1/*
2 * Copyright (c) 1991, 1993
3 * The Regents of the University of California. All rights reserved.
4 *
5 * This code is derived from software contributed to Berkeley by
6 * The Mach Operating System project at Carnegie-Mellon University.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. All advertising materials mentioning features or use of this software
17 * must display the following acknowledgement:
18 * This product includes software developed by the University of
19 * California, Berkeley and its contributors.
20 * 4. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
23 *
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * SUCH DAMAGE.
35 *
36 * from: @(#)vm_map.c 8.3 (Berkeley) 1/12/94
37 *
38 *
39 * Copyright (c) 1987, 1990 Carnegie-Mellon University.
40 * All rights reserved.
41 *
42 * Authors: Avadis Tevanian, Jr., Michael Wayne Young
43 *
44 * Permission to use, copy, modify and distribute this software and
45 * its documentation is hereby granted, provided that both the copyright
46 * notice and this permission notice appear in all copies of the
47 * software, derivative works or modified versions, and any portions
48 * thereof, and that both notices appear in supporting documentation.
49 *
50 * CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
51 * CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
52 * FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
53 *
54 * Carnegie Mellon requests users of this software to return to
55 *
56 * Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
57 * School of Computer Science
58 * Carnegie Mellon University
59 * Pittsburgh PA 15213-3890
60 *
61 * any improvements or extensions that they make and grant Carnegie the
62 * rights to redistribute these changes.
63 *
64 * $FreeBSD: head/sys/vm/vm_map.c 92588 2002-03-18 15:08:09Z green $
65 */
66
67/*
68 * Virtual memory mapping module.
69 */
70
71#include <sys/param.h>
72#include <sys/systm.h>
73#include <sys/ktr.h>
74#include <sys/lock.h>
75#include <sys/mutex.h>
76#include <sys/proc.h>
77#include <sys/vmmeter.h>
78#include <sys/mman.h>
79#include <sys/vnode.h>
80#include <sys/resourcevar.h>
81
82#include <vm/vm.h>
83#include <vm/vm_param.h>
84#include <vm/pmap.h>
85#include <vm/vm_map.h>
86#include <vm/vm_page.h>
87#include <vm/vm_object.h>
88#include <vm/vm_pager.h>
89#include <vm/vm_kern.h>
90#include <vm/vm_extern.h>
91#include <vm/vm_zone.h>
92#include <vm/swap_pager.h>
93
94/*
95 * Virtual memory maps provide for the mapping, protection,
96 * and sharing of virtual memory objects. In addition,
97 * this module provides for an efficient virtual copy of
98 * memory from one map to another.
99 *
100 * Synchronization is required prior to most operations.
101 *
102 * Maps consist of an ordered doubly-linked list of simple
103 * entries; a single hint is used to speed up lookups.
104 *
105 * Since portions of maps are specified by start/end addresses,
106 * which may not align with existing map entries, all
107 * routines merely "clip" entries to these start/end values.
108 * [That is, an entry is split into two, bordering at a
109 * start or end value.] Note that these clippings may not
110 * always be necessary (as the two resulting entries are then
111 * not changed); however, the clipping is done for convenience.
112 *
113 * As mentioned above, virtual copy operations are performed
114 * by copying VM object references from one map to
115 * another, and then marking both regions as copy-on-write.
116 */
117
118/*
119 * vm_map_startup:
120 *
121 * Initialize the vm_map module. Must be called before
122 * any other vm_map routines.
123 *
124 * Map and entry structures are allocated from the general
125 * purpose memory pool with some exceptions:
126 *
127 * - The kernel map and kmem submap are allocated statically.
128 * - Kernel map entries are allocated out of a static pool.
129 *
130 * These restrictions are necessary since malloc() uses the
131 * maps and requires map entries.
132 */
133
134static struct vm_zone kmapentzone_store, mapentzone_store, mapzone_store;
135static vm_zone_t mapentzone, kmapentzone, mapzone, vmspace_zone;
136static struct vm_object kmapentobj, mapentobj, mapobj;
137
138static struct vm_map_entry map_entry_init[MAX_MAPENT];
139static struct vm_map_entry kmap_entry_init[MAX_KMAPENT];
140static struct vm_map map_init[MAX_KMAP];
141
142void
143vm_map_startup(void)
144{
145 mapzone = &mapzone_store;
146 zbootinit(mapzone, "MAP", sizeof (struct vm_map),
147 map_init, MAX_KMAP);
148 kmapentzone = &kmapentzone_store;
149 zbootinit(kmapentzone, "KMAP ENTRY", sizeof (struct vm_map_entry),
150 kmap_entry_init, MAX_KMAPENT);
151 mapentzone = &mapentzone_store;
152 zbootinit(mapentzone, "MAP ENTRY", sizeof (struct vm_map_entry),
153 map_entry_init, MAX_MAPENT);
154}
155
156/*
157 * Allocate a vmspace structure, including a vm_map and pmap,
158 * and initialize those structures. The refcnt is set to 1.
159 * The remaining fields must be initialized by the caller.
160 */
161struct vmspace *
162vmspace_alloc(min, max)
163 vm_offset_t min, max;
164{
165 struct vmspace *vm;
166
167 GIANT_REQUIRED;
168 vm = zalloc(vmspace_zone);
169 CTR1(KTR_VM, "vmspace_alloc: %p", vm);
170 vm_map_init(&vm->vm_map, min, max);
171 pmap_pinit(vmspace_pmap(vm));
172 vm->vm_map.pmap = vmspace_pmap(vm); /* XXX */
173 vm->vm_refcnt = 1;
174 vm->vm_shm = NULL;
175 vm->vm_freer = NULL;
176 return (vm);
177}
178
179void
180vm_init2(void)
181{
182 zinitna(kmapentzone, &kmapentobj,
183 NULL, 0, cnt.v_page_count / 4, ZONE_INTERRUPT, 1);
184 zinitna(mapentzone, &mapentobj,
185 NULL, 0, 0, 0, 1);
186 zinitna(mapzone, &mapobj,
187 NULL, 0, 0, 0, 1);
188 vmspace_zone = zinit("VMSPACE", sizeof (struct vmspace), 0, 0, 3);
189 pmap_init2();
190 vm_object_init2();
191}
192
193static __inline void
194vmspace_dofree(struct vmspace *vm)
195{
196 CTR1(KTR_VM, "vmspace_free: %p", vm);
197 /*
198 * Lock the map, to wait out all other references to it.
199 * Delete all of the mappings and pages they hold, then call
200 * the pmap module to reclaim anything left.
201 */
202 vm_map_lock(&vm->vm_map);
203 (void) vm_map_delete(&vm->vm_map, vm->vm_map.min_offset,
204 vm->vm_map.max_offset);
205 vm_map_unlock(&vm->vm_map);
206 pmap_release(vmspace_pmap(vm));
207 vm_map_destroy(&vm->vm_map);
208 zfree(vmspace_zone, vm);
209}
210
211void
212vmspace_free(struct vmspace *vm)
213{
214 GIANT_REQUIRED;
215
216 if (vm->vm_refcnt == 0)
217 panic("vmspace_free: attempt to free already freed vmspace");
218
219 if (--vm->vm_refcnt == 0)
220 vmspace_dofree(vm);
221}
222
223void
224vmspace_exitfree(struct proc *p)
225{
226 GIANT_REQUIRED;
227
228 if (p == p->p_vmspace->vm_freer)
229 vmspace_dofree(p->p_vmspace);
230}
231
232/*
233 * vmspace_swap_count() - count the approximate swap useage in pages for a
234 * vmspace.
235 *
236 * Swap useage is determined by taking the proportional swap used by
237 * VM objects backing the VM map. To make up for fractional losses,
238 * if the VM object has any swap use at all the associated map entries
239 * count for at least 1 swap page.
240 */
241int
242vmspace_swap_count(struct vmspace *vmspace)
243{
244 vm_map_t map = &vmspace->vm_map;
245 vm_map_entry_t cur;
246 int count = 0;
247
248 for (cur = map->header.next; cur != &map->header; cur = cur->next) {
249 vm_object_t object;
250
251 if ((cur->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 &&
252 (object = cur->object.vm_object) != NULL &&
253 object->type == OBJT_SWAP
254 ) {
255 int n = (cur->end - cur->start) / PAGE_SIZE;
256
257 if (object->un_pager.swp.swp_bcount) {
258 count += object->un_pager.swp.swp_bcount *
259 SWAP_META_PAGES * n / object->size + 1;
260 }
261 }
262 }
263 return (count);
264}
265
266u_char
267vm_map_entry_behavior(struct vm_map_entry *entry)
268{
269 return entry->eflags & MAP_ENTRY_BEHAV_MASK;
270}
271
272void
273vm_map_entry_set_behavior(struct vm_map_entry *entry, u_char behavior)
274{
275 entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) |
276 (behavior & MAP_ENTRY_BEHAV_MASK);
277}
278
279void
280vm_map_lock(vm_map_t map)
281{
282 vm_map_printf("locking map LK_EXCLUSIVE: %p\n", map);
283 if (lockmgr(&map->lock, LK_EXCLUSIVE, NULL, curthread) != 0)
284 panic("vm_map_lock: failed to get lock");
285 map->timestamp++;
286}
287
288void
289vm_map_unlock(vm_map_t map)
290{
291 vm_map_printf("locking map LK_RELEASE: %p\n", map);
292 lockmgr(&(map)->lock, LK_RELEASE, NULL, curthread);
293}
294
295void
296vm_map_lock_read(vm_map_t map)
297{
298 vm_map_printf("locking map LK_SHARED: %p\n", map);
299 lockmgr(&(map)->lock, LK_SHARED, NULL, curthread);
300}
301
302void
303vm_map_unlock_read(vm_map_t map)
304{
305 vm_map_printf("locking map LK_RELEASE: %p\n", map);
306 lockmgr(&(map)->lock, LK_RELEASE, NULL, curthread);
307}
308
309static __inline__ int
310_vm_map_lock_upgrade(vm_map_t map, struct thread *td) {
311 int error;
312
313 vm_map_printf("locking map LK_EXCLUPGRADE: %p\n", map);
314 error = lockmgr(&map->lock, LK_EXCLUPGRADE, NULL, td);
315 if (error == 0)
316 map->timestamp++;
317 return error;
318}
319
320int
321vm_map_lock_upgrade(vm_map_t map)
322{
323 return (_vm_map_lock_upgrade(map, curthread));
324}
325
326void
327vm_map_lock_downgrade(vm_map_t map)
328{
329 vm_map_printf("locking map LK_DOWNGRADE: %p\n", map);
330 lockmgr(&map->lock, LK_DOWNGRADE, NULL, curthread);
331}
332
333void
334vm_map_set_recursive(vm_map_t map)
335{
336 mtx_lock((map)->lock.lk_interlock);
337 map->lock.lk_flags |= LK_CANRECURSE;
338 mtx_unlock((map)->lock.lk_interlock);
339}
340
341void
342vm_map_clear_recursive(vm_map_t map)
343{
344 mtx_lock((map)->lock.lk_interlock);
345 map->lock.lk_flags &= ~LK_CANRECURSE;
346 mtx_unlock((map)->lock.lk_interlock);
347}
348
349vm_offset_t
350vm_map_min(vm_map_t map)
351{
352 return (map->min_offset);
353}
354
355vm_offset_t
356vm_map_max(vm_map_t map)
357{
358 return (map->max_offset);
359}
360
361struct pmap *
362vm_map_pmap(vm_map_t map)
363{
364 return (map->pmap);
365}
366
367struct pmap *
368vmspace_pmap(struct vmspace *vmspace)
369{
370 return &vmspace->vm_pmap;
371}
372
373long
374vmspace_resident_count(struct vmspace *vmspace)
375{
376 return pmap_resident_count(vmspace_pmap(vmspace));
377}
378
379/*
380 * vm_map_create:
381 *
382 * Creates and returns a new empty VM map with
383 * the given physical map structure, and having
384 * the given lower and upper address bounds.
385 */
386vm_map_t
387vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max)
388{
389 vm_map_t result;
390
391 GIANT_REQUIRED;
392
393 result = zalloc(mapzone);
394 CTR1(KTR_VM, "vm_map_create: %p", result);
395 vm_map_init(result, min, max);
396 result->pmap = pmap;
397 return (result);
398}
399
400/*
401 * Initialize an existing vm_map structure
402 * such as that in the vmspace structure.
403 * The pmap is set elsewhere.
404 */
405void
406vm_map_init(vm_map_t map, vm_offset_t min, vm_offset_t max)
407{
408 GIANT_REQUIRED;
409
410 map->header.next = map->header.prev = &map->header;
411 map->nentries = 0;
412 map->size = 0;
413 map->system_map = 0;
414 map->infork = 0;
415 map->min_offset = min;
416 map->max_offset = max;
417 map->first_free = &map->header;
418 map->hint = &map->header;
419 map->timestamp = 0;
420 lockinit(&map->lock, PVM, "thrd_sleep", 0, LK_NOPAUSE);
421}
422
423void
424vm_map_destroy(map)
425 struct vm_map *map;
426{
427 GIANT_REQUIRED;
428 lockdestroy(&map->lock);
429}
430
431/*
432 * vm_map_entry_dispose: [ internal use only ]
433 *
434 * Inverse of vm_map_entry_create.
