FreeBSD/Linux Kernel Cross Reference
sys/vm/vm_map.c
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: releng/5.1/sys/vm/vm_map.c 114317 2003-04-30 19:18:09Z alc $
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 #include <sys/sysent.h>
82 #include <sys/shm.h>
83
84 #include <vm/vm.h>
85 #include <vm/vm_param.h>
86 #include <vm/pmap.h>
87 #include <vm/vm_map.h>
88 #include <vm/vm_page.h>
89 #include <vm/vm_object.h>
90 #include <vm/vm_pager.h>
91 #include <vm/vm_kern.h>
92 #include <vm/vm_extern.h>
93 #include <vm/swap_pager.h>
94 #include <vm/uma.h>
95
96 /*
97 * Virtual memory maps provide for the mapping, protection,
98 * and sharing of virtual memory objects. In addition,
99 * this module provides for an efficient virtual copy of
100 * memory from one map to another.
101 *
102 * Synchronization is required prior to most operations.
103 *
104 * Maps consist of an ordered doubly-linked list of simple
105 * entries; a single hint is used to speed up lookups.
106 *
107 * Since portions of maps are specified by start/end addresses,
108 * which may not align with existing map entries, all
109 * routines merely "clip" entries to these start/end values.
110 * [That is, an entry is split into two, bordering at a
111 * start or end value.] Note that these clippings may not
112 * always be necessary (as the two resulting entries are then
113 * not changed); however, the clipping is done for convenience.
114 *
115 * As mentioned above, virtual copy operations are performed
116 * by copying VM object references from one map to
117 * another, and then marking both regions as copy-on-write.
118 */
119
120 /*
121 * vm_map_startup:
122 *
123 * Initialize the vm_map module. Must be called before
124 * any other vm_map routines.
125 *
126 * Map and entry structures are allocated from the general
127 * purpose memory pool with some exceptions:
128 *
129 * - The kernel map and kmem submap are allocated statically.
130 * - Kernel map entries are allocated out of a static pool.
131 *
132 * These restrictions are necessary since malloc() uses the
133 * maps and requires map entries.
134 */
135
136 static struct mtx map_sleep_mtx;
137 static uma_zone_t mapentzone;
138 static uma_zone_t kmapentzone;
139 static uma_zone_t mapzone;
140 static uma_zone_t vmspace_zone;
141 static struct vm_object kmapentobj;
142 static void vmspace_zinit(void *mem, int size);
143 static void vmspace_zfini(void *mem, int size);
144 static void vm_map_zinit(void *mem, int size);
145 static void vm_map_zfini(void *mem, int size);
146 static void _vm_map_init(vm_map_t map, vm_offset_t min, vm_offset_t max);
147
148 #ifdef INVARIANTS
149 static void vm_map_zdtor(void *mem, int size, void *arg);
150 static void vmspace_zdtor(void *mem, int size, void *arg);
151 #endif
152
153 void
154 vm_map_startup(void)
155 {
156 mtx_init(&map_sleep_mtx, "vm map sleep mutex", NULL, MTX_DEF);
157 mapzone = uma_zcreate("MAP", sizeof(struct vm_map), NULL,
158 #ifdef INVARIANTS
159 vm_map_zdtor,
160 #else
161 NULL,
162 #endif
163 vm_map_zinit, vm_map_zfini, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
164 uma_prealloc(mapzone, MAX_KMAP);
165 kmapentzone = uma_zcreate("KMAP ENTRY", sizeof(struct vm_map_entry),
166 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR,
167 UMA_ZONE_MTXCLASS | UMA_ZONE_VM);
168 uma_prealloc(kmapentzone, MAX_KMAPENT);
169 mapentzone = uma_zcreate("MAP ENTRY", sizeof(struct vm_map_entry),
170 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
171 uma_prealloc(mapentzone, MAX_MAPENT);
172 }
173
174 static void
175 vmspace_zfini(void *mem, int size)
176 {
177 struct vmspace *vm;
178
179 vm = (struct vmspace *)mem;
180
181 vm_map_zfini(&vm->vm_map, sizeof(vm->vm_map));
182 }
183
184 static void
185 vmspace_zinit(void *mem, int size)
186 {
187 struct vmspace *vm;
188
189 vm = (struct vmspace *)mem;
190
191 vm_map_zinit(&vm->vm_map, sizeof(vm->vm_map));
192 }
193
194 static void
195 vm_map_zfini(void *mem, int size)
196 {
197 vm_map_t map;
198
199 map = (vm_map_t)mem;
200 mtx_destroy(&map->system_mtx);
201 lockdestroy(&map->lock);
202 }
203
204 static void
205 vm_map_zinit(void *mem, int size)
206 {
207 vm_map_t map;
208
209 map = (vm_map_t)mem;
210 map->nentries = 0;
211 map->size = 0;
212 map->infork = 0;
213 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK);
214 lockinit(&map->lock, PVM, "thrd_sleep", 0, LK_NOPAUSE);
215 }
216
217 #ifdef INVARIANTS
218 static void
219 vmspace_zdtor(void *mem, int size, void *arg)
220 {
221 struct vmspace *vm;
222
223 vm = (struct vmspace *)mem;
224
225 vm_map_zdtor(&vm->vm_map, sizeof(vm->vm_map), arg);
226 }
227 static void
228 vm_map_zdtor(void *mem, int size, void *arg)
229 {
230 vm_map_t map;
231
232 map = (vm_map_t)mem;
233 KASSERT(map->nentries == 0,
234 ("map %p nentries == %d on free.",
235 map, map->nentries));
236 KASSERT(map->size == 0,
237 ("map %p size == %lu on free.",
238 map, (unsigned long)map->size));
239 KASSERT(map->infork == 0,
240 ("map %p infork == %d on free.",
241 map, map->infork));
242 }
243 #endif /* INVARIANTS */
244
245 /*
246 * Allocate a vmspace structure, including a vm_map and pmap,
247 * and initialize those structures. The refcnt is set to 1.
248 * The remaining fields must be initialized by the caller.
249 */
250 struct vmspace *
251 vmspace_alloc(min, max)
252 vm_offset_t min, max;
253 {
254 struct vmspace *vm;
255
256 GIANT_REQUIRED;
257 vm = uma_zalloc(vmspace_zone, M_WAITOK);
258 CTR1(KTR_VM, "vmspace_alloc: %p", vm);
259 _vm_map_init(&vm->vm_map, min, max);
260 pmap_pinit(vmspace_pmap(vm));
261 vm->vm_map.pmap = vmspace_pmap(vm); /* XXX */
262 vm->vm_refcnt = 1;
263 vm->vm_shm = NULL;
264 vm->vm_exitingcnt = 0;
265 return (vm);
266 }
267
268 void
269 vm_init2(void)
270 {
271 uma_zone_set_obj(kmapentzone, &kmapentobj, lmin(cnt.v_page_count,
272 (VM_MAX_KERNEL_ADDRESS - KERNBASE) / PAGE_SIZE) / 8);
273 vmspace_zone = uma_zcreate("VMSPACE", sizeof(struct vmspace), NULL,
274 #ifdef INVARIANTS
275 vmspace_zdtor,
276 #else
277 NULL,
278 #endif
279 vmspace_zinit, vmspace_zfini, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
280 pmap_init2();
281 }
282
283 static __inline void
284 vmspace_dofree(struct vmspace *vm)
285 {
286 CTR1(KTR_VM, "vmspace_free: %p", vm);
287
288 /*
289 * Make sure any SysV shm is freed, it might not have been in
290 * exit1().
291 */
292 shmexit(vm);
293
294 /*
295 * Lock the map, to wait out all other references to it.
296 * Delete all of the mappings and pages they hold, then call
297 * the pmap module to reclaim anything left.
298 */
299 vm_map_lock(&vm->vm_map);
300 (void) vm_map_delete(&vm->vm_map, vm->vm_map.min_offset,
301 vm->vm_map.max_offset);
302 vm_map_unlock(&vm->vm_map);
303
304 pmap_release(vmspace_pmap(vm));
305 uma_zfree(vmspace_zone, vm);
306 }
307
308 void
309 vmspace_free(struct vmspace *vm)
310 {
311 GIANT_REQUIRED;
312
313 if (vm->vm_refcnt == 0)
314 panic("vmspace_free: attempt to free already freed vmspace");
315
316 if (--vm->vm_refcnt == 0 && vm->vm_exitingcnt == 0)
317 vmspace_dofree(vm);
318 }
319
320 void
321 vmspace_exitfree(struct proc *p)
322 {
323 struct vmspace *vm;
324
325 GIANT_REQUIRED;
326 vm = p->p_vmspace;
327 p->p_vmspace = NULL;
328
329 /*
330 * cleanup by parent process wait()ing on exiting child. vm_refcnt
331 * may not be 0 (e.g. fork() and child exits without exec()ing).
332 * exitingcnt may increment above 0 and drop back down to zero
333 * several times while vm_refcnt is held non-zero. vm_refcnt
334 * may also increment above 0 and drop back down to zero several
335 * times while vm_exitingcnt is held non-zero.
336 *
337 * The last wait on the exiting child's vmspace will clean up
338 * the remainder of the vmspace.
339 */
340 if (--vm->vm_exitingcnt == 0 && vm->vm_refcnt == 0)
341 vmspace_dofree(vm);
342 }
343
344 /*
345 * vmspace_swap_count() - count the approximate swap useage in pages for a
346 * vmspace.
347 *
348 * The map must be locked.
349 *
350 * Swap useage is determined by taking the proportional swap used by
351 * VM objects backing the VM map. To make up for fractional losses,
352 * if the VM object has any swap use at all the associated map entries
353 * count for at least 1 swap page.
354 */
355 int
356 vmspace_swap_count(struct vmspace *vmspace)
357 {
358 vm_map_t map = &vmspace->vm_map;
359 vm_map_entry_t cur;
360 int count = 0;
361
362 for (cur = map->header.next; cur != &map->header; cur = cur->next) {
363 vm_object_t object;
364
365 if ((cur->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 &&
366 (object = cur->object.vm_object) != NULL) {
367 VM_OBJECT_LOCK(object);
368 if (object->type == OBJT_SWAP &&
369 object->un_pager.swp.swp_bcount != 0) {
370 int n = (cur->end - cur->start) / PAGE_SIZE;
371
372 count += object->un_pager.swp.swp_bcount *
373 SWAP_META_PAGES * n / object->size + 1;
374 }
375 VM_OBJECT_UNLOCK(object);
376 }
377 }
378 return (count);
379 }
380
381 void
382 _vm_map_lock(vm_map_t map, const char *file, int line)
383 {
384 int error;
385
386 if (map->system_map)
387 _mtx_lock_flags(&map->system_mtx, 0, file, line);
388 else {
389 error = lockmgr(&map->lock, LK_EXCLUSIVE, NULL, curthread);
390 KASSERT(error == 0, ("%s: failed to get lock", __func__));
391 }
392 map->timestamp++;
393 }
394
395 void
396 _vm_map_unlock(vm_map_t map, const char *file, int line)
397 {
398
399 if (map->system_map)
400 _mtx_unlock_flags(&map->system_mtx, 0, file, line);
401 else
402 lockmgr(&map->lock, LK_RELEASE, NULL, curthread);
403 }
404
405 void
406 _vm_map_lock_read(vm_map_t map, const char *file, int line)
407 {
408 int error;
409
410 if (map->system_map)
411 _mtx_lock_flags(&map->system_mtx, 0, file, line);
412 else {
413 error = lockmgr(&map->lock, LK_EXCLUSIVE, NULL, curthread);
414 KASSERT(error == 0, ("%s: failed to get lock", __func__));
415 }
416 }
417
418 void
419 _vm_map_unlock_read(vm_map_t map, const char *file, int line)
420 {
421
422 if (map->system_map)
423 _mtx_unlock_flags(&map->system_mtx, 0, file, line);
424 else
425 lockmgr(&map->lock, LK_RELEASE, NULL, curthread);
426 }
427
428 int
429 _vm_map_trylock(vm_map_t map, const char *file, int line)
430 {
431 int error;
432
433 error = map->system_map ?
