xref: /dflybsd-src/sys/vm/vm_map.c (revision 330d3c4b487f3fc5d0eb023645b0b2a569f7048e)
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 	lwkt_gettoken(&vm_token);
913 	lwkt_gettoken(&vmobj_token);
914 
915 	if (object) {
916 		/*
917 		 * When object is non-NULL, it could be shared with another
918 		 * process.  We have to set or clear OBJ_ONEMAPPING
919 		 * appropriately.
920 		 */
921 		if ((object->ref_count > 1) || (object->shadow_count != 0)) {
922 			vm_object_clear_flag(object, OBJ_ONEMAPPING);
923 		}
924 	}
925 	else if ((prev_entry != &map->header) &&
926 		 (prev_entry->eflags == protoeflags) &&
927 		 (prev_entry->end == start) &&
928 		 (prev_entry->wired_count == 0) &&
929 		 prev_entry->maptype == maptype &&
930 		 ((prev_entry->object.vm_object == NULL) ||
931 		  vm_object_coalesce(prev_entry->object.vm_object,
932 				     OFF_TO_IDX(prev_entry->offset),
933 				     (vm_size_t)(prev_entry->end - prev_entry->start),
934 				     (vm_size_t)(end - prev_entry->end)))) {
935 		/*
936 		 * We were able to extend the object.  Determine if we
937 		 * can extend the previous map entry to include the
938 		 * new range as well.
939 		 */
940 		if ((prev_entry->inheritance == VM_INHERIT_DEFAULT) &&
941 		    (prev_entry->protection == prot) &&
942 		    (prev_entry->max_protection == max)) {
943 			lwkt_reltoken(&vmobj_token);
944 			lwkt_reltoken(&vm_token);
945 			map->size += (end - prev_entry->end);
946 			prev_entry->end = end;
947 			vm_map_simplify_entry(map, prev_entry, countp);
948 			return (KERN_SUCCESS);
949 		}
950 
951 		/*
952 		 * If we can extend the object but cannot extend the
953 		 * map entry, we have to create a new map entry.  We
954 		 * must bump the ref count on the extended object to
955 		 * account for it.  object may be NULL.
956 		 */
957 		object = prev_entry->object.vm_object;
958 		offset = prev_entry->offset +
959 			(prev_entry->end - prev_entry->start);
960 		vm_object_reference_locked(object);
961 	}
962 
963 	lwkt_reltoken(&vmobj_token);
964 	lwkt_reltoken(&vm_token);
965 
966 	/*
967 	 * NOTE: if conditionals fail, object can be NULL here.  This occurs
968 	 * in things like the buffer map where we manage kva but do not manage
969 	 * backing objects.
970 	 */
971 
972 	/*
973 	 * Create a new entry
974 	 */
975 
976 	new_entry = vm_map_entry_create(map, countp);
977 	new_entry->start = start;
978 	new_entry->end = end;
979 
980 	new_entry->maptype = maptype;
981 	new_entry->eflags = protoeflags;
982 	new_entry->object.vm_object = object;
983 	new_entry->offset = offset;
984 	new_entry->aux.master_pde = 0;
985 
986 	new_entry->inheritance = VM_INHERIT_DEFAULT;
987 	new_entry->protection = prot;
988 	new_entry->max_protection = max;
989 	new_entry->wired_count = 0;
990 
991 	/*
992 	 * Insert the new entry into the list
993 	 */
994 
995 	vm_map_entry_link(map, prev_entry, new_entry);
996 	map->size += new_entry->end - new_entry->start;
997 
998 	/*
999 	 * Update the free space hint.  Entries cannot overlap.
1000 	 * An exact comparison is needed to avoid matching
1001 	 * against the map->header.
1002 	 */
1003 	if ((map->first_free == prev_entry) &&
1004 	    (prev_entry->end == new_entry->start)) {
1005 		map->first_free = new_entry;
1006 	}
1007 
1008 #if 0
1009 	/*
1010 	 * Temporarily removed to avoid MAP_STACK panic, due to
1011 	 * MAP_STACK being a huge hack.  Will be added back in
1012 	 * when MAP_STACK (and the user stack mapping) is fixed.
1013 	 */
1014 	/*
1015 	 * It may be possible to simplify the entry
1016 	 */
1017 	vm_map_simplify_entry(map, new_entry, countp);
1018 #endif
1019 
1020 	/*
1021 	 * Try to pre-populate the page table.  Mappings governed by virtual
1022 	 * page tables cannot be prepopulated without a lot of work, so
1023 	 * don't try.
1024 	 */
1025 	if ((cow & (MAP_PREFAULT|MAP_PREFAULT_PARTIAL)) &&
1026 	    maptype != VM_MAPTYPE_VPAGETABLE) {
1027 		pmap_object_init_pt(map->pmap, start, prot,
1028 				    object, OFF_TO_IDX(offset), end - start,
1029 				    cow & MAP_PREFAULT_PARTIAL);
1030 	}
1031 
1032 	return (KERN_SUCCESS);
1033 }
1034 
1035 /*
1036  * Find sufficient space for `length' bytes in the given map, starting at
1037  * `start'.  Returns 0 on success, 1 on no space.
1038  *
1039  * This function will returned an arbitrarily aligned pointer.  If no
1040  * particular alignment is required you should pass align as 1.  Note that
1041  * the map may return PAGE_SIZE aligned pointers if all the lengths used in
1042  * the map are a multiple of PAGE_SIZE, even if you pass a smaller align
1043  * argument.
1044  *
1045  * 'align' should be a power of 2 but is not required to be.
1046  *
1047  * The map must be exclusively locked.
1048  * No other requirements.
1049  */
1050 int
1051 vm_map_findspace(vm_map_t map, vm_offset_t start, vm_size_t length,
1052 		 vm_size_t align, int flags, vm_offset_t *addr)
1053 {
1054 	vm_map_entry_t entry, next;
1055 	vm_offset_t end;
1056 	vm_offset_t align_mask;
1057 
1058 	if (start < map->min_offset)
1059 		start = map->min_offset;
1060 	if (start > map->max_offset)
1061 		return (1);
1062 
1063 	/*
1064 	 * If the alignment is not a power of 2 we will have to use
1065 	 * a mod/division, set align_mask to a special value.
1066 	 */
1067 	if ((align | (align - 1)) + 1 != (align << 1))
1068 		align_mask = (vm_offset_t)-1;
1069 	else
1070 		align_mask = align - 1;
1071 
1072 	/*
1073 	 * Look for the first possible address; if there's already something
1074 	 * at this address, we have to start after it.
1075 	 */
1076 	if (start == map->min_offset) {
1077 		if ((entry = map->first_free) != &map->header)
1078 			start = entry->end;
1079 	} else {
1080 		vm_map_entry_t tmp;
1081 
1082 		if (vm_map_lookup_entry(map, start, &tmp))
1083 			start = tmp->end;
1084 		entry = tmp;
1085 	}
1086 
1087 	/*
1088 	 * Look through the rest of the map, trying to fit a new region in the
1089 	 * gap between existing regions, or after the very last region.
1090 	 */
1091 	for (;; start = (entry = next)->end) {
1092 		/*
1093 		 * Adjust the proposed start by the requested alignment,
1094 		 * be sure that we didn't wrap the address.
1095 		 */
1096 		if (align_mask == (vm_offset_t)-1)
1097 			end = ((start + align - 1) / align) * align;
1098 		else
1099 			end = (start + align_mask) & ~align_mask;
1100 		if (end < start)
1101 			return (1);
1102 		start = end;
1103 		/*
1104 		 * Find the end of the proposed new region.  Be sure we didn't
1105 		 * go beyond the end of the map, or wrap around the address.
1106 		 * Then check to see if this is the last entry or if the
1107 		 * proposed end fits in the gap between this and the next
1108 		 * entry.
1109 		 */
1110 		end = start + length;
1111 		if (end > map->max_offset || end < start)
1112 			return (1);
1113 		next = entry->next;
1114 
1115 		/*
1116 		 * If the next entry's start address is beyond the desired
1117 		 * end address we may have found a good entry.
1118 		 *
1119 		 * If the next entry is a stack mapping we do not map into
1120 		 * the stack's reserved space.
1121 		 *
1122 		 * XXX continue to allow mapping into the stack's reserved
1123 		 * space if doing a MAP_STACK mapping inside a MAP_STACK
1124 		 * mapping, for backwards compatibility.  But the caller
1125 		 * really should use MAP_STACK | MAP_TRYFIXED if they
1126 		 * want to do that.
1127 		 */
1128 		if (next == &map->header)
1129 			break;
1130 		if (next->start >= end) {
1131 			if ((next->eflags & MAP_ENTRY_STACK) == 0)
1132 				break;
1133 			if (flags & MAP_STACK)
1134 				break;
1135 			if (next->start - next->aux.avail_ssize >= end)
1136 				break;
1137 		}
1138 	}
1139 	map->hint = entry;
1140 
1141 	/*
1142 	 * Grow the kernel_map if necessary.  pmap_growkernel() will panic
1143 	 * if it fails.  The kernel_map is locked and nothing can steal
1144 	 * our address space if pmap_growkernel() blocks.
1145 	 *
1146 	 * NOTE: This may be unconditionally called for kldload areas on
1147 	 *	 x86_64 because these do not bump kernel_vm_end (which would
1148 	 *	 fill 128G worth of page tables!).  Therefore we must not
1149 	 *	 retry.
1150 	 */
1151 	if (map == &kernel_map) {
1152 		vm_offset_t kstop;
1153 
1154 		kstop = round_page(start + length);
1155 		if (kstop > kernel_vm_end)
1156 			pmap_growkernel(start, kstop);
1157 	}
1158 	*addr = start;
1159 	return (0);
1160 }
1161 
1162 /*
1163  * vm_map_find finds an unallocated region in the target address map with
1164  * the given length.  The search is defined to be first-fit from the
1165  * specified address; the region found is returned in the same parameter.
1166  *
1167  * If object is non-NULL, ref count must be bumped by caller
1168  * prior to making call to account for the new entry.
1169  *
1170  * No requirements.  This function will lock the map temporarily.
1171  */
1172 int
1173 vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1174 	    vm_offset_t *addr,	vm_size_t length, vm_size_t align,
1175 	    boolean_t fitit,
1176 	    vm_maptype_t maptype,
1177 	    vm_prot_t prot, vm_prot_t max,
1178 	    int cow)
1179 {
1180 	vm_offset_t start;
1181 	int result;
1182 	int count;
1183 
1184 	start = *addr;
1185 
1186 	count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
1187 	vm_map_lock(map);
1188 	if (fitit) {
1189 		if (vm_map_findspace(map, start, length, align, 0, addr)) {
1190 			vm_map_unlock(map);
1191 			vm_map_entry_release(count);
1192 			return (KERN_NO_SPACE);
1193 		}
1194 		start = *addr;
1195 	}
1196 	result = vm_map_insert(map, &count, object, offset,
1197 			       start, start + length,
1198 			       maptype,
1199 			       prot, max,
1200 			       cow);
1201 	vm_map_unlock(map);
1202 	vm_map_entry_release(count);
1203 
1204 	return (result);
1205 }
1206 
1207 /*
1208  * Simplify the given map entry by merging with either neighbor.  This
1209  * routine also has the ability to merge with both neighbors.
1210  *
1211  * This routine guarentees that the passed entry remains valid (though
1212  * possibly extended).  When merging, this routine may delete one or
1213  * both neighbors.  No action is taken on entries which have their
1214  * in-transition flag set.
1215  *
1216  * The map must be exclusively locked.
1217  */
1218 void
1219 vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry, int *countp)
1220 {
1221 	vm_map_entry_t next, prev;
1222 	vm_size_t prevsize, esize;
1223 
1224 	if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
1225 		++mycpu->gd_cnt.v_intrans_coll;
1226 		return;
1227 	}
1228 
1229 	if (entry->maptype == VM_MAPTYPE_SUBMAP)
1230 		return;
1231 
1232 	prev = entry->prev;
1233 	if (prev != &map->header) {
1234 		prevsize = prev->end - prev->start;
1235 		if ( (prev->end == entry->start) &&
1236 		     (prev->maptype == entry->maptype) &&
1237 		     (prev->object.vm_object == entry->object.vm_object) &&
1238 		     (!prev->object.vm_object ||
1239 			(prev->offset + prevsize == entry->offset)) &&
1240 		     (prev->eflags == entry->eflags) &&
1241 		     (prev->protection == entry->protection) &&
1242 		     (prev->max_protection == entry->max_protection) &&
1243 		     (prev->inheritance == entry->inheritance) &&
1244 		     (prev->wired_count == entry->wired_count)) {
1245 			if (map->first_free == prev)
1246 				map->first_free = entry;
1247 			if (map->hint == prev)
1248 				map->hint = entry;
1249 			vm_map_entry_unlink(map, prev);
1250 			entry->start = prev->start;
1251 			entry->offset = prev->offset;
1252 			if (prev->object.vm_object)
1253 				vm_object_deallocate(prev->object.vm_object);
1254 			vm_map_entry_dispose(map, prev, countp);
1255 		}
1256 	}
1257 
1258 	next = entry->next;
1259 	if (next != &map->header) {
1260 		esize = entry->end - entry->start;
1261 		if ((entry->end == next->start) &&
1262 		    (next->maptype == entry->maptype) &&
1263 		    (next->object.vm_object == entry->object.vm_object) &&
1264 		     (!entry->object.vm_object ||
1265 			(entry->offset + esize == next->offset)) &&
1266 		    (next->eflags == entry->eflags) &&
1267 		    (next->protection == entry->protection) &&
1268 		    (next->max_protection == entry->max_protection) &&
1269 		    (next->inheritance == entry->inheritance) &&
1270 		    (next->wired_count == entry->wired_count)) {
1271 			if (map->first_free == next)
1272 				map->first_free = entry;
1273 			if (map->hint == next)
1274 				map->hint = entry;
1275 			vm_map_entry_unlink(map, next);
1276 			entry->end = next->end;
1277 			if (next->object.vm_object)
1278 				vm_object_deallocate(next->object.vm_object);
1279 			vm_map_entry_dispose(map, next, countp);
1280 	        }
1281 	}
1282 }
1283 
1284 /*
1285  * Asserts that the given entry begins at or after the specified address.
