xref: /dflybsd-src/sys/vm/vm_map.c (revision a9656fbcd49c376aba5e04370d8b0f1fa96e063c)
1 /*
2  * (MPSAFE)
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. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *	This product includes software developed by the University of
21  *	California, Berkeley and its contributors.
22  * 4. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	from: @(#)vm_map.c	8.3 (Berkeley) 1/12/94
39  *
40  *
41  * Copyright (c) 1987, 1990 Carnegie-Mellon University.
42  * All rights reserved.
43  *
44  * Authors: Avadis Tevanian, Jr., Michael Wayne Young
45  *
46  * Permission to use, copy, modify and distribute this software and
47  * its documentation is hereby granted, provided that both the copyright
48  * notice and this permission notice appear in all copies of the
49  * software, derivative works or modified versions, and any portions
50  * thereof, and that both notices appear in supporting documentation.
51  *
52  * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
53  * CONDITION.  CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
54  * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
55  *
56  * Carnegie Mellon requests users of this software to return to
57  *
58  *  Software Distribution Coordinator  or  Software.Distribution@CS.CMU.EDU
59  *  School of Computer Science
60  *  Carnegie Mellon University
61  *  Pittsburgh PA 15213-3890
62  *
63  * any improvements or extensions that they make and grant Carnegie the
64  * rights to redistribute these changes.
65  *
66  * $FreeBSD: src/sys/vm/vm_map.c,v 1.187.2.19 2003/05/27 00:47:02 alc Exp $
67  * $DragonFly: src/sys/vm/vm_map.c,v 1.56 2007/04/29 18:25:41 dillon Exp $
68  */
69 
70 /*
71  *	Virtual memory mapping module.
72  */
73 
74 #include <sys/param.h>
75 #include <sys/systm.h>
76 #include <sys/kernel.h>
77 #include <sys/proc.h>
78 #include <sys/lock.h>
79 #include <sys/vmmeter.h>
80 #include <sys/mman.h>
81 #include <sys/vnode.h>
82 #include <sys/resourcevar.h>
83 #include <sys/shm.h>
84 #include <sys/tree.h>
85 #include <sys/malloc.h>
86 
87 #include <vm/vm.h>
88 #include <vm/vm_param.h>
89 #include <vm/pmap.h>
90 #include <vm/vm_map.h>
91 #include <vm/vm_page.h>
92 #include <vm/vm_object.h>
93 #include <vm/vm_pager.h>
94 #include <vm/vm_kern.h>
95 #include <vm/vm_extern.h>
96 #include <vm/swap_pager.h>
97 #include <vm/vm_zone.h>
98 
99 #include <sys/thread2.h>
100 #include <sys/sysref2.h>
101 
102 /*
103  * Virtual memory maps provide for the mapping, protection, and sharing
104  * of virtual memory objects.  In addition, this module provides for an
105  * efficient virtual copy of memory from one map to another.
106  *
107  * Synchronization is required prior to most operations.
108  *
109  * Maps consist of an ordered doubly-linked list of simple entries.
110  * A hint and a RB tree is used to speed-up lookups.
111  *
112  * Callers looking to modify maps specify start/end addresses which cause
113  * the related map entry to be clipped if necessary, and then later
114  * recombined if the pieces remained compatible.
115  *
116  * Virtual copy operations are performed by copying VM object references
117  * from one map to another, and then marking both regions as copy-on-write.
118  */
119 static void vmspace_terminate(struct vmspace *vm);
120 static void vmspace_lock(struct vmspace *vm);
121 static void vmspace_unlock(struct vmspace *vm);
122 static void vmspace_dtor(void *obj, void *private);
123 
124 MALLOC_DEFINE(M_VMSPACE, "vmspace", "vmspace objcache backingstore");
125 
126 struct sysref_class vmspace_sysref_class = {
127 	.name =		"vmspace",
128 	.mtype =	M_VMSPACE,
129 	.proto =	SYSREF_PROTO_VMSPACE,
130 	.offset =	offsetof(struct vmspace, vm_sysref),
131 	.objsize =	sizeof(struct vmspace),
132 	.mag_capacity =	32,
133 	.flags = SRC_MANAGEDINIT,
134 	.dtor = vmspace_dtor,
135 	.ops = {
136 		.terminate = (sysref_terminate_func_t)vmspace_terminate,
137 		.lock = (sysref_lock_func_t)vmspace_lock,
138 		.unlock = (sysref_lock_func_t)vmspace_unlock
139 	}
140 };
141 
142 #define VMEPERCPU	2
143 
144 static struct vm_zone mapentzone_store, mapzone_store;
145 static vm_zone_t mapentzone, mapzone;
146 static struct vm_object mapentobj, mapobj;
147 
148 static struct vm_map_entry map_entry_init[MAX_MAPENT];
149 static struct vm_map_entry cpu_map_entry_init[MAXCPU][VMEPERCPU];
150 static struct vm_map map_init[MAX_KMAP];
151 
152 static void vm_map_entry_shadow(vm_map_entry_t entry);
153 static vm_map_entry_t vm_map_entry_create(vm_map_t map, int *);
154 static void vm_map_entry_dispose (vm_map_t map, vm_map_entry_t entry, int *);
155 static void _vm_map_clip_end (vm_map_t, vm_map_entry_t, vm_offset_t, int *);
156 static void _vm_map_clip_start (vm_map_t, vm_map_entry_t, vm_offset_t, int *);
157 static void vm_map_entry_delete (vm_map_t, vm_map_entry_t, int *);
158 static void vm_map_entry_unwire (vm_map_t, vm_map_entry_t);
159 static void vm_map_copy_entry (vm_map_t, vm_map_t, vm_map_entry_t,
160 		vm_map_entry_t);
161 static void vm_map_split (vm_map_entry_t);
162 static void vm_map_unclip_range (vm_map_t map, vm_map_entry_t start_entry, vm_offset_t start, vm_offset_t end, int *count, int flags);
163 
164 /*
165  * Initialize the vm_map module.  Must be called before any other vm_map
166  * routines.
167  *
168  * Map and entry structures are allocated from the general purpose
169  * memory pool with some exceptions:
170  *
171  *	- The kernel map is allocated statically.
172  *	- Initial kernel map entries are allocated out of a static pool.
173  *
174  *	These restrictions are necessary since malloc() uses the
175  *	maps and requires map entries.
176  *
177  * Called from the low level boot code only.
178  */
179 void
180 vm_map_startup(void)
181 {
182 	mapzone = &mapzone_store;
183 	zbootinit(mapzone, "MAP", sizeof (struct vm_map),
184 		map_init, MAX_KMAP);
185 	mapentzone = &mapentzone_store;
186 	zbootinit(mapentzone, "MAP ENTRY", sizeof (struct vm_map_entry),
187 		map_entry_init, MAX_MAPENT);
188 }
189 
190 /*
191  * Called prior to any vmspace allocations.
192  *
193  * Called from the low level boot code only.
194  */
195 void
196 vm_init2(void)
197 {
198 	zinitna(mapentzone, &mapentobj, NULL, 0, 0,
199 		ZONE_USE_RESERVE | ZONE_SPECIAL, 1);
200 	zinitna(mapzone, &mapobj, NULL, 0, 0, 0, 1);
201 	pmap_init2();
202 	vm_object_init2();
203 }
204 
205 
206 /*
207  * Red black tree functions
208  *
209  * The caller must hold the related map lock.
210  */
211 static int rb_vm_map_compare(vm_map_entry_t a, vm_map_entry_t b);
212 RB_GENERATE(vm_map_rb_tree, vm_map_entry, rb_entry, rb_vm_map_compare);
213 
214 /* a->start is address, and the only field has to be initialized */
215 static int
216 rb_vm_map_compare(vm_map_entry_t a, vm_map_entry_t b)
217 {
218 	if (a->start < b->start)
219 		return(-1);
220 	else if (a->start > b->start)
221 		return(1);
222 	return(0);
223 }
224 
225 /*
226  * Allocate a vmspace structure, including a vm_map and pmap.
227  * Initialize numerous fields.  While the initial allocation is zerod,
228  * subsequence reuse from the objcache leaves elements of the structure
229  * intact (particularly the pmap), so portions must be zerod.
230  *
231  * The structure is not considered activated until we call sysref_activate().
232  *
233  * No requirements.
234  */
235 struct vmspace *
236 vmspace_alloc(vm_offset_t min, vm_offset_t max)
237 {
238 	struct vmspace *vm;
239 
240 	lwkt_gettoken(&vmspace_token);
241 	vm = sysref_alloc(&vmspace_sysref_class);
242 	bzero(&vm->vm_startcopy,
243 	      (char *)&vm->vm_endcopy - (char *)&vm->vm_startcopy);
244 	vm_map_init(&vm->vm_map, min, max, NULL);
245 	pmap_pinit(vmspace_pmap(vm));		/* (some fields reused) */
246 	vm->vm_map.pmap = vmspace_pmap(vm);		/* XXX */
247 	vm->vm_shm = NULL;
248 	vm->vm_exitingcnt = 0;
249 	cpu_vmspace_alloc(vm);
250 	sysref_activate(&vm->vm_sysref);
251 	lwkt_reltoken(&vmspace_token);
252 
253 	return (vm);
254 }
255 
256 /*
257  * dtor function - Some elements of the pmap are retained in the
258  * free-cached vmspaces to improve performance.  We have to clean them up
259  * here before returning the vmspace to the memory pool.
260  *
261  * No requirements.
262  */
263 static void
264 vmspace_dtor(void *obj, void *private)
265 {
266 	struct vmspace *vm = obj;
267 
268 	pmap_puninit(vmspace_pmap(vm));
269 }
270 
271 /*
272  * Called in two cases:
273  *
274  * (1) When the last sysref is dropped, but exitingcnt might still be
275  *     non-zero.
276  *
277  * (2) When there are no sysrefs (i.e. refcnt is negative) left and the
278  *     exitingcnt becomes zero
279  *
280  * sysref will not scrap the object until we call sysref_put() once more
281  * after the last ref has been dropped.
282  *
283  * Interlocked by the sysref API.
284  */
285 static void
286 vmspace_terminate(struct vmspace *vm)
287 {
288 	int count;
289 
290 	/*
291 	 * If exitingcnt is non-zero we can't get rid of the entire vmspace
292 	 * yet, but we can scrap user memory.
293 	 */
294 	lwkt_gettoken(&vmspace_token);
295 	if (vm->vm_exitingcnt) {
296 		shmexit(vm);
297 		pmap_remove_pages(vmspace_pmap(vm), VM_MIN_USER_ADDRESS,
298 				  VM_MAX_USER_ADDRESS);
299 		vm_map_remove(&vm->vm_map, VM_MIN_USER_ADDRESS,
300 			      VM_MAX_USER_ADDRESS);
301 		lwkt_reltoken(&vmspace_token);
302 		return;
303 	}
304 	cpu_vmspace_free(vm);
305 
306 	/*
307 	 * Make sure any SysV shm is freed, it might not have in
308 	 * exit1()
309 	 */
310 	shmexit(vm);
311 
312 	KKASSERT(vm->vm_upcalls == NULL);
313 
314 	/*
315 	 * Lock the map, to wait out all other references to it.
316 	 * Delete all of the mappings and pages they hold, then call
317 	 * the pmap module to reclaim anything left.
318 	 */
319 	count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
320 	vm_map_lock(&vm->vm_map);
321 	vm_map_delete(&vm->vm_map, vm->vm_map.min_offset,
322 		vm->vm_map.max_offset, &count);
323 	vm_map_unlock(&vm->vm_map);
324 	vm_map_entry_release(count);
325 
326 	pmap_release(vmspace_pmap(vm));
327 	sysref_put(&vm->vm_sysref);
328 	lwkt_reltoken(&vmspace_token);
329 }
330 
331 /*
332  * vmspaces are not currently locked.
333  */
334 static void
335 vmspace_lock(struct vmspace *vm __unused)
336 {
337 }
338 
339 static void
340 vmspace_unlock(struct vmspace *vm __unused)
341 {
342 }
343 
344 /*
345  * This is called during exit indicating that the vmspace is no
346  * longer in used by an exiting process, but the process has not yet
347  * been cleaned up.
348  *
349  * No requirements.
350  */
351 void
352 vmspace_exitbump(struct vmspace *vm)
353 {
354 	lwkt_gettoken(&vmspace_token);
355 	++vm->vm_exitingcnt;
356 	lwkt_reltoken(&vmspace_token);
357 }
358 
359 /*
360  * This is called in the wait*() handling code.  The vmspace can be terminated
361  * after the last wait is finished using it.
362  *
363  * No requirements.
364  */
365 void
366 vmspace_exitfree(struct proc *p)
367 {
368 	struct vmspace *vm;
369 
370 	lwkt_gettoken(&vmspace_token);
371 	vm = p->p_vmspace;
372 	p->p_vmspace = NULL;
373 
374 	if (--vm->vm_exitingcnt == 0 && sysref_isinactive(&vm->vm_sysref))
375 		vmspace_terminate(vm);
376 	lwkt_reltoken(&vmspace_token);
377 }
378 
379 /*
380  * Swap useage is determined by taking the proportional swap used by
381  * VM objects backing the VM map.  To make up for fractional losses,
382  * if the VM object has any swap use at all the associated map entries
383  * count for at least 1 swap page.
384  *
385  * No requirements.
386  */
387 int
388 vmspace_swap_count(struct vmspace *vmspace)
389 {
390 	vm_map_t map = &vmspace->vm_map;
391 	vm_map_entry_t cur;
392 	vm_object_t object;
393 	int count = 0;
394 	int n;
395 
396 	lwkt_gettoken(&vmspace_token);
397 	for (cur = map->header.next; cur != &map->header; cur = cur->next) {
398 		switch(cur->maptype) {
399 		case VM_MAPTYPE_NORMAL:
400 		case VM_MAPTYPE_VPAGETABLE:
401 			if ((object = cur->object.vm_object) == NULL)
402 				break;
403 			if (object->swblock_count) {
404 				n = (cur->end - cur->start) / PAGE_SIZE;
405 				count += object->swblock_count *
406 				    SWAP_META_PAGES * n / object->size + 1;
407 			}
408 			break;
409 		default:
410 			break;
411 		}
412 	}
413 	lwkt_reltoken(&vmspace_token);
414 	return(count);
415 }
416 
417 /*
418  * Calculate the approximate number of anonymous pages in use by
419  * this vmspace.  To make up for fractional losses, we count each
420  * VM object as having at least 1 anonymous page.
421  *
422  * No requirements.
423  */
424 int
425 vmspace_anonymous_count(struct vmspace *vmspace)
426 {
427 	vm_map_t map = &vmspace->vm_map;
428 	vm_map_entry_t cur;
429 	vm_object_t object;
430 	int count = 0;
431 
432 	lwkt_gettoken(&vmspace_token);
433 	for (cur = map->header.next; cur != &map->header; cur = cur->next) {
434 		switch(cur->maptype) {
435 		case VM_MAPTYPE_NORMAL:
436 		case VM_MAPTYPE_VPAGETABLE:
437 			if ((object = cur->object.vm_object) == NULL)
438 				break;
439 			if (object->type != OBJT_DEFAULT &&
440 			    object->type != OBJT_SWAP) {
441 				break;
442 			}
443 			count += object->resident_page_count;
444 			break;
445 		default:
446 			break;
447 		}
448 	}
449 	lwkt_reltoken(&vmspace_token);
450 	return(count);
451 }
452 
453 /*
454  * Creates and returns a new empty VM map with the given physical map
455  * structure, and having the given lower and upper address bounds.
456  *
457  * No requirements.
458  */
459 vm_map_t
460 vm_map_create(vm_map_t result, pmap_t pmap, vm_offset_t min, vm_offset_t max)
461 {
462 	if (result == NULL)
463 		result = zalloc(mapzone);
464 	vm_map_init(result, min, max, pmap);
465 	return (result);
466 }
467 
468 /*
469  * Initialize an existing vm_map structure such as that in the vmspace
470  * structure.  The pmap is initialized elsewhere.
471  *
472  * No requirements.
473  */
474 void
475 vm_map_init(struct vm_map *map, vm_offset_t min, vm_offset_t max, pmap_t pmap)
476 {
477 	map->header.next = map->header.prev = &map->header;
478 	RB_INIT(&map->rb_root);
479 	map->nentries = 0;
480 	map->size = 0;
481 	map->system_map = 0;
482 	map->infork = 0;
483 	map->min_offset = min;
484 	map->max_offset = max;
485 	map->pmap = pmap;
486 	map->first_free = &map->header;
487 	map->hint = &map->header;
488 	map->timestamp = 0;
489 	lockinit(&map->lock, "thrd_sleep", 0, 0);
490 }
491 
492 /*
493  * Shadow the vm_map_entry's object.  This typically needs to be done when
494  * a write fault is taken on an entry which had previously been cloned by
495  * fork().  The shared object (which might be NULL) must become private so
496  * we add a shadow layer above it.