435 */
436static void
437vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry)
438{
439 zfree((map->system_map || !mapentzone) ? kmapentzone : mapentzone, entry);
440}
441
442/*
443 * vm_map_entry_create: [ internal use only ]
444 *
445 * Allocates a VM map entry for insertion.
446 * No entry fields are filled in.
447 */
448static vm_map_entry_t
449vm_map_entry_create(vm_map_t map)
450{
451 vm_map_entry_t new_entry;
452
453 new_entry = zalloc((map->system_map || !mapentzone) ?
454 kmapentzone : mapentzone);
455 if (new_entry == NULL)
456 panic("vm_map_entry_create: kernel resources exhausted");
457 return (new_entry);
458}
459
460/*
461 * vm_map_entry_{un,}link:
462 *
463 * Insert/remove entries from maps.
464 */
465static __inline void
466vm_map_entry_link(vm_map_t map,
467 vm_map_entry_t after_where,
468 vm_map_entry_t entry)
469{
470
471 CTR4(KTR_VM,
472 "vm_map_entry_link: map %p, nentries %d, entry %p, after %p", map,
473 map->nentries, entry, after_where);
474 map->nentries++;
475 entry->prev = after_where;
476 entry->next = after_where->next;
477 entry->next->prev = entry;
478 after_where->next = entry;
479}
480
481static __inline void
482vm_map_entry_unlink(vm_map_t map,
483 vm_map_entry_t entry)
484{
485 vm_map_entry_t prev = entry->prev;
486 vm_map_entry_t next = entry->next;
487
488 next->prev = prev;
489 prev->next = next;
490 map->nentries--;
491 CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map,
492 map->nentries, entry);
493}
494
495/*
496 * SAVE_HINT:
497 *
498 * Saves the specified entry as the hint for
499 * future lookups.
500 */
501#define SAVE_HINT(map,value) \
502 (map)->hint = (value);
503
504/*
505 * vm_map_lookup_entry: [ internal use only ]
506 *
507 * Finds the map entry containing (or
508 * immediately preceding) the specified address
509 * in the given map; the entry is returned
510 * in the "entry" parameter. The boolean
511 * result indicates whether the address is
512 * actually contained in the map.
513 */
514boolean_t
515vm_map_lookup_entry(
516 vm_map_t map,
517 vm_offset_t address,
518 vm_map_entry_t *entry) /* OUT */
519{
520 vm_map_entry_t cur;
521 vm_map_entry_t last;
522
523 GIANT_REQUIRED;
524 /*
525 * Start looking either from the head of the list, or from the hint.
526 */
527 cur = map->hint;
528
529 if (cur == &map->header)
530 cur = cur->next;
531
532 if (address >= cur->start) {
533 /*
534 * Go from hint to end of list.
535 *
536 * But first, make a quick check to see if we are already looking
537 * at the entry we want (which is usually the case). Note also
538 * that we don't need to save the hint here... it is the same
539 * hint (unless we are at the header, in which case the hint
540 * didn't buy us anything anyway).
541 */
542 last = &map->header;
543 if ((cur != last) && (cur->end > address)) {
544 *entry = cur;
545 return (TRUE);
546 }
547 } else {
548 /*
549 * Go from start to hint, *inclusively*
550 */
551 last = cur->next;
552 cur = map->header.next;
553 }
554
555 /*
556 * Search linearly
557 */
558 while (cur != last) {
559 if (cur->end > address) {
560 if (address >= cur->start) {
561 /*
562 * Save this lookup for future hints, and
563 * return
564 */
565 *entry = cur;
566 SAVE_HINT(map, cur);
567 return (TRUE);
568 }
569 break;
570 }
571 cur = cur->next;
572 }
573 *entry = cur->prev;
574 SAVE_HINT(map, *entry);
575 return (FALSE);
576}
577
578/*
579 * vm_map_insert:
580 *
581 * Inserts the given whole VM object into the target
582 * map at the specified address range. The object's
583 * size should match that of the address range.
584 *
585 * Requires that the map be locked, and leaves it so.
586 *
587 * If object is non-NULL, ref count must be bumped by caller
588 * prior to making call to account for the new entry.
589 */
590int
591vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
592 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max,
593 int cow)
594{
595 vm_map_entry_t new_entry;
596 vm_map_entry_t prev_entry;
597 vm_map_entry_t temp_entry;
598 vm_eflags_t protoeflags;
599
600 GIANT_REQUIRED;
601
602 /*
603 * Check that the start and end points are not bogus.
604 */
605 if ((start < map->min_offset) || (end > map->max_offset) ||
606 (start >= end))
607 return (KERN_INVALID_ADDRESS);
608
609 /*
610 * Find the entry prior to the proposed starting address; if it's part
611 * of an existing entry, this range is bogus.
612 */
613 if (vm_map_lookup_entry(map, start, &temp_entry))
614 return (KERN_NO_SPACE);
615
616 prev_entry = temp_entry;
617
618 /*
619 * Assert that the next entry doesn't overlap the end point.
620 */
621 if ((prev_entry->next != &map->header) &&
622 (prev_entry->next->start < end))
623 return (KERN_NO_SPACE);
624
625 protoeflags = 0;
626
627 if (cow & MAP_COPY_ON_WRITE)
628 protoeflags |= MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY;
629
630 if (cow & MAP_NOFAULT) {
631 protoeflags |= MAP_ENTRY_NOFAULT;
632
633 KASSERT(object == NULL,
634 ("vm_map_insert: paradoxical MAP_NOFAULT request"));
635 }
636 if (cow & MAP_DISABLE_SYNCER)
637 protoeflags |= MAP_ENTRY_NOSYNC;
638 if (cow & MAP_DISABLE_COREDUMP)
639 protoeflags |= MAP_ENTRY_NOCOREDUMP;
640
641 if (object) {
642 /*
643 * When object is non-NULL, it could be shared with another
644 * process. We have to set or clear OBJ_ONEMAPPING
645 * appropriately.
646 */
647 if ((object->ref_count > 1) || (object->shadow_count != 0)) {
648 vm_object_clear_flag(object, OBJ_ONEMAPPING);
649 }
650 }
651 else if ((prev_entry != &map->header) &&
652 (prev_entry->eflags == protoeflags) &&
653 (prev_entry->end == start) &&
654 (prev_entry->wired_count == 0) &&
655 ((prev_entry->object.vm_object == NULL) ||
656 vm_object_coalesce(prev_entry->object.vm_object,
657 OFF_TO_IDX(prev_entry->offset),
658 (vm_size_t)(prev_entry->end - prev_entry->start),
659 (vm_size_t)(end - prev_entry->end)))) {
660 /*
661 * We were able to extend the object. Determine if we
662 * can extend the previous map entry to include the
663 * new range as well.
664 */
665 if ((prev_entry->inheritance == VM_INHERIT_DEFAULT) &&
666 (prev_entry->protection == prot) &&
667 (prev_entry->max_protection == max)) {
668 map->size += (end - prev_entry->end);
669 prev_entry->end = end;
670 vm_map_simplify_entry(map, prev_entry);
671 return (KERN_SUCCESS);
672 }
673
674 /*
675 * If we can extend the object but cannot extend the
676 * map entry, we have to create a new map entry. We
677 * must bump the ref count on the extended object to
678 * account for it. object may be NULL.
679 */
680 object = prev_entry->object.vm_object;
681 offset = prev_entry->offset +
682 (prev_entry->end - prev_entry->start);
683 vm_object_reference(object);
684 }
685
686 /*
687 * NOTE: if conditionals fail, object can be NULL here. This occurs
688 * in things like the buffer map where we manage kva but do not manage
689 * backing objects.
690 */
691
692 /*
693 * Create a new entry
694 */
695 new_entry = vm_map_entry_create(map);
696 new_entry->start = start;
697 new_entry->end = end;
698
699 new_entry->eflags = protoeflags;
700 new_entry->object.vm_object = object;
701 new_entry->offset = offset;
702 new_entry->avail_ssize = 0;
703
704 new_entry->inheritance = VM_INHERIT_DEFAULT;
705 new_entry->protection = prot;
706 new_entry->max_protection = max;
707 new_entry->wired_count = 0;
708
709 /*
710 * Insert the new entry into the list
711 */
712 vm_map_entry_link(map, prev_entry, new_entry);
713 map->size += new_entry->end - new_entry->start;
714
715 /*
716 * Update the free space hint
717 */
718 if ((map->first_free == prev_entry) &&
719 (prev_entry->end >= new_entry->start)) {
720 map->first_free = new_entry;
721 }
722
723#if 0
724 /*
725 * Temporarily removed to avoid MAP_STACK panic, due to
726 * MAP_STACK being a huge hack. Will be added back in
727 * when MAP_STACK (and the user stack mapping) is fixed.
728 */
729 /*
730 * It may be possible to simplify the entry
731 */
732 vm_map_simplify_entry(map, new_entry);
733#endif
734
735 if (cow & (MAP_PREFAULT|MAP_PREFAULT_PARTIAL)) {
736 pmap_object_init_pt(map->pmap, start,
737 object, OFF_TO_IDX(offset), end - start,
738 cow & MAP_PREFAULT_PARTIAL);
739 }
740
741 return (KERN_SUCCESS);
742}
743
744/*
745 * Find sufficient space for `length' bytes in the given map, starting at
746 * `start'. The map must be locked. Returns 0 on success, 1 on no space.
747 */
748int
749vm_map_findspace(
750 vm_map_t map,
751 vm_offset_t start,
752 vm_size_t length,
753 vm_offset_t *addr)
754{
755 vm_map_entry_t entry, next;
756 vm_offset_t end;
757
758 GIANT_REQUIRED;
759 if (start < map->min_offset)
760 start = map->min_offset;
761 if (start > map->max_offset)
762 return (1);
763
764 /*
765 * Look for the first possible address; if there's already something
766 * at this address, we have to start after it.
767 */
768 if (start == map->min_offset) {
769 if ((entry = map->first_free) != &map->header)
770 start = entry->end;
771 } else {
772 vm_map_entry_t tmp;
773
774 if (vm_map_lookup_entry(map, start, &tmp))
775 start = tmp->end;
776 entry = tmp;
777 }
778
779 /*
780 * Look through the rest of the map, trying to fit a new region in the
781 * gap between existing regions, or after the very last region.
782 */
783 for (;; start = (entry = next)->end) {
784 /*
785 * Find the end of the proposed new region. Be sure we didn't
786 * go beyond the end of the map, or wrap around the address;
787 * if so, we lose. Otherwise, if this is the last entry, or
788 * if the proposed new region fits before the next entry, we
789 * win.
790 */
791 end = start + length;
792 if (end > map->max_offset || end < start)
793 return (1);
794 next = entry->next;
795 if (next == &map->header || next->start >= end)
796 break;
797 }
798 SAVE_HINT(map, entry);
799 *addr = start;
800 if (map == kernel_map) {
801 vm_offset_t ksize;
802 if ((ksize = round_page(start + length)) > kernel_vm_end) {
803 pmap_growkernel(ksize);
804 }
805 }
806 return (0);
807}
808
809/*
810 * vm_map_find finds an unallocated region in the target address
811 * map with the given length. The search is defined to be
812 * first-fit from the specified address; the region found is
813 * returned in the same parameter.
814 *
815 * If object is non-NULL, ref count must be bumped by caller
816 * prior to making call to account for the new entry.
817 */
818int
819vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
820 vm_offset_t *addr, /* IN/OUT */
821 vm_size_t length, boolean_t find_space, vm_prot_t prot,
822 vm_prot_t max, int cow)
823{
824 vm_offset_t start;
825 int result, s = 0;
826
827 GIANT_REQUIRED;
828
829 start = *addr;
830
831 if (map == kmem_map)
832 s = splvm();
833
834 vm_map_lock(map);
835 if (find_space) {
836 if (vm_map_findspace(map, start, length, addr)) {
837 vm_map_unlock(map);
838 if (map == kmem_map)
839 splx(s);
840 return (KERN_NO_SPACE);
841 }
842 start = *addr;
843 }
844 result = vm_map_insert(map, object, offset,
845 start, start + length, prot, max, cow);
846 vm_map_unlock(map);
847
848 if (map == kmem_map)
849 splx(s);
850
851 return (result);
852}
853
854/*
855 * vm_map_simplify_entry:
856 *
857 * Simplify the given map entry by merging with either neighbor. This
858 * routine also has the ability to merge with both neighbors.
859 *
860 * The map must be locked.
861 *
862 * This routine guarentees that the passed entry remains valid (though
863 * possibly extended). When merging, this routine may delete one or
864 * both neighbors.