1286  * If necessary, it splits the entry into two.
1287  */
1288 #define vm_map_clip_start(map, entry, startaddr, countp)		\
1289 {									\
1290 	if (startaddr > entry->start)					\
1291 		_vm_map_clip_start(map, entry, startaddr, countp);	\
1292 }
1293 
1294 /*
1295  * This routine is called only when it is known that the entry must be split.
1296  *
1297  * The map must be exclusively locked.
1298  */
1299 static void
1300 _vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start,
1301 		   int *countp)
1302 {
1303 	vm_map_entry_t new_entry;
1304 
1305 	/*
1306 	 * Split off the front portion -- note that we must insert the new
1307 	 * entry BEFORE this one, so that this entry has the specified
1308 	 * starting address.
1309 	 */
1310 
1311 	vm_map_simplify_entry(map, entry, countp);
1312 
1313 	/*
1314 	 * If there is no object backing this entry, we might as well create
1315 	 * one now.  If we defer it, an object can get created after the map
1316 	 * is clipped, and individual objects will be created for the split-up
1317 	 * map.  This is a bit of a hack, but is also about the best place to
1318 	 * put this improvement.
1319 	 */
1320 	if (entry->object.vm_object == NULL && !map->system_map) {
1321 		vm_map_entry_allocate_object(entry);
1322 	}
1323 
1324 	new_entry = vm_map_entry_create(map, countp);
1325 	*new_entry = *entry;
1326 
1327 	new_entry->end = start;
1328 	entry->offset += (start - entry->start);
1329 	entry->start = start;
1330 
1331 	vm_map_entry_link(map, entry->prev, new_entry);
1332 
1333 	switch(entry->maptype) {
1334 	case VM_MAPTYPE_NORMAL:
1335 	case VM_MAPTYPE_VPAGETABLE:
1336 		vm_object_reference(new_entry->object.vm_object);
1337 		break;
1338 	default:
1339 		break;
1340 	}
1341 }
1342 
1343 /*
1344  * Asserts that the given entry ends at or before the specified address.
1345  * If necessary, it splits the entry into two.
1346  *
1347  * The map must be exclusively locked.
1348  */
1349 #define vm_map_clip_end(map, entry, endaddr, countp)		\
1350 {								\
1351 	if (endaddr < entry->end)				\
1352 		_vm_map_clip_end(map, entry, endaddr, countp);	\
1353 }
1354 
1355 /*
1356  * This routine is called only when it is known that the entry must be split.
1357  *
1358  * The map must be exclusively locked.
1359  */
1360 static void
1361 _vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end,
1362 		 int *countp)
1363 {
1364 	vm_map_entry_t new_entry;
1365 
1366 	/*
1367 	 * If there is no object backing this entry, we might as well create
1368 	 * one now.  If we defer it, an object can get created after the map
1369 	 * is clipped, and individual objects will be created for the split-up
1370 	 * map.  This is a bit of a hack, but is also about the best place to
1371 	 * put this improvement.
1372 	 */
1373 
1374 	if (entry->object.vm_object == NULL && !map->system_map) {
1375 		vm_map_entry_allocate_object(entry);
1376 	}
1377 
1378 	/*
1379 	 * Create a new entry and insert it AFTER the specified entry
1380 	 */
1381 
1382 	new_entry = vm_map_entry_create(map, countp);
1383 	*new_entry = *entry;
1384 
1385 	new_entry->start = entry->end = end;
1386 	new_entry->offset += (end - entry->start);
1387 
1388 	vm_map_entry_link(map, entry, new_entry);
1389 
1390 	switch(entry->maptype) {
1391 	case VM_MAPTYPE_NORMAL:
1392 	case VM_MAPTYPE_VPAGETABLE:
1393 		vm_object_reference(new_entry->object.vm_object);
1394 		break;
1395 	default:
1396 		break;
1397 	}
1398 }
1399 
1400 /*
1401  * Asserts that the starting and ending region addresses fall within the
1402  * valid range for the map.
1403  */
1404 #define	VM_MAP_RANGE_CHECK(map, start, end)	\
1405 {						\
1406 	if (start < vm_map_min(map))		\
1407 		start = vm_map_min(map);	\
1408 	if (end > vm_map_max(map))		\
1409 		end = vm_map_max(map);		\
1410 	if (start > end)			\
1411 		start = end;			\
1412 }
1413 
1414 /*
1415  * Used to block when an in-transition collison occurs.  The map
1416  * is unlocked for the sleep and relocked before the return.
1417  */
1418 static
1419 void
1420 vm_map_transition_wait(vm_map_t map)
1421 {
1422 	vm_map_unlock(map);
1423 	tsleep(map, 0, "vment", 0);
1424 	vm_map_lock(map);
1425 }
1426 
1427 /*
1428  * When we do blocking operations with the map lock held it is
1429  * possible that a clip might have occured on our in-transit entry,
1430  * requiring an adjustment to the entry in our loop.  These macros
1431  * help the pageable and clip_range code deal with the case.  The
1432  * conditional costs virtually nothing if no clipping has occured.
1433  */
1434 
1435 #define CLIP_CHECK_BACK(entry, save_start)		\
1436     do {						\
1437 	    while (entry->start != save_start) {	\
1438 		    entry = entry->prev;		\
1439 		    KASSERT(entry != &map->header, ("bad entry clip")); \
1440 	    }						\
1441     } while(0)
1442 
1443 #define CLIP_CHECK_FWD(entry, save_end)			\
1444     do {						\
1445 	    while (entry->end != save_end) {		\
1446 		    entry = entry->next;		\
1447 		    KASSERT(entry != &map->header, ("bad entry clip")); \
1448 	    }						\
1449     } while(0)
1450 
1451 
1452 /*
1453  * Clip the specified range and return the base entry.  The
1454  * range may cover several entries starting at the returned base
1455  * and the first and last entry in the covering sequence will be
1456  * properly clipped to the requested start and end address.
1457  *
1458  * If no holes are allowed you should pass the MAP_CLIP_NO_HOLES
1459  * flag.
1460  *
1461  * The MAP_ENTRY_IN_TRANSITION flag will be set for the entries
1462  * covered by the requested range.
1463  *
1464  * The map must be exclusively locked on entry and will remain locked
1465  * on return. If no range exists or the range contains holes and you
1466  * specified that no holes were allowed, NULL will be returned.  This
1467  * routine may temporarily unlock the map in order avoid a deadlock when
1468  * sleeping.
1469  */
1470 static
1471 vm_map_entry_t
1472 vm_map_clip_range(vm_map_t map, vm_offset_t start, vm_offset_t end,
1473 		  int *countp, int flags)
1474 {
1475 	vm_map_entry_t start_entry;
1476 	vm_map_entry_t entry;
1477 
1478 	/*
1479 	 * Locate the entry and effect initial clipping.  The in-transition
1480 	 * case does not occur very often so do not try to optimize it.
1481 	 */
1482 again:
1483 	if (vm_map_lookup_entry(map, start, &start_entry) == FALSE)
1484 		return (NULL);
1485 	entry = start_entry;
1486 	if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
1487 		entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
1488 		++mycpu->gd_cnt.v_intrans_coll;
1489 		++mycpu->gd_cnt.v_intrans_wait;
1490 		vm_map_transition_wait(map);
1491 		/*
1492 		 * entry and/or start_entry may have been clipped while
1493 		 * we slept, or may have gone away entirely.  We have
1494 		 * to restart from the lookup.
1495 		 */
1496 		goto again;
1497 	}
1498 
1499 	/*
1500 	 * Since we hold an exclusive map lock we do not have to restart
1501 	 * after clipping, even though clipping may block in zalloc.
1502 	 */
1503 	vm_map_clip_start(map, entry, start, countp);
1504 	vm_map_clip_end(map, entry, end, countp);
1505 	entry->eflags |= MAP_ENTRY_IN_TRANSITION;
1506 
1507 	/*
1508 	 * Scan entries covered by the range.  When working on the next
1509 	 * entry a restart need only re-loop on the current entry which
1510 	 * we have already locked, since 'next' may have changed.  Also,
1511 	 * even though entry is safe, it may have been clipped so we
1512 	 * have to iterate forwards through the clip after sleeping.
1513 	 */
1514 	while (entry->next != &map->header && entry->next->start < end) {
1515 		vm_map_entry_t next = entry->next;
1516 
1517 		if (flags & MAP_CLIP_NO_HOLES) {
1518 			if (next->start > entry->end) {
1519 				vm_map_unclip_range(map, start_entry,
1520 					start, entry->end, countp, flags);
1521 				return(NULL);
1522 			}
1523 		}
1524 
1525 		if (next->eflags & MAP_ENTRY_IN_TRANSITION) {
1526 			vm_offset_t save_end = entry->end;
1527 			next->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
1528 			++mycpu->gd_cnt.v_intrans_coll;
1529 			++mycpu->gd_cnt.v_intrans_wait;
1530 			vm_map_transition_wait(map);
1531 
1532 			/*
1533 			 * clips might have occured while we blocked.
1534 			 */
1535 			CLIP_CHECK_FWD(entry, save_end);
1536 			CLIP_CHECK_BACK(start_entry, start);
1537 			continue;
1538 		}
1539 		/*
1540 		 * No restart necessary even though clip_end may block, we
1541 		 * are holding the map lock.
1542 		 */
1543 		vm_map_clip_end(map, next, end, countp);
1544 		next->eflags |= MAP_ENTRY_IN_TRANSITION;
1545 		entry = next;
1546 	}
1547 	if (flags & MAP_CLIP_NO_HOLES) {
1548 		if (entry->end != end) {
1549 			vm_map_unclip_range(map, start_entry,
1550 				start, entry->end, countp, flags);
1551 			return(NULL);
1552 		}
1553 	}
1554 	return(start_entry);
1555 }
1556 
1557 /*
1558  * Undo the effect of vm_map_clip_range().  You should pass the same
1559  * flags and the same range that you passed to vm_map_clip_range().
1560  * This code will clear the in-transition flag on the entries and
1561  * wake up anyone waiting.  This code will also simplify the sequence
1562  * and attempt to merge it with entries before and after the sequence.
1563  *
1564  * The map must be locked on entry and will remain locked on return.
1565  *
1566  * Note that you should also pass the start_entry returned by
1567  * vm_map_clip_range().  However, if you block between the two calls
1568  * with the map unlocked please be aware that the start_entry may
1569  * have been clipped and you may need to scan it backwards to find
1570  * the entry corresponding with the original start address.  You are
1571  * responsible for this, vm_map_unclip_range() expects the correct
1572  * start_entry to be passed to it and will KASSERT otherwise.
1573  */
1574 static
1575 void
1576 vm_map_unclip_range(vm_map_t map, vm_map_entry_t start_entry,
1577 		    vm_offset_t start, vm_offset_t end,
1578 		    int *countp, int flags)
1579 {
1580 	vm_map_entry_t entry;
1581 
1582 	entry = start_entry;
1583 
1584 	KASSERT(entry->start == start, ("unclip_range: illegal base entry"));
1585 	while (entry != &map->header && entry->start < end) {
1586 		KASSERT(entry->eflags & MAP_ENTRY_IN_TRANSITION,
1587 			("in-transition flag not set during unclip on: %p",
1588 			entry));
1589 		KASSERT(entry->end <= end,
1590 			("unclip_range: tail wasn't clipped"));
1591 		entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
1592 		if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
1593 			entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
1594 			wakeup(map);
1595 		}
1596 		entry = entry->next;
1597 	}
1598 
1599 	/*
1600 	 * Simplification does not block so there is no restart case.
1601 	 */
1602 	entry = start_entry;
1603 	while (entry != &map->header && entry->start < end) {
1604 		vm_map_simplify_entry(map, entry, countp);
1605 		entry = entry->next;
1606 	}
1607 }
1608 
1609 /*
1610  * Mark the given range as handled by a subordinate map.
1611  *
1612  * This range must have been created with vm_map_find(), and no other
1613  * operations may have been performed on this range prior to calling
1614  * vm_map_submap().
1615  *
1616  * Submappings cannot be removed.
1617  *
1618  * No requirements.