497  *
498  * Object allocation for anonymous mappings is defered as long as possible.
499  * When creating a shadow, however, the underlying object must be instantiated
500  * so it can be shared.
501  *
502  * If the map segment is governed by a virtual page table then it is
503  * possible to address offsets beyond the mapped area.  Just allocate
504  * a maximally sized object for this case.
505  *
506  * The vm_map must be exclusively locked.
507  * No other requirements.
508  */
509 static
510 void
511 vm_map_entry_shadow(vm_map_entry_t entry)
512 {
513 	if (entry->maptype == VM_MAPTYPE_VPAGETABLE) {
514 		vm_object_shadow(&entry->object.vm_object, &entry->offset,
515 				 0x7FFFFFFF);	/* XXX */
516 	} else {
517 		vm_object_shadow(&entry->object.vm_object, &entry->offset,
518 				 atop(entry->end - entry->start));
519 	}
520 	entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
521 }
522 
523 /*
524  * Allocate an object for a vm_map_entry.
525  *
526  * Object allocation for anonymous mappings is defered as long as possible.
527  * This function is called when we can defer no longer, generally when a map
528  * entry might be split or forked or takes a page fault.
529  *
530  * If the map segment is governed by a virtual page table then it is
531  * possible to address offsets beyond the mapped area.  Just allocate
532  * a maximally sized object for this case.
533  *
534  * The vm_map must be exclusively locked.
535  * No other requirements.
536  */
537 void
538 vm_map_entry_allocate_object(vm_map_entry_t entry)
539 {
540 	vm_object_t obj;
541 
542 	if (entry->maptype == VM_MAPTYPE_VPAGETABLE) {
543 		obj = vm_object_allocate(OBJT_DEFAULT, 0x7FFFFFFF); /* XXX */
544 	} else {
545 		obj = vm_object_allocate(OBJT_DEFAULT,
546 					 atop(entry->end - entry->start));
547 	}
548 	entry->object.vm_object = obj;
549 	entry->offset = 0;
550 }
551 
552 /*
553  * Set an initial negative count so the first attempt to reserve
554  * space preloads a bunch of vm_map_entry's for this cpu.  Also
555  * pre-allocate 2 vm_map_entries which will be needed by zalloc() to
556  * map a new page for vm_map_entry structures.  SMP systems are
557  * particularly sensitive.
558  *
559  * This routine is called in early boot so we cannot just call
560  * vm_map_entry_reserve().
561  *
562  * Called from the low level boot code only (for each cpu)
563  */
564 void
565 vm_map_entry_reserve_cpu_init(globaldata_t gd)
566 {
567 	vm_map_entry_t entry;
568 	int i;
569 
570 	gd->gd_vme_avail -= MAP_RESERVE_COUNT * 2;
571 	entry = &cpu_map_entry_init[gd->gd_cpuid][0];
572 	for (i = 0; i < VMEPERCPU; ++i, ++entry) {
573 		entry->next = gd->gd_vme_base;
574 		gd->gd_vme_base = entry;
575 	}
576 }
577 
578 /*
579  * Reserves vm_map_entry structures so code later on can manipulate
580  * map_entry structures within a locked map without blocking trying
581  * to allocate a new vm_map_entry.
582  *
583  * No requirements.
584  */
585 int
586 vm_map_entry_reserve(int count)
587 {
588 	struct globaldata *gd = mycpu;
589 	vm_map_entry_t entry;
590 
591 	/*
592 	 * Make sure we have enough structures in gd_vme_base to handle
593 	 * the reservation request.
594 	 */
595 	crit_enter();
596 	while (gd->gd_vme_avail < count) {
597 		entry = zalloc(mapentzone);
598 		entry->next = gd->gd_vme_base;
599 		gd->gd_vme_base = entry;
600 		++gd->gd_vme_avail;
601 	}
602 	gd->gd_vme_avail -= count;
603 	crit_exit();
604 
605 	return(count);
606 }
607 
608 /*
609  * Releases previously reserved vm_map_entry structures that were not
610  * used.  If we have too much junk in our per-cpu cache clean some of
611  * it out.
612  *
613  * No requirements.
614  */
615 void
616 vm_map_entry_release(int count)
617 {
618 	struct globaldata *gd = mycpu;
619 	vm_map_entry_t entry;
620 
621 	crit_enter();
622 	gd->gd_vme_avail += count;
623 	while (gd->gd_vme_avail > MAP_RESERVE_SLOP) {
624 		entry = gd->gd_vme_base;
625 		KKASSERT(entry != NULL);
626 		gd->gd_vme_base = entry->next;
627 		--gd->gd_vme_avail;
628 		crit_exit();
629 		zfree(mapentzone, entry);
630 		crit_enter();
631 	}
632 	crit_exit();
633 }
634 
635 /*
636  * Reserve map entry structures for use in kernel_map itself.  These
637  * entries have *ALREADY* been reserved on a per-cpu basis when the map
638  * was inited.  This function is used by zalloc() to avoid a recursion
639  * when zalloc() itself needs to allocate additional kernel memory.
640  *
641  * This function works like the normal reserve but does not load the
642  * vm_map_entry cache (because that would result in an infinite
643  * recursion).  Note that gd_vme_avail may go negative.  This is expected.
644  *
645  * Any caller of this function must be sure to renormalize after
646  * potentially eating entries to ensure that the reserve supply
647  * remains intact.
648  *
649  * No requirements.
650  */
651 int
652 vm_map_entry_kreserve(int count)
653 {
654 	struct globaldata *gd = mycpu;
655 
656 	crit_enter();
657 	gd->gd_vme_avail -= count;
658 	crit_exit();
659 	KASSERT(gd->gd_vme_base != NULL,
660 		("no reserved entries left, gd_vme_avail = %d\n",
661 		gd->gd_vme_avail));
662 	return(count);
663 }
664 
665 /*
666  * Release previously reserved map entries for kernel_map.  We do not
667  * attempt to clean up like the normal release function as this would
668  * cause an unnecessary (but probably not fatal) deep procedure call.
669  *
670  * No requirements.
671  */
672 void
673 vm_map_entry_krelease(int count)
674 {
675 	struct globaldata *gd = mycpu;
676 
677 	crit_enter();
678 	gd->gd_vme_avail += count;
679 	crit_exit();
680 }
681 
682 /*
683  * Allocates a VM map entry for insertion.  No entry fields are filled in.
684  *
685  * The entries should have previously been reserved.  The reservation count
686  * is tracked in (*countp).
687  *
688  * No requirements.
689  */
690 static vm_map_entry_t
691 vm_map_entry_create(vm_map_t map, int *countp)
692 {
693 	struct globaldata *gd = mycpu;
694 	vm_map_entry_t entry;
695 
696 	KKASSERT(*countp > 0);
697 	--*countp;
698 	crit_enter();
699 	entry = gd->gd_vme_base;
700 	KASSERT(entry != NULL, ("gd_vme_base NULL! count %d", *countp));
701 	gd->gd_vme_base = entry->next;
702 	crit_exit();
703 
704 	return(entry);
705 }
706 
707 /*
708  * Dispose of a vm_map_entry that is no longer being referenced.
709  *
710  * No requirements.
711  */
712 static void
713 vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry, int *countp)
714 {
715 	struct globaldata *gd = mycpu;
716 
717 	KKASSERT(map->hint != entry);
718 	KKASSERT(map->first_free != entry);
719 
720 	++*countp;
721 	crit_enter();
722 	entry->next = gd->gd_vme_base;
723 	gd->gd_vme_base = entry;
724 	crit_exit();
725 }
726 
727 
728 /*
729  * Insert/remove entries from maps.
730  *
731  * The related map must be exclusively locked.
732  * No other requirements.
733  *
734  * NOTE! We currently acquire the vmspace_token only to avoid races
735  *	 against the pageout daemon's calls to vmspace_*_count(), which
736  *	 are unable to safely lock the vm_map without potentially
737  *	 deadlocking.
738  */
739 static __inline void
740 vm_map_entry_link(vm_map_t map,
741 		  vm_map_entry_t after_where,
742 		  vm_map_entry_t entry)
743 {
744 	ASSERT_VM_MAP_LOCKED(map);
745 
746 	lwkt_gettoken(&vmspace_token);
747 	map->nentries++;
748 	entry->prev = after_where;
749 	entry->next = after_where->next;
750 	entry->next->prev = entry;
751 	after_where->next = entry;
752 	if (vm_map_rb_tree_RB_INSERT(&map->rb_root, entry))
753 		panic("vm_map_entry_link: dup addr map %p ent %p", map, entry);
754 	lwkt_reltoken(&vmspace_token);
755 }
756 
757 static __inline void
758 vm_map_entry_unlink(vm_map_t map,
759 		    vm_map_entry_t entry)
760 {
761 	vm_map_entry_t prev;
762 	vm_map_entry_t next;
763 
764 	ASSERT_VM_MAP_LOCKED(map);
765 
766 	if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
767 		panic("vm_map_entry_unlink: attempt to mess with "
768 		      "locked entry! %p", entry);
769 	}
770 	lwkt_gettoken(&vmspace_token);
771 	prev = entry->prev;
772 	next = entry->next;
773 	next->prev = prev;
774 	prev->next = next;
775 	vm_map_rb_tree_RB_REMOVE(&map->rb_root, entry);
776 	map->nentries--;
777 	lwkt_reltoken(&vmspace_token);
778 }
779 
780 /*
781  * Finds the map entry containing (or immediately preceding) the specified
782  * address in the given map.  The entry is returned in (*entry).
783  *
784  * The boolean result indicates whether the address is actually contained
785  * in the map.
786  *
787  * The related map must be locked.
788  * No other requirements.
789  */
790 boolean_t
791 vm_map_lookup_entry(vm_map_t map, vm_offset_t address, vm_map_entry_t *entry)
792 {
793 	vm_map_entry_t tmp;
794 	vm_map_entry_t last;
795 
796 	ASSERT_VM_MAP_LOCKED(map);
797 #if 0
798 	/*
799 	 * XXX TEMPORARILY DISABLED.  For some reason our attempt to revive
800 	 * the hint code with the red-black lookup meets with system crashes
801 	 * and lockups.  We do not yet know why.
802 	 *
803 	 * It is possible that the problem is related to the setting
804 	 * of the hint during map_entry deletion, in the code specified
805 	 * at the GGG comment later on in this file.
806 	 */
807 	/*
808 	 * Quickly check the cached hint, there's a good chance of a match.
809 	 */
810 	if (map->hint != &map->header) {
811 		tmp = map->hint;
812 		if (address >= tmp->start && address < tmp->end) {
813 			*entry = tmp;
814 			return(TRUE);
815 		}
816 	}
817 #endif
818 
819 	/*
820 	 * Locate the record from the top of the tree.  'last' tracks the
821 	 * closest prior record and is returned if no match is found, which
822 	 * in binary tree terms means tracking the most recent right-branch
823 	 * taken.  If there is no prior record, &map->header is returned.
824 	 */
825 	last = &map->header;
826 	tmp = RB_ROOT(&map->rb_root);
827 
828 	while (tmp) {
829 		if (address >= tmp->start) {
830 			if (address < tmp->end) {
831 				*entry = tmp;
832 				map->hint = tmp;
833 				return(TRUE);
834 			}
835 			last = tmp;
836 			tmp = RB_RIGHT(tmp, rb_entry);
837 		} else {
838 			tmp = RB_LEFT(tmp, rb_entry);
839 		}
840 	}
841 	*entry = last;
842 	return (FALSE);
843 }
844 
845 /*
846  * Inserts the given whole VM object into the target map at the specified
847  * address range.  The object's size should match that of the address range.
848  *
849  * The map must be exclusively locked.
850  * The caller must have reserved sufficient vm_map_entry structures.
851  *
852  * If object is non-NULL, ref count must be bumped by caller
853  * prior to making call to account for the new entry.
854  */
855 int
856 vm_map_insert(vm_map_t map, int *countp,
857 	      vm_object_t object, vm_ooffset_t offset,
858 	      vm_offset_t start, vm_offset_t end,
859 	      vm_maptype_t maptype,
860 	      vm_prot_t prot, vm_prot_t max,
861 	      int cow)
862 {
863 	vm_map_entry_t new_entry;
864 	vm_map_entry_t prev_entry;
865 	vm_map_entry_t temp_entry;
866 	vm_eflags_t protoeflags;
867 
868 	ASSERT_VM_MAP_LOCKED(map);
869 
870 	/*
871 	 * Check that the start and end points are not bogus.
872 	 */
873 	if ((start < map->min_offset) || (end > map->max_offset) ||
874 	    (start >= end))
875 		return (KERN_INVALID_ADDRESS);
876 
877 	/*
878 	 * Find the entry prior to the proposed starting address; if it's part
879 	 * of an existing entry, this range is bogus.
880 	 */
881 	if (vm_map_lookup_entry(map, start, &temp_entry))
882 		return (KERN_NO_SPACE);
883 
884 	prev_entry = temp_entry;
885 
886 	/*
887 	 * Assert that the next entry doesn't overlap the end point.
888 	 */
889 
890 	if ((prev_entry->next != &map->header) &&
891 	    (prev_entry->next->start < end))
892 		return (KERN_NO_SPACE);
893 
894 	protoeflags = 0;
895 
896 	if (cow & MAP_COPY_ON_WRITE)
897 		protoeflags |= MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY;
898 
899 	if (cow & MAP_NOFAULT) {
900 		protoeflags |= MAP_ENTRY_NOFAULT;
901 
902 		KASSERT(object == NULL,
903 			("vm_map_insert: paradoxical MAP_NOFAULT request"));
904 	}
905 	if (cow & MAP_DISABLE_SYNCER)
906 		protoeflags |= MAP_ENTRY_NOSYNC;
907 	if (cow & MAP_DISABLE_COREDUMP)
908 		protoeflags |= MAP_ENTRY_NOCOREDUMP;
909 	if (cow & MAP_IS_STACK)
910 		protoeflags |= MAP_ENTRY_STACK;
911 
912 	if (object) {
913 		/*
914 		 * When object is non-NULL, it could be shared with another
915 		 * process.  We have to set or clear OBJ_ONEMAPPING
916 		 * appropriately.
917 		 */
918 		if ((object->ref_count > 1) || (object->shadow_count != 0)) {
919 			vm_object_clear_flag(object, OBJ_ONEMAPPING);
920 		}
921 	}
922 	else if ((prev_entry != &map->header) &&
923 		 (prev_entry->eflags == protoeflags) &&
924 		 (prev_entry->end == start) &&
925 		 (prev_entry->wired_count == 0) &&
926 		 prev_entry->maptype == maptype &&
927 		 ((prev_entry->object.vm_object == NULL) ||
928 		  vm_object_coalesce(prev_entry->object.vm_object,
929 				     OFF_TO_IDX(prev_entry->offset),
930 				     (vm_size_t)(prev_entry->end - prev_entry->start),
931 				     (vm_size_t)(end - prev_entry->end)))) {
932 		/*
933 		 * We were able to extend the object.  Determine if we
934 		 * can extend the previous map entry to include the
935 		 * new range as well.
936 		 */
937 		if ((prev_entry->inheritance == VM_INHERIT_DEFAULT) &&
938 		    (prev_entry->protection == prot) &&
939 		    (prev_entry->max_protection == max)) {
940 			map->size += (end - prev_entry->end);
941 			prev_entry->end = end;
942 			vm_map_simplify_entry(map, prev_entry, countp);
943 			return (KERN_SUCCESS);
944 		}
945 
946 		/*
947 		 * If we can extend the object but cannot extend the
948 		 * map entry, we have to create a new map entry.  We
949 		 * must bump the ref count on the extended object to
950 		 * account for it.  object may be NULL.
951 		 */
952 		object = prev_entry->object.vm_object;
953 		offset = prev_entry->offset +
954 			(prev_entry->end - prev_entry->start);
955 		vm_object_reference(object);
956 	}
957 
958 	/*
959 	 * NOTE: if conditionals fail, object can be NULL here.  This occurs
960 	 * in things like the buffer map where we manage kva but do not manage
961 	 * backing objects.
962 	 */
963 
964 	/*
965 	 * Create a new entry
966 	 */
967 
968 	new_entry = vm_map_entry_create(map, countp);
969 	new_entry->start = start;
970 	new_entry->end = end;
971 
972 	new_entry->maptype = maptype;
973 	new_entry->eflags = protoeflags;
974 	new_entry->object.vm_object = object;
975 	new_entry->offset = offset;
976 	new_entry->aux.master_pde = 0;
977 
978 	new_entry->inheritance = VM_INHERIT_DEFAULT;
979 	new_entry->protection = prot;
980 	new_entry->max_protection = max;
981 	new_entry->wired_count = 0;
982 
983 	/*
984 	 * Insert the new entry into the list
985 	 */
986 
987 	vm_map_entry_link(map, prev_entry, new_entry);
988 	map->size += new_entry->end - new_entry->start;
989 
990 	/*
991 	 * Update the free space hint.  Entries cannot overlap.
