULL))
4431 		return (KERN_FAILURE);
4432 
4433 	MPASS(!map->system_map);
4434 
4435 	lmemlim = lim_cur(curthread, RLIMIT_MEMLOCK);
4436 	stacklim = lim_cur(curthread, RLIMIT_STACK);
4437 	vmemlim = lim_cur(curthread, RLIMIT_VMEM);
4438 retry:
4439 	/* If addr is not in a hole for a stack grow area, no need to grow. */
4440 	if (gap_entry == NULL && !vm_map_lookup_entry(map, addr, &gap_entry))
4441 		return (KERN_FAILURE);
4442 	if ((gap_entry->eflags & MAP_ENTRY_GUARD) == 0)
4443 		return (KERN_SUCCESS);
4444 	if ((gap_entry->eflags & MAP_ENTRY_STACK_GAP_DN) != 0) {
4445 		stack_entry = vm_map_entry_succ(gap_entry);
4446 		if ((stack_entry->eflags & MAP_ENTRY_GROWS_DOWN) == 0 ||
4447 		    stack_entry->start != gap_entry->end)
4448 			return (KERN_FAILURE);
4449 		grow_amount = round_page(stack_entry->start - addr);
4450 		grow_down = true;
4451 	} else if ((gap_entry->eflags & MAP_ENTRY_STACK_GAP_UP) != 0) {
4452 		stack_entry = vm_map_entry_pred(gap_entry);
4453 		if ((stack_entry->eflags & MAP_ENTRY_GROWS_UP) == 0 ||
4454 		    stack_entry->end != gap_entry->start)
4455 			return (KERN_FAILURE);
4456 		grow_amount = round_page(addr + 1 - stack_entry->end);
4457 		grow_down = false;
4458 	} else {
4459 		return (KERN_FAILURE);
4460 	}
4461 	guard = ((curproc->p_flag2 & P2_STKGAP_DISABLE) != 0 ||
4462 	    (curproc->p_fctl0 & NT_FREEBSD_FCTL_STKGAP_DISABLE) != 0) ? 0 :
4463 	    gap_entry->next_read;
4464 	max_grow = gap_entry->end - gap_entry->start;
4465 	if (guard > max_grow)
4466 		return (KERN_NO_SPACE);
4467 	max_grow -= guard;
4468 	if (grow_amount > max_grow)
4469 		return (KERN_NO_SPACE);
4470 
4471 	/*
4472 	 * If this is the main process stack, see if we're over the stack
4473 	 * limit.
4474 	 */
4475 	is_procstack = addr >= (vm_offset_t)vm->vm_maxsaddr &&
4476 	    addr < (vm_offset_t)p->p_sysent->sv_usrstack;
4477 	if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim))
4478 		return (KERN_NO_SPACE);
4479 
4480 #ifdef RACCT
4481 	if (racct_enable) {
4482 		PROC_LOCK(p);
4483 		if (is_procstack && racct_set(p, RACCT_STACK,
4484 		    ctob(vm->vm_ssize) + grow_amount)) {
4485 			PROC_UNLOCK(p);
4486 			return (KERN_NO_SPACE);
4487 		}
4488 		PROC_UNLOCK(p);
4489 	}
4490 #endif
4491 
4492 	grow_amount = roundup(grow_amount, sgrowsiz);
4493 	if (grow_amount > max_grow)
4494 		grow_amount = max_grow;
4495 	if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > stacklim)) {
4496 		grow_amount = trunc_page((vm_size_t)stacklim) -
4497 		    ctob(vm->vm_ssize);
4498 	}
4499 
4500 #ifdef notyet
4501 	PROC_LOCK(p);
4502 	limit = racct_get_available(p, RACCT_STACK);
4503 	PROC_UNLOCK(p);
4504 	if (is_procstack && (ctob(vm->vm_ssize) + grow_amount > limit))
4505 		grow_amount = limit - ctob(vm->vm_ssize);
4506 #endif
4507 
4508 	if (!old_mlock && (map->flags & MAP_WIREFUTURE) != 0) {
4509 		if (ptoa(pmap_wired_count(map->pmap)) + grow_amount > lmemlim) {
4510 			rv = KERN_NO_SPACE;
4511 			goto out;
4512 		}
4513 #ifdef RACCT
4514 		if (racct_enable) {