434 !_mtx_trylock(&map->system_mtx, 0, file, line) :
435 lockmgr(&map->lock, LK_EXCLUSIVE | LK_NOWAIT, NULL, curthread);
436 if (error == 0)
437 map->timestamp++;
438 return (error == 0);
439 }
440
441 int
442 _vm_map_trylock_read(vm_map_t map, const char *file, int line)
443 {
444 int error;
445
446 error = map->system_map ?
447 !_mtx_trylock(&map->system_mtx, 0, file, line) :
448 lockmgr(&map->lock, LK_EXCLUSIVE | LK_NOWAIT, NULL, curthread);
449 return (error == 0);
450 }
451
452 int
453 _vm_map_lock_upgrade(vm_map_t map, const char *file, int line)
454 {
455
456 if (map->system_map) {
457 #ifdef INVARIANTS
458 _mtx_assert(&map->system_mtx, MA_OWNED, file, line);
459 #endif
460 } else
461 KASSERT(lockstatus(&map->lock, curthread) == LK_EXCLUSIVE,
462 ("%s: lock not held", __func__));
463 map->timestamp++;
464 return (0);
465 }
466
467 void
468 _vm_map_lock_downgrade(vm_map_t map, const char *file, int line)
469 {
470
471 if (map->system_map) {
472 #ifdef INVARIANTS
473 _mtx_assert(&map->system_mtx, MA_OWNED, file, line);
474 #endif
475 } else
476 KASSERT(lockstatus(&map->lock, curthread) == LK_EXCLUSIVE,
477 ("%s: lock not held", __func__));
478 }
479
480 /*
481 * vm_map_unlock_and_wait:
482 */
483 int
484 vm_map_unlock_and_wait(vm_map_t map, boolean_t user_wait)
485 {
486
487 mtx_lock(&map_sleep_mtx);
488 vm_map_unlock(map);
489 return (msleep(&map->root, &map_sleep_mtx, PDROP | PVM, "vmmaps", 0));
490 }
491
492 /*
493 * vm_map_wakeup:
494 */
495 void
496 vm_map_wakeup(vm_map_t map)
497 {
498
499 /*
500 * Acquire and release map_sleep_mtx to prevent a wakeup()
501 * from being performed (and lost) between the vm_map_unlock()
502 * and the msleep() in vm_map_unlock_and_wait().
503 */
504 mtx_lock(&map_sleep_mtx);
505 mtx_unlock(&map_sleep_mtx);
506 wakeup(&map->root);
507 }
508
509 long
510 vmspace_resident_count(struct vmspace *vmspace)
511 {
512 return pmap_resident_count(vmspace_pmap(vmspace));
513 }
514
515 /*
516 * vm_map_create:
517 *
518 * Creates and returns a new empty VM map with
519 * the given physical map structure, and having
520 * the given lower and upper address bounds.
521 */
522 vm_map_t
523 vm_map_create(pmap_t pmap, vm_offset_t min, vm_offset_t max)
524 {
525 vm_map_t result;
526
527 result = uma_zalloc(mapzone, M_WAITOK);
528 CTR1(KTR_VM, "vm_map_create: %p", result);
529 _vm_map_init(result, min, max);
530 result->pmap = pmap;
531 return (result);
532 }
533
534 /*
535 * Initialize an existing vm_map structure
536 * such as that in the vmspace structure.
537 * The pmap is set elsewhere.
538 */
539 static void
540 _vm_map_init(vm_map_t map, vm_offset_t min, vm_offset_t max)
541 {
542
543 map->header.next = map->header.prev = &map->header;
544 map->needs_wakeup = FALSE;
545 map->system_map = 0;
546 map->min_offset = min;
547 map->max_offset = max;
548 map->first_free = &map->header;
549 map->root = NULL;
550 map->timestamp = 0;
551 }
552
553 void
554 vm_map_init(vm_map_t map, vm_offset_t min, vm_offset_t max)
555 {
556 _vm_map_init(map, min, max);
557 mtx_init(&map->system_mtx, "system map", NULL, MTX_DEF | MTX_DUPOK);
558 lockinit(&map->lock, PVM, "thrd_sleep", 0, LK_NOPAUSE);
559 }
560
561 /*
562 * vm_map_entry_dispose: [ internal use only ]
563 *
564 * Inverse of vm_map_entry_create.
565 */
566 static void
567 vm_map_entry_dispose(vm_map_t map, vm_map_entry_t entry)
568 {
569 uma_zfree(map->system_map ? kmapentzone : mapentzone, entry);
570 }
571
572 /*
573 * vm_map_entry_create: [ internal use only ]
574 *
575 * Allocates a VM map entry for insertion.
576 * No entry fields are filled in.
577 */
578 static vm_map_entry_t
579 vm_map_entry_create(vm_map_t map)
580 {
581 vm_map_entry_t new_entry;
582
583 if (map->system_map)
584 new_entry = uma_zalloc(kmapentzone, M_NOWAIT);
585 else
586 new_entry = uma_zalloc(mapentzone, M_WAITOK);
587 if (new_entry == NULL)
588 panic("vm_map_entry_create: kernel resources exhausted");
589 return (new_entry);
590 }
591
592 /*
593 * vm_map_entry_set_behavior:
594 *
595 * Set the expected access behavior, either normal, random, or
596 * sequential.
597 */
598 static __inline void
599 vm_map_entry_set_behavior(vm_map_entry_t entry, u_char behavior)
600 {
601 entry->eflags = (entry->eflags & ~MAP_ENTRY_BEHAV_MASK) |
602 (behavior & MAP_ENTRY_BEHAV_MASK);
603 }
604
605 /*
606 * vm_map_entry_splay:
607 *
608 * Implements Sleator and Tarjan's top-down splay algorithm. Returns
609 * the vm_map_entry containing the given address. If, however, that
610 * address is not found in the vm_map, returns a vm_map_entry that is
611 * adjacent to the address, coming before or after it.
612 */
613 static vm_map_entry_t
614 vm_map_entry_splay(vm_offset_t address, vm_map_entry_t root)
615 {
616 struct vm_map_entry dummy;
617 vm_map_entry_t lefttreemax, righttreemin, y;
618
619 if (root == NULL)
620 return (root);
621 lefttreemax = righttreemin = &dummy;
622 for (;; root = y) {
623 if (address < root->start) {
624 if ((y = root->left) == NULL)
625 break;
626 if (address < y->start) {
627 /* Rotate right. */
628 root->left = y->right;
629 y->right = root;
630 root = y;
631 if ((y = root->left) == NULL)
632 break;
633 }
634 /* Link into the new root's right tree. */
635 righttreemin->left = root;
636 righttreemin = root;
637 } else if (address >= root->end) {
638 if ((y = root->right) == NULL)
639 break;
640 if (address >= y->end) {
641 /* Rotate left. */
642 root->right = y->left;
643 y->left = root;
644 root = y;
645 if ((y = root->right) == NULL)
646 break;
647 }
648 /* Link into the new root's left tree. */
649 lefttreemax->right = root;
650 lefttreemax = root;
651 } else
652 break;
653 }
654 /* Assemble the new root. */
655 lefttreemax->right = root->left;
656 righttreemin->left = root->right;
657 root->left = dummy.right;
658 root->right = dummy.left;
659 return (root);
660 }
661
662 /*
663 * vm_map_entry_{un,}link:
664 *
665 * Insert/remove entries from maps.
666 */
667 static void
668 vm_map_entry_link(vm_map_t map,
669 vm_map_entry_t after_where,
670 vm_map_entry_t entry)
671 {
672
673 CTR4(KTR_VM,
674 "vm_map_entry_link: map %p, nentries %d, entry %p, after %p", map,
675 map->nentries, entry, after_where);
676 map->nentries++;
677 entry->prev = after_where;
678 entry->next = after_where->next;
679 entry->next->prev = entry;
680 after_where->next = entry;
681
682 if (after_where != &map->header) {
683 if (after_where != map->root)
684 vm_map_entry_splay(after_where->start, map->root);
685 entry->right = after_where->right;
686 entry->left = after_where;
687 after_where->right = NULL;
688 } else {
689 entry->right = map->root;
690 entry->left = NULL;
691 }
692 map->root = entry;
693 }
694
695 static void
696 vm_map_entry_unlink(vm_map_t map,
697 vm_map_entry_t entry)
698 {
699 vm_map_entry_t next, prev, root;
700
701 if (entry != map->root)
702 vm_map_entry_splay(entry->start, map->root);
703 if (entry->left == NULL)
704 root = entry->right;
705 else {
706 root = vm_map_entry_splay(entry->start, entry->left);
707 root->right = entry->right;
708 }
709 map->root = root;
710
711 prev = entry->prev;
712 next = entry->next;
713 next->prev = prev;
714 prev->next = next;
715 map->nentries--;
716 CTR3(KTR_VM, "vm_map_entry_unlink: map %p, nentries %d, entry %p", map,
717 map->nentries, entry);
718 }
719
720 /*
721 * vm_map_lookup_entry: [ internal use only ]
722 *
723 * Finds the map entry containing (or
724 * immediately preceding) the specified address
725 * in the given map; the entry is returned
726 * in the "entry" parameter. The boolean
727 * result indicates whether the address is
728 * actually contained in the map.
729 */
730 boolean_t
731 vm_map_lookup_entry(
732 vm_map_t map,
733 vm_offset_t address,
734 vm_map_entry_t *entry) /* OUT */
735 {
736 vm_map_entry_t cur;
737
738 cur = vm_map_entry_splay(address, map->root);
739 if (cur == NULL)
740 *entry = &map->header;
741 else {
742 map->root = cur;
743
744 if (address >= cur->start) {
745 *entry = cur;
746 if (cur->end > address)
747 return (TRUE);
748 } else
749 *entry = cur->prev;
750 }
751 return (FALSE);
752 }
753
754 /*
755 * vm_map_insert:
756 *
757 * Inserts the given whole VM object into the target
758 * map at the specified address range. The object's
759 * size should match that of the address range.
760 *
761 * Requires that the map be locked, and leaves it so.
762 *
763 * If object is non-NULL, ref count must be bumped by caller
764 * prior to making call to account for the new entry.
765 */
766 int
767 vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
768 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max,
769 int cow)
770 {
771 vm_map_entry_t new_entry;
772 vm_map_entry_t prev_entry;
773 vm_map_entry_t temp_entry;
774 vm_eflags_t protoeflags;
775
776 /*
777 * Check that the start and end points are not bogus.
778 */
779 if ((start < map->min_offset) || (end > map->max_offset) ||
780 (start >= end))
781 return (KERN_INVALID_ADDRESS);
782
783 /*
784 * Find the entry prior to the proposed starting address; if it's part
785 * of an existing entry, this range is bogus.
786 */
787 if (vm_map_lookup_entry(map, start, &temp_entry))
788 return (KERN_NO_SPACE);
789
790 prev_entry = temp_entry;
791
792 /*
793 * Assert that the next entry doesn't overlap the end point.
794 */
795 if ((prev_entry->next != &map->header) &&
796 (prev_entry->next->start < end))
797 return (KERN_NO_SPACE);
798
799 protoeflags = 0;
800
801 if (cow & MAP_COPY_ON_WRITE)
802 protoeflags |= MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY;
803
804 if (cow & MAP_NOFAULT) {
805 protoeflags |= MAP_ENTRY_NOFAULT;
806
807 KASSERT(object == NULL,
808 ("vm_map_insert: paradoxical MAP_NOFAULT request"));
809 }
810 if (cow & MAP_DISABLE_SYNCER)
811 protoeflags |= MAP_ENTRY_NOSYNC;
812 if (cow & MAP_DISABLE_COREDUMP)
813 protoeflags |= MAP_ENTRY_NOCOREDUMP;
814
815 if (object != NULL) {
816 /*
817 * OBJ_ONEMAPPING must be cleared unless this mapping
818 * is trivially proven to be the only mapping for any
819 * of the object's pages. (Object granularity
820 * reference counting is insufficient to recognize
821 * aliases with precision.)