992 	 * An exact comparison is needed to avoid matching
993 	 * against the map->header.
994 	 */
995 	if ((map->first_free == prev_entry) &&
996 	    (prev_entry->end == new_entry->start)) {
997 		map->first_free = new_entry;
998 	}
999 
1000 #if 0
1001 	/*
1002 	 * Temporarily removed to avoid MAP_STACK panic, due to
1003 	 * MAP_STACK being a huge hack.  Will be added back in
1004 	 * when MAP_STACK (and the user stack mapping) is fixed.
1005 	 */
1006 	/*
1007 	 * It may be possible to simplify the entry
1008 	 */
1009 	vm_map_simplify_entry(map, new_entry, countp);
1010 #endif
1011 
1012 	/*
1013 	 * Try to pre-populate the page table.  Mappings governed by virtual
1014 	 * page tables cannot be prepopulated without a lot of work, so
1015 	 * don't try.
1016 	 */
1017 	if ((cow & (MAP_PREFAULT|MAP_PREFAULT_PARTIAL)) &&
1018 	    maptype != VM_MAPTYPE_VPAGETABLE) {
1019 		pmap_object_init_pt(map->pmap, start, prot,
1020 				    object, OFF_TO_IDX(offset), end - start,
1021 				    cow & MAP_PREFAULT_PARTIAL);
1022 	}
1023 
1024 	return (KERN_SUCCESS);
1025 }
1026 
1027 /*
1028  * Find sufficient space for `length' bytes in the given map, starting at
1029  * `start'.  Returns 0 on success, 1 on no space.
1030  *
1031  * This function will returned an arbitrarily aligned pointer.  If no
1032  * particular alignment is required you should pass align as 1.  Note that
1033  * the map may return PAGE_SIZE aligned pointers if all the lengths used in
1034  * the map are a multiple of PAGE_SIZE, even if you pass a smaller align
1035  * argument.
1036  *
1037  * 'align' should be a power of 2 but is not required to be.
1038  *
1039  * The map must be exclusively locked.
1040  * No other requirements.
1041  */
1042 int
1043 vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length,
1044 		 vm_size_t align, int flags, vm_offset_t *addr)
1045 {
1046 	vm_map_entry_t entry, next;
1047 	vm_offset_t end;
1048 	vm_offset_t align_mask;
1049 
1050 	if (start < map->min_offset)
1051 		start = map->min_offset;
1052 	if (start > map->max_offset)
1053 		return (1);
1054 
1055 	/*
1056 	 * If the alignment is not a power of 2 we will have to use
1057 	 * a mod/division, set align_mask to a special value.
1058 	 */
1059 	if ((align | (align - 1)) + 1 != (align << 1))
1060 		align_mask = (vm_offset_t)-1;
1061 	else
1062 		align_mask = align - 1;
1063 
1064 retry:
1065 	/*
1066 	 * Look for the first possible address; if there's already something
1067 	 * at this address, we have to start after it.
1068 	 */
1069 	if (start == map->min_offset) {
1070 		if ((entry = map->first_free) != &map->header)
1071 			start = entry->end;
1072 	} else {
1073 		vm_map_entry_t tmp;
1074 
1075 		if (vm_map_lookup_entry(map, start, &tmp))
1076 			start = tmp->end;
1077 		entry = tmp;
1078 	}
1079 
1080 	/*
1081 	 * Look through the rest of the map, trying to fit a new region in the
1082 	 * gap between existing regions, or after the very last region.
1083 	 */
1084 	for (;; start = (entry = next)->end) {
1085 		/*
1086 		 * Adjust the proposed start by the requested alignment,
1087 		 * be sure that we didn't wrap the address.
1088 		 */
1089 		if (align_mask == (vm_offset_t)-1)
1090 			end = ((start + align - 1) / align) * align;
1091 		else
1092 			end = (start + align_mask) & ~align_mask;
1093 		if (end < start)
1094 			return (1);
1095 		start = end;
1096 		/*
1097 		 * Find the end of the proposed new region.  Be sure we didn't
1098 		 * go beyond the end of the map, or wrap around the address.
1099 		 * Then check to see if this is the last entry or if the
1100 		 * proposed end fits in the gap between this and the next
1101 		 * entry.
1102 		 */
1103 		end = start + length;
1104 		if (end > map->max_offset || end < start)
1105 			return (1);
1106 		next = entry->next;
1107 
1108 		/*
1109 		 * If the next entry's start address is beyond the desired
1110 		 * end address we may have found a good entry.
1111 		 *
1112 		 * If the next entry is a stack mapping we do not map into
1113 		 * the stack's reserved space.
1114 		 *
1115 		 * XXX continue to allow mapping into the stack's reserved
1116 		 * space if doing a MAP_STACK mapping inside a MAP_STACK
1117 		 * mapping, for backwards compatibility.  But the caller
1118 		 * really should use MAP_STACK | MAP_TRYFIXED if they
1119 		 * want to do that.
1120 		 */
1121 		if (next == &map->header)
1122 			break;
1123 		if (next->start >= end) {
1124 			if ((next->eflags & MAP_ENTRY_STACK) == 0)
1125 				break;
1126 			if (flags & MAP_STACK)
1127 				break;
1128 			if (next->start - next->aux.avail_ssize >= end)
1129 				break;
1130 		}
1131 	}
1132 	map->hint = entry;
1133 	if (map == &kernel_map) {
1134 		vm_offset_t ksize;
1135 		if ((ksize = round_page(start + length)) > kernel_vm_end) {
1136 			pmap_growkernel(ksize);
1137 			goto retry;
1138 		}
1139 	}
1140 	*addr = start;
1141 	return (0);
1142 }
1143 
1144 /*
1145  * vm_map_find finds an unallocated region in the target address map with
1146  * the given length.  The search is defined to be first-fit from the
1147  * specified address; the region found is returned in the same parameter.
1148  *
1149  * If object is non-NULL, ref count must be bumped by caller
1150  * prior to making call to account for the new entry.
1151  *
1152  * No requirements.  This function will lock the map temporarily.
1153  */
1154 int
1155 vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1156 	    vm_offset_t *addr,	vm_size_t length, vm_size_t align,
1157 	    boolean_t fitit,
1158 	    vm_maptype_t maptype,
1159 	    vm_prot_t prot, vm_prot_t max,
1160 	    int cow)
1161 {
1162 	vm_offset_t start;
1163 	int result;
1164 	int count;
1165 
1166 	start = *addr;
1167 
1168 	count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
1169 	vm_map_lock(map);
1170 	if (fitit) {
1171 		if (vm_map_findspace(map, start, length, align, 0, addr)) {
1172 			vm_map_unlock(map);
1173 			vm_map_entry_release(count);
1174 			return (KERN_NO_SPACE);
1175 		}
1176 		start = *addr;
1177 	}
1178 	result = vm_map_insert(map, &count, object, offset,
1179 			       start, start + length,
1180 			       maptype,
1181 			       prot, max,
1182 			       cow);
1183 	vm_map_unlock(map);
1184 	vm_map_entry_release(count);
1185 
1186 	return (result);
1187 }
1188 
1189 /*
1190  * Simplify the given map entry by merging with either neighbor.  This
1191  * routine also has the ability to merge with both neighbors.
1192  *
1193  * This routine guarentees that the passed entry remains valid (though
1194  * possibly extended).  When merging, this routine may delete one or
1195  * both neighbors.  No action is taken on entries which have their
1196  * in-transition flag set.
1197  *
1198  * The map must be exclusively locked.
1199  */
1200 void
1201 vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry, int *countp)
1202 {
1203 	vm_map_entry_t next, prev;
1204 	vm_size_t prevsize, esize;
1205 
1206 	if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
1207 		++mycpu->gd_cnt.v_intrans_coll;
1208 		return;
1209 	}
1210 
1211 	if (entry->maptype == VM_MAPTYPE_SUBMAP)
1212 		return;
1213 
1214 	prev = entry->prev;
1215 	if (prev != &map->header) {
1216 		prevsize = prev->end - prev->start;
1217 		if ( (prev->end == entry->start) &&
1218 		     (prev->maptype == entry->maptype) &&
1219 		     (prev->object.vm_object == entry->object.vm_object) &&
1220 		     (!prev->object.vm_object ||
1221 			(prev->offset + prevsize == entry->offset)) &&
1222 		     (prev->eflags == entry->eflags) &&
1223 		     (prev->protection == entry->protection) &&
1224 		     (prev->max_protection == entry->max_protection) &&
1225 		     (prev->inheritance == entry->inheritance) &&
1226 		     (prev->wired_count == entry->wired_count)) {
1227 			if (map->first_free == prev)
1228 				map->first_free = entry;
1229 			if (map->hint == prev)
1230 				map->hint = entry;
1231 			vm_map_entry_unlink(map, prev);
1232 			entry->start = prev->start;
1233 			entry->offset = prev->offset;
1234 			if (prev->object.vm_object)
1235 				vm_object_deallocate(prev->object.vm_object);
1236 			vm_map_entry_dispose(map, prev, countp);
1237 		}
1238 	}
1239 
1240 	next = entry->next;
1241 	if (next != &map->header) {
1242 		esize = entry->end - entry->start;
1243 		if ((entry->end == next->start) &&
1244 		    (next->maptype == entry->maptype) &&
1245 		    (next->object.vm_object == entry->object.vm_object) &&
1246 		     (!entry->object.vm_object ||
1247 			(entry->offset + esize == next->offset)) &&
1248 		    (next->eflags == entry->eflags) &&
1249 		    (next->protection == entry->protection) &&
1250 		    (next->max_protection == entry->max_protection) &&
1251 		    (next->inheritance == entry->inheritance) &&
1252 		    (next->wired_count == entry->wired_count)) {
1253 			if (map->first_free == next)
1254 				map->first_free = entry;
1255 			if (map->hint == next)
1256 				map->hint = entry;
1257 			vm_map_entry_unlink(map, next);
1258 			entry->end = next->end;
1259 			if (next->object.vm_object)
1260 				vm_object_deallocate(next->object.vm_object);
1261 			vm_map_entry_dispose(map, next, countp);
1262 	        }
1263 	}
1264 }
1265 
1266 /*
1267  * Asserts that the given entry begins at or after the specified address.
1268  * If necessary, it splits the entry into two.
1269  */
1270 #define vm_map_clip_start(map, entry, startaddr, countp)		\
1271 {									\
1272 	if (startaddr > entry->start)					\
1273 		_vm_map_clip_start(map, entry, startaddr, countp);	\
1274 }
1275 
1276 /*
1277  * This routine is called only when it is known that the entry must be split.
1278  *
1279  * The map must be exclusively locked.
1280  */
1281 static void
1282 _vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start,
1283 		   int *countp)
1284 {
1285 	vm_map_entry_t new_entry;
1286 
1287 	/*
1288 	 * Split off the front portion -- note that we must insert the new
1289 	 * entry BEFORE this one, so that this entry has the specified
1290 	 * starting address.
1291 	 */
1292 
1293 	vm_map_simplify_entry(map, entry, countp);
1294 
1295 	/*
1296 	 * If there is no object backing this entry, we might as well create
1297 	 * one now.  If we defer it, an object can get created after the map
1298 	 * is clipped, and individual objects will be created for the split-up
1299 	 * map.  This is a bit of a hack, but is also about the best place to
1300 	 * put this improvement.
1301 	 */
1302 	if (entry->object.vm_object == NULL && !map->system_map) {
1303 		vm_map_entry_allocate_object(entry);
1304 	}
1305 
1306 	new_entry = vm_map_entry_create(map, countp);
1307 	*new_entry = *entry;
1308 
1309 	new_entry->end = start;
1310 	entry->offset += (start - entry->start);
1311 	entry->start = start;
1312 
1313 	vm_map_entry_link(map, entry->prev, new_entry);
1314 
1315 	switch(entry->maptype) {
1316 	case VM_MAPTYPE_NORMAL:
1317 	case VM_MAPTYPE_VPAGETABLE:
1318 		vm_object_reference(new_entry->object.vm_object);
1319 		break;
1320 	default:
1321 		break;
1322 	}
1323 }
1324 
1325 /*
1326  * Asserts that the given entry ends at or before the specified address.
1327  * If necessary, it splits the entry into two.
1328  *
1329  * The map must be exclusively locked.
1330  */
1331 #define vm_map_clip_end(map, entry, endaddr, countp)		\
1332 {								\
1333 	if (endaddr < entry->end)				\
1334 		_vm_map_clip_end(map, entry, endaddr, countp);	\
1335 }
1336 
1337 /*
1338  * This routine is called only when it is known that the entry must be split.
1339  *
1340  * The map must be exclusively locked.
1341  */
1342 static void
1343 _vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end,
1344 		 int *countp)
1345 {
1346 	vm_map_entry_t new_entry;
1347 
1348 	/*
1349 	 * If there is no object backing this entry, we might as well create
1350 	 * one now.  If we defer it, an object can get created after the map
1351 	 * is clipped, and individual objects will be created for the split-up
1352 	 * map.  This is a bit of a hack, but is also about the best place to
1353 	 * put this improvement.
1354 	 */
1355 
1356 	if (entry->object.vm_object == NULL && !map->system_map) {
1357 		vm_map_entry_allocate_object(entry);
1358 	}
1359 
1360 	/*
1361 	 * Create a new entry and insert it AFTER the specified entry
1362 	 */
1363 
1364 	new_entry = vm_map_entry_create(map, countp);
1365 	*new_entry = *entry;
1366 
1367 	new_entry->start = entry->end = end;
1368 	new_entry->offset += (end - entry->start);
1369 
1370 	vm_map_entry_link(map, entry, new_entry);
1371 
1372 	switch(entry->maptype) {
1373 	case VM_MAPTYPE_NORMAL:
1374 	case VM_MAPTYPE_VPAGETABLE:
1375 		vm_object_reference(new_entry->object.vm_object);
1376 		break;
1377 	default:
1378 		break;
1379 	}
1380 }
1381 
1382 /*
1383  * Asserts that the starting and ending region addresses fall within the
1384  * valid range for the map.
1385  */
1386 #define	VM_MAP_RANGE_CHECK(map, start, end)	\
1387 {						\
1388 	if (start < vm_map_min(map))		\
1389 		start = vm_map_min(map);	\
1390 	if (end > vm_map_max(map))		\
1391 		end = vm_map_max(map);		\
1392 	if (start > end)			\
1393 		start = end;			\
1394 }
1395 
1396 /*
1397  * Used to block when an in-transition collison occurs.  The map
1398  * is unlocked for the sleep and relocked before the return.
1399  */
1400 static
1401 void
1402 vm_map_transition_wait(vm_map_t map)
1403 {
1404 	vm_map_unlock(map);
1405 	tsleep(map, 0, "vment", 0);
1406 	vm_map_lock(map);
1407 }
1408 
1409 /*
1410  * When we do blocking operations with the map lock held it is
1411  * possible that a clip might have occured on our in-transit entry,
1412  * requiring an adjustment to the entry in our loop.  These macros
1413  * help the pageable and clip_range code deal with the case.  The
1414  * conditional costs virtually nothing if no clipping has occured.
1415  */
1416 
1417 #define CLIP_CHECK_BACK(entry, save_start)		\
1418     do {						\
1419 	    while (entry->start != save_start) {	\
1420 		    entry = entry->prev;		\
1421 		    KASSERT(entry != &map->header, ("bad entry clip")); \
1422 	    }						\
1423     } while(0)
1424 
1425 #define CLIP_CHECK_FWD(entry, save_end)			\
1426     do {						\
1427 	    while (entry->end != save_end) {		\
1428 		    entry = entry->next;		\
1429 		    KASSERT(entry != &map->header, ("bad entry clip")); \
1430 	    }						\
1431     } while(0)
1432 
1433 
1434 /*
1435  * Clip the specified range and return the base entry.  The
1436  * range may cover several entries starting at the returned base
1437  * and the first and last entry in the covering sequence will be
1438  * properly clipped to the requested start and end address.
1439  *
1440  * If no holes are allowed you should pass the MAP_CLIP_NO_HOLES
1441  * flag.
1442  *
1443  * The MAP_ENTRY_IN_TRANSITION flag will be set for the entries
1444  * covered by the requested range.
1445  *
1446  * The map must be exclusively locked on entry and will remain locked
1447  * on return. If no range exists or the range contains holes and you
1448  * specified that no holes were allowed, NULL will be returned.  This
1449  * routine may temporarily unlock the map in order avoid a deadlock when
1450  * sleeping.
1451  */
1452 static
1453 vm_map_entry_t
1454 vm_map_clip_range(vm_map_t map, vm_offset_t start, vm_offset_t end,
1455 		  int *countp, int flags)
1456 {
1457 	vm_map_entry_t start_entry;
1458 	vm_map_entry_t entry;
1459 
1460 	/*
1461 	 * Locate the entry and effect initial clipping.  The in-transition
1462 	 * case does not occur very often so do not try to optimize it.