4515 			PROC_LOCK(p);
4516 			if (racct_set(p, RACCT_MEMLOCK,
4517 			    ptoa(pmap_wired_count(map->pmap)) + grow_amount)) {
4518 				PROC_UNLOCK(p);
4519 				rv = KERN_NO_SPACE;
4520 				goto out;
4521 			}
4522 			PROC_UNLOCK(p);
4523 		}
4524 #endif
4525 	}
4526 
4527 	/* If we would blow our VMEM resource limit, no go */
4528 	if (map->size + grow_amount > vmemlim) {
4529 		rv = KERN_NO_SPACE;
4530 		goto out;
4531 	}
4532 #ifdef RACCT
4533 	if (racct_enable) {
4534 		PROC_LOCK(p);
4535 		if (racct_set(p, RACCT_VMEM, map->size + grow_amount)) {
4536 			PROC_UNLOCK(p);
4537 			rv = KERN_NO_SPACE;
4538 			goto out;
4539 		}
4540 		PROC_UNLOCK(p);
4541 	}
4542 #endif
4543 
4544 	if (vm_map_lock_upgrade(map)) {
4545 		gap_entry = NULL;
4546 		vm_map_lock_read(map);
4547 		goto retry;
4548 	}
4549 
4550 	if (grow_down) {
4551 		grow_start = gap_entry->end - grow_amount;
4552 		if (gap_entry->start + grow_amount == gap_entry->end) {
4553 			gap_start = gap_entry->start;
4554 			gap_end = gap_entry->end;
4555 			vm_map_entry_delete(map, gap_entry);
4556 			gap_deleted = true;
4557 		} else {
4558 			MPASS(gap_entry->start < gap_entry->end - grow_amount);
4559 			vm_map_entry_resize(map, gap_entry, -grow_amount);
4560 			gap_deleted = false;
4561 		}
4562 		rv = vm_map_insert(map, NULL, 0, grow_start,
4563 		    grow_start + grow_amount,
4564 		    stack_entry->protection, stack_entry->max_protection,
4565 		    MAP_STACK_GROWS_DOWN);
4566 		if (rv != KERN_SUCCESS) {
4567 			if (gap_deleted) {
4568 				rv1 = vm_map_insert(map, NULL, 0, gap_start,
4569 				    gap_end, VM_PROT_NONE, VM_PROT_NONE,
4570 				    MAP_CREATE_GUARD | MAP_CREATE_STACK_GAP_DN);
4571 				MPASS(rv1 == KERN_SUCCESS);
4572 			} else
4573 				vm_map_entry_resize(map, gap_entry,
4574 				    grow_amount);
4575 		}
4576 	} else {
4577 		grow_start = stack_entry->end;
4578 		cred = stack_entry->cred;
4579 		if (cred == NULL && stack_entry->object.vm_object != NULL)
4580 			cred = stack_entry->object.vm_object->cred;
4581 		if (cred != NULL && !swap_reserve_by_cred(grow_amount, cred))
4582 			rv = KERN_NO_SPACE;
4583 		/* Grow the underlying object if applicable. */
4584 		else if (stack_entry->object.vm_object == NULL ||
4585 		    vm_object_coalesce(stack_entry->object.vm_object,
4586 		    stack_entry->offset,
4587 		    (vm_size_t)(stack_entry->end - stack_entry->start),
4588 		    grow_amount, cred != NULL)) {
4589 			if (gap_entry->start + grow_amount == gap_entry->end) {
4590 				vm_map_entry_delete(map, gap_entry);
4591 				vm_map_entry_resize(map, stack_entry,
4592 				    grow_amount);
4593 			} else {
4594 				gap_entry->start += grow_amount;
4595 				stack_entry->end += grow_amount;
4596 			}
4597 			map->size += grow_amount;
4598 			rv = KERN_SUCCESS;
4599 		} else
4600 			rv = KERN_FAILURE;
4601 	}
4602 	if (rv == KERN_SUCCESS && is_procstack)
4603 		vm->vm_ssize += btoc(grow_amount);
4604 
4605 	/*
4606 	 * Heed the MAP_WIREFUTURE flag if it was set for this process.