865 */
866void
867vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry)
868{
869 vm_map_entry_t next, prev;
870 vm_size_t prevsize, esize;
871
872 GIANT_REQUIRED;
873
874 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP)
875 return;
876
877 prev = entry->prev;
878 if (prev != &map->header) {
879 prevsize = prev->end - prev->start;
880 if ( (prev->end == entry->start) &&
881 (prev->object.vm_object == entry->object.vm_object) &&
882 (!prev->object.vm_object ||
883 (prev->offset + prevsize == entry->offset)) &&
884 (prev->eflags == entry->eflags) &&
885 (prev->protection == entry->protection) &&
886 (prev->max_protection == entry->max_protection) &&
887 (prev->inheritance == entry->inheritance) &&
888 (prev->wired_count == entry->wired_count)) {
889 if (map->first_free == prev)
890 map->first_free = entry;
891 if (map->hint == prev)
892 map->hint = entry;
893 vm_map_entry_unlink(map, prev);
894 entry->start = prev->start;
895 entry->offset = prev->offset;
896 if (prev->object.vm_object)
897 vm_object_deallocate(prev->object.vm_object);
898 vm_map_entry_dispose(map, prev);
899 }
900 }
901
902 next = entry->next;
903 if (next != &map->header) {
904 esize = entry->end - entry->start;
905 if ((entry->end == next->start) &&
906 (next->object.vm_object == entry->object.vm_object) &&
907 (!entry->object.vm_object ||
908 (entry->offset + esize == next->offset)) &&
909 (next->eflags == entry->eflags) &&
910 (next->protection == entry->protection) &&
911 (next->max_protection == entry->max_protection) &&
912 (next->inheritance == entry->inheritance) &&
913 (next->wired_count == entry->wired_count)) {
914 if (map->first_free == next)
915 map->first_free = entry;
916 if (map->hint == next)
917 map->hint = entry;
918 vm_map_entry_unlink(map, next);
919 entry->end = next->end;
920 if (next->object.vm_object)
921 vm_object_deallocate(next->object.vm_object);
922 vm_map_entry_dispose(map, next);
923 }
924 }
925}
926/*
927 * vm_map_clip_start: [ internal use only ]
928 *
929 * Asserts that the given entry begins at or after
930 * the specified address; if necessary,
931 * it splits the entry into two.
932 */
933#define vm_map_clip_start(map, entry, startaddr) \
934{ \
935 if (startaddr > entry->start) \
936 _vm_map_clip_start(map, entry, startaddr); \
937}
938
939/*
940 * This routine is called only when it is known that
941 * the entry must be split.
942 */
943static void
944_vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start)
945{
946 vm_map_entry_t new_entry;
947
948 /*
949 * Split off the front portion -- note that we must insert the new
950 * entry BEFORE this one, so that this entry has the specified
951 * starting address.
952 */
953 vm_map_simplify_entry(map, entry);
954
955 /*
956 * If there is no object backing this entry, we might as well create
957 * one now. If we defer it, an object can get created after the map
958 * is clipped, and individual objects will be created for the split-up
959 * map. This is a bit of a hack, but is also about the best place to
960 * put this improvement.
961 */
962 if (entry->object.vm_object == NULL && !map->system_map) {
963 vm_object_t object;
964 object = vm_object_allocate(OBJT_DEFAULT,
965 atop(entry->end - entry->start));
966 entry->object.vm_object = object;
967 entry->offset = 0;
968 }
969
970 new_entry = vm_map_entry_create(map);
971 *new_entry = *entry;
972
973 new_entry->end = start;
974 entry->offset += (start - entry->start);
975 entry->start = start;
976
977 vm_map_entry_link(map, entry->prev, new_entry);
978
979 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
980 vm_object_reference(new_entry->object.vm_object);
981 }
982}
983
984/*
985 * vm_map_clip_end: [ internal use only ]
986 *
987 * Asserts that the given entry ends at or before
988 * the specified address; if necessary,
989 * it splits the entry into two.
990 */
991#define vm_map_clip_end(map, entry, endaddr) \
992{ \
993 if (endaddr < entry->end) \
994 _vm_map_clip_end(map, entry, endaddr); \
995}
996
997/*
998 * This routine is called only when it is known that
999 * the entry must be split.
1000 */
1001static void
1002_vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end)
1003{
1004 vm_map_entry_t new_entry;
1005
1006 /*
1007 * If there is no object backing this entry, we might as well create
1008 * one now. If we defer it, an object can get created after the map
1009 * is clipped, and individual objects will be created for the split-up
1010 * map. This is a bit of a hack, but is also about the best place to
1011 * put this improvement.
1012 */
1013 if (entry->object.vm_object == NULL && !map->system_map) {
1014 vm_object_t object;
1015 object = vm_object_allocate(OBJT_DEFAULT,
1016 atop(entry->end - entry->start));
1017 entry->object.vm_object = object;
1018 entry->offset = 0;
1019 }
1020
1021 /*
1022 * Create a new entry and insert it AFTER the specified entry
1023 */
1024 new_entry = vm_map_entry_create(map);
1025 *new_entry = *entry;
1026
1027 new_entry->start = entry->end = end;
1028 new_entry->offset += (end - entry->start);
1029
1030 vm_map_entry_link(map, entry, new_entry);
1031
1032 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1033 vm_object_reference(new_entry->object.vm_object);
1034 }
1035}
1036
1037/*
1038 * VM_MAP_RANGE_CHECK: [ internal use only ]
1039 *
1040 * Asserts that the starting and ending region
1041 * addresses fall within the valid range of the map.
1042 */
1043#define VM_MAP_RANGE_CHECK(map, start, end) \
1044 { \
1045 if (start < vm_map_min(map)) \
1046 start = vm_map_min(map); \
1047 if (end > vm_map_max(map)) \
1048 end = vm_map_max(map); \
1049 if (start > end) \
1050 start = end; \
1051 }
1052
1053/*
1054 * vm_map_submap: [ kernel use only ]
1055 *
1056 * Mark the given range as handled by a subordinate map.
1057 *
1058 * This range must have been created with vm_map_find,
1059 * and no other operations may have been performed on this
1060 * range prior to calling vm_map_submap.
1061 *
1062 * Only a limited number of operations can be performed
1063 * within this rage after calling vm_map_submap:
1064 * vm_fault
1065 * [Don't try vm_map_copy!]
1066 *
1067 * To remove a submapping, one must first remove the
1068 * range from the superior map, and then destroy the
1069 * submap (if desired). [Better yet, don't try it.]
1070 */
1071int
1072vm_map_submap(
1073 vm_map_t map,
1074 vm_offset_t start,
1075 vm_offset_t end,
1076 vm_map_t submap)
1077{
1078 vm_map_entry_t entry;
1079 int result = KERN_INVALID_ARGUMENT;
1080
1081 GIANT_REQUIRED;
1082
1083 vm_map_lock(map);
1084
1085 VM_MAP_RANGE_CHECK(map, start, end);
1086
1087 if (vm_map_lookup_entry(map, start, &entry)) {
1088 vm_map_clip_start(map, entry, start);
1089 } else
1090 entry = entry->next;
1091
1092 vm_map_clip_end(map, entry, end);
1093
1094 if ((entry->start == start) && (entry->end == end) &&
1095 ((entry->eflags & MAP_ENTRY_COW) == 0) &&
1096 (entry->object.vm_object == NULL)) {
1097 entry->object.sub_map = submap;
1098 entry->eflags |= MAP_ENTRY_IS_SUB_MAP;
1099 result = KERN_SUCCESS;
1100 }
1101 vm_map_unlock(map);
1102
1103 return (result);
1104}
1105
1106/*
1107 * vm_map_protect:
1108 *
1109 * Sets the protection of the specified address
1110 * region in the target map. If "set_max" is
1111 * specified, the maximum protection is to be set;
1112 * otherwise, only the current protection is affected.
1113 */
1114int
1115vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end,
1116 vm_prot_t new_prot, boolean_t set_max)
1117{
1118 vm_map_entry_t current;
1119 vm_map_entry_t entry;
1120
1121 GIANT_REQUIRED;
1122 vm_map_lock(map);
1123
1124 VM_MAP_RANGE_CHECK(map, start, end);
1125
1126 if (vm_map_lookup_entry(map, start, &entry)) {
1127 vm_map_clip_start(map, entry, start);
1128 } else {
1129 entry = entry->next;
1130 }
1131
1132 /*
1133 * Make a first pass to check for protection violations.
1134 */
1135 current = entry;
1136 while ((current != &map->header) && (current->start < end)) {
1137 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
1138 vm_map_unlock(map);
1139 return (KERN_INVALID_ARGUMENT);
1140 }
1141 if ((new_prot & current->max_protection) != new_prot) {
1142 vm_map_unlock(map);
1143 return (KERN_PROTECTION_FAILURE);
1144 }
1145 current = current->next;
1146 }
1147
1148 /*
1149 * Go back and fix up protections. [Note that clipping is not
1150 * necessary the second time.]
1151 */
1152 current = entry;
1153 while ((current != &map->header) && (current->start < end)) {
1154 vm_prot_t old_prot;
1155
1156 vm_map_clip_end(map, current, end);
1157
1158 old_prot = current->protection;
1159 if (set_max)
1160 current->protection =
1161 (current->max_protection = new_prot) &
1162 old_prot;
1163 else
1164 current->protection = new_prot;
1165
1166 /*
1167 * Update physical map if necessary. Worry about copy-on-write
1168 * here -- CHECK THIS XXX
1169 */
1170 if (current->protection != old_prot) {
1171#define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \
1172 VM_PROT_ALL)
1173 pmap_protect(map->pmap, current->start,
1174 current->end,
1175 current->protection & MASK(current));
1176#undef MASK
1177 }
1178 vm_map_simplify_entry(map, current);
1179 current = current->next;
1180 }
1181 vm_map_unlock(map);
1182 return (KERN_SUCCESS);
1183}
1184
1185/*
1186 * vm_map_madvise:
1187 *
1188 * This routine traverses a processes map handling the madvise
1189 * system call. Advisories are classified as either those effecting
1190 * the vm_map_entry structure, or those effecting the underlying
1191 * objects.
1192 */
1193int
1194vm_map_madvise(
1195 vm_map_t map,
1196 vm_offset_t start,
1197 vm_offset_t end,
1198 int behav)
1199{
1200 vm_map_entry_t current, entry;
1201 int modify_map = 0;
1202
1203 GIANT_REQUIRED;
1204
1205 /*
1206 * Some madvise calls directly modify the vm_map_entry, in which case
1207 * we need to use an exclusive lock on the map and we need to perform
1208 * various clipping operations. Otherwise we only need a read-lock
1209 * on the map.
1210 */
1211 switch(behav) {
1212 case MADV_NORMAL:
1213 case MADV_SEQUENTIAL:
1214 case MADV_RANDOM:
1215 case MADV_NOSYNC:
1216 case MADV_AUTOSYNC:
1217 case MADV_NOCORE:
1218 case MADV_CORE:
1219 modify_map = 1;
1220 vm_map_lock(map);
1221 break;
1222 case MADV_WILLNEED:
1223 case MADV_DONTNEED:
1224 case MADV_FREE:
1225 vm_map_lock_read(map);
1226 break;
1227 default:
1228 return (KERN_INVALID_ARGUMENT);
1229 }
1230
1231 /*
1232 * Locate starting entry and clip if necessary.
1233 */
1234 VM_MAP_RANGE_CHECK(map, start, end);
1235
1236 if (vm_map_lookup_entry(map, start, &entry)) {
1237 if (modify_map)
1238 vm_map_clip_start(map, entry, start);
1239 } else {
1240 entry = entry->next;
1241 }
1242
1243 if (modify_map) {
1244 /*
1245 * madvise behaviors that are implemented in the vm_map_entry.
1246 *
1247 * We clip the vm_map_entry so that behavioral changes are
1248 * limited to the specified address range.
1249 */
1250 for (current = entry;
1251 (current != &map->header) && (current->start < end);
1252 current = current->next
1253 ) {
1254 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
1255 continue;
1256
1257 vm_map_clip_end(map, current, end);
1258
1259 switch (behav) {
1260 case MADV_NORMAL:
1261 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL);
1262 break;
1263 case MADV_SEQUENTIAL:
1264 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL);
1265 break;
1266 case MADV_RANDOM:
1267 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM);
1268 break;
1269 case MADV_NOSYNC:
1270 current->eflags |= MAP_ENTRY_NOSYNC;
1271 break;
1272 case MADV_AUTOSYNC:
1273 current->eflags &= ~MAP_ENTRY_NOSYNC;
1274 break;
1275 case MADV_NOCORE:
1276 current->eflags |= MAP_ENTRY_NOCOREDUMP;
1277 break;
1278 case MADV_CORE:
1279 current->eflags &= ~MAP_ENTRY_NOCOREDUMP;
1280 break;
1281 default:
1282 break;
1283 }
1284 vm_map_simplify_entry(map, current);
1285 }
1286 vm_map_unlock(map);
1287 } else {
1288 vm_pindex_t pindex;
1289 int count;
1290
1291 /*
1292 * madvise behaviors that are implemented in the underlying
1293 * vm_object.