822 */
823 if (object != kmem_object)
824 mtx_lock(&Giant);
825 VM_OBJECT_LOCK(object);
826 if (object->ref_count > 1 || object->shadow_count != 0)
827 vm_object_clear_flag(object, OBJ_ONEMAPPING);
828 VM_OBJECT_UNLOCK(object);
829 if (object != kmem_object)
830 mtx_unlock(&Giant);
831 }
832 else if ((prev_entry != &map->header) &&
833 (prev_entry->eflags == protoeflags) &&
834 (prev_entry->end == start) &&
835 (prev_entry->wired_count == 0) &&
836 ((prev_entry->object.vm_object == NULL) ||
837 vm_object_coalesce(prev_entry->object.vm_object,
838 OFF_TO_IDX(prev_entry->offset),
839 (vm_size_t)(prev_entry->end - prev_entry->start),
840 (vm_size_t)(end - prev_entry->end)))) {
841 /*
842 * We were able to extend the object. Determine if we
843 * can extend the previous map entry to include the
844 * new range as well.
845 */
846 if ((prev_entry->inheritance == VM_INHERIT_DEFAULT) &&
847 (prev_entry->protection == prot) &&
848 (prev_entry->max_protection == max)) {
849 map->size += (end - prev_entry->end);
850 prev_entry->end = end;
851 vm_map_simplify_entry(map, prev_entry);
852 return (KERN_SUCCESS);
853 }
854
855 /*
856 * If we can extend the object but cannot extend the
857 * map entry, we have to create a new map entry. We
858 * must bump the ref count on the extended object to
859 * account for it. object may be NULL.
860 */
861 object = prev_entry->object.vm_object;
862 offset = prev_entry->offset +
863 (prev_entry->end - prev_entry->start);
864 vm_object_reference(object);
865 }
866
867 /*
868 * NOTE: if conditionals fail, object can be NULL here. This occurs
869 * in things like the buffer map where we manage kva but do not manage
870 * backing objects.
871 */
872
873 /*
874 * Create a new entry
875 */
876 new_entry = vm_map_entry_create(map);
877 new_entry->start = start;
878 new_entry->end = end;
879
880 new_entry->eflags = protoeflags;
881 new_entry->object.vm_object = object;
882 new_entry->offset = offset;
883 new_entry->avail_ssize = 0;
884
885 new_entry->inheritance = VM_INHERIT_DEFAULT;
886 new_entry->protection = prot;
887 new_entry->max_protection = max;
888 new_entry->wired_count = 0;
889
890 /*
891 * Insert the new entry into the list
892 */
893 vm_map_entry_link(map, prev_entry, new_entry);
894 map->size += new_entry->end - new_entry->start;
895
896 /*
897 * Update the free space hint
898 */
899 if ((map->first_free == prev_entry) &&
900 (prev_entry->end >= new_entry->start)) {
901 map->first_free = new_entry;
902 }
903
904 #if 0
905 /*
906 * Temporarily removed to avoid MAP_STACK panic, due to
907 * MAP_STACK being a huge hack. Will be added back in
908 * when MAP_STACK (and the user stack mapping) is fixed.
909 */
910 /*
911 * It may be possible to simplify the entry
912 */
913 vm_map_simplify_entry(map, new_entry);
914 #endif
915
916 if (cow & (MAP_PREFAULT|MAP_PREFAULT_PARTIAL)) {
917 mtx_lock(&Giant);
918 pmap_object_init_pt(map->pmap, start,
919 object, OFF_TO_IDX(offset), end - start,
920 cow & MAP_PREFAULT_PARTIAL);
921 mtx_unlock(&Giant);
922 }
923
924 return (KERN_SUCCESS);
925 }
926
927 /*
928 * Find sufficient space for `length' bytes in the given map, starting at
929 * `start'. The map must be locked. Returns 0 on success, 1 on no space.
930 */
931 int
932 vm_map_findspace(
933 vm_map_t map,
934 vm_offset_t start,
935 vm_size_t length,
936 vm_offset_t *addr)
937 {
938 vm_map_entry_t entry, next;
939 vm_offset_t end;
940
941 if (start < map->min_offset)
942 start = map->min_offset;
943 if (start > map->max_offset)
944 return (1);
945
946 /*
947 * Look for the first possible address; if there's already something
948 * at this address, we have to start after it.
949 */
950 if (start == map->min_offset) {
951 if ((entry = map->first_free) != &map->header)
952 start = entry->end;
953 } else {
954 vm_map_entry_t tmp;
955
956 if (vm_map_lookup_entry(map, start, &tmp))
957 start = tmp->end;
958 entry = tmp;
959 }
960
961 /*
962 * Look through the rest of the map, trying to fit a new region in the
963 * gap between existing regions, or after the very last region.
964 */
965 for (;; start = (entry = next)->end) {
966 /*
967 * Find the end of the proposed new region. Be sure we didn't
968 * go beyond the end of the map, or wrap around the address;
969 * if so, we lose. Otherwise, if this is the last entry, or
970 * if the proposed new region fits before the next entry, we
971 * win.
972 */
973 end = start + length;
974 if (end > map->max_offset || end < start)
975 return (1);
976 next = entry->next;
977 if (next == &map->header || next->start >= end)
978 break;
979 }
980 *addr = start;
981 if (map == kernel_map) {
982 vm_offset_t ksize;
983 if ((ksize = round_page(start + length)) > kernel_vm_end) {
984 pmap_growkernel(ksize);
985 }
986 }
987 return (0);
988 }
989
990 /*
991 * vm_map_find finds an unallocated region in the target address
992 * map with the given length. The search is defined to be
993 * first-fit from the specified address; the region found is
994 * returned in the same parameter.
995 *
996 * If object is non-NULL, ref count must be bumped by caller
997 * prior to making call to account for the new entry.
998 */
999 int
1000 vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
1001 vm_offset_t *addr, /* IN/OUT */
1002 vm_size_t length, boolean_t find_space, vm_prot_t prot,
1003 vm_prot_t max, int cow)
1004 {
1005 vm_offset_t start;
1006 int result, s = 0;
1007
1008 start = *addr;
1009
1010 if (map == kmem_map)
1011 s = splvm();
1012
1013 vm_map_lock(map);
1014 if (find_space) {
1015 if (vm_map_findspace(map, start, length, addr)) {
1016 vm_map_unlock(map);
1017 if (map == kmem_map)
1018 splx(s);
1019 return (KERN_NO_SPACE);
1020 }
1021 start = *addr;
1022 }
1023 result = vm_map_insert(map, object, offset,
1024 start, start + length, prot, max, cow);
1025 vm_map_unlock(map);
1026
1027 if (map == kmem_map)
1028 splx(s);
1029
1030 return (result);
1031 }
1032
1033 /*
1034 * vm_map_simplify_entry:
1035 *
1036 * Simplify the given map entry by merging with either neighbor. This
1037 * routine also has the ability to merge with both neighbors.
1038 *
1039 * The map must be locked.
1040 *
1041 * This routine guarentees that the passed entry remains valid (though
1042 * possibly extended). When merging, this routine may delete one or
1043 * both neighbors.
1044 */
1045 void
1046 vm_map_simplify_entry(vm_map_t map, vm_map_entry_t entry)
1047 {
1048 vm_map_entry_t next, prev;
1049 vm_size_t prevsize, esize;
1050
1051 if (entry->eflags & (MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP))
1052 return;
1053
1054 prev = entry->prev;
1055 if (prev != &map->header) {
1056 prevsize = prev->end - prev->start;
1057 if ( (prev->end == entry->start) &&
1058 (prev->object.vm_object == entry->object.vm_object) &&
1059 (!prev->object.vm_object ||
1060 (prev->offset + prevsize == entry->offset)) &&
1061 (prev->eflags == entry->eflags) &&
1062 (prev->protection == entry->protection) &&
1063 (prev->max_protection == entry->max_protection) &&
1064 (prev->inheritance == entry->inheritance) &&
1065 (prev->wired_count == entry->wired_count)) {
1066 if (map->first_free == prev)
1067 map->first_free = entry;
1068 vm_map_entry_unlink(map, prev);
1069 entry->start = prev->start;
1070 entry->offset = prev->offset;
1071 if (prev->object.vm_object)
1072 vm_object_deallocate(prev->object.vm_object);
1073 vm_map_entry_dispose(map, prev);
1074 }
1075 }
1076
1077 next = entry->next;
1078 if (next != &map->header) {
1079 esize = entry->end - entry->start;
1080 if ((entry->end == next->start) &&
1081 (next->object.vm_object == entry->object.vm_object) &&
1082 (!entry->object.vm_object ||
1083 (entry->offset + esize == next->offset)) &&
1084 (next->eflags == entry->eflags) &&
1085 (next->protection == entry->protection) &&
1086 (next->max_protection == entry->max_protection) &&
1087 (next->inheritance == entry->inheritance) &&
1088 (next->wired_count == entry->wired_count)) {
1089 if (map->first_free == next)
1090 map->first_free = entry;
1091 vm_map_entry_unlink(map, next);
1092 entry->end = next->end;
1093 if (next->object.vm_object)
1094 vm_object_deallocate(next->object.vm_object);
1095 vm_map_entry_dispose(map, next);
1096 }
1097 }
1098 }
1099 /*
1100 * vm_map_clip_start: [ internal use only ]
1101 *
1102 * Asserts that the given entry begins at or after
1103 * the specified address; if necessary,
1104 * it splits the entry into two.
1105 */
1106 #define vm_map_clip_start(map, entry, startaddr) \
1107 { \
1108 if (startaddr > entry->start) \
1109 _vm_map_clip_start(map, entry, startaddr); \
1110 }
1111
1112 /*
1113 * This routine is called only when it is known that
1114 * the entry must be split.
1115 */
1116 static void
1117 _vm_map_clip_start(vm_map_t map, vm_map_entry_t entry, vm_offset_t start)
1118 {
1119 vm_map_entry_t new_entry;
1120
1121 /*
1122 * Split off the front portion -- note that we must insert the new
1123 * entry BEFORE this one, so that this entry has the specified
1124 * starting address.
1125 */
1126 vm_map_simplify_entry(map, entry);
1127
1128 /*
1129 * If there is no object backing this entry, we might as well create
1130 * one now. If we defer it, an object can get created after the map
1131 * is clipped, and individual objects will be created for the split-up
1132 * map. This is a bit of a hack, but is also about the best place to
1133 * put this improvement.
1134 */
1135 if (entry->object.vm_object == NULL && !map->system_map) {
1136 vm_object_t object;
1137 object = vm_object_allocate(OBJT_DEFAULT,
1138 atop(entry->end - entry->start));
1139 entry->object.vm_object = object;
1140 entry->offset = 0;
1141 }
1142
1143 new_entry = vm_map_entry_create(map);
1144 *new_entry = *entry;
1145
1146 new_entry->end = start;
1147 entry->offset += (start - entry->start);
1148 entry->start = start;
1149
1150 vm_map_entry_link(map, entry->prev, new_entry);
1151
1152 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1153 vm_object_reference(new_entry->object.vm_object);
1154 }
1155 }
1156
1157 /*
1158 * vm_map_clip_end: [ internal use only ]
1159 *
1160 * Asserts that the given entry ends at or before
1161 * the specified address; if necessary,
1162 * it splits the entry into two.
1163 */
1164 #define vm_map_clip_end(map, entry, endaddr) \
1165 { \
1166 if ((endaddr) < (entry->end)) \
1167 _vm_map_clip_end((map), (entry), (endaddr)); \
1168 }
1169
1170 /*
1171 * This routine is called only when it is known that
1172 * the entry must be split.
1173 */
1174 static void
1175 _vm_map_clip_end(vm_map_t map, vm_map_entry_t entry, vm_offset_t end)
1176 {
1177 vm_map_entry_t new_entry;
1178
1179 /*
1180 * If there is no object backing this entry, we might as well create
1181 * one now. If we defer it, an object can get created after the map
1182 * is clipped, and individual objects will be created for the split-up
1183 * map. This is a bit of a hack, but is also about the best place to
1184 * put this improvement.