1619  */
1620 int
1621 vm_map_submap(vm_map_t map, vm_offset_t start, vm_offset_t end, vm_map_t submap)
1622 {
1623 	vm_map_entry_t entry;
1624 	int result = KERN_INVALID_ARGUMENT;
1625 	int count;
1626 
1627 	count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
1628 	vm_map_lock(map);
1629 
1630 	VM_MAP_RANGE_CHECK(map, start, end);
1631 
1632 	if (vm_map_lookup_entry(map, start, &entry)) {
1633 		vm_map_clip_start(map, entry, start, &count);
1634 	} else {
1635 		entry = entry->next;
1636 	}
1637 
1638 	vm_map_clip_end(map, entry, end, &count);
1639 
1640 	if ((entry->start == start) && (entry->end == end) &&
1641 	    ((entry->eflags & MAP_ENTRY_COW) == 0) &&
1642 	    (entry->object.vm_object == NULL)) {
1643 		entry->object.sub_map = submap;
1644 		entry->maptype = VM_MAPTYPE_SUBMAP;
1645 		result = KERN_SUCCESS;
1646 	}
1647 	vm_map_unlock(map);
1648 	vm_map_entry_release(count);
1649 
1650 	return (result);
1651 }
1652 
1653 /*
1654  * Sets the protection of the specified address region in the target map.
1655  * If "set_max" is specified, the maximum protection is to be set;
1656  * otherwise, only the current protection is affected.
1657  *
1658  * The protection is not applicable to submaps, but is applicable to normal
1659  * maps and maps governed by virtual page tables.  For example, when operating
1660  * on a virtual page table our protection basically controls how COW occurs
1661  * on the backing object, whereas the virtual page table abstraction itself
1662  * is an abstraction for userland.
1663  *
1664  * No requirements.
1665  */
1666 int
1667 vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end,
1668 	       vm_prot_t new_prot, boolean_t set_max)
1669 {
1670 	vm_map_entry_t current;
1671 	vm_map_entry_t entry;
1672 	int count;
1673 
1674 	count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
1675 	vm_map_lock(map);
1676 
1677 	VM_MAP_RANGE_CHECK(map, start, end);
1678 
1679 	if (vm_map_lookup_entry(map, start, &entry)) {
1680 		vm_map_clip_start(map, entry, start, &count);
1681 	} else {
1682 		entry = entry->next;
1683 	}
1684 
1685 	/*
1686 	 * Make a first pass to check for protection violations.
1687 	 */
1688 	current = entry;
1689 	while ((current != &map->header) && (current->start < end)) {
1690 		if (current->maptype == VM_MAPTYPE_SUBMAP) {
1691 			vm_map_unlock(map);
1692 			vm_map_entry_release(count);
1693 			return (KERN_INVALID_ARGUMENT);
1694 		}
1695 		if ((new_prot & current->max_protection) != new_prot) {
1696 			vm_map_unlock(map);
1697 			vm_map_entry_release(count);
1698 			return (KERN_PROTECTION_FAILURE);
1699 		}
1700 		current = current->next;
1701 	}
1702 
1703 	/*
1704 	 * Go back and fix up protections. [Note that clipping is not
1705 	 * necessary the second time.]
1706 	 */
1707 	current = entry;
1708 
1709 	while ((current != &map->header) && (current->start < end)) {
1710 		vm_prot_t old_prot;
1711 
1712 		vm_map_clip_end(map, current, end, &count);
1713 
1714 		old_prot = current->protection;
1715 		if (set_max) {
1716 			current->protection =
1717 			    (current->max_protection = new_prot) &
1718 			    old_prot;
1719 		} else {
1720 			current->protection = new_prot;
1721 		}
1722 
1723 		/*
1724 		 * Update physical map if necessary. Worry about copy-on-write
1725 		 * here -- CHECK THIS XXX
1726 		 */
1727 
1728 		if (current->protection != old_prot) {
1729 #define MASK(entry)	(((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \
1730 							VM_PROT_ALL)
1731 
1732 			pmap_protect(map->pmap, current->start,
1733 			    current->end,
1734 			    current->protection & MASK(current));
1735 #undef	MASK
1736 		}
1737 
1738 		vm_map_simplify_entry(map, current, &count);
1739 
1740 		current = current->next;
1741 	}
1742 
1743 	vm_map_unlock(map);
1744 	vm_map_entry_release(count);
1745 	return (KERN_SUCCESS);
1746 }
1747 
1748 /*
1749  * This routine traverses a processes map handling the madvise
1750  * system call.  Advisories are classified as either those effecting
1751  * the vm_map_entry structure, or those effecting the underlying
1752  * objects.
1753  *
1754  * The <value> argument is used for extended madvise calls.
1755  *
1756  * No requirements.
1757  */
1758 int
1759 vm_map_madvise(vm_map_t map, vm_offset_t start, vm_offset_t end,
1760 	       int behav, off_t value)
1761 {
1762 	vm_map_entry_t current, entry;
1763 	int modify_map = 0;
1764 	int error = 0;
1765 	int count;
1766 
1767 	/*
1768 	 * Some madvise calls directly modify the vm_map_entry, in which case
1769 	 * we need to use an exclusive lock on the map and we need to perform
1770 	 * various clipping operations.  Otherwise we only need a read-lock
1771 	 * on the map.
1772 	 */
1773 
1774 	count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
1775 
1776 	switch(behav) {
1777 	case MADV_NORMAL:
1778 	case MADV_SEQUENTIAL:
1779 	case MADV_RANDOM:
1780 	case MADV_NOSYNC:
1781 	case MADV_AUTOSYNC:
1782 	case MADV_NOCORE:
1783 	case MADV_CORE:
1784 	case MADV_SETMAP:
1785 	case MADV_INVAL:
1786 		modify_map = 1;
1787 		vm_map_lock(map);
1788 		break;
1789 	case MADV_WILLNEED:
1790 	case MADV_DONTNEED:
1791 	case MADV_FREE:
1792 		vm_map_lock_read(map);
1793 		break;
1794 	default:
1795 		vm_map_entry_release(count);
1796 		return (EINVAL);
1797 	}
1798 
1799 	/*
1800 	 * Locate starting entry and clip if necessary.
1801 	 */
1802 
1803 	VM_MAP_RANGE_CHECK(map, start, end);
1804 
1805 	if (vm_map_lookup_entry(map, start, &entry)) {
1806 		if (modify_map)
1807 			vm_map_clip_start(map, entry, start, &count);
1808 	} else {
1809 		entry = entry->next;
1810 	}
1811 
1812 	if (modify_map) {
1813 		/*
1814 		 * madvise behaviors that are implemented in the vm_map_entry.
1815 		 *
1816 		 * We clip the vm_map_entry so that behavioral changes are
1817 		 * limited to the specified address range.
1818 		 */
1819 		for (current = entry;
1820 		     (current != &map->header) && (current->start < end);
1821 		     current = current->next
1822 		) {
1823 			if (current->maptype == VM_MAPTYPE_SUBMAP)
1824 				continue;
1825 
1826 			vm_map_clip_end(map, current, end, &count);
1827 
1828 			switch (behav) {
1829 			case MADV_NORMAL:
1830 				vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL);
1831 				break;
1832 			case MADV_SEQUENTIAL:
1833 				vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL);
1834 				break;
1835 			case MADV_RANDOM:
1836 				vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM);
1837 				break;
1838 			case MADV_NOSYNC:
1839 				current->eflags |= MAP_ENTRY_NOSYNC;
1840 				break;
1841 			case MADV_AUTOSYNC:
1842 				current->eflags &= ~MAP_ENTRY_NOSYNC;
1843 				break;
1844 			case MADV_NOCORE:
1845 				current->eflags |= MAP_ENTRY_NOCOREDUMP;
1846 				break;
1847 			case MADV_CORE:
1848 				current->eflags &= ~MAP_ENTRY_NOCOREDUMP;
1849 				break;
1850 			case MADV_INVAL:
1851 				/*
1852 				 * Invalidate the related pmap entries, used
1853 				 * to flush portions of the real kernel's
1854 				 * pmap when the caller has removed or
1855 				 * modified existing mappings in a virtual
1856 				 * page table.
1857 				 */
1858 				pmap_remove(map->pmap,
1859 					    current->start, current->end);
1860 				break;
1861 			case MADV_SETMAP:
1862 				/*
1863 				 * Set the page directory page for a map
1864 				 * governed by a virtual page table.  Mark
1865 				 * the entry as being governed by a virtual
1866 				 * page table if it is not.
1867 				 *
1868 				 * XXX the page directory page is stored
1869 				 * in the avail_ssize field if the map_entry.
1870 				 *
1871 				 * XXX the map simplification code does not
1872 				 * compare this field so weird things may
1873 				 * happen if you do not apply this function
1874 				 * to the entire mapping governed by the
1875 				 * virtual page table.
1876 				 */
1877 				if (current->maptype != VM_MAPTYPE_VPAGETABLE) {
1878 					error = EINVAL;
1879 					break;
1880 				}
1881 				current->aux.master_pde = value;
1882 				pmap_remove(map->pmap,
1883 					    current->start, current->end);
1884 				break;
1885 			default:
1886 				error = EINVAL;
1887 				break;
1888 			}
1889 			vm_map_simplify_entry(map, current, &count);
1890 		}
1891 		vm_map_unlock(map);
1892 	} else {
1893 		vm_pindex_t pindex;
1894 		int count;
1895 
1896 		/*
1897 		 * madvise behaviors that are implemented in the underlying
1898 		 * vm_object.
1899 		 *
1900 		 * Since we don't clip the vm_map_entry, we have to clip
1901 		 * the vm_object pindex and count.
1902 		 *
1903 		 * NOTE!  We currently do not support these functions on
1904 		 * virtual page tables.
1905 		 */
1906 		for (current = entry;
1907 		     (current != &map->header) && (current->start < end);
1908 		     current = current->next
1909 		) {
1910 			vm_offset_t useStart;
1911 
1912 			if (current->maptype != VM_MAPTYPE_NORMAL)
1913 				continue;
1914 
1915 			pindex = OFF_TO_IDX(current->offset);
1916 			count = atop(current->end - current->start);
1917 			useStart = current->start;
1918 
1919 			if (current->start < start) {
1920 				pindex += atop(start - current->start);
1921 				count -= atop(start - current->start);
1922 				useStart = start;
1923 			}
1924 			if (current->end > end)
1925 				count -= atop(current->end - end);
1926 
1927 			if (count <= 0)
1928 				continue;
1929 
1930 			vm_object_madvise(current->object.vm_object,
1931 					  pindex, count, behav);
1932 
1933 			/*
1934 			 * Try to populate the page table.  Mappings governed
1935 			 * by virtual page tables cannot be pre-populated
1936 			 * without a lot of work so don't try.
1937 			 */
1938 			if (behav == MADV_WILLNEED &&
1939 			    current->maptype != VM_MAPTYPE_VPAGETABLE) {
1940 				pmap_object_init_pt(
1941 				    map->pmap,
1942 				    useStart,
1943 				    current->protection,
1944 				    current->object.vm_object,
1945 				    pindex,
1946 				    (count << PAGE_SHIFT),
1947 				    MAP_PREFAULT_MADVISE
1948 				);
1949 			}
1950 		}
1951 		vm_map_unlock_read(map);
1952 	}
1953 	vm_map_entry_release(count);
1954 	return(error);
1955 }
1956 
1957 
1958 /*
1959  * Sets the inheritance of the specified address range in the target map.
1960  * Inheritance affects how the map will be shared with child maps at the
1961  * time of vm_map_fork.
1962  */
1963 int
1964 vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end,
1965 	       vm_inherit_t new_inheritance)
1966 {
1967 	vm_map_entry_t entry;
1968 	vm_map_entry_t temp_entry;
1969 	int count;
1970 
1971 	switch (new_inheritance) {
1972 	case VM_INHERIT_NONE:
1973 	case VM_INHERIT_COPY:
1974 	case VM_INHERIT_SHARE:
1975 		break;
1976 	default:
1977 		return (KERN_INVALID_ARGUMENT);
1978 	}
1979 
1980 	count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
1981 	vm_map_lock(map);
1982 
1983 	VM_MAP_RANGE_CHECK(map, start, end);
1984 
1985 	if (vm_map_lookup_entry(map, start, &temp_entry)) {
1986 		entry = temp_entry;
1987 		vm_map_clip_start(map, entry, start, &count);
1988 	} else
1989 		entry = temp_entry->next;
1990 
1991 	while ((entry != &map->header) && (entry->start < end)) {
1992 		vm_map_clip_end(map, entry, end, &count);
1993 
1994 		entry->inheritance = new_inheritance;
1995 
1996 		vm_map_simplify_entry(map, entry, &count);
1997 
1998 		entry = entry->next;
1999 	}
2000 	vm_map_unlock(map);
2001 	vm_map_entry_release(count);
2002 	return (KERN_SUCCESS);
2003 }
2004 
2005 /*
2006  * Implement the semantics of mlock
2007  */
2008 int
2009 vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t real_end,
2010 	      boolean_t new_pageable)
2011 {
2012 	vm_map_entry_t entry;
2013 	vm_map_entry_t start_entry;
2014 	vm_offset_t end;
2015 	int rv = KERN_SUCCESS;
2016 	int count;
2017 
2018 	count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
2019 	vm_map_lock(map);
2020 	VM_MAP_RANGE_CHECK(map, start, real_end);
2021 	end = real_end;
2022 
2023 	start_entry = vm_map_clip_range(map, start, end, &count,
2024 					MAP_CLIP_NO_HOLES);
2025 	if (start_entry == NULL) {
2026 		vm_map_unlock(map);
2027 		vm_map_entry_release(count);
2028 		return (KERN_INVALID_ADDRESS);
2029 	}
2030 
2031 	if (new_pageable == 0) {
2032 		entry = start_entry;
2033 		while ((entry != &map->header) && (entry->start < end)) {
2034 			vm_offset_t save_start;
2035 			vm_offset_t save_end;
2036 
2037 			/*
2038 			 * Already user wired or hard wired (trivial cases)
2039 			 */
2040 			if (entry->eflags & MAP_ENTRY_USER_WIRED) {
2041 				entry = entry->next;
2042 				continue;
2043 			}
2044 			if (entry->wired_count != 0) {
2045 				entry->wired_count++;
2046 				entry->eflags |= MAP_ENTRY_USER_WIRED;
2047 				entry = entry->next;
2048 				continue;
2049 			}
2050 
2051 			/*
2052 			 * A new wiring requires instantiation of appropriate
2053 			 * management structures and the faulting in of the
2054 			 * page.