1294 *
1295 * Since we don't clip the vm_map_entry, we have to clip
1296 * the vm_object pindex and count.
1297 */
1298 for (current = entry;
1299 (current != &map->header) && (current->start < end);
1300 current = current->next
1301 ) {
1302 vm_offset_t useStart;
1303
1304 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
1305 continue;
1306
1307 pindex = OFF_TO_IDX(current->offset);
1308 count = atop(current->end - current->start);
1309 useStart = current->start;
1310
1311 if (current->start < start) {
1312 pindex += atop(start - current->start);
1313 count -= atop(start - current->start);
1314 useStart = start;
1315 }
1316 if (current->end > end)
1317 count -= atop(current->end - end);
1318
1319 if (count <= 0)
1320 continue;
1321
1322 vm_object_madvise(current->object.vm_object,
1323 pindex, count, behav);
1324 if (behav == MADV_WILLNEED) {
1325 pmap_object_init_pt(
1326 map->pmap,
1327 useStart,
1328 current->object.vm_object,
1329 pindex,
1330 (count << PAGE_SHIFT),
1331 MAP_PREFAULT_MADVISE
1332 );
1333 }
1334 }
1335 vm_map_unlock_read(map);
1336 }
1337 return (0);
1338}
1339
1340
1341/*
1342 * vm_map_inherit:
1343 *
1344 * Sets the inheritance of the specified address
1345 * range in the target map. Inheritance
1346 * affects how the map will be shared with
1347 * child maps at the time of vm_map_fork.
1348 */
1349int
1350vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end,
1351 vm_inherit_t new_inheritance)
1352{
1353 vm_map_entry_t entry;
1354 vm_map_entry_t temp_entry;
1355
1356 GIANT_REQUIRED;
1357
1358 switch (new_inheritance) {
1359 case VM_INHERIT_NONE:
1360 case VM_INHERIT_COPY:
1361 case VM_INHERIT_SHARE:
1362 break;
1363 default:
1364 return (KERN_INVALID_ARGUMENT);
1365 }
1366
1367 vm_map_lock(map);
1368
1369 VM_MAP_RANGE_CHECK(map, start, end);
1370
1371 if (vm_map_lookup_entry(map, start, &temp_entry)) {
1372 entry = temp_entry;
1373 vm_map_clip_start(map, entry, start);
1374 } else
1375 entry = temp_entry->next;
1376
1377 while ((entry != &map->header) && (entry->start < end)) {
1378 vm_map_clip_end(map, entry, end);
1379
1380 entry->inheritance = new_inheritance;
1381
1382 vm_map_simplify_entry(map, entry);
1383
1384 entry = entry->next;
1385 }
1386
1387 vm_map_unlock(map);
1388 return (KERN_SUCCESS);
1389}
1390
1391/*
1392 * Implement the semantics of mlock
1393 */
1394int
1395vm_map_user_pageable(
1396 vm_map_t map,
1397 vm_offset_t start,
1398 vm_offset_t end,
1399 boolean_t new_pageable)
1400{
1401 vm_map_entry_t entry;
1402 vm_map_entry_t start_entry;
1403 vm_offset_t estart;
1404 vm_offset_t eend;
1405 int rv;
1406
1407 vm_map_lock(map);
1408 VM_MAP_RANGE_CHECK(map, start, end);
1409
1410 if (vm_map_lookup_entry(map, start, &start_entry) == FALSE) {
1411 vm_map_unlock(map);
1412 return (KERN_INVALID_ADDRESS);
1413 }
1414
1415 if (new_pageable) {
1416
1417 entry = start_entry;
1418 vm_map_clip_start(map, entry, start);
1419
1420 /*
1421 * Now decrement the wiring count for each region. If a region
1422 * becomes completely unwired, unwire its physical pages and
1423 * mappings.
1424 */
1425 while ((entry != &map->header) && (entry->start < end)) {
1426 if (entry->eflags & MAP_ENTRY_USER_WIRED) {
1427 vm_map_clip_end(map, entry, end);
1428 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
1429 entry->wired_count--;
1430 if (entry->wired_count == 0)
1431 vm_fault_unwire(map, entry->start, entry->end);
1432 }
1433 vm_map_simplify_entry(map,entry);
1434 entry = entry->next;
1435 }
1436 } else {
1437
1438 entry = start_entry;
1439
1440 while ((entry != &map->header) && (entry->start < end)) {
1441
1442 if (entry->eflags & MAP_ENTRY_USER_WIRED) {
1443 entry = entry->next;
1444 continue;
1445 }
1446
1447 if (entry->wired_count != 0) {
1448 entry->wired_count++;
1449 entry->eflags |= MAP_ENTRY_USER_WIRED;
1450 entry = entry->next;
1451 continue;
1452 }
1453
1454 /* Here on entry being newly wired */
1455
1456 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1457 int copyflag = entry->eflags & MAP_ENTRY_NEEDS_COPY;
1458 if (copyflag && ((entry->protection & VM_PROT_WRITE) != 0)) {
1459
1460 vm_object_shadow(&entry->object.vm_object,
1461 &entry->offset,
1462 atop(entry->end - entry->start));
1463 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
1464
1465 } else if (entry->object.vm_object == NULL &&
1466 !map->system_map) {
1467
1468 entry->object.vm_object =
1469 vm_object_allocate(OBJT_DEFAULT,
1470 atop(entry->end - entry->start));
1471 entry->offset = (vm_offset_t) 0;
1472
1473 }
1474 }
1475
1476 vm_map_clip_start(map, entry, start);
1477 vm_map_clip_end(map, entry, end);
1478
1479 entry->wired_count++;
1480 entry->eflags |= MAP_ENTRY_USER_WIRED;
1481 estart = entry->start;
1482 eend = entry->end;
1483
1484 /* First we need to allow map modifications */
1485 vm_map_set_recursive(map);
1486 vm_map_lock_downgrade(map);
1487 map->timestamp++;
1488
1489 rv = vm_fault_user_wire(map, entry->start, entry->end);
1490 if (rv) {
1491
1492 entry->wired_count--;
1493 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
1494
1495 vm_map_clear_recursive(map);
1496 vm_map_unlock(map);
1497
1498 /*
1499 * At this point, the map is unlocked, and
1500 * entry might no longer be valid. Use copy
1501 * of entry start value obtained while entry
1502 * was valid.
1503 */
1504 (void) vm_map_user_pageable(map, start, estart,
1505 TRUE);
1506 return rv;
1507 }
1508
1509 vm_map_clear_recursive(map);
1510 if (vm_map_lock_upgrade(map)) {
1511 vm_map_lock(map);
1512 if (vm_map_lookup_entry(map, estart, &entry)
1513 == FALSE) {
1514 vm_map_unlock(map);
1515 /*
1516 * vm_fault_user_wire succeded, thus
1517 * the area between start and eend
1518 * is wired and has to be unwired
1519 * here as part of the cleanup.
1520 */
1521 (void) vm_map_user_pageable(map,
1522 start,
1523 eend,
1524 TRUE);
1525 return (KERN_INVALID_ADDRESS);
1526 }
1527 }
1528 vm_map_simplify_entry(map,entry);
1529 }
1530 }
1531 map->timestamp++;
1532 vm_map_unlock(map);
1533 return KERN_SUCCESS;
1534}
1535
1536/*
1537 * vm_map_pageable:
1538 *
1539 * Sets the pageability of the specified address
1540 * range in the target map. Regions specified
1541 * as not pageable require locked-down physical
1542 * memory and physical page maps.
1543 *
1544 * The map must not be locked, but a reference
1545 * must remain to the map throughout the call.
1546 */
1547int
1548vm_map_pageable(
1549 vm_map_t map,
1550 vm_offset_t start,
1551 vm_offset_t end,
1552 boolean_t new_pageable)
1553{
1554 vm_map_entry_t entry;
1555 vm_map_entry_t start_entry;
1556 vm_offset_t failed = 0;
1557 int rv;
1558
1559 GIANT_REQUIRED;
1560
1561 vm_map_lock(map);
1562
1563 VM_MAP_RANGE_CHECK(map, start, end);
1564
1565 /*
1566 * Only one pageability change may take place at one time, since
1567 * vm_fault assumes it will be called only once for each
1568 * wiring/unwiring. Therefore, we have to make sure we're actually
1569 * changing the pageability for the entire region. We do so before
1570 * making any changes.
1571 */
1572 if (vm_map_lookup_entry(map, start, &start_entry) == FALSE) {
1573 vm_map_unlock(map);
1574 return (KERN_INVALID_ADDRESS);
1575 }
1576 entry = start_entry;
1577
1578 /*
1579 * Actions are rather different for wiring and unwiring, so we have
1580 * two separate cases.
1581 */
1582 if (new_pageable) {
1583 vm_map_clip_start(map, entry, start);
1584
1585 /*
1586 * Unwiring. First ensure that the range to be unwired is
1587 * really wired down and that there are no holes.
1588 */
1589 while ((entry != &map->header) && (entry->start < end)) {
1590 if (entry->wired_count == 0 ||
1591 (entry->end < end &&
1592 (entry->next == &map->header ||
1593 entry->next->start > entry->end))) {
1594 vm_map_unlock(map);
1595 return (KERN_INVALID_ARGUMENT);
1596 }
1597 entry = entry->next;
1598 }
1599
1600 /*
1601 * Now decrement the wiring count for each region. If a region
1602 * becomes completely unwired, unwire its physical pages and
1603 * mappings.
1604 */
1605 entry = start_entry;
1606 while ((entry != &map->header) && (entry->start < end)) {
1607 vm_map_clip_end(map, entry, end);
1608
1609 entry->wired_count--;
1610 if (entry->wired_count == 0)
1611 vm_fault_unwire(map, entry->start, entry->end);
1612
1613 vm_map_simplify_entry(map, entry);
1614
1615 entry = entry->next;
1616 }
1617 } else {
1618 /*
1619 * Wiring. We must do this in two passes:
1620 *
1621 * 1. Holding the write lock, we create any shadow or zero-fill
1622 * objects that need to be created. Then we clip each map
1623 * entry to the region to be wired and increment its wiring
1624 * count. We create objects before clipping the map entries
1625 * to avoid object proliferation.
1626 *
1627 * 2. We downgrade to a read lock, and call vm_fault_wire to
1628 * fault in the pages for any newly wired area (wired_count is
1629 * 1).
1630 *
1631 * Downgrading to a read lock for vm_fault_wire avoids a possible
1632 * deadlock with another process that may have faulted on one
1633 * of the pages to be wired (it would mark the page busy,
1634 * blocking us, then in turn block on the map lock that we
1635 * hold). Because of problems in the recursive lock package,
1636 * we cannot upgrade to a write lock in vm_map_lookup. Thus,
1637 * any actions that require the write lock must be done
1638 * beforehand. Because we keep the read lock on the map, the
1639 * copy-on-write status of the entries we modify here cannot
1640 * change.
1641 */
1642
1643 /*
1644 * Pass 1.
1645 */
1646 while ((entry != &map->header) && (entry->start < end)) {
1647 if (entry->wired_count == 0) {
1648
1649 /*
1650 * Perform actions of vm_map_lookup that need
1651 * the write lock on the map: create a shadow
1652 * object for a copy-on-write region, or an
1653 * object for a zero-fill region.
1654 *
1655 * We don't have to do this for entries that
1656 * point to sub maps, because we won't
1657 * hold the lock on the sub map.
1658 */
1659 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1660 int copyflag = entry->eflags & MAP_ENTRY_NEEDS_COPY;
1661 if (copyflag &&
1662 ((entry->protection & VM_PROT_WRITE) != 0)) {
1663
1664 vm_object_shadow(&entry->object.vm_object,
1665 &entry->offset,
1666 atop(entry->end - entry->start));
1667 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
1668 } else if (entry->object.vm_object == NULL &&
1669 !map->system_map) {
1670 entry->object.vm_object =
1671 vm_object_allocate(OBJT_DEFAULT,
1672 atop(entry->end - entry->start));
1673 entry->offset = (vm_offset_t) 0;
1674 }
1675 }
1676 }
1677 vm_map_clip_start(map, entry, start);
1678 vm_map_clip_end(map, entry, end);
1679 entry->wired_count++;
1680
1681 /*
1682 * Check for holes
1683 */
1684 if (entry->end < end &&
1685 (entry->next == &map->header ||
1686 entry->next->start > entry->end)) {
1687 /*
1688 * Found one. Object creation actions do not
1689 * need to be undone, but the wired counts
1690 * need to be restored.
1691 */
1692 while (entry != &map->header && entry->end > start) {
1693 entry->wired_count--;
1694 entry = entry->prev;
1695 }
1696 vm_map_unlock(map);
1697 return (KERN_INVALID_ARGUMENT);
1698 }
1699 entry = entry->next;
1700 }
1701
1702 /*
1703 * Pass 2.