4607 	 */
4608 	if (rv == KERN_SUCCESS && (map->flags & MAP_WIREFUTURE) != 0) {
4609 		rv = vm_map_wire_locked(map, grow_start,
4610 		    grow_start + grow_amount,
4611 		    VM_MAP_WIRE_USER | VM_MAP_WIRE_NOHOLES);
4612 	}
4613 	vm_map_lock_downgrade(map);
4614 
4615 out:
4616 #ifdef RACCT
4617 	if (racct_enable && rv != KERN_SUCCESS) {
4618 		PROC_LOCK(p);
4619 		error = racct_set(p, RACCT_VMEM, map->size);
4620 		KASSERT(error == 0, ("decreasing RACCT_VMEM failed"));
4621 		if (!old_mlock) {
4622 			error = racct_set(p, RACCT_MEMLOCK,
4623 			    ptoa(pmap_wired_count(map->pmap)));
4624 			KASSERT(error == 0, ("decreasing RACCT_MEMLOCK failed"));
4625 		}
4626 	    	error = racct_set(p, RACCT_STACK, ctob(vm->vm_ssize));
4627 		KASSERT(error == 0, ("decreasing RACCT_STACK failed"));
4628 		PROC_UNLOCK(p);
4629 	}
4630 #endif
4631 
4632 	return (rv);
4633 }
4634 
4635 /*
4636  * Unshare the specified VM space for exec.  If other processes are
4637  * mapped to it, then create a new one.  The new vmspace is null.
4638  */
4639 int
4640 vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser)
4641 {
4642 	struct vmspace *oldvmspace = p->p_vmspace;
4643 	struct vmspace *newvmspace;
4644 
4645 	KASSERT((curthread->td_pflags & TDP_EXECVMSPC) == 0,
4646 	    ("vmspace_exec recursed"));
4647 	newvmspace = vmspace_alloc(minuser, maxuser, pmap_pinit);
4648 	if (newvmspace == NULL)
4649 		return (ENOMEM);
4650 	newvmspace->vm_swrss = oldvmspace->vm_swrss;
4651 	/*
4652 	 * This code is written like this for prototype purposes.  The
4653 	 * goal is to avoid running down the vmspace here, but let the
4654 	 * other process's that are still using the vmspace to finally
4655 	 * run it down.  Even though there is little or no chance of blocking
4656 	 * here, it is a good idea to keep this form for future mods.
4657 	 */
4658 	PROC_VMSPACE_LOCK(p);
4659 	p->p_vmspace = newvmspace;
4660 	PROC_VMSPACE_UNLOCK(p);
4661 	if (p == curthread->td_proc)
4662 		pmap_activate(curthread);
4663 	curthread->td_pflags |= TDP_EXECVMSPC;
4664 	return (0);
4665 }
4666 
4667 /*
4668  * Unshare the specified VM space for forcing COW.  This
4669  * is called by rfork, for the (RFMEM|RFPROC) == 0 case.
4670  */
4671 int
4672 vmspace_unshare(struct proc *p)
4673 {
4674 	struct vmspace *oldvmspace = p->p_vmspace;
4675 	struct vmspace *newvmspace;
4676 	vm_ooffset_t fork_charge;
4677 
4678 	if (oldvmspace->vm_refcnt == 1)
4679 		return (0);
4680 	fork_charge = 0;
4681 	newvmspace = vmspace_fork(oldvmspace, &fork_charge);
4682 	if (newvmspace == NULL)
4683 		return (ENOMEM);
4684 	if (!swap_reserve_by_cred(fork_charge, p->p_ucred)) {
4685 		vmspace_free(newvmspace);
4686 		return (ENOMEM);
4687 	}
4688 	PROC_VMSPACE_LOCK(p);
4689 	p->p_vmspace = newvmspace;
4690 	PROC_VMSPACE_UNLOCK(p);
4691 	if (p == curthread->td_proc)
4692 		pmap_activate(curthread);
4693 	vmspace_free(oldvmspace);
4694 	return (0);
4695 }
4696 
4697 /*
4698  *	vm_map_lookup:
4699  *
4700  *	Finds the VM object, offset, and
4701  *	protection for a given virtual address in the
4702  *	specified map, assuming a page fault of the
4703  *	type specified.
4704  *
4705  *	Leaves the map in question locked for read; return
4706  *	values are guaranteed until a vm_map_lookup_done
4707  *	call is performed.  Note that the map argument
4708  *	is in/out; the returned map must be used in
4709  *	the call to vm_map_lookup_done.
4710  *
4711  *	A handle (out_entry) is returned for use in
4712  *	vm_map_lookup_done, to make that fast.