1463 	 */
1464 again:
1465 	if (vm_map_lookup_entry(map, start, &start_entry) == FALSE)
1466 		return (NULL);
1467 	entry = start_entry;
1468 	if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
1469 		entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
1470 		++mycpu->gd_cnt.v_intrans_coll;
1471 		++mycpu->gd_cnt.v_intrans_wait;
1472 		vm_map_transition_wait(map);
1473 		/*
1474 		 * entry and/or start_entry may have been clipped while
1475 		 * we slept, or may have gone away entirely.  We have
1476 		 * to restart from the lookup.
1477 		 */
1478 		goto again;
1479 	}
1480 
1481 	/*
1482 	 * Since we hold an exclusive map lock we do not have to restart
1483 	 * after clipping, even though clipping may block in zalloc.
1484 	 */
1485 	vm_map_clip_start(map, entry, start, countp);
1486 	vm_map_clip_end(map, entry, end, countp);
1487 	entry->eflags |= MAP_ENTRY_IN_TRANSITION;
1488 
1489 	/*
1490 	 * Scan entries covered by the range.  When working on the next
1491 	 * entry a restart need only re-loop on the current entry which
1492 	 * we have already locked, since 'next' may have changed.  Also,
1493 	 * even though entry is safe, it may have been clipped so we
1494 	 * have to iterate forwards through the clip after sleeping.
1495 	 */
1496 	while (entry->next != &map->header && entry->next->start < end) {
1497 		vm_map_entry_t next = entry->next;
1498 
1499 		if (flags & MAP_CLIP_NO_HOLES) {
1500 			if (next->start > entry->end) {
1501 				vm_map_unclip_range(map, start_entry,
1502 					start, entry->end, countp, flags);
1503 				return(NULL);
1504 			}
1505 		}
1506 
1507 		if (next->eflags & MAP_ENTRY_IN_TRANSITION) {
1508 			vm_offset_t save_end = entry->end;
1509 			next->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
1510 			++mycpu->gd_cnt.v_intrans_coll;
1511 			++mycpu->gd_cnt.v_intrans_wait;
1512 			vm_map_transition_wait(map);
1513 
1514 			/*
1515 			 * clips might have occured while we blocked.
1516 			 */
1517 			CLIP_CHECK_FWD(entry, save_end);
1518 			CLIP_CHECK_BACK(start_entry, start);
1519 			continue;
1520 		}
1521 		/*
1522 		 * No restart necessary even though clip_end may block, we
1523 		 * are holding the map lock.
1524 		 */
1525 		vm_map_clip_end(map, next, end, countp);
1526 		next->eflags |= MAP_ENTRY_IN_TRANSITION;
1527 		entry = next;
1528 	}
1529 	if (flags & MAP_CLIP_NO_HOLES) {
1530 		if (entry->end != end) {
1531 			vm_map_unclip_range(map, start_entry,
1532 				start, entry->end, countp, flags);
1533 			return(NULL);
1534 		}
1535 	}
1536 	return(start_entry);
1537 }
1538 
1539 /*
1540  * Undo the effect of vm_map_clip_range().  You should pass the same
1541  * flags and the same range that you passed to vm_map_clip_range().
1542  * This code will clear the in-transition flag on the entries and
1543  * wake up anyone waiting.  This code will also simplify the sequence
1544  * and attempt to merge it with entries before and after the sequence.
1545  *
1546  * The map must be locked on entry and will remain locked on return.
1547  *
1548  * Note that you should also pass the start_entry returned by
1549  * vm_map_clip_range().  However, if you block between the two calls
1550  * with the map unlocked please be aware that the start_entry may
1551  * have been clipped and you may need to scan it backwards to find
1552  * the entry corresponding with the original start address.  You are
1553  * responsible for this, vm_map_unclip_range() expects the correct
1554  * start_entry to be passed to it and will KASSERT otherwise.
1555  */
1556 static
1557 void
1558 vm_map_unclip_range(vm_map_t map, vm_map_entry_t start_entry,
1559 		    vm_offset_t start, vm_offset_t end,
1560 		    int *countp, int flags)
1561 {
1562 	vm_map_entry_t entry;
1563 
1564 	entry = start_entry;
1565 
1566 	KASSERT(entry->start == start, ("unclip_range: illegal base entry"));
1567 	while (entry != &map->header && entry->start < end) {
1568 		KASSERT(entry->eflags & MAP_ENTRY_IN_TRANSITION,
1569 			("in-transition flag not set during unclip on: %p",
1570 			entry));
1571 		KASSERT(entry->end <= end,
1572 			("unclip_range: tail wasn't clipped"));
1573 		entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
1574 		if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
1575 			entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
1576 			wakeup(map);
1577 		}
1578 		entry = entry->next;
1579 	}
1580 
1581 	/*
1582 	 * Simplification does not block so there is no restart case.
1583 	 */
1584 	entry = start_entry;
1585 	while (entry != &map->header && entry->start < end) {
1586 		vm_map_simplify_entry(map, entry, countp);
1587 		entry = entry->next;
1588 	}
1589 }
1590 
1591 /*
1592  * Mark the given range as handled by a subordinate map.
1593  *
1594  * This range must have been created with vm_map_find(), and no other
1595  * operations may have been performed on this range prior to calling
1596  * vm_map_submap().
1597  *
1598  * Submappings cannot be removed.
1599  *
1600  * No requirements.
1601  */
1602 int
1603 vm_map_submap(vm_map_t map, vm_offset_t start, vm_offset_t end, vm_map_t submap)
1604 {
1605 	vm_map_entry_t entry;
1606 	int result = KERN_INVALID_ARGUMENT;
1607 	int count;
1608 
1609 	count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
1610 	vm_map_lock(map);
1611 
1612 	VM_MAP_RANGE_CHECK(map, start, end);
1613 
1614 	if (vm_map_lookup_entry(map, start, &entry)) {
1615 		vm_map_clip_start(map, entry, start, &count);
1616 	} else {
1617 		entry = entry->next;
1618 	}
1619 
1620 	vm_map_clip_end(map, entry, end, &count);
1621 
1622 	if ((entry->start == start) && (entry->end == end) &&
1623 	    ((entry->eflags & MAP_ENTRY_COW) == 0) &&
1624 	    (entry->object.vm_object == NULL)) {
1625 		entry->object.sub_map = submap;
1626 		entry->maptype = VM_MAPTYPE_SUBMAP;
1627 		result = KERN_SUCCESS;
1628 	}
1629 	vm_map_unlock(map);
1630 	vm_map_entry_release(count);
1631 
1632 	return (result);
1633 }
1634 
1635 /*
1636  * Sets the protection of the specified address region in the target map.
1637  * If "set_max" is specified, the maximum protection is to be set;
1638  * otherwise, only the current protection is affected.
1639  *
1640  * The protection is not applicable to submaps, but is applicable to normal
1641  * maps and maps governed by virtual page tables.  For example, when operating
1642  * on a virtual page table our protection basically controls how COW occurs
1643  * on the backing object, whereas the virtual page table abstraction itself
1644  * is an abstraction for userland.
1645  *
1646  * No requirements.
1647  */
1648 int
1649 vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end,
1650 	       vm_prot_t new_prot, boolean_t set_max)
1651 {
1652 	vm_map_entry_t current;
1653 	vm_map_entry_t entry;
1654 	int count;
1655 
1656 	count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
1657 	vm_map_lock(map);
1658 
1659 	VM_MAP_RANGE_CHECK(map, start, end);
1660 
1661 	if (vm_map_lookup_entry(map, start, &entry)) {
1662 		vm_map_clip_start(map, entry, start, &count);
1663 	} else {
1664 		entry = entry->next;
1665 	}
1666 
1667 	/*
1668 	 * Make a first pass to check for protection violations.
1669 	 */
1670 	current = entry;
1671 	while ((current != &map->header) && (current->start < end)) {
1672 		if (current->maptype == VM_MAPTYPE_SUBMAP) {
1673 			vm_map_unlock(map);
1674 			vm_map_entry_release(count);
1675 			return (KERN_INVALID_ARGUMENT);
1676 		}
1677 		if ((new_prot & current->max_protection) != new_prot) {
1678 			vm_map_unlock(map);
1679 			vm_map_entry_release(count);
1680 			return (KERN_PROTECTION_FAILURE);
1681 		}
1682 		current = current->next;
1683 	}
1684 
1685 	/*
1686 	 * Go back and fix up protections. [Note that clipping is not
1687 	 * necessary the second time.]
1688 	 */
1689 	current = entry;
1690 
1691 	while ((current != &map->header) && (current->start < end)) {
1692 		vm_prot_t old_prot;
1693 
1694 		vm_map_clip_end(map, current, end, &count);
1695 
1696 		old_prot = current->protection;
1697 		if (set_max) {
1698 			current->protection =
1699 			    (current->max_protection = new_prot) &
1700 			    old_prot;
1701 		} else {
1702 			current->protection = new_prot;
1703 		}
1704 
1705 		/*
1706 		 * Update physical map if necessary. Worry about copy-on-write
1707 		 * here -- CHECK THIS XXX
1708 		 */
1709 
1710 		if (current->protection != old_prot) {
1711 #define MASK(entry)	(((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \
1712 							VM_PROT_ALL)
1713 
1714 			pmap_protect(map->pmap, current->start,
1715 			    current->end,
1716 			    current->protection & MASK(current));
1717 #undef	MASK
1718 		}
1719 
1720 		vm_map_simplify_entry(map, current, &count);
1721 
1722 		current = current->next;
1723 	}
1724 
1725 	vm_map_unlock(map);
1726 	vm_map_entry_release(count);
1727 	return (KERN_SUCCESS);
1728 }
1729 
1730 /*
1731  * This routine traverses a processes map handling the madvise
1732  * system call.  Advisories are classified as either those effecting
1733  * the vm_map_entry structure, or those effecting the underlying
1734  * objects.
1735  *
1736  * The <value> argument is used for extended madvise calls.
1737  *
1738  * No requirements.
1739  */
1740 int
1741 vm_map_madvise(vm_map_t map, vm_offset_t start, vm_offset_t end,
1742 	       int behav, off_t value)
1743 {
1744 	vm_map_entry_t current, entry;
1745 	int modify_map = 0;
1746 	int error = 0;
1747 	int count;
1748 
1749 	/*
1750 	 * Some madvise calls directly modify the vm_map_entry, in which case
1751 	 * we need to use an exclusive lock on the map and we need to perform
1752 	 * various clipping operations.  Otherwise we only need a read-lock
1753 	 * on the map.
1754 	 */
1755 
1756 	count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
1757 
1758 	switch(behav) {
1759 	case MADV_NORMAL:
1760 	case MADV_SEQUENTIAL:
1761 	case MADV_RANDOM:
1762 	case MADV_NOSYNC:
1763 	case MADV_AUTOSYNC:
1764 	case MADV_NOCORE:
1765 	case MADV_CORE:
1766 	case MADV_SETMAP:
1767 	case MADV_INVAL:
1768 		modify_map = 1;
1769 		vm_map_lock(map);
1770 		break;
1771 	case MADV_WILLNEED:
1772 	case MADV_DONTNEED:
1773 	case MADV_FREE:
1774 		vm_map_lock_read(map);
1775 		break;
1776 	default:
1777 		vm_map_entry_release(count);
1778 		return (EINVAL);
1779 	}
1780 
1781 	/*
1782 	 * Locate starting entry and clip if necessary.
1783 	 */
1784 
1785 	VM_MAP_RANGE_CHECK(map, start, end);
1786 
1787 	if (vm_map_lookup_entry(map, start, &entry)) {
1788 		if (modify_map)
1789 			vm_map_clip_start(map, entry, start, &count);
1790 	} else {
1791 		entry = entry->next;
1792 	}
1793 
1794 	if (modify_map) {
1795 		/*
1796 		 * madvise behaviors that are implemented in the vm_map_entry.
1797 		 *
1798 		 * We clip the vm_map_entry so that behavioral changes are
1799 		 * limited to the specified address range.
1800 		 */
1801 		for (current = entry;
1802 		     (current != &map->header) && (current->start < end);
1803 		     current = current->next
1804 		) {
1805 			if (current->maptype == VM_MAPTYPE_SUBMAP)
1806 				continue;
1807 
1808 			vm_map_clip_end(map, current, end, &count);
1809 
1810 			switch (behav) {
1811 			case MADV_NORMAL:
1812 				vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL);
1813 				break;
1814 			case MADV_SEQUENTIAL:
1815 				vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL);
1816 				break;
1817 			case MADV_RANDOM:
1818 				vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM);
1819 				break;
1820 			case MADV_NOSYNC:
1821 				current->eflags |= MAP_ENTRY_NOSYNC;
1822 				break;
1823 			case MADV_AUTOSYNC:
1824 				current->eflags &= ~MAP_ENTRY_NOSYNC;
1825 				break;
1826 			case MADV_NOCORE:
1827 				current->eflags |= MAP_ENTRY_NOCOREDUMP;
1828 				break;
1829 			case MADV_CORE:
1830 				current->eflags &= ~MAP_ENTRY_NOCOREDUMP;
1831 				break;
1832 			case MADV_INVAL:
1833 				/*
1834 				 * Invalidate the related pmap entries, used
1835 				 * to flush portions of the real kernel's
1836 				 * pmap when the caller has removed or
1837 				 * modified existing mappings in a virtual
1838 				 * page table.
1839 				 */
1840 				pmap_remove(map->pmap,
1841 					    current->start, current->end);
1842 				break;
1843 			case MADV_SETMAP:
1844 				/*
1845 				 * Set the page directory page for a map
1846 				 * governed by a virtual page table.  Mark
1847 				 * the entry as being governed by a virtual
1848 				 * page table if it is not.
1849 				 *
1850 				 * XXX the page directory page is stored
1851 				 * in the avail_ssize field if the map_entry.
1852 				 *
1853 				 * XXX the map simplification code does not
1854 				 * compare this field so weird things may
1855 				 * happen if you do not apply this function
1856 				 * to the entire mapping governed by the
1857 				 * virtual page table.
1858 				 */
1859 				if (current->maptype != VM_MAPTYPE_VPAGETABLE) {
1860 					error = EINVAL;
1861 					break;
1862 				}
1863 				current->aux.master_pde = value;
1864 				pmap_remove(map->pmap,
1865 					    current->start, current->end);
1866 				break;
1867 			default:
1868 				error = EINVAL;
1869 				break;
1870 			}
1871 			vm_map_simplify_entry(map, current, &count);
1872 		}
1873 		vm_map_unlock(map);
1874 	} else {
1875 		vm_pindex_t pindex;
1876 		int count;
1877 
1878 		/*
1879 		 * madvise behaviors that are implemented in the underlying
1880 		 * vm_object.
1881 		 *
1882 		 * Since we don't clip the vm_map_entry, we have to clip
1883 		 * the vm_object pindex and count.
1884 		 *
1885 		 * NOTE!  We currently do not support these functions on
1886 		 * virtual page tables.
1887 		 */
1888 		for (current = entry;
1889 		     (current != &map->header) && (current->start < end);
1890 		     current = current->next
1891 		) {
1892 			vm_offset_t useStart;
1893 
1894 			if (current->maptype != VM_MAPTYPE_NORMAL)
1895 				continue;
1896 
1897 			pindex = OFF_TO_IDX(current->offset);
1898 			count = atop(current->end - current->start);
1899 			useStart = current->start;
1900 
1901 			if (current->start < start) {
1902 				pindex += atop(start - current->start);
1903 				count -= atop(start - current->start);
1904 				useStart = start;
1905 			}
1906 			if (current->end > end)
1907 				count -= atop(current->end - end);
1908 
1909 			if (count <= 0)
1910 				continue;
1911 
1912 			vm_object_madvise(current->object.vm_object,
1913 					  pindex, count, behav);
1914 
1915 			/*
1916 			 * Try to populate the page table.  Mappings governed
1917 			 * by virtual page tables cannot be pre-populated
1918 			 * without a lot of work so don't try.
1919 			 */
1920 			if (behav == MADV_WILLNEED &&
1921 			    current->maptype != VM_MAPTYPE_VPAGETABLE) {
1922 				pmap_object_init_pt(
1923 				    map->pmap,
1924 				    useStart,
1925 				    current->protection,
1926 				    current->object.vm_object,
1927 				    pindex,
1928 				    (count << PAGE_SHIFT),
1929 				    MAP_PREFAULT_MADVISE
1930 				);
1931 			}
1932 		}
1933 		vm_map_unlock_read(map);
1934 	}
1935 	vm_map_entry_release(count);
1936 	return(error);
1937 }
1938 
1939 
1940 /*
1941  * Sets the inheritance of the specified address range in the target map.