2055 			 */
2056 			if (entry->maptype != VM_MAPTYPE_SUBMAP) {
2057 				int copyflag = entry->eflags &
2058 					       MAP_ENTRY_NEEDS_COPY;
2059 				if (copyflag && ((entry->protection &
2060 						  VM_PROT_WRITE) != 0)) {
2061 					vm_map_entry_shadow(entry);
2062 				} else if (entry->object.vm_object == NULL &&
2063 					   !map->system_map) {
2064 					vm_map_entry_allocate_object(entry);
2065 				}
2066 			}
2067 			entry->wired_count++;
2068 			entry->eflags |= MAP_ENTRY_USER_WIRED;
2069 
2070 			/*
2071 			 * Now fault in the area.  Note that vm_fault_wire()
2072 			 * may release the map lock temporarily, it will be
2073 			 * relocked on return.  The in-transition
2074 			 * flag protects the entries.
2075 			 */
2076 			save_start = entry->start;
2077 			save_end = entry->end;
2078 			rv = vm_fault_wire(map, entry, TRUE);
2079 			if (rv) {
2080 				CLIP_CHECK_BACK(entry, save_start);
2081 				for (;;) {
2082 					KASSERT(entry->wired_count == 1, ("bad wired_count on entry"));
2083 					entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2084 					entry->wired_count = 0;
2085 					if (entry->end == save_end)
2086 						break;
2087 					entry = entry->next;
2088 					KASSERT(entry != &map->header, ("bad entry clip during backout"));
2089 				}
2090 				end = save_start;	/* unwire the rest */
2091 				break;
2092 			}
2093 			/*
2094 			 * note that even though the entry might have been
2095 			 * clipped, the USER_WIRED flag we set prevents
2096 			 * duplication so we do not have to do a
2097 			 * clip check.
2098 			 */
2099 			entry = entry->next;
2100 		}
2101 
2102 		/*
2103 		 * If we failed fall through to the unwiring section to
2104 		 * unwire what we had wired so far.  'end' has already
2105 		 * been adjusted.
2106 		 */
2107 		if (rv)
2108 			new_pageable = 1;
2109 
2110 		/*
2111 		 * start_entry might have been clipped if we unlocked the
2112 		 * map and blocked.  No matter how clipped it has gotten
2113 		 * there should be a fragment that is on our start boundary.
2114 		 */
2115 		CLIP_CHECK_BACK(start_entry, start);
2116 	}
2117 
2118 	/*
2119 	 * Deal with the unwiring case.
2120 	 */
2121 	if (new_pageable) {
2122 		/*
2123 		 * This is the unwiring case.  We must first ensure that the
2124 		 * range to be unwired is really wired down.  We know there
2125 		 * are no holes.
2126 		 */
2127 		entry = start_entry;
2128 		while ((entry != &map->header) && (entry->start < end)) {
2129 			if ((entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
2130 				rv = KERN_INVALID_ARGUMENT;
2131 				goto done;
2132 			}
2133 			KASSERT(entry->wired_count != 0, ("wired count was 0 with USER_WIRED set! %p", entry));
2134 			entry = entry->next;
2135 		}
2136 
2137 		/*
2138 		 * Now decrement the wiring count for each region. If a region
2139 		 * becomes completely unwired, unwire its physical pages and
2140 		 * mappings.
2141 		 */
2142 		/*
2143 		 * The map entries are processed in a loop, checking to
2144 		 * make sure the entry is wired and asserting it has a wired
2145 		 * count. However, another loop was inserted more-or-less in
2146 		 * the middle of the unwiring path. This loop picks up the
2147 		 * "entry" loop variable from the first loop without first
2148 		 * setting it to start_entry. Naturally, the secound loop
2149 		 * is never entered and the pages backing the entries are
2150 		 * never unwired. This can lead to a leak of wired pages.
2151 		 */
2152 		entry = start_entry;
2153 		while ((entry != &map->header) && (entry->start < end)) {
2154 			KASSERT(entry->eflags & MAP_ENTRY_USER_WIRED,
2155 				("expected USER_WIRED on entry %p", entry));
2156 			entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2157 			entry->wired_count--;
2158 			if (entry->wired_count == 0)
2159 				vm_fault_unwire(map, entry);
2160 			entry = entry->next;
2161 		}
2162 	}
2163 done:
2164 	vm_map_unclip_range(map, start_entry, start, real_end, &count,
2165 		MAP_CLIP_NO_HOLES);
2166 	map->timestamp++;
2167 	vm_map_unlock(map);
2168 	vm_map_entry_release(count);
2169 	return (rv);
2170 }
2171 
2172 /*
2173  * Sets the pageability of the specified address range in the target map.
2174  * Regions specified as not pageable require locked-down physical
2175  * memory and physical page maps.
2176  *
2177  * The map must not be locked, but a reference must remain to the map
2178  * throughout the call.
2179  *
2180  * This function may be called via the zalloc path and must properly
2181  * reserve map entries for kernel_map.
2182  *
2183  * No requirements.
2184  */
2185 int
2186 vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t real_end, int kmflags)
2187 {
2188 	vm_map_entry_t entry;
2189 	vm_map_entry_t start_entry;
2190 	vm_offset_t end;
2191 	int rv = KERN_SUCCESS;
2192 	int count;
2193 
2194 	if (kmflags & KM_KRESERVE)
2195 		count = vm_map_entry_kreserve(MAP_RESERVE_COUNT);
2196 	else
2197 		count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
2198 	vm_map_lock(map);
2199 	VM_MAP_RANGE_CHECK(map, start, real_end);
2200 	end = real_end;
2201 
2202 	start_entry = vm_map_clip_range(map, start, end, &count,
2203 					MAP_CLIP_NO_HOLES);
2204 	if (start_entry == NULL) {
2205 		vm_map_unlock(map);
2206 		rv = KERN_INVALID_ADDRESS;
2207 		goto failure;
2208 	}
2209 	if ((kmflags & KM_PAGEABLE) == 0) {
2210 		/*
2211 		 * Wiring.
2212 		 *
2213 		 * 1.  Holding the write lock, we create any shadow or zero-fill
2214 		 * objects that need to be created. Then we clip each map
2215 		 * entry to the region to be wired and increment its wiring
2216 		 * count.  We create objects before clipping the map entries
2217 		 * to avoid object proliferation.
2218 		 *
2219 		 * 2.  We downgrade to a read lock, and call vm_fault_wire to
2220 		 * fault in the pages for any newly wired area (wired_count is
2221 		 * 1).
2222 		 *
2223 		 * Downgrading to a read lock for vm_fault_wire avoids a
2224 		 * possible deadlock with another process that may have faulted
2225 		 * on one of the pages to be wired (it would mark the page busy,
2226 		 * blocking us, then in turn block on the map lock that we
2227 		 * hold).  Because of problems in the recursive lock package,
2228 		 * we cannot upgrade to a write lock in vm_map_lookup.  Thus,
2229 		 * any actions that require the write lock must be done
2230 		 * beforehand.  Because we keep the read lock on the map, the
2231 		 * copy-on-write status of the entries we modify here cannot
2232 		 * change.
2233 		 */
2234 		entry = start_entry;
2235 		while ((entry != &map->header) && (entry->start < end)) {
2236 			/*
2237 			 * Trivial case if the entry is already wired
2238 			 */
2239 			if (entry->wired_count) {
2240 				entry->wired_count++;
2241 				entry = entry->next;
2242 				continue;
2243 			}
2244 
2245 			/*
2246 			 * The entry is being newly wired, we have to setup
2247 			 * appropriate management structures.  A shadow
2248 			 * object is required for a copy-on-write region,
2249 			 * or a normal object for a zero-fill region.  We
2250 			 * do not have to do this for entries that point to sub
2251 			 * maps because we won't hold the lock on the sub map.
2252 			 */
2253 			if (entry->maptype != VM_MAPTYPE_SUBMAP) {
2254 				int copyflag = entry->eflags &
2255 					       MAP_ENTRY_NEEDS_COPY;
2256 				if (copyflag && ((entry->protection &
2257 						  VM_PROT_WRITE) != 0)) {
2258 					vm_map_entry_shadow(entry);
2259 				} else if (entry->object.vm_object == NULL &&
2260 					   !map->system_map) {
2261 					vm_map_entry_allocate_object(entry);
2262 				}
2263 			}
2264 
2265 			entry->wired_count++;
2266 			entry = entry->next;
2267 		}
2268 
2269 		/*
2270 		 * Pass 2.
2271 		 */
2272 
2273 		/*
2274 		 * HACK HACK HACK HACK
2275 		 *
2276 		 * vm_fault_wire() temporarily unlocks the map to avoid
2277 		 * deadlocks.  The in-transition flag from vm_map_clip_range
2278 		 * call should protect us from changes while the map is
2279 		 * unlocked.  T
2280 		 *
2281 		 * NOTE: Previously this comment stated that clipping might
2282 		 *	 still occur while the entry is unlocked, but from
2283 		 *	 what I can tell it actually cannot.
2284 		 *
2285 		 *	 It is unclear whether the CLIP_CHECK_*() calls
2286 		 *	 are still needed but we keep them in anyway.
2287 		 *
2288 		 * HACK HACK HACK HACK
2289 		 */
2290 
2291 		entry = start_entry;
2292 		while (entry != &map->header && entry->start < end) {
2293 			/*
2294 			 * If vm_fault_wire fails for any page we need to undo
2295 			 * what has been done.  We decrement the wiring count
2296 			 * for those pages which have not yet been wired (now)
2297 			 * and unwire those that have (later).
2298 			 */
2299 			vm_offset_t save_start = entry->start;
2300 			vm_offset_t save_end = entry->end;
2301 
2302 			if (entry->wired_count == 1)
2303 				rv = vm_fault_wire(map, entry, FALSE);
2304 			if (rv) {
2305 				CLIP_CHECK_BACK(entry, save_start);
2306 				for (;;) {
2307 					KASSERT(entry->wired_count == 1, ("wired_count changed unexpectedly"));
2308 					entry->wired_count = 0;
2309 					if (entry->end == save_end)
2310 						break;
2311 					entry = entry->next;
2312 					KASSERT(entry != &map->header, ("bad entry clip during backout"));
2313 				}
2314 				end = save_start;
2315 				break;
2316 			}
2317 			CLIP_CHECK_FWD(entry, save_end);
2318 			entry = entry->next;
2319 		}
2320 
2321 		/*
2322 		 * If a failure occured undo everything by falling through
2323 		 * to the unwiring code.  'end' has already been adjusted
2324 		 * appropriately.
2325 		 */
2326 		if (rv)
2327 			kmflags |= KM_PAGEABLE;
2328 
2329 		/*
2330 		 * start_entry is still IN_TRANSITION but may have been
2331 		 * clipped since vm_fault_wire() unlocks and relocks the
2332 		 * map.  No matter how clipped it has gotten there should
2333 		 * be a fragment that is on our start boundary.
2334 		 */
2335 		CLIP_CHECK_BACK(start_entry, start);
2336 	}
2337 
2338 	if (kmflags & KM_PAGEABLE) {
2339 		/*
2340 		 * This is the unwiring case.  We must first ensure that the
2341 		 * range to be unwired is really wired down.  We know there
2342 		 * are no holes.
2343 		 */
2344 		entry = start_entry;
2345 		while ((entry != &map->header) && (entry->start < end)) {
2346 			if (entry->wired_count == 0) {
2347 				rv = KERN_INVALID_ARGUMENT;
2348 				goto done;
2349 			}
2350 			entry = entry->next;
2351 		}
2352 
2353 		/*
2354 		 * Now decrement the wiring count for each region. If a region
2355 		 * becomes completely unwired, unwire its physical pages and
2356 		 * mappings.
2357 		 */
2358 		entry = start_entry;
2359 		while ((entry != &map->header) && (entry->start < end)) {
2360 			entry->wired_count--;
2361 			if (entry->wired_count == 0)
2362 				vm_fault_unwire(map, entry);
2363 			entry = entry->next;
2364 		}
2365 	}
2366 done:
2367 	vm_map_unclip_range(map, start_entry, start, real_end,
2368 			    &count, MAP_CLIP_NO_HOLES);
2369 	map->timestamp++;
2370 	vm_map_unlock(map);
2371 failure:
2372 	if (kmflags & KM_KRESERVE)
2373 		vm_map_entry_krelease(count);
2374 	else
2375 		vm_map_entry_release(count);
2376 	return (rv);
2377 }
2378 
2379 /*
2380  * Mark a newly allocated address range as wired but do not fault in
2381  * the pages.  The caller is expected to load the pages into the object.