4713  *
4714  *	If a lookup is requested with "write protection"
4715  *	specified, the map may be changed to perform virtual
4716  *	copying operations, although the data referenced will
4717  *	remain the same.
4718  */
4719 int
4720 vm_map_lookup(vm_map_t *var_map,		/* IN/OUT */
4721 	      vm_offset_t vaddr,
4722 	      vm_prot_t fault_typea,
4723 	      vm_map_entry_t *out_entry,	/* OUT */
4724 	      vm_object_t *object,		/* OUT */
4725 	      vm_pindex_t *pindex,		/* OUT */
4726 	      vm_prot_t *out_prot,		/* OUT */
4727 	      boolean_t *wired)			/* OUT */
4728 {
4729 	vm_map_entry_t entry;
4730 	vm_map_t map = *var_map;
4731 	vm_prot_t prot;
4732 	vm_prot_t fault_type;
4733 	vm_object_t eobject;
4734 	vm_size_t size;
4735 	struct ucred *cred;
4736 
4737 RetryLookup:
4738 
4739 	vm_map_lock_read(map);
4740 
4741 RetryLookupLocked:
4742 	/*
4743 	 * Lookup the faulting address.
4744 	 */
4745 	if (!vm_map_lookup_entry(map, vaddr, out_entry)) {
4746 		vm_map_unlock_read(map);
4747 		return (KERN_INVALID_ADDRESS);
4748 	}
4749 
4750 	entry = *out_entry;
4751 
4752 	/*
4753 	 * Handle submaps.
4754 	 */
4755 	if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
4756 		vm_map_t old_map = map;
4757 
4758 		*var_map = map = entry->object.sub_map;
4759 		vm_map_unlock_read(old_map);
4760 		goto RetryLookup;
4761 	}
4762 
4763 	/*
4764 	 * Check whether this task is allowed to have this page.
4765 	 */
4766 	prot = entry->protection;
4767 	if ((fault_typea & VM_PROT_FAULT_LOOKUP) != 0) {
4768 		fault_typea &= ~VM_PROT_FAULT_LOOKUP;
4769 		if (prot == VM_PROT_NONE && map != kernel_map &&
4770 		    (entry->eflags & MAP_ENTRY_GUARD) != 0 &&
4771 		    (entry->eflags & (MAP_ENTRY_STACK_GAP_DN |
4772 		    MAP_ENTRY_STACK_GAP_UP)) != 0 &&
4773 		    vm_map_growstack(map, vaddr, entry) == KERN_SUCCESS)
4774 			goto RetryLookupLocked;
4775 	}
4776 	fault_type = fault_typea & VM_PROT_ALL;
4777 	if ((fault_type & prot) != fault_type || prot == VM_PROT_NONE) {
4778 		vm_map_unlock_read(map);
4779 		return (KERN_PROTECTION_FAILURE);
4780 	}
4781 	KASSERT((prot & VM_PROT_WRITE) == 0 || (entry->eflags &
4782 	    (MAP_ENTRY_USER_WIRED | MAP_ENTRY_NEEDS_COPY)) !=
4783 	    (MAP_ENTRY_USER_WIRED | MAP_ENTRY_NEEDS_COPY),
4784 	    ("entry %p flags %x", entry, entry->eflags));
4785 	if ((fault_typea & VM_PROT_COPY) != 0 &&
4786 	    (entry->max_protection & VM_PROT_WRITE) == 0 &&
4787 	    (entry->eflags & MAP_ENTRY_COW) == 0) {
4788 		vm_map_unlock_read(map);
4789 		return (KERN_PROTECTION_FAILURE);
4790 	}
4791 
4792 	/*
4793 	 * If this page is not pageable, we have to get it for all possible
4794 	 * accesses.
4795 	 */
4796 	*wired = (entry->wired_count != 0);
4797 	if (*wired)
4798 		fault_type = entry->protection;
4799 	size = entry->end - entry->start;
4800 
4801 	/*
4802 	 * If the entry was copy-on-write, we either ...
4803 	 */
4804 	if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
4805 		/*
4806 		 * If we want to write the page, we may as well handle that
4807 		 * now since we've got the map locked.
4808 		 *
4809 		 * If we don't need to write the page, we just demote the
4810 		 * permissions allowed.
4811 		 */
4812 		if ((fault_type & VM_PROT_WRITE) != 0 ||
4813 		    (fault_typea & VM_PROT_COPY) != 0) {
4814 			/*
4815 			 * Make a new object, and place it in the object
4816 			 * chain.  Note that no new references have appeared
4817 			 * -- one just moved from the map to the new
4818 			 * object.