1942  * Inheritance affects how the map will be shared with child maps at the
1943  * time of vm_map_fork.
1944  */
1945 int
1946 vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end,
1947 	       vm_inherit_t new_inheritance)
1948 {
1949 	vm_map_entry_t entry;
1950 	vm_map_entry_t temp_entry;
1951 	int count;
1952 
1953 	switch (new_inheritance) {
1954 	case VM_INHERIT_NONE:
1955 	case VM_INHERIT_COPY:
1956 	case VM_INHERIT_SHARE:
1957 		break;
1958 	default:
1959 		return (KERN_INVALID_ARGUMENT);
1960 	}
1961 
1962 	count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
1963 	vm_map_lock(map);
1964 
1965 	VM_MAP_RANGE_CHECK(map, start, end);
1966 
1967 	if (vm_map_lookup_entry(map, start, &temp_entry)) {
1968 		entry = temp_entry;
1969 		vm_map_clip_start(map, entry, start, &count);
1970 	} else
1971 		entry = temp_entry->next;
1972 
1973 	while ((entry != &map->header) && (entry->start < end)) {
1974 		vm_map_clip_end(map, entry, end, &count);
1975 
1976 		entry->inheritance = new_inheritance;
1977 
1978 		vm_map_simplify_entry(map, entry, &count);
1979 
1980 		entry = entry->next;
1981 	}
1982 	vm_map_unlock(map);
1983 	vm_map_entry_release(count);
1984 	return (KERN_SUCCESS);
1985 }
1986 
1987 /*
1988  * Implement the semantics of mlock
1989  */
1990 int
1991 vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t real_end,
1992 	      boolean_t new_pageable)
1993 {
1994 	vm_map_entry_t entry;
1995 	vm_map_entry_t start_entry;
1996 	vm_offset_t end;
1997 	int rv = KERN_SUCCESS;
1998 	int count;
1999 
2000 	count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
2001 	vm_map_lock(map);
2002 	VM_MAP_RANGE_CHECK(map, start, real_end);
2003 	end = real_end;
2004 
2005 	start_entry = vm_map_clip_range(map, start, end, &count,
2006 					MAP_CLIP_NO_HOLES);
2007 	if (start_entry == NULL) {
2008 		vm_map_unlock(map);
2009 		vm_map_entry_release(count);
2010 		return (KERN_INVALID_ADDRESS);
2011 	}
2012 
2013 	if (new_pageable == 0) {
2014 		entry = start_entry;
2015 		while ((entry != &map->header) && (entry->start < end)) {
2016 			vm_offset_t save_start;
2017 			vm_offset_t save_end;
2018 
2019 			/*
2020 			 * Already user wired or hard wired (trivial cases)
2021 			 */
2022 			if (entry->eflags & MAP_ENTRY_USER_WIRED) {
2023 				entry = entry->next;
2024 				continue;
2025 			}
2026 			if (entry->wired_count != 0) {
2027 				entry->wired_count++;
2028 				entry->eflags |= MAP_ENTRY_USER_WIRED;
2029 				entry = entry->next;
2030 				continue;
2031 			}
2032 
2033 			/*
2034 			 * A new wiring requires instantiation of appropriate
2035 			 * management structures and the faulting in of the
2036 			 * page.
2037 			 */
2038 			if (entry->maptype != VM_MAPTYPE_SUBMAP) {
2039 				int copyflag = entry->eflags &
2040 					       MAP_ENTRY_NEEDS_COPY;
2041 				if (copyflag && ((entry->protection &
2042 						  VM_PROT_WRITE) != 0)) {
2043 					vm_map_entry_shadow(entry);
2044 				} else if (entry->object.vm_object == NULL &&
2045 					   !map->system_map) {
2046 					vm_map_entry_allocate_object(entry);
2047 				}
2048 			}
2049 			entry->wired_count++;
2050 			entry->eflags |= MAP_ENTRY_USER_WIRED;
2051 
2052 			/*
2053 			 * Now fault in the area.  Note that vm_fault_wire()
2054 			 * may release the map lock temporarily, it will be
2055 			 * relocked on return.  The in-transition
2056 			 * flag protects the entries.
2057 			 */
2058 			save_start = entry->start;
2059 			save_end = entry->end;
2060 			rv = vm_fault_wire(map, entry, TRUE);
2061 			if (rv) {
2062 				CLIP_CHECK_BACK(entry, save_start);
2063 				for (;;) {
2064 					KASSERT(entry->wired_count == 1, ("bad wired_count on entry"));
2065 					entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2066 					entry->wired_count = 0;
2067 					if (entry->end == save_end)
2068 						break;
2069 					entry = entry->next;
2070 					KASSERT(entry != &map->header, ("bad entry clip during backout"));
2071 				}
2072 				end = save_start;	/* unwire the rest */
2073 				break;
2074 			}
2075 			/*
2076 			 * note that even though the entry might have been
2077 			 * clipped, the USER_WIRED flag we set prevents
2078 			 * duplication so we do not have to do a
2079 			 * clip check.
2080 			 */
2081 			entry = entry->next;
2082 		}
2083 
2084 		/*
2085 		 * If we failed fall through to the unwiring section to
2086 		 * unwire what we had wired so far.  'end' has already
2087 		 * been adjusted.
2088 		 */
2089 		if (rv)
2090 			new_pageable = 1;
2091 
2092 		/*
2093 		 * start_entry might have been clipped if we unlocked the
2094 		 * map and blocked.  No matter how clipped it has gotten
2095 		 * there should be a fragment that is on our start boundary.
2096 		 */
2097 		CLIP_CHECK_BACK(start_entry, start);
2098 	}
2099 
2100 	/*
2101 	 * Deal with the unwiring case.
2102 	 */
2103 	if (new_pageable) {
2104 		/*
2105 		 * This is the unwiring case.  We must first ensure that the
2106 		 * range to be unwired is really wired down.  We know there
2107 		 * are no holes.
2108 		 */
2109 		entry = start_entry;
2110 		while ((entry != &map->header) && (entry->start < end)) {
2111 			if ((entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
2112 				rv = KERN_INVALID_ARGUMENT;
2113 				goto done;
2114 			}
2115 			KASSERT(entry->wired_count != 0, ("wired count was 0 with USER_WIRED set! %p", entry));
2116 			entry = entry->next;
2117 		}
2118 
2119 		/*
2120 		 * Now decrement the wiring count for each region. If a region
2121 		 * becomes completely unwired, unwire its physical pages and
2122 		 * mappings.
2123 		 */
2124 		/*
2125 		 * The map entries are processed in a loop, checking to
2126 		 * make sure the entry is wired and asserting it has a wired
2127 		 * count. However, another loop was inserted more-or-less in
2128 		 * the middle of the unwiring path. This loop picks up the
2129 		 * "entry" loop variable from the first loop without first
2130 		 * setting it to start_entry. Naturally, the secound loop
2131 		 * is never entered and the pages backing the entries are
2132 		 * never unwired. This can lead to a leak of wired pages.
2133 		 */
2134 		entry = start_entry;
2135 		while ((entry != &map->header) && (entry->start < end)) {
2136 			KASSERT(entry->eflags & MAP_ENTRY_USER_WIRED,
2137 				("expected USER_WIRED on entry %p", entry));
2138 			entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2139 			entry->wired_count--;
2140 			if (entry->wired_count == 0)
2141 				vm_fault_unwire(map, entry);
2142 			entry = entry->next;
2143 		}
2144 	}
2145 done:
2146 	vm_map_unclip_range(map, start_entry, start, real_end, &count,
2147 		MAP_CLIP_NO_HOLES);
2148 	map->timestamp++;
2149 	vm_map_unlock(map);
2150 	vm_map_entry_release(count);
2151 	return (rv);
2152 }
2153 
2154 /*
2155  * Sets the pageability of the specified address range in the target map.
2156  * Regions specified as not pageable require locked-down physical
2157  * memory and physical page maps.
2158  *
2159  * The map must not be locked, but a reference must remain to the map
2160  * throughout the call.
2161  *
2162  * This function may be called via the zalloc path and must properly
2163  * reserve map entries for kernel_map.
2164  *
2165  * No requirements.
2166  */
2167 int
2168 vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t real_end, int kmflags)
2169 {
2170 	vm_map_entry_t entry;
2171 	vm_map_entry_t start_entry;
2172 	vm_offset_t end;
2173 	int rv = KERN_SUCCESS;
2174 	int count;
2175 
2176 	if (kmflags & KM_KRESERVE)
2177 		count = vm_map_entry_kreserve(MAP_RESERVE_COUNT);
2178 	else
2179 		count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
2180 	vm_map_lock(map);
2181 	VM_MAP_RANGE_CHECK(map, start, real_end);
2182 	end = real_end;
2183 
2184 	start_entry = vm_map_clip_range(map, start, end, &count,
2185 					MAP_CLIP_NO_HOLES);
2186 	if (start_entry == NULL) {
2187 		vm_map_unlock(map);
2188 		rv = KERN_INVALID_ADDRESS;
2189 		goto failure;
2190 	}
2191 	if ((kmflags & KM_PAGEABLE) == 0) {
2192 		/*
2193 		 * Wiring.
2194 		 *
2195 		 * 1.  Holding the write lock, we create any shadow or zero-fill
2196 		 * objects that need to be created. Then we clip each map
2197 		 * entry to the region to be wired and increment its wiring
2198 		 * count.  We create objects before clipping the map entries
2199 		 * to avoid object proliferation.
2200 		 *
2201 		 * 2.  We downgrade to a read lock, and call vm_fault_wire to
2202 		 * fault in the pages for any newly wired area (wired_count is
2203 		 * 1).
2204 		 *
2205 		 * Downgrading to a read lock for vm_fault_wire avoids a
2206 		 * possible deadlock with another process that may have faulted
2207 		 * on one of the pages to be wired (it would mark the page busy,
2208 		 * blocking us, then in turn block on the map lock that we
2209 		 * hold).  Because of problems in the recursive lock package,
2210 		 * we cannot upgrade to a write lock in vm_map_lookup.  Thus,
2211 		 * any actions that require the write lock must be done
2212 		 * beforehand.  Because we keep the read lock on the map, the
2213 		 * copy-on-write status of the entries we modify here cannot
2214 		 * change.
2215 		 */
2216 		entry = start_entry;
2217 		while ((entry != &map->header) && (entry->start < end)) {
2218 			/*
2219 			 * Trivial case if the entry is already wired
2220 			 */
2221 			if (entry->wired_count) {
2222 				entry->wired_count++;
2223 				entry = entry->next;
2224 				continue;
2225 			}
2226 
2227 			/*
2228 			 * The entry is being newly wired, we have to setup
2229 			 * appropriate management structures.  A shadow
2230 			 * object is required for a copy-on-write region,
2231 			 * or a normal object for a zero-fill region.  We
2232 			 * do not have to do this for entries that point to sub
2233 			 * maps because we won't hold the lock on the sub map.
2234 			 */
2235 			if (entry->maptype != VM_MAPTYPE_SUBMAP) {
2236 				int copyflag = entry->eflags &
2237 					       MAP_ENTRY_NEEDS_COPY;
2238 				if (copyflag && ((entry->protection &
2239 						  VM_PROT_WRITE) != 0)) {
2240 					vm_map_entry_shadow(entry);
2241 				} else if (entry->object.vm_object == NULL &&
2242 					   !map->system_map) {
2243 					vm_map_entry_allocate_object(entry);
2244 				}
2245 			}
2246 
2247 			entry->wired_count++;
2248 			entry = entry->next;
2249 		}
2250 
2251 		/*
2252 		 * Pass 2.
2253 		 */
2254 
2255 		/*
2256 		 * HACK HACK HACK HACK
2257 		 *
2258 		 * vm_fault_wire() temporarily unlocks the map to avoid
2259 		 * deadlocks.  The in-transition flag from vm_map_clip_range
2260 		 * call should protect us from changes while the map is
2261 		 * unlocked.  T
2262 		 *
2263 		 * NOTE: Previously this comment stated that clipping might
2264 		 *	 still occur while the entry is unlocked, but from
2265 		 *	 what I can tell it actually cannot.
2266 		 *
2267 		 *	 It is unclear whether the CLIP_CHECK_*() calls
2268 		 *	 are still needed but we keep them in anyway.
2269 		 *
2270 		 * HACK HACK HACK HACK
2271 		 */
2272 
2273 		entry = start_entry;
2274 		while (entry != &map->header && entry->start < end) {
2275 			/*
2276 			 * If vm_fault_wire fails for any page we need to undo
2277 			 * what has been done.  We decrement the wiring count
2278 			 * for those pages which have not yet been wired (now)
2279 			 * and unwire those that have (later).
2280 			 */
2281 			vm_offset_t save_start = entry->start;
2282 			vm_offset_t save_end = entry->end;
2283 
2284 			if (entry->wired_count == 1)
2285 				rv = vm_fault_wire(map, entry, FALSE);
2286 			if (rv) {
2287 				CLIP_CHECK_BACK(entry, save_start);
2288 				for (;;) {
2289 					KASSERT(entry->wired_count == 1, ("wired_count changed unexpectedly"));
2290 					entry->wired_count = 0;
2291 					if (entry->end == save_end)
2292 						break;
2293 					entry = entry->next;
2294 					KASSERT(entry != &map->header, ("bad entry clip during backout"));
2295 				}
2296 				end = save_start;
2297 				break;
2298 			}
2299 			CLIP_CHECK_FWD(entry, save_end);
2300 			entry = entry->next;
2301 		}
2302 
2303 		/*
2304 		 * If a failure occured undo everything by falling through
2305 		 * to the unwiring code.  'end' has already been adjusted
2306 		 * appropriately.
2307 		 */
2308 		if (rv)
2309 			kmflags |= KM_PAGEABLE;
2310 
2311 		/*
2312 		 * start_entry is still IN_TRANSITION but may have been
2313 		 * clipped since vm_fault_wire() unlocks and relocks the
2314 		 * map.  No matter how clipped it has gotten there should
2315 		 * be a fragment that is on our start boundary.
2316 		 */
2317 		CLIP_CHECK_BACK(start_entry, start);
2318 	}
2319 
2320 	if (kmflags & KM_PAGEABLE) {
2321 		/*
2322 		 * This is the unwiring case.  We must first ensure that the
2323 		 * range to be unwired is really wired down.  We know there
2324 		 * are no holes.
2325 		 */
2326 		entry = start_entry;
2327 		while ((entry != &map->header) && (entry->start < end)) {
2328 			if (entry->wired_count == 0) {
2329 				rv = KERN_INVALID_ARGUMENT;
2330 				goto done;
2331 			}
2332 			entry = entry->next;
2333 		}
2334 
2335 		/*
2336 		 * Now decrement the wiring count for each region. If a region
2337 		 * becomes completely unwired, unwire its physical pages and
2338 		 * mappings.
2339 		 */
2340 		entry = start_entry;
2341 		while ((entry != &map->header) && (entry->start < end)) {
2342 			entry->wired_count--;
2343 			if (entry->wired_count == 0)
2344 				vm_fault_unwire(map, entry);
2345 			entry = entry->next;
2346 		}
2347 	}
2348 done:
2349 	vm_map_unclip_range(map, start_entry, start, real_end,
2350 			    &count, MAP_CLIP_NO_HOLES);
2351 	map->timestamp++;
2352 	vm_map_unlock(map);
2353 failure:
2354 	if (kmflags & KM_KRESERVE)
2355 		vm_map_entry_krelease(count);
2356 	else
2357 		vm_map_entry_release(count);
2358 	return (rv);
2359 }
2360 
2361 /*
2362  * Mark a newly allocated address range as wired but do not fault in
2363  * the pages.  The caller is expected to load the pages into the object.
2364  *
2365  * The map must be locked on entry and will remain locked on return.
2366  * No other requirements.
2367  */
2368 void
2369 vm_map_set_wired_quick(vm_map_t map, vm_offset_t addr, vm_size_t size,
2370 		       int *countp)
2371 {
2372 	vm_map_entry_t scan;
2373 	vm_map_entry_t entry;
2374 
2375 	entry = vm_map_clip_range(map, addr, addr + size,
2376 				  countp, MAP_CLIP_NO_HOLES);
2377 	for (scan = entry;
2378 	     scan != &map->header && scan->start < addr + size;
2379 	     scan = scan->next) {
2380 	    KKASSERT(entry->wired_count == 0);
2381 	    entry->wired_count = 1;
2382 	}
2383 	vm_map_unclip_range(map, entry, addr, addr + size,
2384 			    countp, MAP_CLIP_NO_HOLES);
2385 }
2386 
2387 /*
2388  * Push any dirty cached pages in the address range to their pager.
2389  * If syncio is TRUE, dirty pages are written synchronously.
2390  * If invalidate is TRUE, any cached pages are freed as well.
2391  *
2392  * This routine is called by sys_msync()
2393  *
2394  * Returns an error if any part of the specified range is not mapped.