1185 */
1186 if (entry->object.vm_object == NULL && !map->system_map) {
1187 vm_object_t object;
1188 object = vm_object_allocate(OBJT_DEFAULT,
1189 atop(entry->end - entry->start));
1190 entry->object.vm_object = object;
1191 entry->offset = 0;
1192 }
1193
1194 /*
1195 * Create a new entry and insert it AFTER the specified entry
1196 */
1197 new_entry = vm_map_entry_create(map);
1198 *new_entry = *entry;
1199
1200 new_entry->start = entry->end = end;
1201 new_entry->offset += (end - entry->start);
1202
1203 vm_map_entry_link(map, entry, new_entry);
1204
1205 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1206 vm_object_reference(new_entry->object.vm_object);
1207 }
1208 }
1209
1210 /*
1211 * VM_MAP_RANGE_CHECK: [ internal use only ]
1212 *
1213 * Asserts that the starting and ending region
1214 * addresses fall within the valid range of the map.
1215 */
1216 #define VM_MAP_RANGE_CHECK(map, start, end) \
1217 { \
1218 if (start < vm_map_min(map)) \
1219 start = vm_map_min(map); \
1220 if (end > vm_map_max(map)) \
1221 end = vm_map_max(map); \
1222 if (start > end) \
1223 start = end; \
1224 }
1225
1226 /*
1227 * vm_map_submap: [ kernel use only ]
1228 *
1229 * Mark the given range as handled by a subordinate map.
1230 *
1231 * This range must have been created with vm_map_find,
1232 * and no other operations may have been performed on this
1233 * range prior to calling vm_map_submap.
1234 *
1235 * Only a limited number of operations can be performed
1236 * within this rage after calling vm_map_submap:
1237 * vm_fault
1238 * [Don't try vm_map_copy!]
1239 *
1240 * To remove a submapping, one must first remove the
1241 * range from the superior map, and then destroy the
1242 * submap (if desired). [Better yet, don't try it.]
1243 */
1244 int
1245 vm_map_submap(
1246 vm_map_t map,
1247 vm_offset_t start,
1248 vm_offset_t end,
1249 vm_map_t submap)
1250 {
1251 vm_map_entry_t entry;
1252 int result = KERN_INVALID_ARGUMENT;
1253
1254 vm_map_lock(map);
1255
1256 VM_MAP_RANGE_CHECK(map, start, end);
1257
1258 if (vm_map_lookup_entry(map, start, &entry)) {
1259 vm_map_clip_start(map, entry, start);
1260 } else
1261 entry = entry->next;
1262
1263 vm_map_clip_end(map, entry, end);
1264
1265 if ((entry->start == start) && (entry->end == end) &&
1266 ((entry->eflags & MAP_ENTRY_COW) == 0) &&
1267 (entry->object.vm_object == NULL)) {
1268 entry->object.sub_map = submap;
1269 entry->eflags |= MAP_ENTRY_IS_SUB_MAP;
1270 result = KERN_SUCCESS;
1271 }
1272 vm_map_unlock(map);
1273
1274 return (result);
1275 }
1276
1277 /*
1278 * vm_map_protect:
1279 *
1280 * Sets the protection of the specified address
1281 * region in the target map. If "set_max" is
1282 * specified, the maximum protection is to be set;
1283 * otherwise, only the current protection is affected.
1284 */
1285 int
1286 vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end,
1287 vm_prot_t new_prot, boolean_t set_max)
1288 {
1289 vm_map_entry_t current;
1290 vm_map_entry_t entry;
1291
1292 vm_map_lock(map);
1293
1294 VM_MAP_RANGE_CHECK(map, start, end);
1295
1296 if (vm_map_lookup_entry(map, start, &entry)) {
1297 vm_map_clip_start(map, entry, start);
1298 } else {
1299 entry = entry->next;
1300 }
1301
1302 /*
1303 * Make a first pass to check for protection violations.
1304 */
1305 current = entry;
1306 while ((current != &map->header) && (current->start < end)) {
1307 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
1308 vm_map_unlock(map);
1309 return (KERN_INVALID_ARGUMENT);
1310 }
1311 if ((new_prot & current->max_protection) != new_prot) {
1312 vm_map_unlock(map);
1313 return (KERN_PROTECTION_FAILURE);
1314 }
1315 current = current->next;
1316 }
1317
1318 /*
1319 * Go back and fix up protections. [Note that clipping is not
1320 * necessary the second time.]
1321 */
1322 current = entry;
1323 while ((current != &map->header) && (current->start < end)) {
1324 vm_prot_t old_prot;
1325
1326 vm_map_clip_end(map, current, end);
1327
1328 old_prot = current->protection;
1329 if (set_max)
1330 current->protection =
1331 (current->max_protection = new_prot) &
1332 old_prot;
1333 else
1334 current->protection = new_prot;
1335
1336 /*
1337 * Update physical map if necessary. Worry about copy-on-write
1338 * here -- CHECK THIS XXX
1339 */
1340 if (current->protection != old_prot) {
1341 mtx_lock(&Giant);
1342 vm_page_lock_queues();
1343 #define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \
1344 VM_PROT_ALL)
1345 pmap_protect(map->pmap, current->start,
1346 current->end,
1347 current->protection & MASK(current));
1348 #undef MASK
1349 vm_page_unlock_queues();
1350 mtx_unlock(&Giant);
1351 }
1352 vm_map_simplify_entry(map, current);
1353 current = current->next;
1354 }
1355 vm_map_unlock(map);
1356 return (KERN_SUCCESS);
1357 }
1358
1359 /*
1360 * vm_map_madvise:
1361 *
1362 * This routine traverses a processes map handling the madvise
1363 * system call. Advisories are classified as either those effecting
1364 * the vm_map_entry structure, or those effecting the underlying
1365 * objects.
1366 */
1367 int
1368 vm_map_madvise(
1369 vm_map_t map,
1370 vm_offset_t start,
1371 vm_offset_t end,
1372 int behav)
1373 {
1374 vm_map_entry_t current, entry;
1375 int modify_map = 0;
1376
1377 /*
1378 * Some madvise calls directly modify the vm_map_entry, in which case
1379 * we need to use an exclusive lock on the map and we need to perform
1380 * various clipping operations. Otherwise we only need a read-lock
1381 * on the map.
1382 */
1383 switch(behav) {
1384 case MADV_NORMAL:
1385 case MADV_SEQUENTIAL:
1386 case MADV_RANDOM:
1387 case MADV_NOSYNC:
1388 case MADV_AUTOSYNC:
1389 case MADV_NOCORE:
1390 case MADV_CORE:
1391 modify_map = 1;
1392 vm_map_lock(map);
1393 break;
1394 case MADV_WILLNEED:
1395 case MADV_DONTNEED:
1396 case MADV_FREE:
1397 vm_map_lock_read(map);
1398 break;
1399 default:
1400 return (KERN_INVALID_ARGUMENT);
1401 }
1402
1403 /*
1404 * Locate starting entry and clip if necessary.
1405 */
1406 VM_MAP_RANGE_CHECK(map, start, end);
1407
1408 if (vm_map_lookup_entry(map, start, &entry)) {
1409 if (modify_map)
1410 vm_map_clip_start(map, entry, start);
1411 } else {
1412 entry = entry->next;
1413 }
1414
1415 if (modify_map) {
1416 /*
1417 * madvise behaviors that are implemented in the vm_map_entry.
1418 *
1419 * We clip the vm_map_entry so that behavioral changes are
1420 * limited to the specified address range.
1421 */
1422 for (current = entry;
1423 (current != &map->header) && (current->start < end);
1424 current = current->next
1425 ) {
1426 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
1427 continue;
1428
1429 vm_map_clip_end(map, current, end);
1430
1431 switch (behav) {
1432 case MADV_NORMAL:
1433 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL);
1434 break;
1435 case MADV_SEQUENTIAL:
1436 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL);
1437 break;
1438 case MADV_RANDOM:
1439 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM);
1440 break;
1441 case MADV_NOSYNC:
1442 current->eflags |= MAP_ENTRY_NOSYNC;
1443 break;
1444 case MADV_AUTOSYNC:
1445 current->eflags &= ~MAP_ENTRY_NOSYNC;
1446 break;
1447 case MADV_NOCORE:
1448 current->eflags |= MAP_ENTRY_NOCOREDUMP;
1449 break;
1450 case MADV_CORE:
1451 current->eflags &= ~MAP_ENTRY_NOCOREDUMP;
1452 break;
1453 default:
1454 break;
1455 }
1456 vm_map_simplify_entry(map, current);
1457 }
1458 vm_map_unlock(map);
1459 } else {
1460 vm_pindex_t pindex;
1461 int count;
1462
1463 /*
1464 * madvise behaviors that are implemented in the underlying
1465 * vm_object.
1466 *
1467 * Since we don't clip the vm_map_entry, we have to clip
1468 * the vm_object pindex and count.
1469 */
1470 for (current = entry;
1471 (current != &map->header) && (current->start < end);
1472 current = current->next
1473 ) {
1474 vm_offset_t useStart;
1475
1476 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
1477 continue;
1478
1479 pindex = OFF_TO_IDX(current->offset);
1480 count = atop(current->end - current->start);
1481 useStart = current->start;
1482
1483 if (current->start < start) {
1484 pindex += atop(start - current->start);
1485 count -= atop(start - current->start);
1486 useStart = start;
1487 }
1488 if (current->end > end)
1489 count -= atop(current->end - end);
1490
1491 if (count <= 0)
1492 continue;
1493
1494 vm_object_madvise(current->object.vm_object,
1495 pindex, count, behav);
1496 if (behav == MADV_WILLNEED) {
1497 mtx_lock(&Giant);
1498 pmap_object_init_pt(
1499 map->pmap,
1500 useStart,
1501 current->object.vm_object,
1502 pindex,
1503 (count << PAGE_SHIFT),
1504 MAP_PREFAULT_MADVISE
1505 );
1506 mtx_unlock(&Giant);
1507 }
1508 }
1509 vm_map_unlock_read(map);
1510 }
1511 return (0);
1512 }
1513
1514
1515 /*
1516 * vm_map_inherit:
1517 *
1518 * Sets the inheritance of the specified address
1519 * range in the target map. Inheritance
1520 * affects how the map will be shared with
1521 * child maps at the time of vm_map_fork.
1522 */
1523 int
1524 vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end,
1525 vm_inherit_t new_inheritance)
1526 {
1527 vm_map_entry_t entry;
1528 vm_map_entry_t temp_entry;
1529
1530 switch (new_inheritance) {
1531 case VM_INHERIT_NONE:
1532 case VM_INHERIT_COPY:
1533 case VM_INHERIT_SHARE:
1534 break;
1535 default:
1536 return (KERN_INVALID_ARGUMENT);
1537 }
1538 vm_map_lock(map);
1539 VM_MAP_RANGE_CHECK(map, start, end);
1540 if (vm_map_lookup_entry(map, start, &temp_entry)) {
1541 entry = temp_entry;
1542 vm_map_clip_start(map, entry, start);
1543 } else
1544 entry = temp_entry->next;
1545 while ((entry != &map->header) && (entry->start < end)) {
1546 vm_map_clip_end(map, entry, end);
1547 entry->inheritance = new_inheritance;
1548 vm_map_simplify_entry(map, entry);
1549 entry = entry->next;
1550 }
1551 vm_map_unlock(map);
1552 return (KERN_SUCCESS);
1553 }
1554
1555 /*
1556 * vm_map_unwire:
1557 *
1558 * Implements both kernel and user unwiring.
1559 */
1560 int
1561 vm_map_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end,
1562 boolean_t user_unwire)
1563 {
1564 vm_map_entry_t entry, first_entry, tmp_entry;
1565 vm_offset_t saved_start;
1566 unsigned int last_timestamp;
1567 int rv;
1568 boolean_t need_wakeup, result;
1569
1570 vm_map_lock(map);
1571 VM_MAP_RANGE_CHECK(map, start, end);
1572 if (!vm_map_lookup_entry(map, start, &first_entry)) {
1573 vm_map_unlock(map);
1574 return (KERN_INVALID_ADDRESS);
1575 }
1576 last_timestamp = map->timestamp;
1577 entry = first_entry;
1578 while (entry != &map->header && entry->start < end) {
1579 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
1580 /*
1581 * We have not yet clipped the entry.