4819 			 */
4820 			if (vm_map_lock_upgrade(map))
4821 				goto RetryLookup;
4822 
4823 			if (entry->cred == NULL) {
4824 				/*
4825 				 * The debugger owner is charged for
4826 				 * the memory.
4827 				 */
4828 				cred = curthread->td_ucred;
4829 				crhold(cred);
4830 				if (!swap_reserve_by_cred(size, cred)) {
4831 					crfree(cred);
4832 					vm_map_unlock(map);
4833 					return (KERN_RESOURCE_SHORTAGE);
4834 				}
4835 				entry->cred = cred;
4836 			}
4837 			eobject = entry->object.vm_object;
4838 			vm_object_shadow(&entry->object.vm_object,
4839 			    &entry->offset, size, entry->cred, false);
4840 			if (eobject == entry->object.vm_object) {
4841 				/*
4842 				 * The object was not shadowed.
4843 				 */
4844 				swap_release_by_cred(size, entry->cred);
4845 				crfree(entry->cred);
4846 			}
4847 			entry->cred = NULL;
4848 			entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
4849 
4850 			vm_map_lock_downgrade(map);
4851 		} else {
4852 			/*
4853 			 * We're attempting to read a copy-on-write page --
4854 			 * don't allow writes.
4855 			 */
4856 			prot &= ~VM_PROT_WRITE;
4857 		}
4858 	}
4859 
4860 	/*
4861 	 * Create an object if necessary.
4862 	 */
4863 	if (entry->object.vm_object == NULL && !map->system_map) {
4864 		if (vm_map_lock_upgrade(map))
4865 			goto RetryLookup;
4866 		entry->object.vm_object = vm_object_allocate_anon(atop(size),
4867 		    NULL, entry->cred, entry->cred != NULL ? size : 0);
4868 		entry->offset = 0;
4869 		entry->cred = NULL;
4870 		vm_map_lock_downgrade(map);
4871 	}
4872 
4873 	/*
4874 	 * Return the object/offset from this entry.  If the entry was
4875 	 * copy-on-write or empty, it has been fixed up.
4876 	 */
4877 	*pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
4878 	*object = entry->object.vm_object;
4879 
4880 	*out_prot = prot;
4881 	return (KERN_SUCCESS);
4882 }
4883 
4884 /*
4885  *	vm_map_lookup_locked:
4886  *
4887  *	Lookup the faulting address.  A version of vm_map_lookup that returns
4888  *      KERN_FAILURE instead of blocking on map lock or memory allocation.
4889  */
4890 int
4891 vm_map_lookup_locked(vm_map_t *var_map,		/* IN/OUT */
4892 		     vm_offset_t vaddr,
4893 		     vm_prot_t fault_typea,
4894 		     vm_map_entry_t *out_entry,	/* OUT */
4895 		     vm_object_t *object,	/* OUT */
4896 		     vm_pindex_t *pindex,	/* OUT */
4897 		     vm_prot_t *out_prot,	/* OUT */
4898 		     boolean_t *wired)		/* OUT */
4899 {
4900 	vm_map_entry_t entry;
4901 	vm_map_t map = *var_map;
4902 	vm_prot_t prot;
4903 	vm_prot_t fault_type = fault_typea;
4904 
4905 	/*
4906 	 * Lookup the faulting address.
4907 	 */
4908 	if (!vm_map_lookup_entry(map, vaddr, out_entry))
4909 		return (KERN_INVALID_ADDRESS);
4910 
4911 	entry = *out_entry;
4912 
4913 	/*
4914 	 * Fail if the entry refers to a submap.
4915 	 */
4916 	if (entry->eflags & MAP_ENTRY_IS_SUB_MAP)
4917 		return (KERN_FAILURE);
4918 
4919 	/*
4920 	 * Check whether this task is allowed to have this page.
4921 	 */
4922 	prot = entry->protection;
4923 	fault_type &= VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXECUTE;
4924 	if ((fault_type & prot) != fault_type)
4925 		return (KERN_PROTECTION_FAILURE);
4926 
4927 	/*
4928 	 * If this page is not pageable, we have to get it for all possible
4929 	 * accesses.