2395  *
2396  * No requirements.
2397  */
2398 int
2399 vm_map_clean(vm_map_t map, vm_offset_t start, vm_offset_t end,
2400 	     boolean_t syncio, boolean_t invalidate)
2401 {
2402 	vm_map_entry_t current;
2403 	vm_map_entry_t entry;
2404 	vm_size_t size;
2405 	vm_object_t object;
2406 	vm_ooffset_t offset;
2407 
2408 	vm_map_lock_read(map);
2409 	VM_MAP_RANGE_CHECK(map, start, end);
2410 	if (!vm_map_lookup_entry(map, start, &entry)) {
2411 		vm_map_unlock_read(map);
2412 		return (KERN_INVALID_ADDRESS);
2413 	}
2414 	/*
2415 	 * Make a first pass to check for holes.
2416 	 */
2417 	for (current = entry; current->start < end; current = current->next) {
2418 		if (current->maptype == VM_MAPTYPE_SUBMAP) {
2419 			vm_map_unlock_read(map);
2420 			return (KERN_INVALID_ARGUMENT);
2421 		}
2422 		if (end > current->end &&
2423 		    (current->next == &map->header ||
2424 			current->end != current->next->start)) {
2425 			vm_map_unlock_read(map);
2426 			return (KERN_INVALID_ADDRESS);
2427 		}
2428 	}
2429 
2430 	if (invalidate)
2431 		pmap_remove(vm_map_pmap(map), start, end);
2432 
2433 	/*
2434 	 * Make a second pass, cleaning/uncaching pages from the indicated
2435 	 * objects as we go.
2436 	 *
2437 	 * Hold vm_token to avoid blocking in vm_object_reference()
2438 	 */
2439 	lwkt_gettoken(&vm_token);
2440 	for (current = entry; current->start < end; current = current->next) {
2441 		offset = current->offset + (start - current->start);
2442 		size = (end <= current->end ? end : current->end) - start;
2443 		if (current->maptype == VM_MAPTYPE_SUBMAP) {
2444 			vm_map_t smap;
2445 			vm_map_entry_t tentry;
2446 			vm_size_t tsize;
2447 
2448 			smap = current->object.sub_map;
2449 			vm_map_lock_read(smap);
2450 			vm_map_lookup_entry(smap, offset, &tentry);
2451 			tsize = tentry->end - offset;
2452 			if (tsize < size)
2453 				size = tsize;
2454 			object = tentry->object.vm_object;
2455 			offset = tentry->offset + (offset - tentry->start);
2456 			vm_map_unlock_read(smap);
2457 		} else {
2458 			object = current->object.vm_object;
2459 		}
2460 		/*
2461 		 * Note that there is absolutely no sense in writing out
2462 		 * anonymous objects, so we track down the vnode object
2463 		 * to write out.
2464 		 * We invalidate (remove) all pages from the address space
2465 		 * anyway, for semantic correctness.
2466 		 *
2467 		 * note: certain anonymous maps, such as MAP_NOSYNC maps,
2468 		 * may start out with a NULL object.
2469 		 */
2470 		while (object && object->backing_object) {
2471 			offset += object->backing_object_offset;
2472 			object = object->backing_object;
2473 			if (object->size < OFF_TO_IDX( offset + size))
2474 				size = IDX_TO_OFF(object->size) - offset;
2475 		}
2476 		if (object && (object->type == OBJT_VNODE) &&
2477 		    (current->protection & VM_PROT_WRITE) &&
2478 		    (object->flags & OBJ_NOMSYNC) == 0) {
2479 			/*
2480 			 * Flush pages if writing is allowed, invalidate them
2481 			 * if invalidation requested.  Pages undergoing I/O
2482 			 * will be ignored by vm_object_page_remove().
2483 			 *
2484 			 * We cannot lock the vnode and then wait for paging
2485 			 * to complete without deadlocking against vm_fault.
2486 			 * Instead we simply call vm_object_page_remove() and
2487 			 * allow it to block internally on a page-by-page
2488 			 * basis when it encounters pages undergoing async
2489 			 * I/O.
2490 			 */
2491 			int flags;
2492 
2493 			vm_object_reference(object);
2494 			vn_lock(object->handle, LK_EXCLUSIVE | LK_RETRY);
2495 			flags = (syncio || invalidate) ? OBJPC_SYNC : 0;
2496 			flags |= invalidate ? OBJPC_INVAL : 0;
2497 
2498 			/*
2499 			 * When operating on a virtual page table just
2500 			 * flush the whole object.  XXX we probably ought
2501 			 * to
2502 			 */
2503 			switch(current->maptype) {
2504 			case VM_MAPTYPE_NORMAL:
2505 				vm_object_page_clean(object,
2506 				    OFF_TO_IDX(offset),
2507 				    OFF_TO_IDX(offset + size + PAGE_MASK),
2508 				    flags);
2509 				break;
2510 			case VM_MAPTYPE_VPAGETABLE:
2511 				vm_object_page_clean(object, 0, 0, flags);
2512 				break;
2513 			}
2514 			vn_unlock(((struct vnode *)object->handle));
2515 			vm_object_deallocate(object);
2516 		}
2517 		if (object && invalidate &&
2518 		   ((object->type == OBJT_VNODE) ||
2519 		    (object->type == OBJT_DEVICE))) {
2520 			int clean_only =
2521 				(object->type == OBJT_DEVICE) ? FALSE : TRUE;
2522 			vm_object_reference(object);
2523 			switch(current->maptype) {
2524 			case VM_MAPTYPE_NORMAL:
2525 				vm_object_page_remove(object,
2526 				    OFF_TO_IDX(offset),
2527 				    OFF_TO_IDX(offset + size + PAGE_MASK),
2528 				    clean_only);
2529 				break;
2530 			case VM_MAPTYPE_VPAGETABLE:
2531 				vm_object_page_remove(object, 0, 0, clean_only);
2532 				break;
2533 			}
2534 			vm_object_deallocate(object);
2535 		}
2536 		start += size;
2537 	}
2538 	vm_map_unlock_read(map);
2539 	lwkt_reltoken(&vm_token);
2540 
2541 	return (KERN_SUCCESS);
2542 }
2543 
2544 /*
2545  * Make the region specified by this entry pageable.
2546  *
2547  * The vm_map must be exclusively locked.
2548  */
2549 static void
2550 vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry)
2551 {
2552 	entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2553 	entry->wired_count = 0;
2554 	vm_fault_unwire(map, entry);
2555 }
2556 
2557 /*
2558  * Deallocate the given entry from the target map.
2559  *
2560  * The vm_map must be exclusively locked.
2561  */
2562 static void
2563 vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry, int *countp)
2564 {
2565 	vm_map_entry_unlink(map, entry);
2566 	map->size -= entry->end - entry->start;
2567 
2568 	switch(entry->maptype) {
2569 	case VM_MAPTYPE_NORMAL:
2570 	case VM_MAPTYPE_VPAGETABLE:
2571 		vm_object_deallocate(entry->object.vm_object);
2572 		break;
2573 	default:
2574 		break;
2575 	}
2576 
2577 	vm_map_entry_dispose(map, entry, countp);
2578 }
2579 
2580 /*
2581  * Deallocates the given address range from the target map.
2582  *
2583  * The vm_map must be exclusively locked.
2584  */
2585 int
2586 vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end, int *countp)
2587 {
2588 	vm_object_t object;
2589 	vm_map_entry_t entry;
2590 	vm_map_entry_t first_entry;
2591 
2592 	ASSERT_VM_MAP_LOCKED(map);
2593 again:
2594 	/*
2595 	 * Find the start of the region, and clip it.  Set entry to point
2596 	 * at the first record containing the requested address or, if no
2597 	 * such record exists, the next record with a greater address.  The
2598 	 * loop will run from this point until a record beyond the termination
2599 	 * address is encountered.
2600 	 *
2601 	 * map->hint must be adjusted to not point to anything we delete,
2602 	 * so set it to the entry prior to the one being deleted.
2603 	 *
2604 	 * GGG see other GGG comment.
2605 	 */
2606 	if (vm_map_lookup_entry(map, start, &first_entry)) {
2607 		entry = first_entry;
2608 		vm_map_clip_start(map, entry, start, countp);
2609 		map->hint = entry->prev;	/* possible problem XXX */
2610 	} else {
2611 		map->hint = first_entry;	/* possible problem XXX */
2612 		entry = first_entry->next;
2613 	}
2614 
2615 	/*
2616 	 * If a hole opens up prior to the current first_free then
2617 	 * adjust first_free.  As with map->hint, map->first_free
2618 	 * cannot be left set to anything we might delete.
2619 	 */
2620 	if (entry == &map->header) {
2621 		map->first_free = &map->header;
2622 	} else if (map->first_free->start >= start) {
2623 		map->first_free = entry->prev;
2624 	}
2625 
2626 	/*
2627 	 * Step through all entries in this region
2628 	 */
2629 	while ((entry != &map->header) && (entry->start < end)) {
2630 		vm_map_entry_t next;
2631 		vm_offset_t s, e;
2632 		vm_pindex_t offidxstart, offidxend, count;
2633 
2634 		/*
2635 		 * If we hit an in-transition entry we have to sleep and
2636 		 * retry.  It's easier (and not really slower) to just retry
2637 		 * since this case occurs so rarely and the hint is already
2638 		 * pointing at the right place.  We have to reset the
2639 		 * start offset so as not to accidently delete an entry
2640 		 * another process just created in vacated space.
2641 		 */
2642 		if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2643 			entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2644 			start = entry->start;
2645 			++mycpu->gd_cnt.v_intrans_coll;
2646 			++mycpu->gd_cnt.v_intrans_wait;
2647 			vm_map_transition_wait(map);
2648 			goto again;
2649 		}
2650 		vm_map_clip_end(map, entry, end, countp);
2651 
2652 		s = entry->start;
2653 		e = entry->end;
2654 		next = entry->next;
2655 
2656 		offidxstart = OFF_TO_IDX(entry->offset);
2657 		count = OFF_TO_IDX(e - s);
2658 		object = entry->object.vm_object;
2659 
2660 		/*
2661 		 * Unwire before removing addresses from the pmap; otherwise,
2662 		 * unwiring will put the entries back in the pmap.
2663 		 */
2664 		if (entry->wired_count != 0)
2665 			vm_map_entry_unwire(map, entry);
2666 
2667 		offidxend = offidxstart + count;
2668 
2669 		/*
2670 		 * Hold vm_token when manipulating vm_objects.
2671 		 */
2672 		lwkt_gettoken(&vm_token);
2673 		if (object == &kernel_object) {
2674 			vm_object_page_remove(object, offidxstart,
2675 					      offidxend, FALSE);
2676 		} else {
2677 			pmap_remove(map->pmap, s, e);
2678 			if (object != NULL &&
2679 			    object->ref_count != 1 &&
2680 			    (object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) ==
2681 			     OBJ_ONEMAPPING &&
2682 			    (object->type == OBJT_DEFAULT ||
2683 			     object->type == OBJT_SWAP)) {
2684 				vm_object_collapse(object);
2685 				vm_object_page_remove(object, offidxstart,
2686 						      offidxend, FALSE);
2687 				if (object->type == OBJT_SWAP) {
2688 					swap_pager_freespace(object,
2689 							     offidxstart,
2690 							     count);
2691 				}
2692 				if (offidxend >= object->size &&
2693 				    offidxstart < object->size) {
2694 					object->size = offidxstart;
2695 				}
2696 			}
2697 		}
2698 		lwkt_reltoken(&vm_token);
2699 
2700 		/*
2701 		 * Delete the entry (which may delete the object) only after
2702 		 * removing all pmap entries pointing to its pages.
2703 		 * (Otherwise, its page frames may be reallocated, and any
2704 		 * modify bits will be set in the wrong object!)
2705 		 */
2706 		vm_map_entry_delete(map, entry, countp);
2707 		entry = next;
2708 	}
2709 	return (KERN_SUCCESS);
2710 }
2711 
2712 /*
2713  * Remove the given address range from the target map.
2714  * This is the exported form of vm_map_delete.
2715  *
2716  * No requirements.
2717  */
2718 int
2719 vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end)
2720 {
2721 	int result;
2722 	int count;
2723 
2724 	count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
2725 	vm_map_lock(map);
2726 	VM_MAP_RANGE_CHECK(map, start, end);
2727 	result = vm_map_delete(map, start, end, &count);
2728 	vm_map_unlock(map);
2729 	vm_map_entry_release(count);
2730 
2731 	return (result);
2732 }
2733 
2734 /*
2735  * Assert that the target map allows the specified privilege on the
2736  * entire address region given.  The entire region must be allocated.
2737  *
2738  * The caller must specify whether the vm_map is already locked or not.
2739  */
2740 boolean_t
2741 vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end,
2742 			vm_prot_t protection, boolean_t have_lock)
2743 {
2744 	vm_map_entry_t entry;
2745 	vm_map_entry_t tmp_entry;
2746 	boolean_t result;
2747 
2748 	if (have_lock == FALSE)
2749 		vm_map_lock_read(map);
2750 
2751 	if (!vm_map_lookup_entry(map, start, &tmp_entry)) {
2752 		if (have_lock == FALSE)
2753 			vm_map_unlock_read(map);
2754 		return (FALSE);
2755 	}
2756 	entry = tmp_entry;
2757 
2758 	result = TRUE;
2759 	while (start < end) {
2760 		if (entry == &map->header) {
2761 			result = FALSE;
2762 			break;
2763 		}
2764 		/*
2765 		 * No holes allowed!
2766 		 */
2767 
2768 		if (start < entry->start) {
2769 			result = FALSE;
2770 			break;
2771 		}
2772 		/*
2773 		 * Check protection associated with entry.
2774 		 */
2775 
2776 		if ((entry->protection & protection) != protection) {
2777 			result = FALSE;
2778 			break;
2779 		}
2780 		/* go to next entry */
2781 
2782 		start = entry->end;
2783 		entry = entry->next;
2784 	}
2785 	if (have_lock == FALSE)
2786 		vm_map_unlock_read(map);
2787 	return (result);
2788 }
2789 
2790 /*
2791  * Split the pages in a map entry into a new object.  This affords
2792  * easier removal of unused pages, and keeps object inheritance from
2793  * being a negative impact on memory usage.
2794  *
2795  * The vm_map must be exclusively locked.
2796  */
2797 static void
2798 vm_map_split(vm_map_entry_t entry)
2799 {
2800 	vm_page_t m;
2801 	vm_object_t orig_object, new_object, source;
2802 	vm_offset_t s, e;
2803 	vm_pindex_t offidxstart, offidxend, idx;
2804 	vm_size_t size;
2805 	vm_ooffset_t offset;
2806 
2807 	orig_object = entry->object.vm_object;
2808 	if (orig_object->type != OBJT_DEFAULT && orig_object->type != OBJT_SWAP)
2809 		return;
2810 	if (orig_object->ref_count <= 1)
2811 		return;
2812 
2813 	offset = entry->offset;
2814 	s = entry->start;
2815 	e = entry->end;
2816 
2817 	offidxstart = OFF_TO_IDX(offset);
2818 	offidxend = offidxstart + OFF_TO_IDX(e - s);
2819 	size = offidxend - offidxstart;
2820 
2821 	switch(orig_object->type) {
2822 	case OBJT_DEFAULT:
2823 		new_object = default_pager_alloc(NULL, IDX_TO_OFF(size),
2824 						 VM_PROT_ALL, 0);
2825 		break;
2826 	case OBJT_SWAP:
2827 		new_object = swap_pager_alloc(NULL, IDX_TO_OFF(size),
2828 					      VM_PROT_ALL, 0);
2829 		break;
2830 	default:
2831 		/* not reached */
2832 		new_object = NULL;
2833 		KKASSERT(0);
2834 	}
2835 	if (new_object == NULL)
2836 		return;
2837 
2838 	/*
2839 	 * vm_token required when manipulating vm_objects.
2840 	 */
2841 	lwkt_gettoken(&vm_token);
2842 
2843 	source = orig_object->backing_object;
2844 	if (source != NULL) {
2845 		vm_object_reference(source);	/* Referenced by new_object */
2846 		LIST_INSERT_HEAD(&source->shadow_head,
2847 				  new_object, shadow_list);
2848 		vm_object_clear_flag(source, OBJ_ONEMAPPING);
2849 		new_object->backing_object_offset =
2850 			orig_object->backing_object_offset + IDX_TO_OFF(offidxstart);
2851 		new_object->backing_object = source;
2852 		source->shadow_count++;
2853 		source->generation++;
2854 	}
2855 
2856 	for (idx = 0; idx < size; idx++) {
2857 		vm_page_t m;
2858 
2859 		crit_enter();
2860 	retry:
2861 		m = vm_page_lookup(orig_object, offidxstart + idx);
2862 		if (m == NULL) {
2863 			crit_exit();
2864 			continue;
2865 		}
2866 
2867 		/*
2868 		 * We must wait for pending I/O to complete before we can
2869 		 * rename the page.