1704 */
1705
1706 /*
1707 * HACK HACK HACK HACK
1708 *
1709 * If we are wiring in the kernel map or a submap of it,
1710 * unlock the map to avoid deadlocks. We trust that the
1711 * kernel is well-behaved, and therefore will not do
1712 * anything destructive to this region of the map while
1713 * we have it unlocked. We cannot trust user processes
1714 * to do the same.
1715 *
1716 * HACK HACK HACK HACK
1717 */
1718 if (vm_map_pmap(map) == kernel_pmap) {
1719 vm_map_unlock(map); /* trust me ... */
1720 } else {
1721 vm_map_lock_downgrade(map);
1722 }
1723
1724 rv = 0;
1725 entry = start_entry;
1726 while (entry != &map->header && entry->start < end) {
1727 /*
1728 * If vm_fault_wire fails for any page we need to undo
1729 * what has been done. We decrement the wiring count
1730 * for those pages which have not yet been wired (now)
1731 * and unwire those that have (later).
1732 *
1733 * XXX this violates the locking protocol on the map,
1734 * needs to be fixed.
1735 */
1736 if (rv)
1737 entry->wired_count--;
1738 else if (entry->wired_count == 1) {
1739 rv = vm_fault_wire(map, entry->start, entry->end);
1740 if (rv) {
1741 failed = entry->start;
1742 entry->wired_count--;
1743 }
1744 }
1745 entry = entry->next;
1746 }
1747
1748 if (vm_map_pmap(map) == kernel_pmap) {
1749 vm_map_lock(map);
1750 }
1751 if (rv) {
1752 vm_map_unlock(map);
1753 (void) vm_map_pageable(map, start, failed, TRUE);
1754 return (rv);
1755 }
1756 /*
1757 * An exclusive lock on the map is needed in order to call
1758 * vm_map_simplify_entry(). If the current lock on the map
1759 * is only a shared lock, an upgrade is needed.
1760 */
1761 if (vm_map_pmap(map) != kernel_pmap &&
1762 vm_map_lock_upgrade(map)) {
1763 vm_map_lock(map);
1764 if (vm_map_lookup_entry(map, start, &start_entry) ==
1765 FALSE) {
1766 vm_map_unlock(map);
1767 return KERN_SUCCESS;
1768 }
1769 }
1770 vm_map_simplify_entry(map, start_entry);
1771 }
1772
1773 vm_map_unlock(map);
1774
1775 return (KERN_SUCCESS);
1776}
1777
1778/*
1779 * vm_map_clean
1780 *
1781 * Push any dirty cached pages in the address range to their pager.
1782 * If syncio is TRUE, dirty pages are written synchronously.
1783 * If invalidate is TRUE, any cached pages are freed as well.
1784 *
1785 * Returns an error if any part of the specified range is not mapped.
1786 */
1787int
1788vm_map_clean(
1789 vm_map_t map,
1790 vm_offset_t start,
1791 vm_offset_t end,
1792 boolean_t syncio,
1793 boolean_t invalidate)
1794{
1795 vm_map_entry_t current;
1796 vm_map_entry_t entry;
1797 vm_size_t size;
1798 vm_object_t object;
1799 vm_ooffset_t offset;
1800
1801 GIANT_REQUIRED;
1802
1803 vm_map_lock_read(map);
1804 VM_MAP_RANGE_CHECK(map, start, end);
1805 if (!vm_map_lookup_entry(map, start, &entry)) {
1806 vm_map_unlock_read(map);
1807 return (KERN_INVALID_ADDRESS);
1808 }
1809 /*
1810 * Make a first pass to check for holes.
1811 */
1812 for (current = entry; current->start < end; current = current->next) {
1813 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
1814 vm_map_unlock_read(map);
1815 return (KERN_INVALID_ARGUMENT);
1816 }
1817 if (end > current->end &&
1818 (current->next == &map->header ||
1819 current->end != current->next->start)) {
1820 vm_map_unlock_read(map);
1821 return (KERN_INVALID_ADDRESS);
1822 }
1823 }
1824
1825 if (invalidate)
1826 pmap_remove(vm_map_pmap(map), start, end);
1827 /*
1828 * Make a second pass, cleaning/uncaching pages from the indicated
1829 * objects as we go.
1830 */
1831 for (current = entry; current->start < end; current = current->next) {
1832 offset = current->offset + (start - current->start);
1833 size = (end <= current->end ? end : current->end) - start;
1834 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
1835 vm_map_t smap;
1836 vm_map_entry_t tentry;
1837 vm_size_t tsize;
1838
1839 smap = current->object.sub_map;
1840 vm_map_lock_read(smap);
1841 (void) vm_map_lookup_entry(smap, offset, &tentry);
1842 tsize = tentry->end - offset;
1843 if (tsize < size)
1844 size = tsize;
1845 object = tentry->object.vm_object;
1846 offset = tentry->offset + (offset - tentry->start);
1847 vm_map_unlock_read(smap);
1848 } else {
1849 object = current->object.vm_object;
1850 }
1851 /*
1852 * Note that there is absolutely no sense in writing out
1853 * anonymous objects, so we track down the vnode object
1854 * to write out.
1855 * We invalidate (remove) all pages from the address space
1856 * anyway, for semantic correctness.
1857 *
1858 * note: certain anonymous maps, such as MAP_NOSYNC maps,
1859 * may start out with a NULL object.
1860 */
1861 while (object && object->backing_object) {
1862 object = object->backing_object;
1863 offset += object->backing_object_offset;
1864 if (object->size < OFF_TO_IDX(offset + size))
1865 size = IDX_TO_OFF(object->size) - offset;
1866 }
1867 if (object && (object->type == OBJT_VNODE) &&
1868 (current->protection & VM_PROT_WRITE)) {
1869 /*
1870 * Flush pages if writing is allowed, invalidate them
1871 * if invalidation requested. Pages undergoing I/O
1872 * will be ignored by vm_object_page_remove().
1873 *
1874 * We cannot lock the vnode and then wait for paging
1875 * to complete without deadlocking against vm_fault.
1876 * Instead we simply call vm_object_page_remove() and
1877 * allow it to block internally on a page-by-page
1878 * basis when it encounters pages undergoing async
1879 * I/O.
1880 */
1881 int flags;
1882
1883 vm_object_reference(object);
1884 vn_lock(object->handle, LK_EXCLUSIVE | LK_RETRY, curthread);
1885 flags = (syncio || invalidate) ? OBJPC_SYNC : 0;
1886 flags |= invalidate ? OBJPC_INVAL : 0;
1887 vm_object_page_clean(object,
1888 OFF_TO_IDX(offset),
1889 OFF_TO_IDX(offset + size + PAGE_MASK),
1890 flags);
1891 if (invalidate) {
1892 /*vm_object_pip_wait(object, "objmcl");*/
1893 vm_object_page_remove(object,
1894 OFF_TO_IDX(offset),
1895 OFF_TO_IDX(offset + size + PAGE_MASK),
1896 FALSE);
1897 }
1898 VOP_UNLOCK(object->handle, 0, curthread);
1899 vm_object_deallocate(object);
1900 }
1901 start += size;
1902 }
1903
1904 vm_map_unlock_read(map);
1905 return (KERN_SUCCESS);
1906}
1907
1908/*
1909 * vm_map_entry_unwire: [ internal use only ]
1910 *
1911 * Make the region specified by this entry pageable.
1912 *
1913 * The map in question should be locked.
1914 * [This is the reason for this routine's existence.]
1915 */
1916static void
1917vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry)
1918{
1919 vm_fault_unwire(map, entry->start, entry->end);
1920 entry->wired_count = 0;
1921}
1922
1923/*
1924 * vm_map_entry_delete: [ internal use only ]
1925 *
1926 * Deallocate the given entry from the target map.
1927 */
1928static void
1929vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry)
1930{
1931 vm_map_entry_unlink(map, entry);
1932 map->size -= entry->end - entry->start;
1933
1934 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1935 vm_object_deallocate(entry->object.vm_object);
1936 }
1937
1938 vm_map_entry_dispose(map, entry);
1939}
1940
1941/*
1942 * vm_map_delete: [ internal use only ]
1943 *
1944 * Deallocates the given address range from the target
1945 * map.
1946 */
1947int
1948vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end)
1949{
1950 vm_object_t object;
1951 vm_map_entry_t entry;
1952 vm_map_entry_t first_entry;
1953
1954 GIANT_REQUIRED;
1955
1956 /*
1957 * Find the start of the region, and clip it
1958 */
1959 if (!vm_map_lookup_entry(map, start, &first_entry))
1960 entry = first_entry->next;
1961 else {
1962 entry = first_entry;
1963 vm_map_clip_start(map, entry, start);
1964 /*
1965 * Fix the lookup hint now, rather than each time though the
1966 * loop.
1967 */
1968 SAVE_HINT(map, entry->prev);
1969 }
1970
1971 /*
1972 * Save the free space hint
1973 */
1974 if (entry == &map->header) {
1975 map->first_free = &map->header;
1976 } else if (map->first_free->start >= start) {
1977 map->first_free = entry->prev;
1978 }
1979
1980 /*
1981 * Step through all entries in this region
1982 */
1983 while ((entry != &map->header) && (entry->start < end)) {
1984 vm_map_entry_t next;
1985 vm_offset_t s, e;
1986 vm_pindex_t offidxstart, offidxend, count;
1987
1988 vm_map_clip_end(map, entry, end);
1989
1990 s = entry->start;
1991 e = entry->end;
1992 next = entry->next;
1993
1994 offidxstart = OFF_TO_IDX(entry->offset);
1995 count = OFF_TO_IDX(e - s);
1996 object = entry->object.vm_object;
1997
1998 /*
1999 * Unwire before removing addresses from the pmap; otherwise,
2000 * unwiring will put the entries back in the pmap.
2001 */
2002 if (entry->wired_count != 0) {
2003 vm_map_entry_unwire(map, entry);
2004 }
2005
2006 offidxend = offidxstart + count;
2007
2008 if ((object == kernel_object) || (object == kmem_object)) {
2009 vm_object_page_remove(object, offidxstart, offidxend, FALSE);
2010 } else {
2011 pmap_remove(map->pmap, s, e);
2012 if (object != NULL &&
2013 object->ref_count != 1 &&
2014 (object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING &&
2015 (object->type == OBJT_DEFAULT || object->type == OBJT_SWAP)) {
2016 vm_object_collapse(object);
2017 vm_object_page_remove(object, offidxstart, offidxend, FALSE);
2018 if (object->type == OBJT_SWAP) {
2019 swap_pager_freespace(object, offidxstart, count);
2020 }
2021 if (offidxend >= object->size &&
2022 offidxstart < object->size) {
2023 object->size = offidxstart;
2024 }
2025 }
2026 }
2027
2028 /*
2029 * Delete the entry (which may delete the object) only after
2030 * removing all pmap entries pointing to its pages.
2031 * (Otherwise, its page frames may be reallocated, and any
2032 * modify bits will be set in the wrong object!)
2033 */
2034 vm_map_entry_delete(map, entry);
2035 entry = next;
2036 }
2037 return (KERN_SUCCESS);
2038}
2039
2040/*
2041 * vm_map_remove:
2042 *
2043 * Remove the given address range from the target map.
2044 * This is the exported form of vm_map_delete.
2045 */
2046int
2047vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end)
2048{
2049 int result, s = 0;
2050
2051 GIANT_REQUIRED;
2052
2053 if (map == kmem_map)
2054 s = splvm();
2055
2056 vm_map_lock(map);
2057 VM_MAP_RANGE_CHECK(map, start, end);
2058 result = vm_map_delete(map, start, end);
2059 vm_map_unlock(map);
2060
2061 if (map == kmem_map)
2062 splx(s);
2063
2064 return (result);
2065}
2066
2067/*
2068 * vm_map_check_protection:
2069 *
2070 * Assert that the target map allows the specified
2071 * privilege on the entire address region given.
2072 * The entire region must be allocated.
2073 */
2074boolean_t
2075vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end,
2076 vm_prot_t protection)
2077{
2078 vm_map_entry_t entry;
2079 vm_map_entry_t tmp_entry;
2080
2081 GIANT_REQUIRED;
2082
2083 vm_map_lock_read(map);
2084 if (!vm_map_lookup_entry(map, start, &tmp_entry)) {
2085 vm_map_unlock_read(map);
2086 return (FALSE);
2087 }
2088 entry = tmp_entry;
2089
2090 while (start < end) {
2091 if (entry == &map->header) {
2092 vm_map_unlock_read(map);
2093 return (FALSE);
2094 }
2095 /*
2096 * No holes allowed!
2097 */
2098 if (start < entry->start) {
2099 vm_map_unlock_read(map);
2100 return (FALSE);
2101 }
2102 /*
2103 * Check protection associated with entry.