2382  *
2383  * The map must be locked on entry and will remain locked on return.
2384  * No other requirements.
2385  */
2386 void
2387 vm_map_set_wired_quick(vm_map_t map, vm_offset_t addr, vm_size_t size,
2388 		       int *countp)
2389 {
2390 	vm_map_entry_t scan;
2391 	vm_map_entry_t entry;
2392 
2393 	entry = vm_map_clip_range(map, addr, addr + size,
2394 				  countp, MAP_CLIP_NO_HOLES);
2395 	for (scan = entry;
2396 	     scan != &map->header && scan->start < addr + size;
2397 	     scan = scan->next) {
2398 	    KKASSERT(entry->wired_count == 0);
2399 	    entry->wired_count = 1;
2400 	}
2401 	vm_map_unclip_range(map, entry, addr, addr + size,
2402 			    countp, MAP_CLIP_NO_HOLES);
2403 }
2404 
2405 /*
2406  * Push any dirty cached pages in the address range to their pager.
2407  * If syncio is TRUE, dirty pages are written synchronously.
2408  * If invalidate is TRUE, any cached pages are freed as well.
2409  *
2410  * This routine is called by sys_msync()
2411  *
2412  * Returns an error if any part of the specified range is not mapped.
2413  *
2414  * No requirements.
2415  */
2416 int
2417 vm_map_clean(vm_map_t map, vm_offset_t start, vm_offset_t end,
2418 	     boolean_t syncio, boolean_t invalidate)
2419 {
2420 	vm_map_entry_t current;
2421 	vm_map_entry_t entry;
2422 	vm_size_t size;
2423 	vm_object_t object;
2424 	vm_ooffset_t offset;
2425 
2426 	vm_map_lock_read(map);
2427 	VM_MAP_RANGE_CHECK(map, start, end);
2428 	if (!vm_map_lookup_entry(map, start, &entry)) {
2429 		vm_map_unlock_read(map);
2430 		return (KERN_INVALID_ADDRESS);
2431 	}
2432 	/*
2433 	 * Make a first pass to check for holes.
2434 	 */
2435 	for (current = entry; current->start < end; current = current->next) {
2436 		if (current->maptype == VM_MAPTYPE_SUBMAP) {
2437 			vm_map_unlock_read(map);
2438 			return (KERN_INVALID_ARGUMENT);
2439 		}
2440 		if (end > current->end &&
2441 		    (current->next == &map->header ||
2442 			current->end != current->next->start)) {
2443 			vm_map_unlock_read(map);
2444 			return (KERN_INVALID_ADDRESS);
2445 		}
2446 	}
2447 
2448 	if (invalidate)
2449 		pmap_remove(vm_map_pmap(map), start, end);
2450 
2451 	/*
2452 	 * Make a second pass, cleaning/uncaching pages from the indicated
2453 	 * objects as we go.
2454 	 *
2455 	 * Hold vm_token to avoid blocking in vm_object_reference()
2456 	 */
2457 	lwkt_gettoken(&vm_token);
2458 	lwkt_gettoken(&vmobj_token);
2459 
2460 	for (current = entry; current->start < end; current = current->next) {
2461 		offset = current->offset + (start - current->start);
2462 		size = (end <= current->end ? end : current->end) - start;
2463 		if (current->maptype == VM_MAPTYPE_SUBMAP) {
2464 			vm_map_t smap;
2465 			vm_map_entry_t tentry;
2466 			vm_size_t tsize;
2467 
2468 			smap = current->object.sub_map;
2469 			vm_map_lock_read(smap);
2470 			vm_map_lookup_entry(smap, offset, &tentry);
2471 			tsize = tentry->end - offset;
2472 			if (tsize < size)
2473 				size = tsize;
2474 			object = tentry->object.vm_object;
2475 			offset = tentry->offset + (offset - tentry->start);
2476 			vm_map_unlock_read(smap);
2477 		} else {
2478 			object = current->object.vm_object;
2479 		}
2480 		/*
2481 		 * Note that there is absolutely no sense in writing out
2482 		 * anonymous objects, so we track down the vnode object
2483 		 * to write out.
2484 		 * We invalidate (remove) all pages from the address space
2485 		 * anyway, for semantic correctness.
2486 		 *
2487 		 * note: certain anonymous maps, such as MAP_NOSYNC maps,
2488 		 * may start out with a NULL object.
2489 		 */
2490 		while (object && object->backing_object) {
2491 			offset += object->backing_object_offset;
2492 			object = object->backing_object;
2493 			if (object->size < OFF_TO_IDX( offset + size))
2494 				size = IDX_TO_OFF(object->size) - offset;
2495 		}
2496 		if (object && (object->type == OBJT_VNODE) &&
2497 		    (current->protection & VM_PROT_WRITE) &&
2498 		    (object->flags & OBJ_NOMSYNC) == 0) {
2499 			/*
2500 			 * Flush pages if writing is allowed, invalidate them
2501 			 * if invalidation requested.  Pages undergoing I/O
2502 			 * will be ignored by vm_object_page_remove().
2503 			 *
2504 			 * We cannot lock the vnode and then wait for paging
2505 			 * to complete without deadlocking against vm_fault.
2506 			 * Instead we simply call vm_object_page_remove() and
2507 			 * allow it to block internally on a page-by-page
2508 			 * basis when it encounters pages undergoing async
2509 			 * I/O.
2510 			 */
2511 			int flags;
2512 
2513 			vm_object_reference_locked(object);
2514 			vn_lock(object->handle, LK_EXCLUSIVE | LK_RETRY);
2515 			flags = (syncio || invalidate) ? OBJPC_SYNC : 0;
2516 			flags |= invalidate ? OBJPC_INVAL : 0;
2517 
2518 			/*
2519 			 * When operating on a virtual page table just
2520 			 * flush the whole object.  XXX we probably ought
2521 			 * to
2522 			 */
2523 			switch(current->maptype) {
2524 			case VM_MAPTYPE_NORMAL:
2525 				vm_object_page_clean(object,
2526 				    OFF_TO_IDX(offset),
2527 				    OFF_TO_IDX(offset + size + PAGE_MASK),
2528 				    flags);
2529 				break;
2530 			case VM_MAPTYPE_VPAGETABLE:
2531 				vm_object_page_clean(object, 0, 0, flags);
2532 				break;
2533 			}
2534 			vn_unlock(((struct vnode *)object->handle));
2535 			vm_object_deallocate_locked(object);
2536 		}
2537 		if (object && invalidate &&
2538 		   ((object->type == OBJT_VNODE) ||
2539 		    (object->type == OBJT_DEVICE))) {
2540 			int clean_only =
2541 				(object->type == OBJT_DEVICE) ? FALSE : TRUE;
2542 			vm_object_reference_locked(object);
2543 			switch(current->maptype) {
2544 			case VM_MAPTYPE_NORMAL:
2545 				vm_object_page_remove(object,
2546 				    OFF_TO_IDX(offset),
2547 				    OFF_TO_IDX(offset + size + PAGE_MASK),
2548 				    clean_only);
2549 				break;
2550 			case VM_MAPTYPE_VPAGETABLE:
2551 				vm_object_page_remove(object, 0, 0, clean_only);
2552 				break;
2553 			}
2554 			vm_object_deallocate_locked(object);
2555 		}
2556 		start += size;
2557 	}
2558 
2559 	lwkt_reltoken(&vmobj_token);
2560 	lwkt_reltoken(&vm_token);
2561 	vm_map_unlock_read(map);
2562 
2563 	return (KERN_SUCCESS);
2564 }
2565 
2566 /*
2567  * Make the region specified by this entry pageable.
2568  *
2569  * The vm_map must be exclusively locked.
2570  */
2571 static void
2572 vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry)
2573 {
2574 	entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2575 	entry->wired_count = 0;
2576 	vm_fault_unwire(map, entry);
2577 }
2578 
2579 /*
2580  * Deallocate the given entry from the target map.
2581  *
2582  * The vm_map must be exclusively locked.
2583  */
2584 static void
2585 vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry, int *countp)
2586 {
2587 	vm_map_entry_unlink(map, entry);
2588 	map->size -= entry->end - entry->start;
2589 
2590 	switch(entry->maptype) {
2591 	case VM_MAPTYPE_NORMAL:
2592 	case VM_MAPTYPE_VPAGETABLE:
2593 		vm_object_deallocate(entry->object.vm_object);
2594 		break;
2595 	default:
2596 		break;
2597 	}
2598 
2599 	vm_map_entry_dispose(map, entry, countp);
2600 }
2601 
2602 /*
2603  * Deallocates the given address range from the target map.
2604  *
2605  * The vm_map must be exclusively locked.
2606  */
2607 int
2608 vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end, int *countp)
2609 {
2610 	vm_object_t object;
2611 	vm_map_entry_t entry;
2612 	vm_map_entry_t first_entry;
2613 
2614 	ASSERT_VM_MAP_LOCKED(map);
2615 again:
2616 	/*
2617 	 * Find the start of the region, and clip it.  Set entry to point
2618 	 * at the first record containing the requested address or, if no
2619 	 * such record exists, the next record with a greater address.  The
2620 	 * loop will run from this point until a record beyond the termination
2621 	 * address is encountered.
2622 	 *
2623 	 * map->hint must be adjusted to not point to anything we delete,
2624 	 * so set it to the entry prior to the one being deleted.
2625 	 *
2626 	 * GGG see other GGG comment.
2627 	 */
2628 	if (vm_map_lookup_entry(map, start, &first_entry)) {
2629 		entry = first_entry;
2630 		vm_map_clip_start(map, entry, start, countp);
2631 		map->hint = entry->prev;	/* possible problem XXX */
2632 	} else {
2633 		map->hint = first_entry;	/* possible problem XXX */
2634 		entry = first_entry->next;
2635 	}
2636 
2637 	/*
2638 	 * If a hole opens up prior to the current first_free then
2639 	 * adjust first_free.  As with map->hint, map->first_free
2640 	 * cannot be left set to anything we might delete.
2641 	 */
2642 	if (entry == &map->header) {
2643 		map->first_free = &map->header;
2644 	} else if (map->first_free->start >= start) {
2645 		map->first_free = entry->prev;
2646 	}
2647 
2648 	/*
2649 	 * Step through all entries in this region
2650 	 */
2651 	while ((entry != &map->header) && (entry->start < end)) {
2652 		vm_map_entry_t next;
2653 		vm_offset_t s, e;
2654 		vm_pindex_t offidxstart, offidxend, count;
2655 
2656 		/*
2657 		 * If we hit an in-transition entry we have to sleep and
2658 		 * retry.  It's easier (and not really slower) to just retry
2659 		 * since this case occurs so rarely and the hint is already
2660 		 * pointing at the right place.  We have to reset the
2661 		 * start offset so as not to accidently delete an entry
2662 		 * another process just created in vacated space.
2663 		 */
2664 		if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2665 			entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2666 			start = entry->start;
2667 			++mycpu->gd_cnt.v_intrans_coll;
2668 			++mycpu->gd_cnt.v_intrans_wait;
2669 			vm_map_transition_wait(map);
2670 			goto again;
2671 		}
2672 		vm_map_clip_end(map, entry, end, countp);
2673 
2674 		s = entry->start;
2675 		e = entry->end;
2676 		next = entry->next;
2677 
2678 		offidxstart = OFF_TO_IDX(entry->offset);
2679 		count = OFF_TO_IDX(e - s);
2680 		object = entry->object.vm_object;
2681 
2682 		/*
2683 		 * Unwire before removing addresses from the pmap; otherwise,
2684 		 * unwiring will put the entries back in the pmap.
2685 		 */
2686 		if (entry->wired_count != 0)
2687 			vm_map_entry_unwire(map, entry);
2688 
2689 		offidxend = offidxstart + count;
2690 
2691 		/*
2692 		 * Hold vm_token when manipulating vm_objects,
2693 		 *
2694 		 * Hold vmobj_token when potentially adding or removing
2695 		 * objects (collapse requires both).
2696 		 */
2697 		lwkt_gettoken(&vm_token);
2698 		lwkt_gettoken(&vmobj_token);
2699 
2700 		if (object == &kernel_object) {
2701 			vm_object_page_remove(object, offidxstart,
2702 					      offidxend, FALSE);
2703 		} else {
2704 			pmap_remove(map->pmap, s, e);
2705 
2706 			if (object != NULL &&
2707 			    object->ref_count != 1 &&
2708 			    (object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) ==
2709 			     OBJ_ONEMAPPING &&
2710 			    (object->type == OBJT_DEFAULT ||
2711 			     object->type == OBJT_SWAP)) {
2712 				vm_object_collapse(object);
2713 				vm_object_page_remove(object, offidxstart,
2714 						      offidxend, FALSE);
2715 				if (object->type == OBJT_SWAP) {
2716 					swap_pager_freespace(object,
2717 							     offidxstart,
2718 							     count);
2719 				}
2720 				if (offidxend >= object->size &&
2721 				    offidxstart < object->size) {
2722 					object->size = offidxstart;
2723 				}
2724 			}
2725 		}
2726 		lwkt_reltoken(&vmobj_token);
2727 		lwkt_reltoken(&vm_token);
2728 
2729 		/*
2730 		 * Delete the entry (which may delete the object) only after
2731 		 * removing all pmap entries pointing to its pages.