1582 */
1583 saved_start = (start >= entry->start) ? start :
1584 entry->start;
1585 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
1586 if (vm_map_unlock_and_wait(map, user_unwire)) {
1587 /*
1588 * Allow interruption of user unwiring?
1589 */
1590 }
1591 vm_map_lock(map);
1592 if (last_timestamp+1 != map->timestamp) {
1593 /*
1594 * Look again for the entry because the map was
1595 * modified while it was unlocked.
1596 * Specifically, the entry may have been
1597 * clipped, merged, or deleted.
1598 */
1599 if (!vm_map_lookup_entry(map, saved_start,
1600 &tmp_entry)) {
1601 if (saved_start == start) {
1602 /*
1603 * First_entry has been deleted.
1604 */
1605 vm_map_unlock(map);
1606 return (KERN_INVALID_ADDRESS);
1607 }
1608 end = saved_start;
1609 rv = KERN_INVALID_ADDRESS;
1610 goto done;
1611 }
1612 if (entry == first_entry)
1613 first_entry = tmp_entry;
1614 else
1615 first_entry = NULL;
1616 entry = tmp_entry;
1617 }
1618 last_timestamp = map->timestamp;
1619 continue;
1620 }
1621 vm_map_clip_start(map, entry, start);
1622 vm_map_clip_end(map, entry, end);
1623 /*
1624 * Mark the entry in case the map lock is released. (See
1625 * above.)
1626 */
1627 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
1628 /*
1629 * Check the map for holes in the specified region.
1630 */
1631 if (entry->end < end && (entry->next == &map->header ||
1632 entry->next->start > entry->end)) {
1633 end = entry->end;
1634 rv = KERN_INVALID_ADDRESS;
1635 goto done;
1636 }
1637 /*
1638 * Require that the entry is wired.
1639 */
1640 if (entry->wired_count == 0 || (user_unwire &&
1641 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0)) {
1642 end = entry->end;
1643 rv = KERN_INVALID_ARGUMENT;
1644 goto done;
1645 }
1646 entry = entry->next;
1647 }
1648 rv = KERN_SUCCESS;
1649 done:
1650 need_wakeup = FALSE;
1651 if (first_entry == NULL) {
1652 result = vm_map_lookup_entry(map, start, &first_entry);
1653 KASSERT(result, ("vm_map_unwire: lookup failed"));
1654 }
1655 entry = first_entry;
1656 while (entry != &map->header && entry->start < end) {
1657 if (rv == KERN_SUCCESS) {
1658 if (user_unwire)
1659 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
1660 entry->wired_count--;
1661 if (entry->wired_count == 0) {
1662 /*
1663 * Retain the map lock.
1664 */
1665 vm_fault_unwire(map, entry->start, entry->end);
1666 }
1667 }
1668 KASSERT(entry->eflags & MAP_ENTRY_IN_TRANSITION,
1669 ("vm_map_unwire: in-transition flag missing"));
1670 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
1671 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
1672 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
1673 need_wakeup = TRUE;
1674 }
1675 vm_map_simplify_entry(map, entry);
1676 entry = entry->next;
1677 }
1678 vm_map_unlock(map);
1679 if (need_wakeup)
1680 vm_map_wakeup(map);
1681 return (rv);
1682 }
1683
1684 /*
1685 * vm_map_wire:
1686 *
1687 * Implements both kernel and user wiring.
1688 */
1689 int
1690 vm_map_wire(vm_map_t map, vm_offset_t start, vm_offset_t end,
1691 boolean_t user_wire)
1692 {
1693 vm_map_entry_t entry, first_entry, tmp_entry;
1694 vm_offset_t saved_end, saved_start;
1695 unsigned int last_timestamp;
1696 int rv;
1697 boolean_t need_wakeup, result;
1698
1699 vm_map_lock(map);
1700 VM_MAP_RANGE_CHECK(map, start, end);
1701 if (!vm_map_lookup_entry(map, start, &first_entry)) {
1702 vm_map_unlock(map);
1703 return (KERN_INVALID_ADDRESS);
1704 }
1705 last_timestamp = map->timestamp;
1706 entry = first_entry;
1707 while (entry != &map->header && entry->start < end) {
1708 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
1709 /*
1710 * We have not yet clipped the entry.
1711 */
1712 saved_start = (start >= entry->start) ? start :
1713 entry->start;
1714 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
1715 if (vm_map_unlock_and_wait(map, user_wire)) {
1716 /*
1717 * Allow interruption of user wiring?
1718 */
1719 }
1720 vm_map_lock(map);
1721 if (last_timestamp + 1 != map->timestamp) {
1722 /*
1723 * Look again for the entry because the map was
1724 * modified while it was unlocked.
1725 * Specifically, the entry may have been
1726 * clipped, merged, or deleted.
1727 */
1728 if (!vm_map_lookup_entry(map, saved_start,
1729 &tmp_entry)) {
1730 if (saved_start == start) {
1731 /*
1732 * first_entry has been deleted.
1733 */
1734 vm_map_unlock(map);
1735 return (KERN_INVALID_ADDRESS);
1736 }
1737 end = saved_start;
1738 rv = KERN_INVALID_ADDRESS;
1739 goto done;
1740 }
1741 if (entry == first_entry)
1742 first_entry = tmp_entry;
1743 else
1744 first_entry = NULL;
1745 entry = tmp_entry;
1746 }
1747 last_timestamp = map->timestamp;
1748 continue;
1749 }
1750 vm_map_clip_start(map, entry, start);
1751 vm_map_clip_end(map, entry, end);
1752 /*
1753 * Mark the entry in case the map lock is released. (See
1754 * above.)
1755 */
1756 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
1757 /*
1758 *
1759 */
1760 if (entry->wired_count == 0) {
1761 entry->wired_count++;
1762 saved_start = entry->start;
1763 saved_end = entry->end;
1764 /*
1765 * Release the map lock, relying on the in-transition
1766 * mark.
1767 */
1768 vm_map_unlock(map);
1769 rv = vm_fault_wire(map, saved_start, saved_end,
1770 user_wire);
1771 vm_map_lock(map);
1772 if (last_timestamp + 1 != map->timestamp) {
1773 /*
1774 * Look again for the entry because the map was
1775 * modified while it was unlocked. The entry
1776 * may have been clipped, but NOT merged or
1777 * deleted.
1778 */
1779 result = vm_map_lookup_entry(map, saved_start,
1780 &tmp_entry);
1781 KASSERT(result, ("vm_map_wire: lookup failed"));
1782 if (entry == first_entry)
1783 first_entry = tmp_entry;
1784 else
1785 first_entry = NULL;
1786 entry = tmp_entry;
1787 while (entry->end < saved_end) {
1788 if (rv != KERN_SUCCESS) {
1789 KASSERT(entry->wired_count == 1,
1790 ("vm_map_wire: bad count"));
1791 entry->wired_count = -1;
1792 }
1793 entry = entry->next;
1794 }
1795 }
1796 last_timestamp = map->timestamp;
1797 if (rv != KERN_SUCCESS) {
1798 KASSERT(entry->wired_count == 1,
1799 ("vm_map_wire: bad count"));
1800 /*
1801 * Assign an out-of-range value to represent
1802 * the failure to wire this entry.
1803 */
1804 entry->wired_count = -1;
1805 end = entry->end;
1806 goto done;
1807 }
1808 } else if (!user_wire ||
1809 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
1810 entry->wired_count++;
1811 }
1812 /*
1813 * Check the map for holes in the specified region.
1814 */
1815 if (entry->end < end && (entry->next == &map->header ||
1816 entry->next->start > entry->end)) {
1817 end = entry->end;
1818 rv = KERN_INVALID_ADDRESS;
1819 goto done;
1820 }
1821 entry = entry->next;
1822 }
1823 rv = KERN_SUCCESS;
1824 done:
1825 need_wakeup = FALSE;
1826 if (first_entry == NULL) {
1827 result = vm_map_lookup_entry(map, start, &first_entry);
1828 KASSERT(result, ("vm_map_wire: lookup failed"));
1829 }
1830 entry = first_entry;
1831 while (entry != &map->header && entry->start < end) {
1832 if (rv == KERN_SUCCESS) {
1833 if (user_wire)
1834 entry->eflags |= MAP_ENTRY_USER_WIRED;
1835 } else if (entry->wired_count == -1) {
1836 /*
1837 * Wiring failed on this entry. Thus, unwiring is
1838 * unnecessary.
1839 */
1840 entry->wired_count = 0;
1841 } else {
1842 if (!user_wire ||
1843 (entry->eflags & MAP_ENTRY_USER_WIRED) == 0)
1844 entry->wired_count--;
1845 if (entry->wired_count == 0) {
1846 /*
1847 * Retain the map lock.
1848 */
1849 vm_fault_unwire(map, entry->start, entry->end);
1850 }
1851 }
1852 KASSERT(entry->eflags & MAP_ENTRY_IN_TRANSITION,
1853 ("vm_map_wire: in-transition flag missing"));
1854 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
1855 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
1856 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
1857 need_wakeup = TRUE;
1858 }
1859 vm_map_simplify_entry(map, entry);
1860 entry = entry->next;
1861 }
1862 vm_map_unlock(map);
1863 if (need_wakeup)
1864 vm_map_wakeup(map);
1865 return (rv);
1866 }
1867
1868 /*
1869 * vm_map_clean
1870 *
1871 * Push any dirty cached pages in the address range to their pager.
1872 * If syncio is TRUE, dirty pages are written synchronously.
1873 * If invalidate is TRUE, any cached pages are freed as well.
1874 *
1875 * Returns an error if any part of the specified range is not mapped.
1876 */
1877 int
1878 vm_map_clean(
1879 vm_map_t map,
1880 vm_offset_t start,
1881 vm_offset_t end,
1882 boolean_t syncio,
1883 boolean_t invalidate)
1884 {
1885 vm_map_entry_t current;
1886 vm_map_entry_t entry;
1887 vm_size_t size;
1888 vm_object_t object;
1889 vm_ooffset_t offset;
1890
1891 GIANT_REQUIRED;
1892
1893 vm_map_lock_read(map);
1894 VM_MAP_RANGE_CHECK(map, start, end);
1895 if (!vm_map_lookup_entry(map, start, &entry)) {
1896 vm_map_unlock_read(map);
1897 return (KERN_INVALID_ADDRESS);
1898 }
1899 /*
1900 * Make a first pass to check for holes.
1901 */
1902 for (current = entry; current->start < end; current = current->next) {
1903 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
1904 vm_map_unlock_read(map);
1905 return (KERN_INVALID_ARGUMENT);
1906 }
1907 if (end > current->end &&
1908 (current->next == &map->header ||
1909 current->end != current->next->start)) {
1910 vm_map_unlock_read(map);
1911 return (KERN_INVALID_ADDRESS);
1912 }
1913 }
1914
1915 if (invalidate) {
1916 vm_page_lock_queues();
1917 pmap_remove(map->pmap, start, end);
1918 vm_page_unlock_queues();
1919 }
1920 /*
1921 * Make a second pass, cleaning/uncaching pages from the indicated
1922 * objects as we go.
1923 */
1924 for (current = entry; current->start < end; current = current->next) {
1925 offset = current->offset + (start - current->start);
1926 size = (end <= current->end ? end : current->end) - start;
1927 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
1928 vm_map_t smap;
1929 vm_map_entry_t tentry;
1930 vm_size_t tsize;
1931
1932 smap = current->object.sub_map;
1933 vm_map_lock_read(smap);
1934 (void) vm_map_lookup_entry(smap, offset, &tentry);
1935 tsize = tentry->end - offset;
1936 if (tsize < size)
1937 size = tsize;
1938 object = tentry->object.vm_object;
1939 offset = tentry->offset + (offset - tentry->start);
1940 vm_map_unlock_read(smap);
1941 } else {
1942 object = current->object.vm_object;
1943 }
1944 /*
1945 * Note that there is absolutely no sense in writing out
1946 * anonymous objects, so we track down the vnode object
1947 * to write out.
1948 * We invalidate (remove) all pages from the address space
1949 * anyway, for semantic correctness.