2870 		 *
2871 		 * We do not have to VM_PROT_NONE the page as mappings should
2872 		 * not be changed by this operation.
2873 		 */
2874 		if (vm_page_sleep_busy(m, TRUE, "spltwt"))
2875 			goto retry;
2876 		vm_page_busy(m);
2877 		vm_page_rename(m, new_object, idx);
2878 		/* page automatically made dirty by rename and cache handled */
2879 		vm_page_busy(m);
2880 		crit_exit();
2881 	}
2882 
2883 	if (orig_object->type == OBJT_SWAP) {
2884 		vm_object_pip_add(orig_object, 1);
2885 		/*
2886 		 * copy orig_object pages into new_object
2887 		 * and destroy unneeded pages in
2888 		 * shadow object.
2889 		 */
2890 		swap_pager_copy(orig_object, new_object, offidxstart, 0);
2891 		vm_object_pip_wakeup(orig_object);
2892 	}
2893 
2894 	/*
2895 	 * Wakeup the pages we played with.  No spl protection is needed
2896 	 * for a simple wakeup.
2897 	 */
2898 	for (idx = 0; idx < size; idx++) {
2899 		m = vm_page_lookup(new_object, idx);
2900 		if (m)
2901 			vm_page_wakeup(m);
2902 	}
2903 
2904 	entry->object.vm_object = new_object;
2905 	entry->offset = 0LL;
2906 	vm_object_deallocate(orig_object);
2907 	lwkt_reltoken(&vm_token);
2908 }
2909 
2910 /*
2911  * Copies the contents of the source entry to the destination
2912  * entry.  The entries *must* be aligned properly.
2913  *
2914  * The vm_map must be exclusively locked.
2915  */
2916 static void
2917 vm_map_copy_entry(vm_map_t src_map, vm_map_t dst_map,
2918 	vm_map_entry_t src_entry, vm_map_entry_t dst_entry)
2919 {
2920 	vm_object_t src_object;
2921 
2922 	if (dst_entry->maptype == VM_MAPTYPE_SUBMAP)
2923 		return;
2924 	if (src_entry->maptype == VM_MAPTYPE_SUBMAP)
2925 		return;
2926 
2927 	lwkt_gettoken(&vm_token);
2928 	if (src_entry->wired_count == 0) {
2929 		/*
2930 		 * If the source entry is marked needs_copy, it is already
2931 		 * write-protected.
2932 		 */
2933 		if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) {
2934 			pmap_protect(src_map->pmap,
2935 			    src_entry->start,
2936 			    src_entry->end,
2937 			    src_entry->protection & ~VM_PROT_WRITE);
2938 		}
2939 
2940 		/*
2941 		 * Make a copy of the object.
2942 		 */
2943 		if ((src_object = src_entry->object.vm_object) != NULL) {
2944 			if ((src_object->handle == NULL) &&
2945 				(src_object->type == OBJT_DEFAULT ||
2946 				 src_object->type == OBJT_SWAP)) {
2947 				vm_object_collapse(src_object);
2948 				if ((src_object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) {
2949 					vm_map_split(src_entry);
2950 					src_object = src_entry->object.vm_object;
2951 				}
2952 			}
2953 
2954 			vm_object_reference(src_object);
2955 			vm_object_clear_flag(src_object, OBJ_ONEMAPPING);
2956 			dst_entry->object.vm_object = src_object;
2957 			src_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
2958 			dst_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
2959 			dst_entry->offset = src_entry->offset;
2960 		} else {
2961 			dst_entry->object.vm_object = NULL;
2962 			dst_entry->offset = 0;
2963 		}
2964 
2965 		pmap_copy(dst_map->pmap, src_map->pmap, dst_entry->start,
2966 		    dst_entry->end - dst_entry->start, src_entry->start);
2967 	} else {
2968 		/*
2969 		 * Of course, wired down pages can't be set copy-on-write.
2970 		 * Cause wired pages to be copied into the new map by
2971 		 * simulating faults (the new pages are pageable)
2972 		 */
2973 		vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry);
2974 	}
2975 	lwkt_reltoken(&vm_token);
2976 }
2977 
2978 /*
2979  * vmspace_fork:
2980  * Create a new process vmspace structure and vm_map
2981  * based on those of an existing process.  The new map
2982  * is based on the old map, according to the inheritance
2983  * values on the regions in that map.
2984  *
2985  * The source map must not be locked.
2986  * No requirements.
2987  */
2988 struct vmspace *
2989 vmspace_fork(struct vmspace *vm1)
2990 {
2991 	struct vmspace *vm2;
2992 	vm_map_t old_map = &vm1->vm_map;
2993 	vm_map_t new_map;
2994 	vm_map_entry_t old_entry;
2995 	vm_map_entry_t new_entry;
2996 	vm_object_t object;
2997 	int count;
2998 
2999 	lwkt_gettoken(&vm_token);
3000 	lwkt_gettoken(&vmspace_token);
3001 	vm_map_lock(old_map);
3002 	old_map->infork = 1;
3003 
3004 	/*
3005 	 * XXX Note: upcalls are not copied.
3006 	 */
3007 	vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset);
3008 	bcopy(&vm1->vm_startcopy, &vm2->vm_startcopy,
3009 	    (caddr_t)&vm1->vm_endcopy - (caddr_t)&vm1->vm_startcopy);
3010 	new_map = &vm2->vm_map;	/* XXX */
3011 	new_map->timestamp = 1;
3012 
3013 	vm_map_lock(new_map);
3014 
3015 	count = 0;
3016 	old_entry = old_map->header.next;
3017 	while (old_entry != &old_map->header) {
3018 		++count;
3019 		old_entry = old_entry->next;
3020 	}
3021 
3022 	count = vm_map_entry_reserve(count + MAP_RESERVE_COUNT);
3023 
3024 	old_entry = old_map->header.next;
3025 	while (old_entry != &old_map->header) {
3026 		if (old_entry->maptype == VM_MAPTYPE_SUBMAP)
3027 			panic("vm_map_fork: encountered a submap");
3028 
3029 		switch (old_entry->inheritance) {
3030 		case VM_INHERIT_NONE:
3031 			break;
3032 		case VM_INHERIT_SHARE:
3033 			/*
3034 			 * Clone the entry, creating the shared object if
3035 			 * necessary.
3036 			 */
3037 			object = old_entry->object.vm_object;
3038 			if (object == NULL) {
3039 				vm_map_entry_allocate_object(old_entry);
3040 				object = old_entry->object.vm_object;
3041 			}
3042 
3043 			/*
3044 			 * Add the reference before calling vm_map_entry_shadow
3045 			 * to insure that a shadow object is created.
3046 			 */
3047 			vm_object_reference(object);
3048 			if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3049 				vm_map_entry_shadow(old_entry);
3050 				/* Transfer the second reference too. */
3051 				vm_object_reference(
3052 				    old_entry->object.vm_object);
3053 				vm_object_deallocate(object);
3054 				object = old_entry->object.vm_object;
3055 			}
3056 			vm_object_clear_flag(object, OBJ_ONEMAPPING);
3057 
3058 			/*
3059 			 * Clone the entry, referencing the shared object.
3060 			 */
3061 			new_entry = vm_map_entry_create(new_map, &count);
3062 			*new_entry = *old_entry;
3063 			new_entry->eflags &= ~MAP_ENTRY_USER_WIRED;
3064 			new_entry->wired_count = 0;
3065 
3066 			/*
3067 			 * Insert the entry into the new map -- we know we're
3068 			 * inserting at the end of the new map.
3069 			 */
3070 
3071 			vm_map_entry_link(new_map, new_map->header.prev,
3072 					  new_entry);
3073 
3074 			/*
3075 			 * Update the physical map
3076 			 */
3077 			pmap_copy(new_map->pmap, old_map->pmap,
3078 			    new_entry->start,
3079 			    (old_entry->end - old_entry->start),
3080 			    old_entry->start);
3081 			break;
3082 		case VM_INHERIT_COPY:
3083 			/*
3084 			 * Clone the entry and link into the map.
3085 			 */
3086 			new_entry = vm_map_entry_create(new_map, &count);
3087 			*new_entry = *old_entry;
3088 			new_entry->eflags &= ~MAP_ENTRY_USER_WIRED;
3089 			new_entry->wired_count = 0;
3090 			new_entry->object.vm_object = NULL;
3091 			vm_map_entry_link(new_map, new_map->header.prev,
3092 					  new_entry);
3093 			vm_map_copy_entry(old_map, new_map, old_entry,
3094 					  new_entry);
3095 			break;
3096 		}
3097 		old_entry = old_entry->next;
3098 	}
3099 
3100 	new_map->size = old_map->size;
3101 	old_map->infork = 0;
3102 	vm_map_unlock(old_map);
3103 	vm_map_unlock(new_map);
3104 	vm_map_entry_release(count);
3105 	lwkt_reltoken(&vmspace_token);
3106 	lwkt_reltoken(&vm_token);
3107 
3108 	return (vm2);
3109 }
3110 
3111 /*
3112  * Create an auto-grow stack entry
3113  *
3114  * No requirements.
3115  */
3116 int
3117 vm_map_stack (vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
3118 	      int flags, vm_prot_t prot, vm_prot_t max, int cow)
3119 {
3120 	vm_map_entry_t	prev_entry;
3121 	vm_map_entry_t	new_stack_entry;
3122 	vm_size_t	init_ssize;
3123 	int		rv;
3124 	int		count;
3125 	vm_offset_t	tmpaddr;
3126 
3127 	cow |= MAP_IS_STACK;
3128 
3129 	if (max_ssize < sgrowsiz)
3130 		init_ssize = max_ssize;
3131 	else
3132 		init_ssize = sgrowsiz;
3133 
3134 	count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
3135 	vm_map_lock(map);
3136 
3137 	/*
3138 	 * Find space for the mapping
3139 	 */
3140 	if ((flags & (MAP_FIXED | MAP_TRYFIXED)) == 0) {
3141 		if (vm_map_findspace(map, addrbos, max_ssize, 1,
3142 				     flags, &tmpaddr)) {
3143 			vm_map_unlock(map);
3144 			vm_map_entry_release(count);
3145 			return (KERN_NO_SPACE);
3146 		}
3147 		addrbos = tmpaddr;
3148 	}
3149 
3150 	/* If addr is already mapped, no go */
3151 	if (vm_map_lookup_entry(map, addrbos, &prev_entry)) {
3152 		vm_map_unlock(map);
3153 		vm_map_entry_release(count);
3154 		return (KERN_NO_SPACE);
3155 	}
3156 
3157 #if 0
3158 	/* XXX already handled by kern_mmap() */
3159 	/* If we would blow our VMEM resource limit, no go */
3160 	if (map->size + init_ssize >
3161 	    curproc->p_rlimit[RLIMIT_VMEM].rlim_cur) {
3162 		vm_map_unlock(map);
3163 		vm_map_entry_release(count);
3164 		return (KERN_NO_SPACE);
3165 	}
3166 #endif
3167 
3168 	/*
3169 	 * If we can't accomodate max_ssize in the current mapping,
3170 	 * no go.  However, we need to be aware that subsequent user
3171 	 * mappings might map into the space we have reserved for
3172 	 * stack, and currently this space is not protected.
3173 	 *
3174 	 * Hopefully we will at least detect this condition
3175 	 * when we try to grow the stack.
3176 	 */
3177 	if ((prev_entry->next != &map->header) &&
3178 	    (prev_entry->next->start < addrbos + max_ssize)) {
3179 		vm_map_unlock(map);
3180 		vm_map_entry_release(count);
3181 		return (KERN_NO_SPACE);
3182 	}
3183 
3184 	/*
3185 	 * We initially map a stack of only init_ssize.  We will
3186 	 * grow as needed later.  Since this is to be a grow
3187 	 * down stack, we map at the top of the range.
3188 	 *
3189 	 * Note: we would normally expect prot and max to be
3190 	 * VM_PROT_ALL, and cow to be 0.  Possibly we should
3191 	 * eliminate these as input parameters, and just
3192 	 * pass these values here in the insert call.
3193 	 */
3194 	rv = vm_map_insert(map, &count,
3195 			   NULL, 0, addrbos + max_ssize - init_ssize,
3196 	                   addrbos + max_ssize,
3197 			   VM_MAPTYPE_NORMAL,
3198 			   prot, max,
3199 			   cow);
3200 
3201 	/* Now set the avail_ssize amount */
3202 	if (rv == KERN_SUCCESS) {
3203 		if (prev_entry != &map->header)
3204 			vm_map_clip_end(map, prev_entry, addrbos + max_ssize - init_ssize, &count);
3205 		new_stack_entry = prev_entry->next;
3206 		if (new_stack_entry->end   != addrbos + max_ssize ||
3207 		    new_stack_entry->start != addrbos + max_ssize - init_ssize)
3208 			panic ("Bad entry start/end for new stack entry");
3209 		else
3210 			new_stack_entry->aux.avail_ssize = max_ssize - init_ssize;
3211 	}
3212 
3213 	vm_map_unlock(map);
3214 	vm_map_entry_release(count);
3215 	return (rv);
3216 }
3217 
3218 /*
3219  * Attempts to grow a vm stack entry.  Returns KERN_SUCCESS if the
3220  * desired address is already mapped, or if we successfully grow
3221  * the stack.  Also returns KERN_SUCCESS if addr is outside the
3222  * stack range (this is strange, but preserves compatibility with
3223  * the grow function in vm_machdep.c).
3224  *
3225  * No requirements.
3226  */
3227 int
3228 vm_map_growstack (struct proc *p, vm_offset_t addr)
3229 {
3230 	vm_map_entry_t prev_entry;
3231 	vm_map_entry_t stack_entry;
3232 	vm_map_entry_t new_stack_entry;
3233 	struct vmspace *vm = p->p_vmspace;
3234 	vm_map_t map = &vm->vm_map;
3235 	vm_offset_t    end;
3236 	int grow_amount;
3237 	int rv = KERN_SUCCESS;
3238 	int is_procstack;
3239 	int use_read_lock = 1;
3240 	int count;
3241 
3242 	count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
3243 Retry:
3244 	if (use_read_lock)
3245 		vm_map_lock_read(map);
3246 	else
3247 		vm_map_lock(map);
3248 
3249 	/* If addr is already in the entry range, no need to grow.*/
3250 	if (vm_map_lookup_entry(map, addr, &prev_entry))
3251 		goto done;
3252 
3253 	if ((stack_entry = prev_entry->next) == &map->header)
3254 		goto done;
3255 	if (prev_entry == &map->header)
3256 		end = stack_entry->start - stack_entry->aux.avail_ssize;
3257 	else
3258 		end = prev_entry->end;
3259 
3260 	/*
3261 	 * This next test mimics the old grow function in vm_machdep.c.
3262 	 * It really doesn't quite make sense, but we do it anyway
3263 	 * for compatibility.
3264 	 *
3265 	 * If not growable stack, return success.  This signals the
3266 	 * caller to proceed as he would normally with normal vm.
3267 	 */
3268 	if (stack_entry->aux.avail_ssize < 1 ||
3269 	    addr >= stack_entry->start ||
3270 	    addr <  stack_entry->start - stack_entry->aux.avail_ssize) {
3271 		goto done;
3272 	}
3273 
3274 	/* Find the minimum grow amount */
3275 	grow_amount = roundup (stack_entry->start - addr, PAGE_SIZE);
3276 	if (grow_amount > stack_entry->aux.avail_ssize) {
3277 		rv = KERN_NO_SPACE;
3278 		goto done;
3279 	}
3280 
3281 	/*
3282 	 * If there is no longer enough space between the entries
3283 	 * nogo, and adjust the available space.  Note: this
3284 	 * should only happen if the user has mapped into the
3285 	 * stack area after the stack was created, and is
3286 	 * probably an error.
3287 	 *
3288 	 * This also effectively destroys any guard page the user
3289 	 * might have intended by limiting the stack size.
3290 	 */
3291 	if (grow_amount > stack_entry->start - end) {
3292 		if (use_read_lock && vm_map_lock_upgrade(map)) {
3293 			use_read_lock = 0;
3294 			goto Retry;
3295 		}
3296 		use_read_lock = 0;
3297 		stack_entry->aux.avail_ssize = stack_entry->start - end;
3298 		rv = KERN_NO_SPACE;
3299 		goto done;
3300 	}
3301 
3302 	is_procstack = addr >= (vm_offset_t)vm->vm_maxsaddr;
3303 
3304 	/* If this is the main process stack, see if we're over the
3305 	 * stack limit.