2104 */
2105 if ((entry->protection & protection) != protection) {
2106 vm_map_unlock_read(map);
2107 return (FALSE);
2108 }
2109 /* go to next entry */
2110 start = entry->end;
2111 entry = entry->next;
2112 }
2113 vm_map_unlock_read(map);
2114 return (TRUE);
2115}
2116
2117/*
2118 * Split the pages in a map entry into a new object. This affords
2119 * easier removal of unused pages, and keeps object inheritance from
2120 * being a negative impact on memory usage.
2121 */
2122static void
2123vm_map_split(vm_map_entry_t entry)
2124{
2125 vm_page_t m;
2126 vm_object_t orig_object, new_object, source;
2127 vm_offset_t s, e;
2128 vm_pindex_t offidxstart, offidxend, idx;
2129 vm_size_t size;
2130 vm_ooffset_t offset;
2131
2132 GIANT_REQUIRED;
2133
2134 orig_object = entry->object.vm_object;
2135 if (orig_object->type != OBJT_DEFAULT && orig_object->type != OBJT_SWAP)
2136 return;
2137 if (orig_object->ref_count <= 1)
2138 return;
2139
2140 offset = entry->offset;
2141 s = entry->start;
2142 e = entry->end;
2143
2144 offidxstart = OFF_TO_IDX(offset);
2145 offidxend = offidxstart + OFF_TO_IDX(e - s);
2146 size = offidxend - offidxstart;
2147
2148 new_object = vm_pager_allocate(orig_object->type,
2149 NULL, IDX_TO_OFF(size), VM_PROT_ALL, 0LL);
2150 if (new_object == NULL)
2151 return;
2152
2153 source = orig_object->backing_object;
2154 if (source != NULL) {
2155 vm_object_reference(source); /* Referenced by new_object */
2156 TAILQ_INSERT_TAIL(&source->shadow_head,
2157 new_object, shadow_list);
2158 vm_object_clear_flag(source, OBJ_ONEMAPPING);
2159 new_object->backing_object_offset =
2160 orig_object->backing_object_offset + IDX_TO_OFF(offidxstart);
2161 new_object->backing_object = source;
2162 source->shadow_count++;
2163 source->generation++;
2164 }
2165
2166 for (idx = 0; idx < size; idx++) {
2167 vm_page_t m;
2168
2169 retry:
2170 m = vm_page_lookup(orig_object, offidxstart + idx);
2171 if (m == NULL)
2172 continue;
2173
2174 /*
2175 * We must wait for pending I/O to complete before we can
2176 * rename the page.
2177 *
2178 * We do not have to VM_PROT_NONE the page as mappings should
2179 * not be changed by this operation.
2180 */
2181 if (vm_page_sleep_busy(m, TRUE, "spltwt"))
2182 goto retry;
2183
2184 vm_page_busy(m);
2185 vm_page_rename(m, new_object, idx);
2186 /* page automatically made dirty by rename and cache handled */
2187 vm_page_busy(m);
2188 }
2189
2190 if (orig_object->type == OBJT_SWAP) {
2191 vm_object_pip_add(orig_object, 1);
2192 /*
2193 * copy orig_object pages into new_object
2194 * and destroy unneeded pages in
2195 * shadow object.
2196 */
2197 swap_pager_copy(orig_object, new_object, offidxstart, 0);
2198 vm_object_pip_wakeup(orig_object);
2199 }
2200
2201 for (idx = 0; idx < size; idx++) {
2202 m = vm_page_lookup(new_object, idx);
2203 if (m) {
2204 vm_page_wakeup(m);
2205 }
2206 }
2207
2208 entry->object.vm_object = new_object;
2209 entry->offset = 0LL;
2210 vm_object_deallocate(orig_object);
2211}
2212
2213/*
2214 * vm_map_copy_entry:
2215 *
2216 * Copies the contents of the source entry to the destination
2217 * entry. The entries *must* be aligned properly.
2218 */
2219static void
2220vm_map_copy_entry(
2221 vm_map_t src_map,
2222 vm_map_t dst_map,
2223 vm_map_entry_t src_entry,
2224 vm_map_entry_t dst_entry)
2225{
2226 vm_object_t src_object;
2227
2228 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP)
2229 return;
2230
2231 if (src_entry->wired_count == 0) {
2232
2233 /*
2234 * If the source entry is marked needs_copy, it is already
2235 * write-protected.
2236 */
2237 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) {
2238 pmap_protect(src_map->pmap,
2239 src_entry->start,
2240 src_entry->end,
2241 src_entry->protection & ~VM_PROT_WRITE);
2242 }
2243
2244 /*
2245 * Make a copy of the object.
2246 */
2247 if ((src_object = src_entry->object.vm_object) != NULL) {
2248
2249 if ((src_object->handle == NULL) &&
2250 (src_object->type == OBJT_DEFAULT ||
2251 src_object->type == OBJT_SWAP)) {
2252 vm_object_collapse(src_object);
2253 if ((src_object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) {
2254 vm_map_split(src_entry);
2255 src_object = src_entry->object.vm_object;
2256 }
2257 }
2258
2259 vm_object_reference(src_object);
2260 vm_object_clear_flag(src_object, OBJ_ONEMAPPING);
2261 dst_entry->object.vm_object = src_object;
2262 src_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
2263 dst_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
2264 dst_entry->offset = src_entry->offset;
2265 } else {
2266 dst_entry->object.vm_object = NULL;
2267 dst_entry->offset = 0;
2268 }
2269
2270 pmap_copy(dst_map->pmap, src_map->pmap, dst_entry->start,
2271 dst_entry->end - dst_entry->start, src_entry->start);
2272 } else {
2273 /*
2274 * Of course, wired down pages can't be set copy-on-write.
2275 * Cause wired pages to be copied into the new map by
2276 * simulating faults (the new pages are pageable)
2277 */
2278 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry);
2279 }
2280}
2281
2282/*
2283 * vmspace_fork:
2284 * Create a new process vmspace structure and vm_map
2285 * based on those of an existing process. The new map
2286 * is based on the old map, according to the inheritance
2287 * values on the regions in that map.
2288 *
2289 * The source map must not be locked.
2290 */
2291struct vmspace *
2292vmspace_fork(struct vmspace *vm1)
2293{
2294 struct vmspace *vm2;
2295 vm_map_t old_map = &vm1->vm_map;
2296 vm_map_t new_map;
2297 vm_map_entry_t old_entry;
2298 vm_map_entry_t new_entry;
2299 vm_object_t object;
2300
2301 GIANT_REQUIRED;
2302
2303 vm_map_lock(old_map);
2304 old_map->infork = 1;
2305
2306 vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset);
2307 bcopy(&vm1->vm_startcopy, &vm2->vm_startcopy,
2308 (caddr_t) &vm1->vm_endcopy - (caddr_t) &vm1->vm_startcopy);
2309 new_map = &vm2->vm_map; /* XXX */
2310 new_map->timestamp = 1;
2311
2312 old_entry = old_map->header.next;
2313
2314 while (old_entry != &old_map->header) {
2315 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP)
2316 panic("vm_map_fork: encountered a submap");
2317
2318 switch (old_entry->inheritance) {
2319 case VM_INHERIT_NONE:
2320 break;
2321
2322 case VM_INHERIT_SHARE:
2323 /*
2324 * Clone the entry, creating the shared object if necessary.
2325 */
2326 object = old_entry->object.vm_object;
2327 if (object == NULL) {
2328 object = vm_object_allocate(OBJT_DEFAULT,
2329 atop(old_entry->end - old_entry->start));
2330 old_entry->object.vm_object = object;
2331 old_entry->offset = (vm_offset_t) 0;
2332 }
2333
2334 /*
2335 * Add the reference before calling vm_object_shadow
2336 * to insure that a shadow object is created.
2337 */
2338 vm_object_reference(object);
2339 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) {
2340 vm_object_shadow(&old_entry->object.vm_object,
2341 &old_entry->offset,
2342 atop(old_entry->end - old_entry->start));
2343 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
2344 /* Transfer the second reference too. */
2345 vm_object_reference(
2346 old_entry->object.vm_object);
2347 vm_object_deallocate(object);
2348 object = old_entry->object.vm_object;
2349 }
2350 vm_object_clear_flag(object, OBJ_ONEMAPPING);
2351
2352 /*
2353 * Clone the entry, referencing the shared object.
2354 */
2355 new_entry = vm_map_entry_create(new_map);
2356 *new_entry = *old_entry;
2357 new_entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2358 new_entry->wired_count = 0;
2359
2360 /*
2361 * Insert the entry into the new map -- we know we're
2362 * inserting at the end of the new map.
2363 */
2364 vm_map_entry_link(new_map, new_map->header.prev,
2365 new_entry);
2366
2367 /*
2368 * Update the physical map
2369 */
2370 pmap_copy(new_map->pmap, old_map->pmap,
2371 new_entry->start,
2372 (old_entry->end - old_entry->start),
2373 old_entry->start);
2374 break;
2375
2376 case VM_INHERIT_COPY:
2377 /*
2378 * Clone the entry and link into the map.
2379 */
2380 new_entry = vm_map_entry_create(new_map);
2381 *new_entry = *old_entry;
2382 new_entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2383 new_entry->wired_count = 0;
2384 new_entry->object.vm_object = NULL;
2385 vm_map_entry_link(new_map, new_map->header.prev,
2386 new_entry);
2387 vm_map_copy_entry(old_map, new_map, old_entry,
2388 new_entry);
2389 break;
2390 }
2391 old_entry = old_entry->next;
2392 }
2393
2394 new_map->size = old_map->size;
2395 old_map->infork = 0;
2396 vm_map_unlock(old_map);
2397
2398 return (vm2);
2399}
2400
2401int
2402vm_map_stack (vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
2403 vm_prot_t prot, vm_prot_t max, int cow)
2404{
2405 vm_map_entry_t prev_entry;
2406 vm_map_entry_t new_stack_entry;
2407 vm_size_t init_ssize;
2408 int rv;
2409
2410 GIANT_REQUIRED;
2411
2412 if (VM_MIN_ADDRESS > 0 && addrbos < VM_MIN_ADDRESS)
2413 return (KERN_NO_SPACE);
2414
2415 if (max_ssize < sgrowsiz)
2416 init_ssize = max_ssize;
2417 else
2418 init_ssize = sgrowsiz;
2419
2420 vm_map_lock(map);
2421
2422 /* If addr is already mapped, no go */
2423 if (vm_map_lookup_entry(map, addrbos, &prev_entry)) {
2424 vm_map_unlock(map);
2425 return (KERN_NO_SPACE);
2426 }
2427
2428 /* If we can't accomodate max_ssize in the current mapping,
2429 * no go. However, we need to be aware that subsequent user
2430 * mappings might map into the space we have reserved for
2431 * stack, and currently this space is not protected.
2432 *
2433 * Hopefully we will at least detect this condition
2434 * when we try to grow the stack.
2435 */
2436 if ((prev_entry->next != &map->header) &&
2437 (prev_entry->next->start < addrbos + max_ssize)) {
2438 vm_map_unlock(map);
2439 return (KERN_NO_SPACE);
2440 }
2441
2442 /* We initially map a stack of only init_ssize. We will
2443 * grow as needed later. Since this is to be a grow
2444 * down stack, we map at the top of the range.
2445 *
2446 * Note: we would normally expect prot and max to be
2447 * VM_PROT_ALL, and cow to be 0. Possibly we should
2448 * eliminate these as input parameters, and just
2449 * pass these values here in the insert call.
2450 */
2451 rv = vm_map_insert(map, NULL, 0, addrbos + max_ssize - init_ssize,
2452 addrbos + max_ssize, prot, max, cow);
2453
2454 /* Now set the avail_ssize amount */
2455 if (rv == KERN_SUCCESS){
2456 if (prev_entry != &map->header)
2457 vm_map_clip_end(map, prev_entry, addrbos + max_ssize - init_ssize);
2458 new_stack_entry = prev_entry->next;
2459 if (new_stack_entry->end != addrbos + max_ssize ||
2460 new_stack_entry->start != addrbos + max_ssize - init_ssize)
2461 panic ("Bad entry start/end for new stack entry");
2462 else
2463 new_stack_entry->avail_ssize = max_ssize - init_ssize;
2464 }
2465
2466 vm_map_unlock(map);
2467 return (rv);
2468}
2469
2470/* Attempts to grow a vm stack entry. Returns KERN_SUCCESS if the
2471 * desired address is already mapped, or if we successfully grow
2472 * the stack. Also returns KERN_SUCCESS if addr is outside the
2473 * stack range (this is strange, but preserves compatibility with
2474 * the grow function in vm_machdep.c).