1950 *
1951 * note: certain anonymous maps, such as MAP_NOSYNC maps,
1952 * may start out with a NULL object.
1953 */
1954 while (object && object->backing_object) {
1955 object = object->backing_object;
1956 offset += object->backing_object_offset;
1957 if (object->size < OFF_TO_IDX(offset + size))
1958 size = IDX_TO_OFF(object->size) - offset;
1959 }
1960 if (object && (object->type == OBJT_VNODE) &&
1961 (current->protection & VM_PROT_WRITE)) {
1962 /*
1963 * Flush pages if writing is allowed, invalidate them
1964 * if invalidation requested. Pages undergoing I/O
1965 * will be ignored by vm_object_page_remove().
1966 *
1967 * We cannot lock the vnode and then wait for paging
1968 * to complete without deadlocking against vm_fault.
1969 * Instead we simply call vm_object_page_remove() and
1970 * allow it to block internally on a page-by-page
1971 * basis when it encounters pages undergoing async
1972 * I/O.
1973 */
1974 int flags;
1975
1976 vm_object_reference(object);
1977 vn_lock(object->handle, LK_EXCLUSIVE | LK_RETRY, curthread);
1978 flags = (syncio || invalidate) ? OBJPC_SYNC : 0;
1979 flags |= invalidate ? OBJPC_INVAL : 0;
1980 VM_OBJECT_LOCK(object);
1981 vm_object_page_clean(object,
1982 OFF_TO_IDX(offset),
1983 OFF_TO_IDX(offset + size + PAGE_MASK),
1984 flags);
1985 VM_OBJECT_UNLOCK(object);
1986 VOP_UNLOCK(object->handle, 0, curthread);
1987 vm_object_deallocate(object);
1988 }
1989 if (object && invalidate &&
1990 ((object->type == OBJT_VNODE) ||
1991 (object->type == OBJT_DEVICE))) {
1992 VM_OBJECT_LOCK(object);
1993 vm_object_page_remove(object,
1994 OFF_TO_IDX(offset),
1995 OFF_TO_IDX(offset + size + PAGE_MASK),
1996 FALSE);
1997 VM_OBJECT_UNLOCK(object);
1998 }
1999 start += size;
2000 }
2001
2002 vm_map_unlock_read(map);
2003 return (KERN_SUCCESS);
2004 }
2005
2006 /*
2007 * vm_map_entry_unwire: [ internal use only ]
2008 *
2009 * Make the region specified by this entry pageable.
2010 *
2011 * The map in question should be locked.
2012 * [This is the reason for this routine's existence.]
2013 */
2014 static void
2015 vm_map_entry_unwire(vm_map_t map, vm_map_entry_t entry)
2016 {
2017 vm_fault_unwire(map, entry->start, entry->end);
2018 entry->wired_count = 0;
2019 }
2020
2021 /*
2022 * vm_map_entry_delete: [ internal use only ]
2023 *
2024 * Deallocate the given entry from the target map.
2025 */
2026 static void
2027 vm_map_entry_delete(vm_map_t map, vm_map_entry_t entry)
2028 {
2029 vm_map_entry_unlink(map, entry);
2030 map->size -= entry->end - entry->start;
2031
2032 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
2033 vm_object_deallocate(entry->object.vm_object);
2034 }
2035
2036 vm_map_entry_dispose(map, entry);
2037 }
2038
2039 /*
2040 * vm_map_delete: [ internal use only ]
2041 *
2042 * Deallocates the given address range from the target
2043 * map.
2044 */
2045 int
2046 vm_map_delete(vm_map_t map, vm_offset_t start, vm_offset_t end)
2047 {
2048 vm_object_t object;
2049 vm_map_entry_t entry;
2050 vm_map_entry_t first_entry;
2051
2052 /*
2053 * Find the start of the region, and clip it
2054 */
2055 if (!vm_map_lookup_entry(map, start, &first_entry))
2056 entry = first_entry->next;
2057 else {
2058 entry = first_entry;
2059 vm_map_clip_start(map, entry, start);
2060 }
2061
2062 /*
2063 * Save the free space hint
2064 */
2065 if (entry == &map->header) {
2066 map->first_free = &map->header;
2067 } else if (map->first_free->start >= start) {
2068 map->first_free = entry->prev;
2069 }
2070
2071 /*
2072 * Step through all entries in this region
2073 */
2074 while ((entry != &map->header) && (entry->start < end)) {
2075 vm_map_entry_t next;
2076 vm_offset_t s, e;
2077 vm_pindex_t offidxstart, offidxend, count;
2078
2079 /*
2080 * Wait for wiring or unwiring of an entry to complete.
2081 */
2082 if ((entry->eflags & MAP_ENTRY_IN_TRANSITION) != 0) {
2083 unsigned int last_timestamp;
2084 vm_offset_t saved_start;
2085 vm_map_entry_t tmp_entry;
2086
2087 saved_start = entry->start;
2088 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2089 last_timestamp = map->timestamp;
2090 (void) vm_map_unlock_and_wait(map, FALSE);
2091 vm_map_lock(map);
2092 if (last_timestamp + 1 != map->timestamp) {
2093 /*
2094 * Look again for the entry because the map was
2095 * modified while it was unlocked.
2096 * Specifically, the entry may have been
2097 * clipped, merged, or deleted.
2098 */
2099 if (!vm_map_lookup_entry(map, saved_start,
2100 &tmp_entry))
2101 entry = tmp_entry->next;
2102 else {
2103 entry = tmp_entry;
2104 vm_map_clip_start(map, entry,
2105 saved_start);
2106 }
2107 }
2108 continue;
2109 }
2110 vm_map_clip_end(map, entry, end);
2111
2112 s = entry->start;
2113 e = entry->end;
2114 next = entry->next;
2115
2116 offidxstart = OFF_TO_IDX(entry->offset);
2117 count = OFF_TO_IDX(e - s);
2118 object = entry->object.vm_object;
2119
2120 /*
2121 * Unwire before removing addresses from the pmap; otherwise,
2122 * unwiring will put the entries back in the pmap.
2123 */
2124 if (entry->wired_count != 0) {
2125 vm_map_entry_unwire(map, entry);
2126 }
2127
2128 offidxend = offidxstart + count;
2129
2130 if (object == kernel_object || object == kmem_object) {
2131 if (object == kernel_object)
2132 GIANT_REQUIRED;
2133 VM_OBJECT_LOCK(object);
2134 vm_object_page_remove(object, offidxstart, offidxend, FALSE);
2135 VM_OBJECT_UNLOCK(object);
2136 } else {
2137 mtx_lock(&Giant);
2138 vm_page_lock_queues();
2139 pmap_remove(map->pmap, s, e);
2140 vm_page_unlock_queues();
2141 if (object != NULL) {
2142 VM_OBJECT_LOCK(object);
2143 if (object->ref_count != 1 &&
2144 (object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING &&
2145 (object->type == OBJT_DEFAULT || object->type == OBJT_SWAP)) {
2146 VM_OBJECT_UNLOCK(object);
2147 vm_object_collapse(object);
2148 VM_OBJECT_LOCK(object);
2149 vm_object_page_remove(object, offidxstart, offidxend, FALSE);
2150 if (object->type == OBJT_SWAP)
2151 swap_pager_freespace(object, offidxstart, count);
2152 if (offidxend >= object->size &&
2153 offidxstart < object->size)
2154 object->size = offidxstart;
2155 }
2156 VM_OBJECT_UNLOCK(object);
2157 }
2158 mtx_unlock(&Giant);
2159 }
2160
2161 /*
2162 * Delete the entry (which may delete the object) only after
2163 * removing all pmap entries pointing to its pages.
2164 * (Otherwise, its page frames may be reallocated, and any
2165 * modify bits will be set in the wrong object!)
2166 */
2167 vm_map_entry_delete(map, entry);
2168 entry = next;
2169 }
2170 return (KERN_SUCCESS);
2171 }
2172
2173 /*
2174 * vm_map_remove:
2175 *
2176 * Remove the given address range from the target map.
2177 * This is the exported form of vm_map_delete.
2178 */
2179 int
2180 vm_map_remove(vm_map_t map, vm_offset_t start, vm_offset_t end)
2181 {
2182 int result, s = 0;
2183
2184 if (map == kmem_map)
2185 s = splvm();
2186
2187 vm_map_lock(map);
2188 VM_MAP_RANGE_CHECK(map, start, end);
2189 result = vm_map_delete(map, start, end);
2190 vm_map_unlock(map);
2191
2192 if (map == kmem_map)
2193 splx(s);
2194
2195 return (result);
2196 }
2197
2198 /*
2199 * vm_map_check_protection:
2200 *
2201 * Assert that the target map allows the specified privilege on the
2202 * entire address region given. The entire region must be allocated.
2203 *
2204 * WARNING! This code does not and should not check whether the
2205 * contents of the region is accessible. For example a smaller file
2206 * might be mapped into a larger address space.
2207 *
2208 * NOTE! This code is also called by munmap().
2209 */
2210 boolean_t
2211 vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end,
2212 vm_prot_t protection)
2213 {
2214 vm_map_entry_t entry;
2215 vm_map_entry_t tmp_entry;
2216
2217 vm_map_lock_read(map);
2218 if (!vm_map_lookup_entry(map, start, &tmp_entry)) {
2219 vm_map_unlock_read(map);
2220 return (FALSE);
2221 }
2222 entry = tmp_entry;
2223
2224 while (start < end) {
2225 if (entry == &map->header) {
2226 vm_map_unlock_read(map);
2227 return (FALSE);
2228 }
2229 /*
2230 * No holes allowed!
2231 */
2232 if (start < entry->start) {
2233 vm_map_unlock_read(map);
2234 return (FALSE);
2235 }
2236 /*
2237 * Check protection associated with entry.
2238 */
2239 if ((entry->protection & protection) != protection) {
2240 vm_map_unlock_read(map);
2241 return (FALSE);
2242 }
2243 /* go to next entry */
2244 start = entry->end;
2245 entry = entry->next;
2246 }
2247 vm_map_unlock_read(map);
2248 return (TRUE);
2249 }
2250
2251 /*
2252 * vm_map_copy_entry:
2253 *
2254 * Copies the contents of the source entry to the destination
2255 * entry. The entries *must* be aligned properly.
2256 */
2257 static void
2258 vm_map_copy_entry(
2259 vm_map_t src_map,
2260 vm_map_t dst_map,
2261 vm_map_entry_t src_entry,
2262 vm_map_entry_t dst_entry)
2263 {
2264 vm_object_t src_object;
2265
2266 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP)
2267 return;
2268
2269 if (src_entry->wired_count == 0) {
2270
2271 /*
2272 * If the source entry is marked needs_copy, it is already
2273 * write-protected.
2274 */
2275 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) {
2276 vm_page_lock_queues();
2277 pmap_protect(src_map->pmap,
2278 src_entry->start,
2279 src_entry->end,
2280 src_entry->protection & ~VM_PROT_WRITE);
2281 vm_page_unlock_queues();
2282 }
2283
2284 /*
2285 * Make a copy of the object.
2286 */
2287 if ((src_object = src_entry->object.vm_object) != NULL) {
2288
2289 if ((src_object->handle == NULL) &&
2290 (src_object->type == OBJT_DEFAULT ||
2291 src_object->type == OBJT_SWAP)) {
2292 vm_object_collapse(src_object);
2293 if ((src_object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) {
2294 vm_object_split(src_entry);
2295 src_object = src_entry->object.vm_object;
2296 }
2297 }
2298
2299 vm_object_reference(src_object);
2300 VM_OBJECT_LOCK(src_object);
2301 vm_object_clear_flag(src_object, OBJ_ONEMAPPING);
2302 VM_OBJECT_UNLOCK(src_object);
2303 dst_entry->object.vm_object = src_object;
2304 src_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
2305 dst_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
2306 dst_entry->offset = src_entry->offset;
2307 } else {
2308 dst_entry->object.vm_object = NULL;
2309 dst_entry->offset = 0;
2310 }
2311
2312 pmap_copy(dst_map->pmap, src_map->pmap, dst_entry->start,
2313 dst_entry->end - dst_entry->start, src_entry->start);
2314 } else {
2315 /*
2316 * Of course, wired down pages can't be set copy-on-write.