2732 		 * (Otherwise, its page frames may be reallocated, and any
2733 		 * modify bits will be set in the wrong object!)
2734 		 */
2735 		vm_map_entry_delete(map, entry, countp);
2736 		entry = next;
2737 	}
2738 	return (KERN_SUCCESS);
2739 }
2740 
2741 /*
2742  * Remove the given address range from the target map.
2743  * This is the exported form of vm_map_delete.
2744  *
2745  * No requirements.
2746  */
2747 int
2748 vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end)
2749 {
2750 	int result;
2751 	int count;
2752 
2753 	count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
2754 	vm_map_lock(map);
2755 	VM_MAP_RANGE_CHECK(map, start, end);
2756 	result = vm_map_delete(map, start, end, &count);
2757 	vm_map_unlock(map);
2758 	vm_map_entry_release(count);
2759 
2760 	return (result);
2761 }
2762 
2763 /*
2764  * Assert that the target map allows the specified privilege on the
2765  * entire address region given.  The entire region must be allocated.
2766  *
2767  * The caller must specify whether the vm_map is already locked or not.
2768  */
2769 boolean_t
2770 vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end,
2771 			vm_prot_t protection, boolean_t have_lock)
2772 {
2773 	vm_map_entry_t entry;
2774 	vm_map_entry_t tmp_entry;
2775 	boolean_t result;
2776 
2777 	if (have_lock == FALSE)
2778 		vm_map_lock_read(map);
2779 
2780 	if (!vm_map_lookup_entry(map, start, &tmp_entry)) {
2781 		if (have_lock == FALSE)
2782 			vm_map_unlock_read(map);
2783 		return (FALSE);
2784 	}
2785 	entry = tmp_entry;
2786 
2787 	result = TRUE;
2788 	while (start < end) {
2789 		if (entry == &map->header) {
2790 			result = FALSE;
2791 			break;
2792 		}
2793 		/*
2794 		 * No holes allowed!
2795 		 */
2796 
2797 		if (start < entry->start) {
2798 			result = FALSE;
2799 			break;
2800 		}
2801 		/*
2802 		 * Check protection associated with entry.
2803 		 */
2804 
2805 		if ((entry->protection & protection) != protection) {
2806 			result = FALSE;
2807 			break;
2808 		}
2809 		/* go to next entry */
2810 
2811 		start = entry->end;
2812 		entry = entry->next;
2813 	}
2814 	if (have_lock == FALSE)
2815 		vm_map_unlock_read(map);
2816 	return (result);
2817 }
2818 
2819 /*
2820  * Split the pages in a map entry into a new object.  This affords
2821  * easier removal of unused pages, and keeps object inheritance from
2822  * being a negative impact on memory usage.
2823  *
2824  * The vm_map must be exclusively locked.
2825  */
2826 static void
2827 vm_map_split(vm_map_entry_t entry)
2828 {
2829 	vm_page_t m;
2830 	vm_object_t orig_object, new_object, source;
2831 	vm_offset_t s, e;
2832 	vm_pindex_t offidxstart, offidxend, idx;
2833 	vm_size_t size;
2834 	vm_ooffset_t offset;
2835 
2836 	orig_object = entry->object.vm_object;
2837 	if (orig_object->type != OBJT_DEFAULT && orig_object->type != OBJT_SWAP)
2838 		return;
2839 	if (orig_object->ref_count <= 1)
2840 		return;
2841 
2842 	offset = entry->offset;
2843 	s = entry->start;
2844 	e = entry->end;
2845 
2846 	offidxstart = OFF_TO_IDX(offset);
2847 	offidxend = offidxstart + OFF_TO_IDX(e - s);
2848 	size = offidxend - offidxstart;
2849 
2850 	switch(orig_object->type) {
2851 	case OBJT_DEFAULT:
2852 		new_object = default_pager_alloc(NULL, IDX_TO_OFF(size),
2853 						 VM_PROT_ALL, 0);
2854 		break;
2855 	case OBJT_SWAP:
2856 		new_object = swap_pager_alloc(NULL, IDX_TO_OFF(size),
2857 					      VM_PROT_ALL, 0);
2858 		break;
2859 	default:
2860 		/* not reached */
2861 		new_object = NULL;
2862 		KKASSERT(0);
2863 	}
2864 	if (new_object == NULL)
2865 		return;
2866 
2867 	/*
2868 	 * vm_token required when manipulating vm_objects.
2869 	 */
2870 	lwkt_gettoken(&vm_token);
2871 	lwkt_gettoken(&vmobj_token);
2872 
2873 	source = orig_object->backing_object;
2874 	if (source != NULL) {
2875 		/* Referenced by new_object */
2876 		vm_object_reference_locked(source);
2877 		LIST_INSERT_HEAD(&source->shadow_head,
2878 				 new_object, shadow_list);
2879 		vm_object_clear_flag(source, OBJ_ONEMAPPING);
2880 		new_object->backing_object_offset =
2881 			orig_object->backing_object_offset +
2882 			IDX_TO_OFF(offidxstart);
2883 		new_object->backing_object = source;
2884 		source->shadow_count++;
2885 		source->generation++;
2886 	}
2887 
2888 	for (idx = 0; idx < size; idx++) {
2889 		vm_page_t m;
2890 
2891 	retry:
2892 		m = vm_page_lookup(orig_object, offidxstart + idx);
2893 		if (m == NULL)
2894 			continue;
2895 
2896 		/*
2897 		 * We must wait for pending I/O to complete before we can
2898 		 * rename the page.
2899 		 *
2900 		 * We do not have to VM_PROT_NONE the page as mappings should
2901 		 * not be changed by this operation.
2902 		 */
2903 		if (vm_page_sleep_busy(m, TRUE, "spltwt"))
2904 			goto retry;
2905 		vm_page_busy(m);
2906 		vm_page_rename(m, new_object, idx);
2907 		/* page automatically made dirty by rename and cache handled */
2908 		vm_page_busy(m);
2909 	}
2910 
2911 	if (orig_object->type == OBJT_SWAP) {
2912 		vm_object_pip_add(orig_object, 1);
2913 		/*
2914 		 * copy orig_object pages into new_object
2915 		 * and destroy unneeded pages in
2916 		 * shadow object.
2917 		 */
2918 		swap_pager_copy(orig_object, new_object, offidxstart, 0);
2919 		vm_object_pip_wakeup(orig_object);
2920 	}
2921 
2922 	/*
2923 	 * Wakeup the pages we played with.  No spl protection is needed
2924 	 * for a simple wakeup.
2925 	 */
2926 	for (idx = 0; idx < size; idx++) {
2927 		m = vm_page_lookup(new_object, idx);
2928 		if (m)
2929 			vm_page_wakeup(m);
2930 	}
2931 
2932 	entry->object.vm_object = new_object;
2933 	entry->offset = 0LL;
2934 	vm_object_deallocate_locked(orig_object);
2935 	lwkt_reltoken(&vmobj_token);
2936 	lwkt_reltoken(&vm_token);
2937 }
2938 
2939 /*
2940  * Copies the contents of the source entry to the destination
2941  * entry.  The entries *must* be aligned properly.
2942  *
2943  * The vm_map must be exclusively locked.
2944  * vm_token must be held
2945  */
2946 static void
2947 vm_map_copy_entry(vm_map_t src_map, vm_map_t dst_map,
2948 	vm_map_entry_t src_entry, vm_map_entry_t dst_entry)
2949 {
2950 	vm_object_t src_object;
2951 
2952 	if (dst_entry->maptype == VM_MAPTYPE_SUBMAP)
2953 		return;
2954 	if (src_entry->maptype == VM_MAPTYPE_SUBMAP)
2955 		return;
2956 
2957 	ASSERT_LWKT_TOKEN_HELD(&vm_token);
2958 	lwkt_gettoken(&vmobj_token);		/* required for collapse */
2959 
2960 	if (src_entry->wired_count == 0) {
2961 		/*
2962 		 * If the source entry is marked needs_copy, it is already
2963 		 * write-protected.
2964 		 */
2965 		if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) {
2966 			pmap_protect(src_map->pmap,
2967 			    src_entry->start,
2968 			    src_entry->end,
2969 			    src_entry->protection & ~VM_PROT_WRITE);
2970 		}
2971 
2972 		/*
2973 		 * Make a copy of the object.
2974 		 */
2975 		if ((src_object = src_entry->object.vm_object) != NULL) {
2976 			if ((src_object->handle == NULL) &&
2977 				(src_object->type == OBJT_DEFAULT ||
2978 				 src_object->type == OBJT_SWAP)) {
2979 				vm_object_collapse(src_object);
2980 				if ((src_object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) {
2981 					vm_map_split(src_entry);
2982 					src_object = src_entry->object.vm_object;
2983 				}
2984 			}
2985 
2986 			vm_object_reference_locked(src_object);
2987 			vm_object_clear_flag(src_object, OBJ_ONEMAPPING);
2988 			dst_entry->object.vm_object = src_object;
2989 			src_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
2990 			dst_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
2991 			dst_entry->offset = src_entry->offset;
2992 		} else {
2993 			dst_entry->object.vm_object = NULL;
2994 			dst_entry->offset = 0;
2995 		}
2996 
2997 		pmap_copy(dst_map->pmap, src_map->pmap, dst_entry->start,
2998 		    dst_entry->end - dst_entry->start, src_entry->start);
2999 	} else {
3000 		/*
3001 		 * Of course, wired down pages can't be set copy-on-write.
3002 		 * Cause wired pages to be copied into the new map by
3003 		 * simulating faults (the new pages are pageable)
3004 		 */
3005 		vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry);
3006 	}
3007 	lwkt_reltoken(&vmobj_token);
3008 }
3009 
3010 /*
3011  * vmspace_fork:
3012  * Create a new process vmspace structure and vm_map
3013  * based on those of an existing process.  The new map
3014  * is based on the old map, according to the inheritance
3015  * values on the regions in that map.
3016  *
3017  * The source map must not be locked.
3018  * No requirements.
3019  */
3020 struct vmspace *
3021 vmspace_fork(struct vmspace *vm1)
3022 {
3023 	struct vmspace *vm2;
3024 	vm_map_t old_map = &vm1->vm_map;
3025 	vm_map_t new_map;
3026 	vm_map_entry_t old_entry;
3027 	vm_map_entry_t new_entry;
3028 	vm_object_t object;
3029 	int count;
3030 
3031 	lwkt_gettoken(&vm_token);
3032 	lwkt_gettoken(&vmspace_token);
3033 	lwkt_gettoken(&vmobj_token);
3034 	vm_map_lock(old_map);
3035 	old_map->infork = 1;
3036 
3037 	/*
3038 	 * XXX Note: upcalls are not copied.
3039 	 */
3040 	vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset);
3041 	bcopy(&vm1->vm_startcopy, &vm2->vm_startcopy,
3042 	    (caddr_t)&vm1->vm_endcopy - (caddr_t)&vm1->vm_startcopy);
3043 	new_map = &vm2->vm_map;	/* XXX */
3044 	new_map->timestamp = 1;
3045 
3046 	vm_map_lock(new_map);
3047 
3048 	count = 0;
3049 	old_entry = old_map->header.next;
3050 	while (old_entry != &old_map->header) {
3051 		++count;
3052 		old_entry = old_entry->next;
3053 	}
3054 
3055 	count = vm_map_entry_reserve(count + MAP_RESERVE_COUNT);
3056 
3057 	old_entry = old_map->header.next;
3058 	while (old_entry != &old_map->header) {
3059 		if (old_entry->maptype == VM_MAPTYPE_SUBMAP)
3060 			panic("vm_map_fork: encountered a submap");
3061 
3062 		switch (old_entry->inheritance) {
3063 		case VM_INHERIT_NONE:
3064 			break;
3065 		case VM_INHERIT_SHARE:
3066 			/*
3067 			 * Clone the entry, creating the shared object if
3068 			 * necessary.
3069 			 */
3070 			object = old_entry->object.vm_object;
3071 			if (object == NULL) {
3072 				vm_map_entry_allocate_object(old_entry);
3073 				object = old_entry->object.vm_object;
3074 			}
3075 
3076 			/*
3077 			 * Add the reference before calling vm_map_entry_shadow
3078 			 * to insure that a shadow object is created.
3079 			 */
3080 			vm_object_reference_locked(object);
3081 			if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3082 				vm_map_entry_shadow(old_entry);
3083 				/* Transfer the second reference too. */
3084 				vm_object_reference_locked(
3085 				    old_entry->object.vm_object);
3086 				vm_object_deallocate_locked(object);
3087 				object = old_entry->object.vm_object;
3088 			}
3089 			vm_object_clear_flag(object, OBJ_ONEMAPPING);
3090 
3091 			/*
3092 			 * Clone the entry, referencing the shared object.
3093 			 */
3094 			new_entry = vm_map_entry_create(new_map, &count);
3095 			*new_entry = *old_entry;
3096 			new_entry->eflags &= ~MAP_ENTRY_USER_WIRED;
3097 			new_entry->wired_count = 0;
3098 
3099 			/*
3100 			 * Insert the entry into the new map -- we know we're
3101 			 * inserting at the end of the new map.