3306 	 */
3307 	if (is_procstack && (ctob(vm->vm_ssize) + grow_amount >
3308 			     p->p_rlimit[RLIMIT_STACK].rlim_cur)) {
3309 		rv = KERN_NO_SPACE;
3310 		goto done;
3311 	}
3312 
3313 	/* Round up the grow amount modulo SGROWSIZ */
3314 	grow_amount = roundup (grow_amount, sgrowsiz);
3315 	if (grow_amount > stack_entry->aux.avail_ssize) {
3316 		grow_amount = stack_entry->aux.avail_ssize;
3317 	}
3318 	if (is_procstack && (ctob(vm->vm_ssize) + grow_amount >
3319 	                     p->p_rlimit[RLIMIT_STACK].rlim_cur)) {
3320 		grow_amount = p->p_rlimit[RLIMIT_STACK].rlim_cur -
3321 		              ctob(vm->vm_ssize);
3322 	}
3323 
3324 	/* If we would blow our VMEM resource limit, no go */
3325 	if (map->size + grow_amount > p->p_rlimit[RLIMIT_VMEM].rlim_cur) {
3326 		rv = KERN_NO_SPACE;
3327 		goto done;
3328 	}
3329 
3330 	if (use_read_lock && vm_map_lock_upgrade(map)) {
3331 		use_read_lock = 0;
3332 		goto Retry;
3333 	}
3334 	use_read_lock = 0;
3335 
3336 	/* Get the preliminary new entry start value */
3337 	addr = stack_entry->start - grow_amount;
3338 
3339 	/* If this puts us into the previous entry, cut back our growth
3340 	 * to the available space.  Also, see the note above.
3341 	 */
3342 	if (addr < end) {
3343 		stack_entry->aux.avail_ssize = stack_entry->start - end;
3344 		addr = end;
3345 	}
3346 
3347 	rv = vm_map_insert(map, &count,
3348 			   NULL, 0, addr, stack_entry->start,
3349 			   VM_MAPTYPE_NORMAL,
3350 			   VM_PROT_ALL, VM_PROT_ALL,
3351 			   0);
3352 
3353 	/* Adjust the available stack space by the amount we grew. */
3354 	if (rv == KERN_SUCCESS) {
3355 		if (prev_entry != &map->header)
3356 			vm_map_clip_end(map, prev_entry, addr, &count);
3357 		new_stack_entry = prev_entry->next;
3358 		if (new_stack_entry->end   != stack_entry->start  ||
3359 		    new_stack_entry->start != addr)
3360 			panic ("Bad stack grow start/end in new stack entry");
3361 		else {
3362 			new_stack_entry->aux.avail_ssize =
3363 				stack_entry->aux.avail_ssize -
3364 				(new_stack_entry->end - new_stack_entry->start);
3365 			if (is_procstack)
3366 				vm->vm_ssize += btoc(new_stack_entry->end -
3367 						     new_stack_entry->start);
3368 		}
3369 	}
3370 
3371 done:
3372 	if (use_read_lock)
3373 		vm_map_unlock_read(map);
3374 	else
3375 		vm_map_unlock(map);
3376 	vm_map_entry_release(count);
3377 	return (rv);
3378 }
3379 
3380 /*
3381  * Unshare the specified VM space for exec.  If other processes are
3382  * mapped to it, then create a new one.  The new vmspace is null.
3383  *
3384  * No requirements.
3385  */
3386 void
3387 vmspace_exec(struct proc *p, struct vmspace *vmcopy)
3388 {
3389 	struct vmspace *oldvmspace = p->p_vmspace;
3390 	struct vmspace *newvmspace;
3391 	vm_map_t map = &p->p_vmspace->vm_map;
3392 
3393 	/*
3394 	 * If we are execing a resident vmspace we fork it, otherwise
3395 	 * we create a new vmspace.  Note that exitingcnt and upcalls
3396 	 * are not copied to the new vmspace.
3397 	 */
3398 	lwkt_gettoken(&vmspace_token);
3399 	if (vmcopy)  {
3400 		newvmspace = vmspace_fork(vmcopy);
3401 	} else {
3402 		newvmspace = vmspace_alloc(map->min_offset, map->max_offset);
3403 		bcopy(&oldvmspace->vm_startcopy, &newvmspace->vm_startcopy,
3404 		      (caddr_t)&oldvmspace->vm_endcopy -
3405 		       (caddr_t)&oldvmspace->vm_startcopy);
3406 	}
3407 
3408 	/*
3409 	 * Finish initializing the vmspace before assigning it
3410 	 * to the process.  The vmspace will become the current vmspace
3411 	 * if p == curproc.
3412 	 */
3413 	pmap_pinit2(vmspace_pmap(newvmspace));
3414 	pmap_replacevm(p, newvmspace, 0);
3415 	sysref_put(&oldvmspace->vm_sysref);
3416 	lwkt_reltoken(&vmspace_token);
3417 }
3418 
3419 /*
3420  * Unshare the specified VM space for forcing COW.  This
3421  * is called by rfork, for the (RFMEM|RFPROC) == 0 case.
3422  *
3423  * The exitingcnt test is not strictly necessary but has been
3424  * included for code sanity (to make the code a bit more deterministic).
3425  */
3426 void
3427 vmspace_unshare(struct proc *p)
3428 {
3429 	struct vmspace *oldvmspace = p->p_vmspace;
3430 	struct vmspace *newvmspace;
3431 
3432 	lwkt_gettoken(&vmspace_token);
3433 	if (oldvmspace->vm_sysref.refcnt == 1 && oldvmspace->vm_exitingcnt == 0)
3434 		return;
3435 	newvmspace = vmspace_fork(oldvmspace);
3436 	pmap_pinit2(vmspace_pmap(newvmspace));
3437 	pmap_replacevm(p, newvmspace, 0);
3438 	sysref_put(&oldvmspace->vm_sysref);
3439 	lwkt_reltoken(&vmspace_token);
3440 }
3441 
3442 /*
3443  * Finds the VM object, offset, and protection for a given virtual address
3444  * in the specified map, assuming a page fault of the type specified.
3445  *
3446  * Leaves the map in question locked for read; return values are guaranteed
3447  * until a vm_map_lookup_done call is performed.  Note that the map argument
3448  * is in/out; the returned map must be used in the call to vm_map_lookup_done.
3449  *
3450  * A handle (out_entry) is returned for use in vm_map_lookup_done, to make
3451  * that fast.
3452  *
3453  * If a lookup is requested with "write protection" specified, the map may
3454  * be changed to perform virtual copying operations, although the data
3455  * referenced will remain the same.
3456  *
3457  * No requirements.
3458  */
3459 int
3460 vm_map_lookup(vm_map_t *var_map,		/* IN/OUT */
3461 	      vm_offset_t vaddr,
3462 	      vm_prot_t fault_typea,
3463 	      vm_map_entry_t *out_entry,	/* OUT */
3464 	      vm_object_t *object,		/* OUT */
3465 	      vm_pindex_t *pindex,		/* OUT */
3466 	      vm_prot_t *out_prot,		/* OUT */
3467 	      boolean_t *wired)			/* OUT */
3468 {
3469 	vm_map_entry_t entry;
3470 	vm_map_t map = *var_map;
3471 	vm_prot_t prot;
3472 	vm_prot_t fault_type = fault_typea;
3473 	int use_read_lock = 1;
3474 	int rv = KERN_SUCCESS;
3475 
3476 RetryLookup:
3477 	if (use_read_lock)
3478 		vm_map_lock_read(map);
3479 	else
3480 		vm_map_lock(map);
3481 
3482 	/*
3483 	 * If the map has an interesting hint, try it before calling full
3484 	 * blown lookup routine.
3485 	 */
3486 	entry = map->hint;
3487 	*out_entry = entry;
3488 
3489 	if ((entry == &map->header) ||
3490 	    (vaddr < entry->start) || (vaddr >= entry->end)) {
3491 		vm_map_entry_t tmp_entry;
3492 
3493 		/*
3494 		 * Entry was either not a valid hint, or the vaddr was not
3495 		 * contained in the entry, so do a full lookup.
3496 		 */
3497 		if (!vm_map_lookup_entry(map, vaddr, &tmp_entry)) {
3498 			rv = KERN_INVALID_ADDRESS;
3499 			goto done;
3500 		}
3501 
3502 		entry = tmp_entry;
3503 		*out_entry = entry;
3504 	}
3505 
3506 	/*
3507 	 * Handle submaps.
3508 	 */
3509 	if (entry->maptype == VM_MAPTYPE_SUBMAP) {
3510 		vm_map_t old_map = map;
3511 
3512 		*var_map = map = entry->object.sub_map;
3513 		if (use_read_lock)
3514 			vm_map_unlock_read(old_map);
3515 		else
3516 			vm_map_unlock(old_map);
3517 		use_read_lock = 1;
3518 		goto RetryLookup;
3519 	}
3520 
3521 	/*
3522 	 * Check whether this task is allowed to have this page.
3523 	 * Note the special case for MAP_ENTRY_COW
3524 	 * pages with an override.  This is to implement a forced
3525 	 * COW for debuggers.
3526 	 */
3527 
3528 	if (fault_type & VM_PROT_OVERRIDE_WRITE)
3529 		prot = entry->max_protection;
3530 	else
3531 		prot = entry->protection;
3532 
3533 	fault_type &= (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE);
3534 	if ((fault_type & prot) != fault_type) {
3535 		rv = KERN_PROTECTION_FAILURE;
3536 		goto done;
3537 	}
3538 
3539 	if ((entry->eflags & MAP_ENTRY_USER_WIRED) &&
3540 	    (entry->eflags & MAP_ENTRY_COW) &&
3541 	    (fault_type & VM_PROT_WRITE) &&
3542 	    (fault_typea & VM_PROT_OVERRIDE_WRITE) == 0) {
3543 		rv = KERN_PROTECTION_FAILURE;
3544 		goto done;
3545 	}
3546 
3547 	/*
3548 	 * If this page is not pageable, we have to get it for all possible
3549 	 * accesses.
3550 	 */
3551 	*wired = (entry->wired_count != 0);
3552 	if (*wired)
3553 		prot = fault_type = entry->protection;
3554 
3555 	/*
3556 	 * Virtual page tables may need to update the accessed (A) bit
3557 	 * in a page table entry.  Upgrade the fault to a write fault for
3558 	 * that case if the map will support it.  If the map does not support
3559 	 * it the page table entry simply will not be updated.
3560 	 */
3561 	if (entry->maptype == VM_MAPTYPE_VPAGETABLE) {
3562 		if (prot & VM_PROT_WRITE)
3563 			fault_type |= VM_PROT_WRITE;
3564 	}
3565 
3566 	/*
3567 	 * If the entry was copy-on-write, we either ...
3568 	 */
3569 	if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3570 		/*
3571 		 * If we want to write the page, we may as well handle that
3572 		 * now since we've got the map locked.
3573 		 *
3574 		 * If we don't need to write the page, we just demote the
3575 		 * permissions allowed.
3576 		 */
3577 
3578 		if (fault_type & VM_PROT_WRITE) {
3579 			/*
3580 			 * Make a new object, and place it in the object
3581 			 * chain.  Note that no new references have appeared
3582 			 * -- one just moved from the map to the new
3583 			 * object.
3584 			 */
3585 
3586 			if (use_read_lock && vm_map_lock_upgrade(map)) {
3587 				use_read_lock = 0;
3588 				goto RetryLookup;
3589 			}
3590 			use_read_lock = 0;
3591 
3592 			vm_map_entry_shadow(entry);
3593 		} else {
3594 			/*
3595 			 * We're attempting to read a copy-on-write page --
3596 			 * don't allow writes.
3597 			 */
3598 
3599 			prot &= ~VM_PROT_WRITE;
3600 		}
3601 	}
3602 
3603 	/*
3604 	 * Create an object if necessary.
3605 	 */
3606 	if (entry->object.vm_object == NULL &&
3607 	    !map->system_map) {
3608 		if (use_read_lock && vm_map_lock_upgrade(map))  {
3609 			use_read_lock = 0;
3610 			goto RetryLookup;
3611 		}
3612 		use_read_lock = 0;
3613 		vm_map_entry_allocate_object(entry);
3614 	}
3615 
3616 	/*
3617 	 * Return the object/offset from this entry.  If the entry was
3618 	 * copy-on-write or empty, it has been fixed up.
3619 	 */
3620 
3621 	*pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
3622 	*object = entry->object.vm_object;
3623 
3624 	/*
3625 	 * Return whether this is the only map sharing this data.  On
3626 	 * success we return with a read lock held on the map.  On failure
3627 	 * we return with the map unlocked.
3628 	 */
3629 	*out_prot = prot;
3630 done:
3631 	if (rv == KERN_SUCCESS) {
3632 		if (use_read_lock == 0)
3633 			vm_map_lock_downgrade(map);
3634 	} else if (use_read_lock) {
3635 		vm_map_unlock_read(map);
3636 	} else {
3637 		vm_map_unlock(map);
3638 	}
3639 	return (rv);
3640 }
3641 
3642 /*
3643  * Releases locks acquired by a vm_map_lookup()
3644  * (according to the handle returned by that lookup).
3645  *
3646  * No other requirements.
3647  */
3648 void
3649 vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry, int count)
3650 {
3651 	/*
3652 	 * Unlock the main-level map
3653 	 */
3654 	vm_map_unlock_read(map);
3655 	if (count)
3656 		vm_map_entry_release(count);
3657 }
3658 
3659 #include "opt_ddb.h"
3660 #ifdef DDB
3661 #include <sys/kernel.h>
3662 
3663 #include <ddb/ddb.h>
3664 
3665 /*
3666  * Debugging only
3667  */
3668 DB_SHOW_COMMAND(map, vm_map_print)
3669 {
3670 	static int nlines;
3671 	/* XXX convert args. */
3672 	vm_map_t map = (vm_map_t)addr;
3673 	boolean_t full = have_addr;
3674 
3675 	vm_map_entry_t entry;
3676 
3677 	db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n",
3678 	    (void *)map,
3679 	    (void *)map->pmap, map->nentries, map->timestamp);
3680 	nlines++;
3681 
3682 	if (!full && db_indent)
3683 		return;
3684 
3685 	db_indent += 2;
3686 	for (entry = map->header.next; entry != &map->header;
3687 	    entry = entry->next) {
3688 		db_iprintf("map entry %p: start=%p, end=%p\n",
3689 		    (void *)entry, (void *)entry->start, (void *)entry->end);
3690 		nlines++;
3691 		{
3692 			static char *inheritance_name[4] =
3693 			{"share", "copy", "none", "donate_copy"};
3694 
3695 			db_iprintf(" prot=%x/%x/%s",
3696 			    entry->protection,
3697 			    entry->max_protection,
3698 			    inheritance_name[(int)(unsigned char)entry->inheritance]);
3699 			if (entry->wired_count != 0)
3700 				db_printf(", wired");
3701 		}
3702 		if (entry->maptype == VM_MAPTYPE_SUBMAP) {
3703 			/* XXX no %qd in kernel.  Truncate entry->offset. */
3704 			db_printf(", share=%p, offset=0x%lx\n",
3705 			    (void *)entry->object.sub_map,
3706 			    (long)entry->offset);
3707 			nlines++;
3708 			if ((entry->prev == &map->header) ||
3709 			    (entry->prev->object.sub_map !=
3710 				entry->object.sub_map)) {
3711 				db_indent += 2;
3712 				vm_map_print((db_expr_t)(intptr_t)
3713 					     entry->object.sub_map,
3714 					     full, 0, NULL);
3715 				db_indent -= 2;
3716 			}
3717 		} else {
3718 			/* XXX no %qd in kernel.  Truncate entry->offset. */
3719 			db_printf(", object=%p, offset=0x%lx",
3720 			    (void *)entry->object.vm_object,
3721 			    (long)entry->offset);
3722 			if (entry->eflags & MAP_ENTRY_COW)
3723 				db_printf(", copy (%s)",
3724 				    (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done");
3725 			db_printf("\n");
3726 			nlines++;
3727 
3728 			if ((entry->prev == &map->header) ||
3729 			    (entry->prev->object.vm_object !=
3730 				entry->object.vm_object)) {
3731 				db_indent += 2;
3732 				vm_object_print((db_expr_t)(intptr_t)
3733 						entry->object.vm_object,
3734 						full, 0, NULL);
3735 				nlines += 4;
3736 				db_indent -= 2;
3737 			}
3738 		}
3739 	}
3740 	db_indent -= 2;
3741 	if (db_indent == 0)
3742 		nlines = 0;
3743 }
3744 
3745 /*
3746  * Debugging only
3747  */
3748 DB_SHOW_COMMAND(procvm, procvm)
3749 {
3750 	struct proc *p;
3751 
3752 	if (have_addr) {
3753 		p = (struct proc *) addr;
3754 	} else {
3755 		p = curproc;
3756 	}
3757 
3758 	db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n",
3759 	    (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map,
3760 	    (void *)vmspace_pmap(p->p_vmspace));
3761 
3762 	vm_map_print((db_expr_t)(intptr_t)&p->p_vmspace->vm_map, 1, 0, NULL);
3763 }
3764 
3765 #endif /* DDB */
3766