2317 * Cause wired pages to be copied into the new map by
2318 * simulating faults (the new pages are pageable)
2319 */
2320 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry);
2321 }
2322 }
2323
2324 /*
2325 * vmspace_fork:
2326 * Create a new process vmspace structure and vm_map
2327 * based on those of an existing process. The new map
2328 * is based on the old map, according to the inheritance
2329 * values on the regions in that map.
2330 *
2331 * The source map must not be locked.
2332 */
2333 struct vmspace *
2334 vmspace_fork(struct vmspace *vm1)
2335 {
2336 struct vmspace *vm2;
2337 vm_map_t old_map = &vm1->vm_map;
2338 vm_map_t new_map;
2339 vm_map_entry_t old_entry;
2340 vm_map_entry_t new_entry;
2341 vm_object_t object;
2342
2343 GIANT_REQUIRED;
2344
2345 vm_map_lock(old_map);
2346 old_map->infork = 1;
2347
2348 vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset);
2349 bcopy(&vm1->vm_startcopy, &vm2->vm_startcopy,
2350 (caddr_t) &vm1->vm_endcopy - (caddr_t) &vm1->vm_startcopy);
2351 new_map = &vm2->vm_map; /* XXX */
2352 new_map->timestamp = 1;
2353
2354 old_entry = old_map->header.next;
2355
2356 while (old_entry != &old_map->header) {
2357 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP)
2358 panic("vm_map_fork: encountered a submap");
2359
2360 switch (old_entry->inheritance) {
2361 case VM_INHERIT_NONE:
2362 break;
2363
2364 case VM_INHERIT_SHARE:
2365 /*
2366 * Clone the entry, creating the shared object if necessary.
2367 */
2368 object = old_entry->object.vm_object;
2369 if (object == NULL) {
2370 object = vm_object_allocate(OBJT_DEFAULT,
2371 atop(old_entry->end - old_entry->start));
2372 old_entry->object.vm_object = object;
2373 old_entry->offset = (vm_offset_t) 0;
2374 }
2375
2376 /*
2377 * Add the reference before calling vm_object_shadow
2378 * to insure that a shadow object is created.
2379 */
2380 vm_object_reference(object);
2381 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) {
2382 vm_object_shadow(&old_entry->object.vm_object,
2383 &old_entry->offset,
2384 atop(old_entry->end - old_entry->start));
2385 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
2386 /* Transfer the second reference too. */
2387 vm_object_reference(
2388 old_entry->object.vm_object);
2389 vm_object_deallocate(object);
2390 object = old_entry->object.vm_object;
2391 }
2392 VM_OBJECT_LOCK(object);
2393 vm_object_clear_flag(object, OBJ_ONEMAPPING);
2394 VM_OBJECT_UNLOCK(object);
2395
2396 /*
2397 * Clone the entry, referencing the shared object.
2398 */
2399 new_entry = vm_map_entry_create(new_map);
2400 *new_entry = *old_entry;
2401 new_entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2402 new_entry->wired_count = 0;
2403
2404 /*
2405 * Insert the entry into the new map -- we know we're
2406 * inserting at the end of the new map.
2407 */
2408 vm_map_entry_link(new_map, new_map->header.prev,
2409 new_entry);
2410
2411 /*
2412 * Update the physical map
2413 */
2414 pmap_copy(new_map->pmap, old_map->pmap,
2415 new_entry->start,
2416 (old_entry->end - old_entry->start),
2417 old_entry->start);
2418 break;
2419
2420 case VM_INHERIT_COPY:
2421 /*
2422 * Clone the entry and link into the map.
2423 */
2424 new_entry = vm_map_entry_create(new_map);
2425 *new_entry = *old_entry;
2426 new_entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2427 new_entry->wired_count = 0;
2428 new_entry->object.vm_object = NULL;
2429 vm_map_entry_link(new_map, new_map->header.prev,
2430 new_entry);
2431 vm_map_copy_entry(old_map, new_map, old_entry,
2432 new_entry);
2433 break;
2434 }
2435 old_entry = old_entry->next;
2436 }
2437
2438 new_map->size = old_map->size;
2439 old_map->infork = 0;
2440 vm_map_unlock(old_map);
2441
2442 return (vm2);
2443 }
2444
2445 int
2446 vm_map_stack (vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
2447 vm_prot_t prot, vm_prot_t max, int cow)
2448 {
2449 vm_map_entry_t prev_entry;
2450 vm_map_entry_t new_stack_entry;
2451 vm_size_t init_ssize;
2452 int rv;
2453
2454 if (addrbos < vm_map_min(map))
2455 return (KERN_NO_SPACE);
2456
2457 if (max_ssize < sgrowsiz)
2458 init_ssize = max_ssize;
2459 else
2460 init_ssize = sgrowsiz;
2461
2462 vm_map_lock(map);
2463
2464 /* If addr is already mapped, no go */
2465 if (vm_map_lookup_entry(map, addrbos, &prev_entry)) {
2466 vm_map_unlock(map);
2467 return (KERN_NO_SPACE);
2468 }
2469
2470 /* If we would blow our VMEM resource limit, no go */
2471 if (map->size + init_ssize >
2472 curthread->td_proc->p_rlimit[RLIMIT_VMEM].rlim_cur) {
2473 vm_map_unlock(map);
2474 return (KERN_NO_SPACE);
2475 }
2476
2477 /* If we can't accomodate max_ssize in the current mapping,
2478 * no go. However, we need to be aware that subsequent user
2479 * mappings might map into the space we have reserved for
2480 * stack, and currently this space is not protected.
2481 *
2482 * Hopefully we will at least detect this condition
2483 * when we try to grow the stack.
2484 */
2485 if ((prev_entry->next != &map->header) &&
2486 (prev_entry->next->start < addrbos + max_ssize)) {
2487 vm_map_unlock(map);
2488 return (KERN_NO_SPACE);
2489 }
2490
2491 /* We initially map a stack of only init_ssize. We will
2492 * grow as needed later. Since this is to be a grow
2493 * down stack, we map at the top of the range.
2494 *
2495 * Note: we would normally expect prot and max to be
2496 * VM_PROT_ALL, and cow to be 0. Possibly we should
2497 * eliminate these as input parameters, and just
2498 * pass these values here in the insert call.
2499 */
2500 rv = vm_map_insert(map, NULL, 0, addrbos + max_ssize - init_ssize,
2501 addrbos + max_ssize, prot, max, cow);
2502
2503 /* Now set the avail_ssize amount */
2504 if (rv == KERN_SUCCESS){
2505 if (prev_entry != &map->header)
2506 vm_map_clip_end(map, prev_entry, addrbos + max_ssize - init_ssize);
2507 new_stack_entry = prev_entry->next;
2508 if (new_stack_entry->end != addrbos + max_ssize ||
2509 new_stack_entry->start != addrbos + max_ssize - init_ssize)
2510 panic ("Bad entry start/end for new stack entry");
2511 else
2512 new_stack_entry->avail_ssize = max_ssize - init_ssize;
2513 }
2514
2515 vm_map_unlock(map);
2516 return (rv);
2517 }
2518
2519 /* Attempts to grow a vm stack entry. Returns KERN_SUCCESS if the
2520 * desired address is already mapped, or if we successfully grow
2521 * the stack. Also returns KERN_SUCCESS if addr is outside the
2522 * stack range (this is strange, but preserves compatibility with
2523 * the grow function in vm_machdep.c).
2524 */
2525 int
2526 vm_map_growstack (struct proc *p, vm_offset_t addr)
2527 {
2528 vm_map_entry_t prev_entry;
2529 vm_map_entry_t stack_entry;
2530 vm_map_entry_t new_stack_entry;
2531 struct vmspace *vm = p->p_vmspace;
2532 vm_map_t map = &vm->vm_map;
2533 vm_offset_t end;
2534 int grow_amount;
2535 int rv;
2536 int is_procstack;
2537
2538 GIANT_REQUIRED;
2539
2540 Retry:
2541 vm_map_lock_read(map);
2542
2543 /* If addr is already in the entry range, no need to grow.*/
2544 if (vm_map_lookup_entry(map, addr, &prev_entry)) {
2545 vm_map_unlock_read(map);
2546 return (KERN_SUCCESS);
2547 }
2548
2549 if ((stack_entry = prev_entry->next) == &map->header) {
2550 vm_map_unlock_read(map);
2551 return (KERN_SUCCESS);
2552 }
2553 if (prev_entry == &map->header)
2554 end = stack_entry->start - stack_entry->avail_ssize;
2555 else
2556 end = prev_entry->end;
2557
2558 /* This next test mimics the old grow function in vm_machdep.c.
2559 * It really doesn't quite make sense, but we do it anyway
2560 * for compatibility.
2561 *
2562 * If not growable stack, return success. This signals the
2563 * caller to proceed as he would normally with normal vm.
2564 */
2565 if (stack_entry->avail_ssize < 1 ||
2566 addr >= stack_entry->start ||
2567 addr < stack_entry->start - stack_entry->avail_ssize) {
2568 vm_map_unlock_read(map);
2569 return (KERN_SUCCESS);
2570 }
2571
2572 /* Find the minimum grow amount */
2573 grow_amount = roundup (stack_entry->start - addr, PAGE_SIZE);
2574 if (grow_amount > stack_entry->avail_ssize) {
2575 vm_map_unlock_read(map);
2576 return (KERN_NO_SPACE);
2577 }
2578
2579 /* If there is no longer enough space between the entries
2580 * nogo, and adjust the available space. Note: this
2581 * should only happen if the user has mapped into the
2582 * stack area after the stack was created, and is
2583 * probably an error.
2584 *
2585 * This also effectively destroys any guard page the user
2586 * might have intended by limiting the stack size.
2587 */
2588 if (grow_amount > stack_entry->start - end) {
2589 if (vm_map_lock_upgrade(map))
2590 goto Retry;
2591
2592 stack_entry->avail_ssize = stack_entry->start - end;
2593
2594 vm_map_unlock(map);
2595 return (KERN_NO_SPACE);
2596 }
2597
2598 is_procstack = addr >= (vm_offset_t)vm->vm_maxsaddr;
2599
2600 /* If this is the main process stack, see if we're over the
2601 * stack limit.
2602 */
2603 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount >
2604 p->p_rlimit[RLIMIT_STACK].rlim_cur)) {
2605 vm_map_unlock_read(map);
2606 return (KERN_NO_SPACE);
2607 }
2608
2609 /* Round up the grow amount modulo SGROWSIZ */
2610 grow_amount = roundup (grow_amount, sgrowsiz);
2611 if (grow_amount > stack_entry->avail_ssize) {
2612 grow_amount = stack_entry->avail_ssize;
2613 }
2614 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount >
2615 p->p_rlimit[RLIMIT_STACK].rlim_cur)) {
2616 grow_amount = p->p_rlimit[RLIMIT_STACK].rlim_cur -
2617 ctob(vm->vm_ssize);
2618 }
2619
2620 /* If we would blow our VMEM resource limit, no go */
2621 if (map->size + grow_amount >
2622 curthread->td_proc->p_rlimit[RLIMIT_VMEM].rlim_cur) {
2623 vm_map_unlock_read(map);
2624 return (KERN_NO_SPACE);
2625 }
2626
2627 if (vm_map_lock_upgrade(map))
2628 goto Retry;
2629
2630 /* Get the preliminary new entry start value */
2631 addr = stack_entry->start - grow_amount;
2632
2633 /* If this puts us into the previous entry, cut back our growth
2634 * to the available space. Also, see the note above.