3102 			 */
3103 
3104 			vm_map_entry_link(new_map, new_map->header.prev,
3105 					  new_entry);
3106 
3107 			/*
3108 			 * Update the physical map
3109 			 */
3110 			pmap_copy(new_map->pmap, old_map->pmap,
3111 			    new_entry->start,
3112 			    (old_entry->end - old_entry->start),
3113 			    old_entry->start);
3114 			break;
3115 		case VM_INHERIT_COPY:
3116 			/*
3117 			 * Clone the entry and link into the map.
3118 			 */
3119 			new_entry = vm_map_entry_create(new_map, &count);
3120 			*new_entry = *old_entry;
3121 			new_entry->eflags &= ~MAP_ENTRY_USER_WIRED;
3122 			new_entry->wired_count = 0;
3123 			new_entry->object.vm_object = NULL;
3124 			vm_map_entry_link(new_map, new_map->header.prev,
3125 					  new_entry);
3126 			vm_map_copy_entry(old_map, new_map, old_entry,
3127 					  new_entry);
3128 			break;
3129 		}
3130 		old_entry = old_entry->next;
3131 	}
3132 
3133 	new_map->size = old_map->size;
3134 	old_map->infork = 0;
3135 	vm_map_unlock(old_map);
3136 	vm_map_unlock(new_map);
3137 	vm_map_entry_release(count);
3138 
3139 	lwkt_reltoken(&vmobj_token);
3140 	lwkt_reltoken(&vmspace_token);
3141 	lwkt_reltoken(&vm_token);
3142 
3143 	return (vm2);
3144 }
3145 
3146 /*
3147  * Create an auto-grow stack entry
3148  *
3149  * No requirements.
3150  */
3151 int
3152 vm_map_stack (vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
3153 	      int flags, vm_prot_t prot, vm_prot_t max, int cow)
3154 {
3155 	vm_map_entry_t	prev_entry;
3156 	vm_map_entry_t	new_stack_entry;
3157 	vm_size_t	init_ssize;
3158 	int		rv;
3159 	int		count;
3160 	vm_offset_t	tmpaddr;
3161 
3162 	cow |= MAP_IS_STACK;
3163 
3164 	if (max_ssize < sgrowsiz)
3165 		init_ssize = max_ssize;
3166 	else
3167 		init_ssize = sgrowsiz;
3168 
3169 	count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
3170 	vm_map_lock(map);
3171 
3172 	/*
3173 	 * Find space for the mapping
3174 	 */
3175 	if ((flags & (MAP_FIXED | MAP_TRYFIXED)) == 0) {
3176 		if (vm_map_findspace(map, addrbos, max_ssize, 1,
3177 				     flags, &tmpaddr)) {
3178 			vm_map_unlock(map);
3179 			vm_map_entry_release(count);
3180 			return (KERN_NO_SPACE);
3181 		}
3182 		addrbos = tmpaddr;
3183 	}
3184 
3185 	/* If addr is already mapped, no go */
3186 	if (vm_map_lookup_entry(map, addrbos, &prev_entry)) {
3187 		vm_map_unlock(map);
3188 		vm_map_entry_release(count);
3189 		return (KERN_NO_SPACE);
3190 	}
3191 
3192 #if 0
3193 	/* XXX already handled by kern_mmap() */
3194 	/* If we would blow our VMEM resource limit, no go */
3195 	if (map->size + init_ssize >
3196 	    curproc->p_rlimit[RLIMIT_VMEM].rlim_cur) {
3197 		vm_map_unlock(map);
3198 		vm_map_entry_release(count);
3199 		return (KERN_NO_SPACE);
3200 	}
3201 #endif
3202 
3203 	/*
3204 	 * If we can't accomodate max_ssize in the current mapping,
3205 	 * no go.  However, we need to be aware that subsequent user
3206 	 * mappings might map into the space we have reserved for
3207 	 * stack, and currently this space is not protected.
3208 	 *
3209 	 * Hopefully we will at least detect this condition
3210 	 * when we try to grow the stack.
3211 	 */
3212 	if ((prev_entry->next != &map->header) &&
3213 	    (prev_entry->next->start < addrbos + max_ssize)) {
3214 		vm_map_unlock(map);
3215 		vm_map_entry_release(count);
3216 		return (KERN_NO_SPACE);
3217 	}
3218 
3219 	/*
3220 	 * We initially map a stack of only init_ssize.  We will
3221 	 * grow as needed later.  Since this is to be a grow
3222 	 * down stack, we map at the top of the range.
3223 	 *
3224 	 * Note: we would normally expect prot and max to be
3225 	 * VM_PROT_ALL, and cow to be 0.  Possibly we should
3226 	 * eliminate these as input parameters, and just
3227 	 * pass these values here in the insert call.
3228 	 */
3229 	rv = vm_map_insert(map, &count,
3230 			   NULL, 0, addrbos + max_ssize - init_ssize,
3231 	                   addrbos + max_ssize,
3232 			   VM_MAPTYPE_NORMAL,
3233 			   prot, max,
3234 			   cow);
3235 
3236 	/* Now set the avail_ssize amount */
3237 	if (rv == KERN_SUCCESS) {
3238 		if (prev_entry != &map->header)
3239 			vm_map_clip_end(map, prev_entry, addrbos + max_ssize - init_ssize, &count);
3240 		new_stack_entry = prev_entry->next;
3241 		if (new_stack_entry->end   != addrbos + max_ssize ||
3242 		    new_stack_entry->start != addrbos + max_ssize - init_ssize)
3243 			panic ("Bad entry start/end for new stack entry");
3244 		else
3245 			new_stack_entry->aux.avail_ssize = max_ssize - init_ssize;
3246 	}
3247 
3248 	vm_map_unlock(map);
3249 	vm_map_entry_release(count);
3250 	return (rv);
3251 }
3252 
3253 /*
3254  * Attempts to grow a vm stack entry.  Returns KERN_SUCCESS if the
3255  * desired address is already mapped, or if we successfully grow
3256  * the stack.  Also returns KERN_SUCCESS if addr is outside the
3257  * stack range (this is strange, but preserves compatibility with
3258  * the grow function in vm_machdep.c).
3259  *
3260  * No requirements.
3261  */
3262 int
3263 vm_map_growstack (struct proc *p, vm_offset_t addr)
3264 {
3265 	vm_map_entry_t prev_entry;
3266 	vm_map_entry_t stack_entry;
3267 	vm_map_entry_t new_stack_entry;
3268 	struct vmspace *vm = p->p_vmspace;
3269 	vm_map_t map = &vm->vm_map;
3270 	vm_offset_t    end;
3271 	int grow_amount;
3272 	int rv = KERN_SUCCESS;
3273 	int is_procstack;
3274 	int use_read_lock = 1;
3275 	int count;
3276 
3277 	count = vm_map_entry_reserve(MAP_RESERVE_COUNT);
3278 Retry:
3279 	if (use_read_lock)
3280 		vm_map_lock_read(map);
3281 	else
3282 		vm_map_lock(map);
3283 
3284 	/* If addr is already in the entry range, no need to grow.*/
3285 	if (vm_map_lookup_entry(map, addr, &prev_entry))
3286 		goto done;
3287 
3288 	if ((stack_entry = prev_entry->next) == &map->header)
3289 		goto done;
3290 	if (prev_entry == &map->header)
3291 		end = stack_entry->start - stack_entry->aux.avail_ssize;
3292 	else
3293 		end = prev_entry->end;
3294 
3295 	/*
3296 	 * This next test mimics the old grow function in vm_machdep.c.
3297 	 * It really doesn't quite make sense, but we do it anyway
3298 	 * for compatibility.
3299 	 *
3300 	 * If not growable stack, return success.  This signals the
3301 	 * caller to proceed as he would normally with normal vm.
3302 	 */
3303 	if (stack_entry->aux.avail_ssize < 1 ||
3304 	    addr >= stack_entry->start ||
3305 	    addr <  stack_entry->start - stack_entry->aux.avail_ssize) {
3306 		goto done;
3307 	}
3308 
3309 	/* Find the minimum grow amount */
3310 	grow_amount = roundup (stack_entry->start - addr, PAGE_SIZE);
3311 	if (grow_amount > stack_entry->aux.avail_ssize) {
3312 		rv = KERN_NO_SPACE;
3313 		goto done;
3314 	}
3315 
3316 	/*
3317 	 * If there is no longer enough space between the entries
3318 	 * nogo, and adjust the available space.  Note: this
3319 	 * should only happen if the user has mapped into the
3320 	 * stack area after the stack was created, and is
3321 	 * probably an error.
3322 	 *
3323 	 * This also effectively destroys any guard page the user
3324 	 * might have intended by limiting the stack size.
3325 	 */
3326 	if (grow_amount > stack_entry->start - end) {
3327 		if (use_read_lock && vm_map_lock_upgrade(map)) {
3328 			use_read_lock = 0;
3329 			goto Retry;
3330 		}
3331 		use_read_lock = 0;
3332 		stack_entry->aux.avail_ssize = stack_entry->start - end;
3333 		rv = KERN_NO_SPACE;
3334 		goto done;
3335 	}
3336 
3337 	is_procstack = addr >= (vm_offset_t)vm->vm_maxsaddr;
3338 
3339 	/* If this is the main process stack, see if we're over the
3340 	 * stack limit.
3341 	 */
3342 	if (is_procstack && (ctob(vm->vm_ssize) + grow_amount >
3343 			     p->p_rlimit[RLIMIT_STACK].rlim_cur)) {
3344 		rv = KERN_NO_SPACE;
3345 		goto done;
3346 	}
3347 
3348 	/* Round up the grow amount modulo SGROWSIZ */
3349 	grow_amount = roundup (grow_amount, sgrowsiz);
3350 	if (grow_amount > stack_entry->aux.avail_ssize) {
3351 		grow_amount = stack_entry->aux.avail_ssize;
3352 	}
3353 	if (is_procstack && (ctob(vm->vm_ssize) + grow_amount >
3354 	                     p->p_rlimit[RLIMIT_STACK].rlim_cur)) {
3355 		grow_amount = p->p_rlimit[RLIMIT_STACK].rlim_cur -
3356 		              ctob(vm->vm_ssize);
3357 	}
3358 
3359 	/* If we would blow our VMEM resource limit, no go */
3360 	if (map->size + grow_amount > p->p_rlimit[RLIMIT_VMEM].rlim_cur) {
3361 		rv = KERN_NO_SPACE;
3362 		goto done;
3363 	}
3364 
3365 	if (use_read_lock && vm_map_lock_upgrade(map)) {
3366 		use_read_lock = 0;
3367 		goto Retry;
3368 	}
3369 	use_read_lock = 0;
3370 
3371 	/* Get the preliminary new entry start value */
3372 	addr = stack_entry->start - grow_amount;
3373 
3374 	/* If this puts us into the previous entry, cut back our growth
3375 	 * to the available space.  Also, see the note above.
3376 	 */
3377 	if (addr < end) {
3378 		stack_entry->aux.avail_ssize = stack_entry->start - end;
3379 		addr = end;
3380 	}
3381 
3382 	rv = vm_map_insert(map, &count,
3383 			   NULL, 0, addr, stack_entry->start,
3384 			   VM_MAPTYPE_NORMAL,
3385 			   VM_PROT_ALL, VM_PROT_ALL,
3386 			   0);
3387 
3388 	/* Adjust the available stack space by the amount we grew. */
3389 	if (rv == KERN_SUCCESS) {
3390 		if (prev_entry != &map->header)
3391 			vm_map_clip_end(map, prev_entry, addr, &count);
3392 		new_stack_entry = prev_entry->next;
3393 		if (new_stack_entry->end   != stack_entry->start  ||
3394 		    new_stack_entry->start != addr)
3395 			panic ("Bad stack grow start/end in new stack entry");
3396 		else {
3397 			new_stack_entry->aux.avail_ssize =
3398 				stack_entry->aux.avail_ssize -
3399 				(new_stack_entry->end - new_stack_entry->start);
3400 			if (is_procstack)
3401 				vm->vm_ssize += btoc(new_stack_entry->end -
3402 						     new_stack_entry->start);
3403 		}
3404 	}
3405 
3406 done:
3407 	if (use_read_lock)
3408 		vm_map_unlock_read(map);
3409 	else
3410 		vm_map_unlock(map);
3411 	vm_map_entry_release(count);
3412 	return (rv);
3413 }
3414 
3415 /*
3416  * Unshare the specified VM space for exec.  If other processes are
3417  * mapped to it, then create a new one.  The new vmspace is null.
3418  *
3419  * No requirements.
3420  */
3421 void
3422 vmspace_exec(struct proc *p, struct vmspace *vmcopy)
3423 {
3424 	struct vmspace *oldvmspace = p->p_vmspace;
3425 	struct vmspace *newvmspace;
3426 	vm_map_t map = &p->p_vmspace->vm_map;
3427 
3428 	/*
3429 	 * If we are execing a resident vmspace we fork it, otherwise
3430 	 * we create a new vmspace.  Note that exitingcnt and upcalls
3431 	 * are not copied to the new vmspace.