4930 	 */
4931 	*wired = (entry->wired_count != 0);
4932 	if (*wired)
4933 		fault_type = entry->protection;
4934 
4935 	if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
4936 		/*
4937 		 * Fail if the entry was copy-on-write for a write fault.
4938 		 */
4939 		if (fault_type & VM_PROT_WRITE)
4940 			return (KERN_FAILURE);
4941 		/*
4942 		 * We're attempting to read a copy-on-write page --
4943 		 * don't allow writes.
4944 		 */
4945 		prot &= ~VM_PROT_WRITE;
4946 	}
4947 
4948 	/*
4949 	 * Fail if an object should be created.
4950 	 */
4951 	if (entry->object.vm_object == NULL && !map->system_map)
4952 		return (KERN_FAILURE);
4953 
4954 	/*
4955 	 * Return the object/offset from this entry.  If the entry was
4956 	 * copy-on-write or empty, it has been fixed up.
4957 	 */
4958 	*pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
4959 	*object = entry->object.vm_object;
4960 
4961 	*out_prot = prot;
4962 	return (KERN_SUCCESS);
4963 }
4964 
4965 /*
4966  *	vm_map_lookup_done:
4967  *
4968  *	Releases locks acquired by a vm_map_lookup
4969  *	(according to the handle returned by that lookup).
4970  */
4971 void
4972 vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry)
4973 {
4974 	/*
4975 	 * Unlock the main-level map
4976 	 */
4977 	vm_map_unlock_read(map);
4978 }
4979 
4980 vm_offset_t
4981 vm_map_max_KBI(const struct vm_map *map)
4982 {
4983 
4984 	return (vm_map_max(map));
4985 }
4986 
4987 vm_offset_t
4988 vm_map_min_KBI(const struct vm_map *map)
4989 {
4990 
4991 	return (vm_map_min(map));
4992 }
4993 
4994 pmap_t
4995 vm_map_pmap_KBI(vm_map_t map)
4996 {
4997 
4998 	return (map->pmap);
4999 }
5000 
5001 #ifdef INVARIANTS
5002 static void
5003 _vm_map_assert_consistent(vm_map_t map, int check)
5004 {
5005 	vm_map_entry_t entry, prev;
5006 	vm_map_entry_t cur, header, lbound, ubound;
5007 	vm_size_t max_left, max_right;
5008 
5009 #ifdef DIAGNOSTIC
5010 	++map->nupdates;
5011 #endif
5012 	if (enable_vmmap_check != check)
5013 		return;
5014 
5015 	header = prev = &map->header;
5016 	VM_MAP_ENTRY_FOREACH(entry, map) {
5017 		KASSERT(prev->end <= entry->start,
5018 		    ("map %p prev->end = %jx, start = %jx", map,
5019 		    (uintmax_t)prev->end, (uintmax_t)entry->start));
5020 		KASSERT(entry->start < entry->end,
5021 		    ("map %p start = %jx, end = %jx", map,
5022 		    (uintmax_t)entry->start, (uintmax_t)entry->end));
5023 		KASSERT(entry->left == header ||
5024 		    entry->left->start < entry->start,
5025 		    ("map %p left->start = %jx, start = %jx", map,
5026 		    (uintmax_t)entry->left->start, (uintmax_t)entry->start));
5027 		KASSERT(entry->right == header ||
5028 		    entry->start < entry->right->start,
5029 		    ("map %p start = %jx, right->start = %jx", map,
5030 		    (uintmax_t)entry->start, (uintmax_t)entry->right->start));
5031 		cur = map->root;
5032 		lbound = ubound = header;
5033 		for (;;) {
5034 			if (entry->start < cur->start) {
5035 				ubound = cur;
5036 				cur = cur->left;
5037 				KASSERT(cur != lbound,
5038 				    ("map %p cannot find %jx",
5039 				    map, (uintmax_t)entry->start));
5040 			} else if (cur->end <= entry->start) {
5041 				lbound = cur;
5042 				cur = cur->right;
5043 				KASSERT(cur != ubound,
5044 				    ("map %p cannot find %jx",
5045 				    map, (uintmax_t)entry->start));
5046 			} else {
5047 				KASSERT(cur == entry,
5048 				    ("map %p cannot find %jx",
5049 				    map, (uintmax_t)entry->start));
5050 				break;
5051 			}
5052 		}
5053 		max_left = vm_map_entry_max_free_left(entry, lbound);