2475 */
2476int
2477vm_map_growstack (struct proc *p, vm_offset_t addr)
2478{
2479 vm_map_entry_t prev_entry;
2480 vm_map_entry_t stack_entry;
2481 vm_map_entry_t new_stack_entry;
2482 struct vmspace *vm = p->p_vmspace;
2483 vm_map_t map = &vm->vm_map;
2484 vm_offset_t end;
2485 int grow_amount;
2486 int rv;
2487 int is_procstack;
2488
2489 GIANT_REQUIRED;
2490
2491Retry:
2492 vm_map_lock_read(map);
2493
2494 /* If addr is already in the entry range, no need to grow.*/
2495 if (vm_map_lookup_entry(map, addr, &prev_entry)) {
2496 vm_map_unlock_read(map);
2497 return (KERN_SUCCESS);
2498 }
2499
2500 if ((stack_entry = prev_entry->next) == &map->header) {
2501 vm_map_unlock_read(map);
2502 return (KERN_SUCCESS);
2503 }
2504 if (prev_entry == &map->header)
2505 end = stack_entry->start - stack_entry->avail_ssize;
2506 else
2507 end = prev_entry->end;
2508
2509 /* This next test mimics the old grow function in vm_machdep.c.
2510 * It really doesn't quite make sense, but we do it anyway
2511 * for compatibility.
2512 *
2513 * If not growable stack, return success. This signals the
2514 * caller to proceed as he would normally with normal vm.
2515 */
2516 if (stack_entry->avail_ssize < 1 ||
2517 addr >= stack_entry->start ||
2518 addr < stack_entry->start - stack_entry->avail_ssize) {
2519 vm_map_unlock_read(map);
2520 return (KERN_SUCCESS);
2521 }
2522
2523 /* Find the minimum grow amount */
2524 grow_amount = roundup (stack_entry->start - addr, PAGE_SIZE);
2525 if (grow_amount > stack_entry->avail_ssize) {
2526 vm_map_unlock_read(map);
2527 return (KERN_NO_SPACE);
2528 }
2529
2530 /* If there is no longer enough space between the entries
2531 * nogo, and adjust the available space. Note: this
2532 * should only happen if the user has mapped into the
2533 * stack area after the stack was created, and is
2534 * probably an error.
2535 *
2536 * This also effectively destroys any guard page the user
2537 * might have intended by limiting the stack size.
2538 */
2539 if (grow_amount > stack_entry->start - end) {
2540 if (vm_map_lock_upgrade(map))
2541 goto Retry;
2542
2543 stack_entry->avail_ssize = stack_entry->start - end;
2544
2545 vm_map_unlock(map);
2546 return (KERN_NO_SPACE);
2547 }
2548
2549 is_procstack = addr >= (vm_offset_t)vm->vm_maxsaddr;
2550
2551 /* If this is the main process stack, see if we're over the
2552 * stack limit.
2553 */
2554 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount >
2555 p->p_rlimit[RLIMIT_STACK].rlim_cur)) {
2556 vm_map_unlock_read(map);
2557 return (KERN_NO_SPACE);
2558 }
2559
2560 /* Round up the grow amount modulo SGROWSIZ */
2561 grow_amount = roundup (grow_amount, sgrowsiz);
2562 if (grow_amount > stack_entry->avail_ssize) {
2563 grow_amount = stack_entry->avail_ssize;
2564 }
2565 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount >
2566 p->p_rlimit[RLIMIT_STACK].rlim_cur)) {
2567 grow_amount = p->p_rlimit[RLIMIT_STACK].rlim_cur -
2568 ctob(vm->vm_ssize);
2569 }
2570
2571 if (vm_map_lock_upgrade(map))
2572 goto Retry;
2573
2574 /* Get the preliminary new entry start value */
2575 addr = stack_entry->start - grow_amount;
2576
2577 /* If this puts us into the previous entry, cut back our growth
2578 * to the available space. Also, see the note above.
2579 */
2580 if (addr < end) {
2581 stack_entry->avail_ssize = stack_entry->start - end;
2582 addr = end;
2583 }
2584
2585 rv = vm_map_insert(map, NULL, 0, addr, stack_entry->start,
2586 VM_PROT_ALL,
2587 VM_PROT_ALL,
2588 0);
2589
2590 /* Adjust the available stack space by the amount we grew. */
2591 if (rv == KERN_SUCCESS) {
2592 if (prev_entry != &map->header)
2593 vm_map_clip_end(map, prev_entry, addr);
2594 new_stack_entry = prev_entry->next;
2595 if (new_stack_entry->end != stack_entry->start ||
2596 new_stack_entry->start != addr)
2597 panic ("Bad stack grow start/end in new stack entry");
2598 else {
2599 new_stack_entry->avail_ssize = stack_entry->avail_ssize -
2600 (new_stack_entry->end -
2601 new_stack_entry->start);
2602 if (is_procstack)
2603 vm->vm_ssize += btoc(new_stack_entry->end -
2604 new_stack_entry->start);
2605 }
2606 }
2607
2608 vm_map_unlock(map);
2609 return (rv);
2610}
2611
2612/*
2613 * Unshare the specified VM space for exec. If other processes are
2614 * mapped to it, then create a new one. The new vmspace is null.
2615 */
2616void
2617vmspace_exec(struct proc *p)
2618{
2619 struct vmspace *oldvmspace = p->p_vmspace;
2620 struct vmspace *newvmspace;
2621 vm_map_t map = &p->p_vmspace->vm_map;
2622
2623 GIANT_REQUIRED;
2624 newvmspace = vmspace_alloc(map->min_offset, map->max_offset);
2625 bcopy(&oldvmspace->vm_startcopy, &newvmspace->vm_startcopy,
2626 (caddr_t) (newvmspace + 1) - (caddr_t) &newvmspace->vm_startcopy);
2627 /*
2628 * This code is written like this for prototype purposes. The
2629 * goal is to avoid running down the vmspace here, but let the
2630 * other process's that are still using the vmspace to finally
2631 * run it down. Even though there is little or no chance of blocking
2632 * here, it is a good idea to keep this form for future mods.
2633 */
2634 p->p_vmspace = newvmspace;
2635 pmap_pinit2(vmspace_pmap(newvmspace));
2636 vmspace_free(oldvmspace);
2637 if (p == curthread->td_proc) /* XXXKSE ? */
2638 pmap_activate(curthread);
2639}
2640
2641/*
2642 * Unshare the specified VM space for forcing COW. This
2643 * is called by rfork, for the (RFMEM|RFPROC) == 0 case.
2644 */
2645void
2646vmspace_unshare(struct proc *p)
2647{
2648 struct vmspace *oldvmspace = p->p_vmspace;
2649 struct vmspace *newvmspace;
2650
2651 GIANT_REQUIRED;
2652 if (oldvmspace->vm_refcnt == 1)
2653 return;
2654 newvmspace = vmspace_fork(oldvmspace);
2655 p->p_vmspace = newvmspace;
2656 pmap_pinit2(vmspace_pmap(newvmspace));
2657 vmspace_free(oldvmspace);
2658 if (p == curthread->td_proc) /* XXXKSE ? */
2659 pmap_activate(curthread);
2660}
2661
2662/*
2663 * vm_map_lookup:
2664 *
2665 * Finds the VM object, offset, and
2666 * protection for a given virtual address in the
2667 * specified map, assuming a page fault of the
2668 * type specified.
2669 *
2670 * Leaves the map in question locked for read; return
2671 * values are guaranteed until a vm_map_lookup_done
2672 * call is performed. Note that the map argument
2673 * is in/out; the returned map must be used in
2674 * the call to vm_map_lookup_done.
2675 *
2676 * A handle (out_entry) is returned for use in
2677 * vm_map_lookup_done, to make that fast.
2678 *
2679 * If a lookup is requested with "write protection"
2680 * specified, the map may be changed to perform virtual
2681 * copying operations, although the data referenced will
2682 * remain the same.
2683 */
2684int
2685vm_map_lookup(vm_map_t *var_map, /* IN/OUT */
2686 vm_offset_t vaddr,
2687 vm_prot_t fault_typea,
2688 vm_map_entry_t *out_entry, /* OUT */
2689 vm_object_t *object, /* OUT */
2690 vm_pindex_t *pindex, /* OUT */
2691 vm_prot_t *out_prot, /* OUT */
2692 boolean_t *wired) /* OUT */
2693{
2694 vm_map_entry_t entry;
2695 vm_map_t map = *var_map;
2696 vm_prot_t prot;
2697 vm_prot_t fault_type = fault_typea;
2698
2699 GIANT_REQUIRED;
2700RetryLookup:;
2701 /*
2702 * Lookup the faulting address.
2703 */
2704
2705 vm_map_lock_read(map);
2706#define RETURN(why) \
2707 { \
2708 vm_map_unlock_read(map); \
2709 return (why); \
2710 }
2711
2712 /*
2713 * If the map has an interesting hint, try it before calling full
2714 * blown lookup routine.
2715 */
2716 entry = map->hint;
2717 *out_entry = entry;
2718 if ((entry == &map->header) ||
2719 (vaddr < entry->start) || (vaddr >= entry->end)) {
2720 vm_map_entry_t tmp_entry;
2721
2722 /*
2723 * Entry was either not a valid hint, or the vaddr was not
2724 * contained in the entry, so do a full lookup.
2725 */
2726 if (!vm_map_lookup_entry(map, vaddr, &tmp_entry))
2727 RETURN(KERN_INVALID_ADDRESS);
2728
2729 entry = tmp_entry;
2730 *out_entry = entry;
2731 }
2732
2733 /*
2734 * Handle submaps.
2735 */
2736 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
2737 vm_map_t old_map = map;
2738
2739 *var_map = map = entry->object.sub_map;
2740 vm_map_unlock_read(old_map);
2741 goto RetryLookup;
2742 }
2743
2744 /*
2745 * Check whether this task is allowed to have this page.
2746 * Note the special case for MAP_ENTRY_COW
2747 * pages with an override. This is to implement a forced
2748 * COW for debuggers.
2749 */
2750 if (fault_type & VM_PROT_OVERRIDE_WRITE)
2751 prot = entry->max_protection;
2752 else
2753 prot = entry->protection;
2754 fault_type &= (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE);
2755 if ((fault_type & prot) != fault_type) {
2756 RETURN(KERN_PROTECTION_FAILURE);
2757 }
2758 if ((entry->eflags & MAP_ENTRY_USER_WIRED) &&
2759 (entry->eflags & MAP_ENTRY_COW) &&
2760 (fault_type & VM_PROT_WRITE) &&
2761 (fault_typea & VM_PROT_OVERRIDE_WRITE) == 0) {
2762 RETURN(KERN_PROTECTION_FAILURE);
2763 }
2764
2765 /*
2766 * If this page is not pageable, we have to get it for all possible
2767 * accesses.
2768 */
2769 *wired = (entry->wired_count != 0);
2770 if (*wired)
2771 prot = fault_type = entry->protection;
2772
2773 /*
2774 * If the entry was copy-on-write, we either ...
2775 */
2776 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
2777 /*
2778 * If we want to write the page, we may as well handle that
2779 * now since we've got the map locked.
2780 *
2781 * If we don't need to write the page, we just demote the
2782 * permissions allowed.
2783 */
2784 if (fault_type & VM_PROT_WRITE) {
2785 /*
2786 * Make a new object, and place it in the object
2787 * chain. Note that no new references have appeared
2788 * -- one just moved from the map to the new
2789 * object.
2790 */
2791 if (vm_map_lock_upgrade(map))
2792 goto RetryLookup;
2793 vm_object_shadow(
2794 &entry->object.vm_object,
2795 &entry->offset,
2796 atop(entry->end - entry->start));
2797 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
2798 vm_map_lock_downgrade(map);
2799 } else {
2800 /*
2801 * We're attempting to read a copy-on-write page --
2802 * don't allow writes.
2803 */
2804 prot &= ~VM_PROT_WRITE;
2805 }
2806 }
2807
2808 /*
2809 * Create an object if necessary.
2810 */
2811 if (entry->object.vm_object == NULL &&
2812 !map->system_map) {
2813 if (vm_map_lock_upgrade(map))
2814 goto RetryLookup;
2815 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT,
2816 atop(entry->end - entry->start));
2817 entry->offset = 0;
2818 vm_map_lock_downgrade(map);
2819 }
2820
2821 /*
2822 * Return the object/offset from this entry. If the entry was
2823 * copy-on-write or empty, it has been fixed up.
2824 */
2825 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
2826 *object = entry->object.vm_object;
2827
2828 /*
2829 * Return whether this is the only map sharing this data.
2830 */
2831 *out_prot = prot;
2832 return (KERN_SUCCESS);
2833
2834#undef RETURN
2835}
2836
2837/*
2838 * vm_map_lookup_done:
2839 *
2840 * Releases locks acquired by a vm_map_lookup
2841 * (according to the handle returned by that lookup).
2842 */
2843void
2844vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry)
2845{
2846 /*
2847 * Unlock the main-level map
2848 */
2849 GIANT_REQUIRED;
2850 vm_map_unlock_read(map);
2851}
2852
2853/*
2854 * Implement uiomove with VM operations. This handles (and collateral changes)
2855 * support every combination of source object modification, and COW type
2856 * operations.