2635 */
2636 if (addr < end) {
2637 stack_entry->avail_ssize = stack_entry->start - end;
2638 addr = end;
2639 }
2640
2641 rv = vm_map_insert(map, NULL, 0, addr, stack_entry->start,
2642 p->p_sysent->sv_stackprot, VM_PROT_ALL, 0);
2643
2644 /* Adjust the available stack space by the amount we grew. */
2645 if (rv == KERN_SUCCESS) {
2646 if (prev_entry != &map->header)
2647 vm_map_clip_end(map, prev_entry, addr);
2648 new_stack_entry = prev_entry->next;
2649 if (new_stack_entry->end != stack_entry->start ||
2650 new_stack_entry->start != addr)
2651 panic ("Bad stack grow start/end in new stack entry");
2652 else {
2653 new_stack_entry->avail_ssize = stack_entry->avail_ssize -
2654 (new_stack_entry->end -
2655 new_stack_entry->start);
2656 if (is_procstack)
2657 vm->vm_ssize += btoc(new_stack_entry->end -
2658 new_stack_entry->start);
2659 }
2660 }
2661
2662 vm_map_unlock(map);
2663 return (rv);
2664 }
2665
2666 /*
2667 * Unshare the specified VM space for exec. If other processes are
2668 * mapped to it, then create a new one. The new vmspace is null.
2669 */
2670 void
2671 vmspace_exec(struct proc *p, vm_offset_t minuser, vm_offset_t maxuser)
2672 {
2673 struct vmspace *oldvmspace = p->p_vmspace;
2674 struct vmspace *newvmspace;
2675
2676 GIANT_REQUIRED;
2677 newvmspace = vmspace_alloc(minuser, maxuser);
2678 bcopy(&oldvmspace->vm_startcopy, &newvmspace->vm_startcopy,
2679 (caddr_t) (newvmspace + 1) - (caddr_t) &newvmspace->vm_startcopy);
2680 /*
2681 * This code is written like this for prototype purposes. The
2682 * goal is to avoid running down the vmspace here, but let the
2683 * other process's that are still using the vmspace to finally
2684 * run it down. Even though there is little or no chance of blocking
2685 * here, it is a good idea to keep this form for future mods.
2686 */
2687 p->p_vmspace = newvmspace;
2688 pmap_pinit2(vmspace_pmap(newvmspace));
2689 vmspace_free(oldvmspace);
2690 if (p == curthread->td_proc) /* XXXKSE ? */
2691 pmap_activate(curthread);
2692 }
2693
2694 /*
2695 * Unshare the specified VM space for forcing COW. This
2696 * is called by rfork, for the (RFMEM|RFPROC) == 0 case.
2697 */
2698 void
2699 vmspace_unshare(struct proc *p)
2700 {
2701 struct vmspace *oldvmspace = p->p_vmspace;
2702 struct vmspace *newvmspace;
2703
2704 GIANT_REQUIRED;
2705 if (oldvmspace->vm_refcnt == 1)
2706 return;
2707 newvmspace = vmspace_fork(oldvmspace);
2708 p->p_vmspace = newvmspace;
2709 pmap_pinit2(vmspace_pmap(newvmspace));
2710 vmspace_free(oldvmspace);
2711 if (p == curthread->td_proc) /* XXXKSE ? */
2712 pmap_activate(curthread);
2713 }
2714
2715 /*
2716 * vm_map_lookup:
2717 *
2718 * Finds the VM object, offset, and
2719 * protection for a given virtual address in the
2720 * specified map, assuming a page fault of the
2721 * type specified.
2722 *
2723 * Leaves the map in question locked for read; return
2724 * values are guaranteed until a vm_map_lookup_done
2725 * call is performed. Note that the map argument
2726 * is in/out; the returned map must be used in
2727 * the call to vm_map_lookup_done.
2728 *
2729 * A handle (out_entry) is returned for use in
2730 * vm_map_lookup_done, to make that fast.
2731 *
2732 * If a lookup is requested with "write protection"
2733 * specified, the map may be changed to perform virtual
2734 * copying operations, although the data referenced will
2735 * remain the same.
2736 */
2737 int
2738 vm_map_lookup(vm_map_t *var_map, /* IN/OUT */
2739 vm_offset_t vaddr,
2740 vm_prot_t fault_typea,
2741 vm_map_entry_t *out_entry, /* OUT */
2742 vm_object_t *object, /* OUT */
2743 vm_pindex_t *pindex, /* OUT */
2744 vm_prot_t *out_prot, /* OUT */
2745 boolean_t *wired) /* OUT */
2746 {
2747 vm_map_entry_t entry;
2748 vm_map_t map = *var_map;
2749 vm_prot_t prot;
2750 vm_prot_t fault_type = fault_typea;
2751
2752 RetryLookup:;
2753 /*
2754 * Lookup the faulting address.
2755 */
2756
2757 vm_map_lock_read(map);
2758 #define RETURN(why) \
2759 { \
2760 vm_map_unlock_read(map); \
2761 return (why); \
2762 }
2763
2764 /*
2765 * If the map has an interesting hint, try it before calling full
2766 * blown lookup routine.
2767 */
2768 entry = map->root;
2769 *out_entry = entry;
2770 if (entry == NULL ||
2771 (vaddr < entry->start) || (vaddr >= entry->end)) {
2772 /*
2773 * Entry was either not a valid hint, or the vaddr was not
2774 * contained in the entry, so do a full lookup.
2775 */
2776 if (!vm_map_lookup_entry(map, vaddr, out_entry))
2777 RETURN(KERN_INVALID_ADDRESS);
2778
2779 entry = *out_entry;
2780 }
2781
2782 /*
2783 * Handle submaps.
2784 */
2785 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
2786 vm_map_t old_map = map;
2787
2788 *var_map = map = entry->object.sub_map;
2789 vm_map_unlock_read(old_map);
2790 goto RetryLookup;
2791 }
2792
2793 /*
2794 * Check whether this task is allowed to have this page.
2795 * Note the special case for MAP_ENTRY_COW
2796 * pages with an override. This is to implement a forced
2797 * COW for debuggers.
2798 */
2799 if (fault_type & VM_PROT_OVERRIDE_WRITE)
2800 prot = entry->max_protection;
2801 else
2802 prot = entry->protection;
2803 fault_type &= (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE);
2804 if ((fault_type & prot) != fault_type) {
2805 RETURN(KERN_PROTECTION_FAILURE);
2806 }
2807 if ((entry->eflags & MAP_ENTRY_USER_WIRED) &&
2808 (entry->eflags & MAP_ENTRY_COW) &&
2809 (fault_type & VM_PROT_WRITE) &&
2810 (fault_typea & VM_PROT_OVERRIDE_WRITE) == 0) {
2811 RETURN(KERN_PROTECTION_FAILURE);
2812 }
2813
2814 /*
2815 * If this page is not pageable, we have to get it for all possible
2816 * accesses.
2817 */
2818 *wired = (entry->wired_count != 0);
2819 if (*wired)
2820 prot = fault_type = entry->protection;
2821
2822 /*
2823 * If the entry was copy-on-write, we either ...
2824 */
2825 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
2826 /*
2827 * If we want to write the page, we may as well handle that
2828 * now since we've got the map locked.
2829 *
2830 * If we don't need to write the page, we just demote the
2831 * permissions allowed.
2832 */
2833 if (fault_type & VM_PROT_WRITE) {
2834 /*
2835 * Make a new object, and place it in the object
2836 * chain. Note that no new references have appeared
2837 * -- one just moved from the map to the new
2838 * object.
2839 */
2840 if (vm_map_lock_upgrade(map))
2841 goto RetryLookup;
2842
2843 vm_object_shadow(
2844 &entry->object.vm_object,
2845 &entry->offset,
2846 atop(entry->end - entry->start));
2847 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
2848
2849 vm_map_lock_downgrade(map);
2850 } else {
2851 /*
2852 * We're attempting to read a copy-on-write page --
2853 * don't allow writes.
2854 */
2855 prot &= ~VM_PROT_WRITE;
2856 }
2857 }
2858
2859 /*
2860 * Create an object if necessary.
2861 */
2862 if (entry->object.vm_object == NULL &&
2863 !map->system_map) {
2864 if (vm_map_lock_upgrade(map))
2865 goto RetryLookup;
2866 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT,
2867 atop(entry->end - entry->start));
2868 entry->offset = 0;
2869 vm_map_lock_downgrade(map);
2870 }
2871
2872 /*
2873 * Return the object/offset from this entry. If the entry was
2874 * copy-on-write or empty, it has been fixed up.
2875 */
2876 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
2877 *object = entry->object.vm_object;
2878
2879 /*
2880 * Return whether this is the only map sharing this data.
2881 */
2882 *out_prot = prot;
2883 return (KERN_SUCCESS);
2884
2885 #undef RETURN
2886 }
2887
2888 /*
2889 * vm_map_lookup_done:
2890 *
2891 * Releases locks acquired by a vm_map_lookup
2892 * (according to the handle returned by that lookup).
2893 */
2894 void
2895 vm_map_lookup_done(vm_map_t map, vm_map_entry_t entry)
2896 {
2897 /*
2898 * Unlock the main-level map
2899 */
2900 vm_map_unlock_read(map);
2901 }
2902
2903 #include "opt_ddb.h"
2904 #ifdef DDB
2905 #include <sys/kernel.h>
2906
2907 #include <ddb/ddb.h>
2908
2909 /*
2910 * vm_map_print: [ debug ]
2911 */
2912 DB_SHOW_COMMAND(map, vm_map_print)
2913 {
2914 static int nlines;
2915 /* XXX convert args. */
2916 vm_map_t map = (vm_map_t)addr;
2917 boolean_t full = have_addr;
2918
2919 vm_map_entry_t entry;
2920
2921 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n",
2922 (void *)map,
2923 (void *)map->pmap, map->nentries, map->timestamp);
2924 nlines++;
2925
2926 if (!full && db_indent)
2927 return;
2928
2929 db_indent += 2;
2930 for (entry = map->header.next; entry != &map->header;
2931 entry = entry->next) {
2932 db_iprintf("map entry %p: start=%p, end=%p\n",
2933 (void *)entry, (void *)entry->start, (void *)entry->end);
2934 nlines++;
2935 {
2936 static char *inheritance_name[4] =
2937 {"share", "copy", "none", "donate_copy"};
2938
2939 db_iprintf(" prot=%x/%x/%s",
2940 entry->protection,
2941 entry->max_protection,
2942 inheritance_name[(int)(unsigned char)entry->inheritance]);
2943 if (entry->wired_count != 0)
2944 db_printf(", wired");
2945 }
2946 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
2947 db_printf(", share=%p, offset=0x%jx\n",
2948 (void *)entry->object.sub_map,
2949 (uintmax_t)entry->offset);
2950 nlines++;
2951 if ((entry->prev == &map->header) ||
2952 (entry->prev->object.sub_map !=
2953 entry->object.sub_map)) {
2954 db_indent += 2;
2955 vm_map_print((db_expr_t)(intptr_t)
2956 entry->object.sub_map,
2957 full, 0, (char *)0);
2958 db_indent -= 2;
2959 }
2960 } else {
2961 db_printf(", object=%p, offset=0x%jx",
2962 (void *)entry->object.vm_object,
2963 (uintmax_t)entry->offset);
2964 if (entry->eflags & MAP_ENTRY_COW)
2965 db_printf(", copy (%s)",
2966 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done");
2967 db_printf("\n");
2968 nlines++;
2969
2970 if ((entry->prev == &map->header) ||
2971 (entry->prev->object.vm_object !=
2972 entry->object.vm_object)) {
2973 db_indent += 2;
2974 vm_object_print((db_expr_t)(intptr_t)
2975 entry->object.vm_object,
2976 full, 0, (char *)0);
2977 nlines += 4;
2978 db_indent -= 2;
2979 }
2980 }
2981 }
2982 db_indent -= 2;
2983 if (db_indent == 0)
2984 nlines = 0;
2985 }
2986
2987
2988 DB_SHOW_COMMAND(procvm, procvm)
2989 {
2990 struct proc *p;
2991
2992 if (have_addr) {
2993 p = (struct proc *) addr;
2994 } else {
2995 p = curproc;
2996 }
2997
2998 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n",
2999 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map,
3000 (void *)vmspace_pmap(p->p_vmspace));
3001
3002 vm_map_print((db_expr_t)(intptr_t)&p->p_vmspace->vm_map, 1, 0, NULL);
3003 }
3004
3005 #endif /* DDB */
Cache object: acb573b25fdc0608a2056c30b7a7c914
|