5054 		max_right = vm_map_entry_max_free_right(entry, ubound);
5055 		KASSERT(entry->max_free == vm_size_max(max_left, max_right),
5056 		    ("map %p max = %jx, max_left = %jx, max_right = %jx", map,
5057 		    (uintmax_t)entry->max_free,
5058 		    (uintmax_t)max_left, (uintmax_t)max_right));
5059 		prev = entry;
5060 	}
5061 	KASSERT(prev->end <= entry->start,
5062 	    ("map %p prev->end = %jx, start = %jx", map,
5063 	    (uintmax_t)prev->end, (uintmax_t)entry->start));
5064 }
5065 #endif
5066 
5067 #include "opt_ddb.h"
5068 #ifdef DDB
5069 #include <sys/kernel.h>
5070 
5071 #include <ddb/ddb.h>
5072 
5073 static void
5074 vm_map_print(vm_map_t map)
5075 {
5076 	vm_map_entry_t entry, prev;
5077 
5078 	db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n",
5079 	    (void *)map,
5080 	    (void *)map->pmap, map->nentries, map->timestamp);
5081 
5082 	db_indent += 2;
5083 	prev = &map->header;
5084 	VM_MAP_ENTRY_FOREACH(entry, map) {
5085 		db_iprintf("map entry %p: start=%p, end=%p, eflags=%#x, \n",
5086 		    (void *)entry, (void *)entry->start, (void *)entry->end,
5087 		    entry->eflags);
5088 		{
5089 			static const char * const inheritance_name[4] =
5090 			{"share", "copy", "none", "donate_copy"};
5091 
5092 			db_iprintf(" prot=%x/%x/%s",
5093 			    entry->protection,
5094 			    entry->max_protection,
5095 			    inheritance_name[(int)(unsigned char)
5096 			    entry->inheritance]);
5097 			if (entry->wired_count != 0)
5098 				db_printf(", wired");
5099 		}
5100 		if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
5101 			db_printf(", share=%p, offset=0x%jx\n",
5102 			    (void *)entry->object.sub_map,
5103 			    (uintmax_t)entry->offset);
5104 			if (prev == &map->header ||
5105 			    prev->object.sub_map !=
5106 				entry->object.sub_map) {
5107 				db_indent += 2;
5108 				vm_map_print((vm_map_t)entry->object.sub_map);
5109 				db_indent -= 2;
5110 			}
5111 		} else {
5112 			if (entry->cred != NULL)
5113 				db_printf(", ruid %d", entry->cred->cr_ruid);
5114 			db_printf(", object=%p, offset=0x%jx",
5115 			    (void *)entry->object.vm_object,
5116 			    (uintmax_t)entry->offset);
5117 			if (entry->object.vm_object && entry->object.vm_object->cred)
5118 				db_printf(", obj ruid %d charge %jx",
5119 				    entry->object.vm_object->cred->cr_ruid,
5120 				    (uintmax_t)entry->object.vm_object->charge);
5121 			if (entry->eflags & MAP_ENTRY_COW)
5122 				db_printf(", copy (%s)",
5123 				    (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done");
5124 			db_printf("\n");
5125 
5126 			if (prev == &map->header ||
5127 			    prev->object.vm_object !=
5128 				entry->object.vm_object) {
5129 				db_indent += 2;
5130 				vm_object_print((db_expr_t)(intptr_t)
5131 						entry->object.vm_object,
5132 						0, 0, (char *)0);
5133 				db_indent -= 2;
5134 			}
5135 		}
5136 		prev = entry;
5137 	}
5138 	db_indent -= 2;
5139 }
5140 
5141 DB_SHOW_COMMAND(map, map)
5142 {
5143 
5144 	if (!have_addr) {
5145 		db_printf("usage: show map <addr>\n");
5146 		return;
5147 	}
5148 	vm_map_print((vm_map_t)addr);
5149 }
5150 
5151 DB_SHOW_COMMAND(procvm, procvm)
5152 {
5153 	struct proc *p;
5154 
5155 	if (have_addr) {
5156 		p = db_lookup_proc(addr);
5157 	} else {
5158 		p = curproc;
5159 	}
5160 
5161 	db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n",
5162 	    (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map,
5163 	    (void *)vmspace_pmap(p->p_vmspace));
5164 
5165 	vm_map_print((vm_map_t)&p->p_vmspace->vm_map);
5166 }
5167 
5168 #endif /* DDB */
5169