2857 */
2858int
2859vm_uiomove(
2860 vm_map_t mapa,
2861 vm_object_t srcobject,
2862 off_t cp,
2863 int cnta,
2864 vm_offset_t uaddra,
2865 int *npages)
2866{
2867 vm_map_t map;
2868 vm_object_t first_object, oldobject, object;
2869 vm_map_entry_t entry;
2870 vm_prot_t prot;
2871 boolean_t wired;
2872 int tcnt, rv;
2873 vm_offset_t uaddr, start, end, tend;
2874 vm_pindex_t first_pindex, osize, oindex;
2875 off_t ooffset;
2876 int cnt;
2877
2878 GIANT_REQUIRED;
2879
2880 if (npages)
2881 *npages = 0;
2882
2883 cnt = cnta;
2884 uaddr = uaddra;
2885
2886 while (cnt > 0) {
2887 map = mapa;
2888
2889 if ((vm_map_lookup(&map, uaddr,
2890 VM_PROT_READ, &entry, &first_object,
2891 &first_pindex, &prot, &wired)) != KERN_SUCCESS) {
2892 return EFAULT;
2893 }
2894
2895 vm_map_clip_start(map, entry, uaddr);
2896
2897 tcnt = cnt;
2898 tend = uaddr + tcnt;
2899 if (tend > entry->end) {
2900 tcnt = entry->end - uaddr;
2901 tend = entry->end;
2902 }
2903
2904 vm_map_clip_end(map, entry, tend);
2905
2906 start = entry->start;
2907 end = entry->end;
2908
2909 osize = atop(tcnt);
2910
2911 oindex = OFF_TO_IDX(cp);
2912 if (npages) {
2913 vm_pindex_t idx;
2914 for (idx = 0; idx < osize; idx++) {
2915 vm_page_t m;
2916 if ((m = vm_page_lookup(srcobject, oindex + idx)) == NULL) {
2917 vm_map_lookup_done(map, entry);
2918 return 0;
2919 }
2920 /*
2921 * disallow busy or invalid pages, but allow
2922 * m->busy pages if they are entirely valid.
2923 */
2924 if ((m->flags & PG_BUSY) ||
2925 ((m->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL)) {
2926 vm_map_lookup_done(map, entry);
2927 return 0;
2928 }
2929 }
2930 }
2931
2932/*
2933 * If we are changing an existing map entry, just redirect
2934 * the object, and change mappings.
2935 */
2936 if ((first_object->type == OBJT_VNODE) &&
2937 ((oldobject = entry->object.vm_object) == first_object)) {
2938
2939 if ((entry->offset != cp) || (oldobject != srcobject)) {
2940 /*
2941 * Remove old window into the file
2942 */
2943 pmap_remove (map->pmap, uaddr, tend);
2944
2945 /*
2946 * Force copy on write for mmaped regions
2947 */
2948 vm_object_pmap_copy_1 (srcobject, oindex, oindex + osize);
2949
2950 /*
2951 * Point the object appropriately
2952 */
2953 if (oldobject != srcobject) {
2954
2955 /*
2956 * Set the object optimization hint flag
2957 */
2958 vm_object_set_flag(srcobject, OBJ_OPT);
2959 vm_object_reference(srcobject);
2960 entry->object.vm_object = srcobject;
2961
2962 if (oldobject) {
2963 vm_object_deallocate(oldobject);
2964 }
2965 }
2966
2967 entry->offset = cp;
2968 map->timestamp++;
2969 } else {
2970 pmap_remove (map->pmap, uaddr, tend);
2971 }
2972
2973 } else if ((first_object->ref_count == 1) &&
2974 (first_object->size == osize) &&
2975 ((first_object->type == OBJT_DEFAULT) ||
2976 (first_object->type == OBJT_SWAP)) ) {
2977
2978 oldobject = first_object->backing_object;
2979
2980 if ((first_object->backing_object_offset != cp) ||
2981 (oldobject != srcobject)) {
2982 /*
2983 * Remove old window into the file
2984 */
2985 pmap_remove (map->pmap, uaddr, tend);
2986
2987 /*
2988 * Remove unneeded old pages
2989 */
2990 vm_object_page_remove(first_object, 0, 0, 0);
2991
2992 /*
2993 * Invalidate swap space
2994 */
2995 if (first_object->type == OBJT_SWAP) {
2996 swap_pager_freespace(first_object,
2997 0,
2998 first_object->size);
2999 }
3000
3001 /*
3002 * Force copy on write for mmaped regions
3003 */
3004 vm_object_pmap_copy_1 (srcobject, oindex, oindex + osize);
3005
3006 /*
3007 * Point the object appropriately
3008 */
3009 if (oldobject != srcobject) {
3010 /*
3011 * Set the object optimization hint flag
3012 */
3013 vm_object_set_flag(srcobject, OBJ_OPT);
3014 vm_object_reference(srcobject);
3015
3016 if (oldobject) {
3017 TAILQ_REMOVE(&oldobject->shadow_head,
3018 first_object, shadow_list);
3019 oldobject->shadow_count--;
3020 /* XXX bump generation? */
3021 vm_object_deallocate(oldobject);
3022 }
3023
3024 TAILQ_INSERT_TAIL(&srcobject->shadow_head,
3025 first_object, shadow_list);
3026 srcobject->shadow_count++;
3027 /* XXX bump generation? */
3028
3029 first_object->backing_object = srcobject;
3030 }
3031 first_object->backing_object_offset = cp;
3032 map->timestamp++;
3033 } else {
3034 pmap_remove (map->pmap, uaddr, tend);
3035 }
3036/*
3037 * Otherwise, we have to do a logical mmap.
3038 */
3039 } else {
3040
3041 vm_object_set_flag(srcobject, OBJ_OPT);
3042 vm_object_reference(srcobject);
3043
3044 pmap_remove (map->pmap, uaddr, tend);
3045
3046 vm_object_pmap_copy_1 (srcobject, oindex, oindex + osize);
3047 vm_map_lock_upgrade(map);
3048
3049 if (entry == &map->header) {
3050 map->first_free = &map->header;
3051 } else if (map->first_free->start >= start) {
3052 map->first_free = entry->prev;
3053 }
3054
3055 SAVE_HINT(map, entry->prev);
3056 vm_map_entry_delete(map, entry);
3057
3058 object = srcobject;
3059 ooffset = cp;
3060
3061 rv = vm_map_insert(map, object, ooffset, start, tend,
3062 VM_PROT_ALL, VM_PROT_ALL, MAP_COPY_ON_WRITE);
3063
3064 if (rv != KERN_SUCCESS)
3065 panic("vm_uiomove: could not insert new entry: %d", rv);
3066 }
3067
3068/*
3069 * Map the window directly, if it is already in memory
3070 */
3071 pmap_object_init_pt(map->pmap, uaddr,
3072 srcobject, oindex, tcnt, 0);
3073
3074 map->timestamp++;
3075 vm_map_unlock(map);
3076
3077 cnt -= tcnt;
3078 uaddr += tcnt;
3079 cp += tcnt;
3080 if (npages)
3081 *npages += osize;
3082 }
3083 return 0;
3084}
3085
3086/*
3087 * Performs the copy_on_write operations necessary to allow the virtual copies
3088 * into user space to work. This has to be called for write(2) system calls
3089 * from other processes, file unlinking, and file size shrinkage.
3090 */
3091void
3092vm_freeze_copyopts(vm_object_t object, vm_pindex_t froma, vm_pindex_t toa)
3093{
3094 int rv;
3095 vm_object_t robject;
3096 vm_pindex_t idx;
3097
3098 GIANT_REQUIRED;
3099 if ((object == NULL) ||
3100 ((object->flags & OBJ_OPT) == 0))
3101 return;
3102
3103 if (object->shadow_count > object->ref_count)
3104 panic("vm_freeze_copyopts: sc > rc");
3105
3106 while ((robject = TAILQ_FIRST(&object->shadow_head)) != NULL) {
3107 vm_pindex_t bo_pindex;
3108 vm_page_t m_in, m_out;
3109
3110 bo_pindex = OFF_TO_IDX(robject->backing_object_offset);
3111
3112 vm_object_reference(robject);
3113
3114 vm_object_pip_wait(robject, "objfrz");
3115
3116 if (robject->ref_count == 1) {
3117 vm_object_deallocate(robject);
3118 continue;
3119 }
3120
3121 vm_object_pip_add(robject, 1);
3122
3123 for (idx = 0; idx < robject->size; idx++) {
3124
3125 m_out = vm_page_grab(robject, idx,
3126 VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
3127
3128 if (m_out->valid == 0) {
3129 m_in = vm_page_grab(object, bo_pindex + idx,
3130 VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
3131 if (m_in->valid == 0) {
3132 rv = vm_pager_get_pages(object, &m_in, 1, 0);
3133 if (rv != VM_PAGER_OK) {
3134 printf("vm_freeze_copyopts: cannot read page from file: %lx\n", (long)m_in->pindex);
3135 continue;
3136 }
3137 vm_page_deactivate(m_in);
3138 }
3139
3140 vm_page_protect(m_in, VM_PROT_NONE);
3141 pmap_copy_page(VM_PAGE_TO_PHYS(m_in), VM_PAGE_TO_PHYS(m_out));
3142 m_out->valid = m_in->valid;
3143 vm_page_dirty(m_out);
3144 vm_page_activate(m_out);
3145 vm_page_wakeup(m_in);
3146 }
3147 vm_page_wakeup(m_out);
3148 }
3149
3150 object->shadow_count--;
3151 object->ref_count--;
3152 TAILQ_REMOVE(&object->shadow_head, robject, shadow_list);
3153 robject->backing_object = NULL;
3154 robject->backing_object_offset = 0;
3155
3156 vm_object_pip_wakeup(robject);
3157 vm_object_deallocate(robject);
3158 }
3159
3160 vm_object_clear_flag(object, OBJ_OPT);
3161}
3162
3163#include "opt_ddb.h"
3164#ifdef DDB
3165#include <sys/kernel.h>
3166
3167#include <ddb/ddb.h>
3168
3169/*
3170 * vm_map_print: [ debug ]
3171 */
3172DB_SHOW_COMMAND(map, vm_map_print)
3173{
3174 static int nlines;
3175 /* XXX convert args. */
3176 vm_map_t map = (vm_map_t)addr;
3177 boolean_t full = have_addr;
3178
3179 vm_map_entry_t entry;
3180
3181 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n",
3182 (void *)map,
3183 (void *)map->pmap, map->nentries, map->timestamp);
3184 nlines++;
3185
3186 if (!full && db_indent)
3187 return;
3188
3189 db_indent += 2;
3190 for (entry = map->header.next; entry != &map->header;
3191 entry = entry->next) {
3192 db_iprintf("map entry %p: start=%p, end=%p\n",
3193 (void *)entry, (void *)entry->start, (void *)entry->end);
3194 nlines++;
3195 {
3196 static char *inheritance_name[4] =
3197 {"share", "copy", "none", "donate_copy"};
3198
3199 db_iprintf(" prot=%x/%x/%s",
3200 entry->protection,
3201 entry->max_protection,
3202 inheritance_name[(int)(unsigned char)entry->inheritance]);
3203 if (entry->wired_count != 0)
3204 db_printf(", wired");
3205 }
3206 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
3207 /* XXX no %qd in kernel. Truncate entry->offset. */
3208 db_printf(", share=%p, offset=0x%lx\n",
3209 (void *)entry->object.sub_map,
3210 (long)entry->offset);
3211 nlines++;
3212 if ((entry->prev == &map->header) ||
3213 (entry->prev->object.sub_map !=
3214 entry->object.sub_map)) {
3215 db_indent += 2;
3216 vm_map_print((db_expr_t)(intptr_t)
3217 entry->object.sub_map,
3218 full, 0, (char *)0);
3219 db_indent -= 2;
3220 }
3221 } else {
3222 /* XXX no %qd in kernel. Truncate entry->offset. */
3223 db_printf(", object=%p, offset=0x%lx",
3224 (void *)entry->object.vm_object,
3225 (long)entry->offset);
3226 if (entry->eflags & MAP_ENTRY_COW)
3227 db_printf(", copy (%s)",
3228 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done");
3229 db_printf("\n");
3230 nlines++;
3231
3232 if ((entry->prev == &map->header) ||
3233 (entry->prev->object.vm_object !=
3234 entry->object.vm_object)) {
3235 db_indent += 2;
3236 vm_object_print((db_expr_t)(intptr_t)
3237 entry->object.vm_object,
3238 full, 0, (char *)0);
3239 nlines += 4;
3240 db_indent -= 2;
3241 }
3242 }
3243 }
3244 db_indent -= 2;
3245 if (db_indent == 0)
3246 nlines = 0;
3247}
3248
3249
3250DB_SHOW_COMMAND(procvm, procvm)
3251{
3252 struct proc *p;
3253
3254 if (have_addr) {
3255 p = (struct proc *) addr;
3256 } else {
3257 p = curproc;
3258 }
3259
3260 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n",
3261 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map,
3262 (void *)vmspace_pmap(p->p_vmspace));
3263
3264 vm_map_print((db_expr_t)(intptr_t)&p->p_vmspace->vm_map, 1, 0, NULL);
3265}
3266
3267#endif /* DDB */