3432 	 */
3433 	lwkt_gettoken(&vmspace_token);
3434 	if (vmcopy)  {
3435 		newvmspace = vmspace_fork(vmcopy);
3436 	} else {
3437 		newvmspace = vmspace_alloc(map->min_offset, map->max_offset);
3438 		bcopy(&oldvmspace->vm_startcopy, &newvmspace->vm_startcopy,
3439 		      (caddr_t)&oldvmspace->vm_endcopy -
3440 		       (caddr_t)&oldvmspace->vm_startcopy);
3441 	}
3442 
3443 	/*
3444 	 * Finish initializing the vmspace before assigning it
3445 	 * to the process.  The vmspace will become the current vmspace
3446 	 * if p == curproc.
3447 	 */
3448 	pmap_pinit2(vmspace_pmap(newvmspace));
3449 	pmap_replacevm(p, newvmspace, 0);
3450 	sysref_put(&oldvmspace->vm_sysref);
3451 	lwkt_reltoken(&vmspace_token);
3452 }
3453 
3454 /*
3455  * Unshare the specified VM space for forcing COW.  This
3456  * is called by rfork, for the (RFMEM|RFPROC) == 0 case.
3457  *
3458  * The exitingcnt test is not strictly necessary but has been
3459  * included for code sanity (to make the code a bit more deterministic).
3460  */
3461 void
3462 vmspace_unshare(struct proc *p)
3463 {
3464 	struct vmspace *oldvmspace = p->p_vmspace;
3465 	struct vmspace *newvmspace;
3466 
3467 	lwkt_gettoken(&vmspace_token);
3468 	if (oldvmspace->vm_sysref.refcnt == 1 && oldvmspace->vm_exitingcnt == 0)
3469 		return;
3470 	newvmspace = vmspace_fork(oldvmspace);
3471 	pmap_pinit2(vmspace_pmap(newvmspace));
3472 	pmap_replacevm(p, newvmspace, 0);
3473 	sysref_put(&oldvmspace->vm_sysref);
3474 	lwkt_reltoken(&vmspace_token);
3475 }
3476 
3477 /*
3478  * Finds the VM object, offset, and protection for a given virtual address
3479  * in the specified map, assuming a page fault of the type specified.
3480  *
3481  * Leaves the map in question locked for read; return values are guaranteed
3482  * until a vm_map_lookup_done call is performed.  Note that the map argument
3483  * is in/out; the returned map must be used in the call to vm_map_lookup_done.
3484  *
3485  * A handle (out_entry) is returned for use in vm_map_lookup_done, to make
3486  * that fast.
3487  *
3488  * If a lookup is requested with "write protection" specified, the map may
3489  * be changed to perform virtual copying operations, although the data
3490  * referenced will remain the same.
3491  *
3492  * No requirements.
3493  */
3494 int
3495 vm_map_lookup(vm_map_t *var_map,		/* IN/OUT */
3496 	      vm_offset_t vaddr,
3497 	      vm_prot_t fault_typea,
3498 	      vm_map_entry_t *out_entry,	/* OUT */
3499 	      vm_object_t *object,		/* OUT */
3500 	      vm_pindex_t *pindex,		/* OUT */
3501 	      vm_prot_t *out_prot,		/* OUT */
3502 	      boolean_t *wired)			/* OUT */
3503 {
3504 	vm_map_entry_t entry;
3505 	vm_map_t map = *var_map;
3506 	vm_prot_t prot;
3507 	vm_prot_t fault_type = fault_typea;
3508 	int use_read_lock = 1;
3509 	int rv = KERN_SUCCESS;
3510 
3511 RetryLookup:
3512 	if (use_read_lock)
3513 		vm_map_lock_read(map);
3514 	else
3515 		vm_map_lock(map);
3516 
3517 	/*
3518 	 * If the map has an interesting hint, try it before calling full
3519 	 * blown lookup routine.
3520 	 */
3521 	entry = map->hint;
3522 	*out_entry = entry;
3523 
3524 	if ((entry == &map->header) ||
3525 	    (vaddr < entry->start) || (vaddr >= entry->end)) {
3526 		vm_map_entry_t tmp_entry;
3527 
3528 		/*
3529 		 * Entry was either not a valid hint, or the vaddr was not
3530 		 * contained in the entry, so do a full lookup.
3531 		 */
3532 		if (!vm_map_lookup_entry(map, vaddr, &tmp_entry)) {
3533 			rv = KERN_INVALID_ADDRESS;
3534 			goto done;
3535 		}
3536 
3537 		entry = tmp_entry;
3538 		*out_entry = entry;
3539 	}
3540 
3541 	/*
3542 	 * Handle submaps.
3543 	 */
3544 	if (entry->maptype == VM_MAPTYPE_SUBMAP) {
3545 		vm_map_t old_map = map;
3546 
3547 		*var_map = map = entry->object.sub_map;
3548 		if (use_read_lock)
3549 			vm_map_unlock_read(old_map);
3550 		else
3551 			vm_map_unlock(old_map);
3552 		use_read_lock = 1;
3553 		goto RetryLookup;
3554 	}
3555 
3556 	/*
3557 	 * Check whether this task is allowed to have this page.
3558 	 * Note the special case for MAP_ENTRY_COW
3559 	 * pages with an override.  This is to implement a forced
3560 	 * COW for debuggers.
3561 	 */
3562 
3563 	if (fault_type & VM_PROT_OVERRIDE_WRITE)
3564 		prot = entry->max_protection;
3565 	else
3566 		prot = entry->protection;
3567 
3568 	fault_type &= (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE);
3569 	if ((fault_type & prot) != fault_type) {
3570 		rv = KERN_PROTECTION_FAILURE;
3571 		goto done;
3572 	}
3573 
3574 	if ((entry->eflags & MAP_ENTRY_USER_WIRED) &&
3575 	    (entry->eflags & MAP_ENTRY_COW) &&
3576 	    (fault_type & VM_PROT_WRITE) &&
3577 	    (fault_typea & VM_PROT_OVERRIDE_WRITE) == 0) {
3578 		rv = KERN_PROTECTION_FAILURE;
3579 		goto done;
3580 	}
3581 
3582 	/*
3583 	 * If this page is not pageable, we have to get it for all possible
3584 	 * accesses.
3585 	 */
3586 	*wired = (entry->wired_count != 0);
3587 	if (*wired)
3588 		prot = fault_type = entry->protection;
3589 
3590 	/*
3591 	 * Virtual page tables may need to update the accessed (A) bit
3592 	 * in a page table entry.  Upgrade the fault to a write fault for
3593 	 * that case if the map will support it.  If the map does not support
3594 	 * it the page table entry simply will not be updated.
3595 	 */
3596 	if (entry->maptype == VM_MAPTYPE_VPAGETABLE) {
3597 		if (prot & VM_PROT_WRITE)
3598 			fault_type |= VM_PROT_WRITE;
3599 	}
3600 
3601 	/*
3602 	 * If the entry was copy-on-write, we either ...
3603 	 */
3604 	if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
3605 		/*
3606 		 * If we want to write the page, we may as well handle that
3607 		 * now since we've got the map locked.
3608 		 *
3609 		 * If we don't need to write the page, we just demote the
3610 		 * permissions allowed.
3611 		 */
3612 
3613 		if (fault_type & VM_PROT_WRITE) {
3614 			/*
3615 			 * Make a new object, and place it in the object
3616 			 * chain.  Note that no new references have appeared
3617 			 * -- one just moved from the map to the new
3618 			 * object.
3619 			 */
3620 
3621 			if (use_read_lock && vm_map_lock_upgrade(map)) {
3622 				use_read_lock = 0;
3623 				goto RetryLookup;
3624 			}
3625 			use_read_lock = 0;
3626 
3627 			vm_map_entry_shadow(entry);
3628 		} else {
3629 			/*
3630 			 * We're attempting to read a copy-on-write page --
3631 			 * don't allow writes.
3632 			 */
3633 
3634 			prot &= ~VM_PROT_WRITE;
3635 		}
3636 	}
3637 
3638 	/*
3639 	 * Create an object if necessary.
3640 	 */
3641 	if (entry->object.vm_object == NULL &&
3642 	    !map->system_map) {
3643 		if (use_read_lock && vm_map_lock_upgrade(map))  {
3644 			use_read_lock = 0;
3645 			goto RetryLookup;
3646 		}
3647 		use_read_lock = 0;
3648 		vm_map_entry_allocate_object(entry);
3649 	}
3650 
3651 	/*
3652 	 * Return the object/offset from this entry.  If the entry was
3653 	 * copy-on-write or empty, it has been fixed up.
3654 	 */
3655 
3656 	*pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
3657 	*object = entry->object.vm_object;
3658 
3659 	/*
3660 	 * Return whether this is the only map sharing this data.  On
3661 	 * success we return with a read lock held on the map.  On failure
3662 	 * we return with the map unlocked.
3663 	 */
3664 	*out_prot = prot;
3665 done:
3666 	if (rv == KERN_SUCCESS) {
3667 		if (use_read_lock == 0)
3668 			vm_map_lock_downgrade(map);
3669 	} else if (use_read_lock) {
3670 		vm_map_unlock_read(map);
3671 	} else {
3672 		vm_map_unlock(map);
3673 	}
3674 	return (rv);
3675 }
3676 
3677 /*
3678  * Releases locks acquired by a vm_map_lookup()
3679  * (according to the handle returned by that lookup).
3680  *
3681  * No other requirements.
3682  */
3683 void
3684 vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry, int count)
3685 {
3686 	/*
3687 	 * Unlock the main-level map
3688 	 */
3689 	vm_map_unlock_read(map);
3690 	if (count)
3691 		vm_map_entry_release(count);
3692 }
3693 
3694 #include "opt_ddb.h"
3695 #ifdef DDB
3696 #include <sys/kernel.h>
3697 
3698 #include <ddb/ddb.h>
3699 
3700 /*
3701  * Debugging only
3702  */
3703 DB_SHOW_COMMAND(map, vm_map_print)
3704 {
3705 	static int nlines;
3706 	/* XXX convert args. */
3707 	vm_map_t map = (vm_map_t)addr;
3708 	boolean_t full = have_addr;
3709 
3710 	vm_map_entry_t entry;
3711 
3712 	db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n",
3713 	    (void *)map,
3714 	    (void *)map->pmap, map->nentries, map->timestamp);
3715 	nlines++;
3716 
3717 	if (!full && db_indent)
3718 		return;
3719 
3720 	db_indent += 2;
3721 	for (entry = map->header.next; entry != &map->header;
3722 	    entry = entry->next) {
3723 		db_iprintf("map entry %p: start=%p, end=%p\n",
3724 		    (void *)entry, (void *)entry->start, (void *)entry->end);
3725 		nlines++;
3726 		{
3727 			static char *inheritance_name[4] =
3728 			{"share", "copy", "none", "donate_copy"};
3729 
3730 			db_iprintf(" prot=%x/%x/%s",
3731 			    entry->protection,
3732 			    entry->max_protection,
3733 			    inheritance_name[(int)(unsigned char)entry->inheritance]);
3734 			if (entry->wired_count != 0)
3735 				db_printf(", wired");
3736 		}
3737 		if (entry->maptype == VM_MAPTYPE_SUBMAP) {
3738 			/* XXX no %qd in kernel.  Truncate entry->offset. */
3739 			db_printf(", share=%p, offset=0x%lx\n",
3740 			    (void *)entry->object.sub_map,
3741 			    (long)entry->offset);
3742 			nlines++;
3743 			if ((entry->prev == &map->header) ||
3744 			    (entry->prev->object.sub_map !=
3745 				entry->object.sub_map)) {
3746 				db_indent += 2;
3747 				vm_map_print((db_expr_t)(intptr_t)
3748 					     entry->object.sub_map,
3749 					     full, 0, NULL);
3750 				db_indent -= 2;
3751 			}
3752 		} else {
3753 			/* XXX no %qd in kernel.  Truncate entry->offset. */
3754 			db_printf(", object=%p, offset=0x%lx",
3755 			    (void *)entry->object.vm_object,
3756 			    (long)entry->offset);
3757 			if (entry->eflags & MAP_ENTRY_COW)
3758 				db_printf(", copy (%s)",
3759 				    (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done");
3760 			db_printf("\n");
3761 			nlines++;
3762 
3763 			if ((entry->prev == &map->header) ||
3764 			    (entry->prev->object.vm_object !=
3765 				entry->object.vm_object)) {
3766 				db_indent += 2;
3767 				vm_object_print((db_expr_t)(intptr_t)
3768 						entry->object.vm_object,
3769 						full, 0, NULL);
3770 				nlines += 4;
3771 				db_indent -= 2;
3772 			}
3773 		}
3774 	}
3775 	db_indent -= 2;
3776 	if (db_indent == 0)
3777 		nlines = 0;
3778 }
3779 
3780 /*
3781  * Debugging only
3782  */
3783 DB_SHOW_COMMAND(procvm, procvm)
3784 {
3785 	struct proc *p;
3786 
3787 	if (have_addr) {
3788 		p = (struct proc *) addr;
3789 	} else {
3790 		p = curproc;
3791 	}
3792 
3793 	db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n",
3794 	    (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map,
3795 	    (void *)vmspace_pmap(p->p_vmspace));
3796 
3797 	vm_map_print((db_expr_t)(intptr_t)&p->p_vmspace->vm_map, 1, 0, NULL);
3798 }
3799 
3800 #endif /* DDB */
3801