Add the 'eatmem' utility. eatmem is a memory stressor.
[dragonfly.git] / sys / vm / vm_map.c
CommitLineData
984263bc
MD
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: src/sys/vm/vm_map.c,v 1.187.2.19 2003/05/27 00:47:02 alc Exp $
e9bb90e8 65 * $DragonFly: src/sys/vm/vm_map.c,v 1.9 2003/08/25 17:01:13 dillon Exp $
984263bc
MD
66 */
67
68/*
69 * Virtual memory mapping module.
70 */
71
72#include <sys/param.h>
73#include <sys/systm.h>
74#include <sys/proc.h>
fef0fdf2 75#include <sys/lock.h>
984263bc
MD
76#include <sys/vmmeter.h>
77#include <sys/mman.h>
78#include <sys/vnode.h>
79#include <sys/resourcevar.h>
fef0fdf2 80#include <sys/shm.h>
984263bc
MD
81
82#include <vm/vm.h>
83#include <vm/vm_param.h>
984263bc
MD
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/swap_pager.h>
92#include <vm/vm_zone.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
1388df65
RG
142static void _vm_map_clip_end (vm_map_t, vm_map_entry_t, vm_offset_t);
143static void _vm_map_clip_start (vm_map_t, vm_map_entry_t, vm_offset_t);
144static vm_map_entry_t vm_map_entry_create (vm_map_t);
145static void vm_map_entry_delete (vm_map_t, vm_map_entry_t);
146static void vm_map_entry_dispose (vm_map_t, vm_map_entry_t);
147static void vm_map_entry_unwire (vm_map_t, vm_map_entry_t);
148static void vm_map_copy_entry (vm_map_t, vm_map_t, vm_map_entry_t,
149 vm_map_entry_t);
150static void vm_map_split (vm_map_entry_t);
151static void vm_map_unclip_range (vm_map_t map, vm_map_entry_t start_entry, vm_offset_t start, vm_offset_t end, int flags);
984263bc
MD
152
153void
154vm_map_startup()
155{
156 mapzone = &mapzone_store;
157 zbootinit(mapzone, "MAP", sizeof (struct vm_map),
158 map_init, MAX_KMAP);
159 kmapentzone = &kmapentzone_store;
160 zbootinit(kmapentzone, "KMAP ENTRY", sizeof (struct vm_map_entry),
161 kmap_entry_init, MAX_KMAPENT);
162 mapentzone = &mapentzone_store;
163 zbootinit(mapentzone, "MAP ENTRY", sizeof (struct vm_map_entry),
164 map_entry_init, MAX_MAPENT);
165}
166
167/*
168 * Allocate a vmspace structure, including a vm_map and pmap,
169 * and initialize those structures. The refcnt is set to 1.
170 * The remaining fields must be initialized by the caller.
171 */
172struct vmspace *
173vmspace_alloc(min, max)
174 vm_offset_t min, max;
175{
176 struct vmspace *vm;
177
178 vm = zalloc(vmspace_zone);
179 vm_map_init(&vm->vm_map, min, max);
180 pmap_pinit(vmspace_pmap(vm));
181 vm->vm_map.pmap = vmspace_pmap(vm); /* XXX */
182 vm->vm_refcnt = 1;
183 vm->vm_shm = NULL;
184 vm->vm_exitingcnt = 0;
185 return (vm);
186}
187
188void
189vm_init2(void) {
190 zinitna(kmapentzone, &kmapentobj,
191 NULL, 0, lmin((VM_MAX_KERNEL_ADDRESS - KERNBASE) / PAGE_SIZE,
12e4aaff 192 vmstats.v_page_count) / 8, ZONE_INTERRUPT, 1);
984263bc
MD
193 zinitna(mapentzone, &mapentobj,
194 NULL, 0, 0, 0, 1);
195 zinitna(mapzone, &mapobj,
196 NULL, 0, 0, 0, 1);
197 vmspace_zone = zinit("VMSPACE", sizeof (struct vmspace), 0, 0, 3);
198 pmap_init2();
199 vm_object_init2();
200}
201
202static __inline void
203vmspace_dofree(struct vmspace *vm)
204{
fef0fdf2
MD
205 /*
206 * Make sure any SysV shm is freed, it might not have in
207 * exit1()
208 */
209 shmexit(vm);
210
984263bc
MD
211 /*
212 * Lock the map, to wait out all other references to it.
213 * Delete all of the mappings and pages they hold, then call
214 * the pmap module to reclaim anything left.
215 */
216 vm_map_lock(&vm->vm_map);
217 (void) vm_map_delete(&vm->vm_map, vm->vm_map.min_offset,
218 vm->vm_map.max_offset);
219 vm_map_unlock(&vm->vm_map);
220
221 pmap_release(vmspace_pmap(vm));
222 zfree(vmspace_zone, vm);
223}
224
225void
226vmspace_free(struct vmspace *vm)
227{
228 if (vm->vm_refcnt == 0)
229 panic("vmspace_free: attempt to free already freed vmspace");
230
231 if (--vm->vm_refcnt == 0 && vm->vm_exitingcnt == 0)
232 vmspace_dofree(vm);
233}
234
235void
236vmspace_exitfree(struct proc *p)
237{
238 struct vmspace *vm;
239
240 vm = p->p_vmspace;
241 p->p_vmspace = NULL;
242
243 /*
244 * cleanup by parent process wait()ing on exiting child. vm_refcnt
245 * may not be 0 (e.g. fork() and child exits without exec()ing).
246 * exitingcnt may increment above 0 and drop back down to zero
247 * several times while vm_refcnt is held non-zero. vm_refcnt
248 * may also increment above 0 and drop back down to zero several
249 * times while vm_exitingcnt is held non-zero.
250 *
251 * The last wait on the exiting child's vmspace will clean up
252 * the remainder of the vmspace.
253 */
254 if (--vm->vm_exitingcnt == 0 && vm->vm_refcnt == 0)
255 vmspace_dofree(vm);
256}
257
258/*
259 * vmspace_swap_count() - count the approximate swap useage in pages for a
260 * vmspace.
261 *
262 * Swap useage is determined by taking the proportional swap used by
263 * VM objects backing the VM map. To make up for fractional losses,
264 * if the VM object has any swap use at all the associated map entries
265 * count for at least 1 swap page.
266 */
267int
268vmspace_swap_count(struct vmspace *vmspace)
269{
270 vm_map_t map = &vmspace->vm_map;
271 vm_map_entry_t cur;
272 int count = 0;
273
274 for (cur = map->header.next; cur != &map->header; cur = cur->next) {
275 vm_object_t object;
276
277 if ((cur->eflags & MAP_ENTRY_IS_SUB_MAP) == 0 &&
278 (object = cur->object.vm_object) != NULL &&
279 object->type == OBJT_SWAP
280 ) {
281 int n = (cur->end - cur->start) / PAGE_SIZE;
282
283 if (object->un_pager.swp.swp_bcount) {
284 count += object->un_pager.swp.swp_bcount *
285 SWAP_META_PAGES * n / object->size + 1;
286 }
287 }
288 }
289 return(count);
290}
291
292
293/*
294 * vm_map_create:
295 *
296 * Creates and returns a new empty VM map with
297 * the given physical map structure, and having
298 * the given lower and upper address bounds.
299 */
300vm_map_t
301vm_map_create(pmap, min, max)
302 pmap_t pmap;
303 vm_offset_t min, max;
304{
305 vm_map_t result;
306
307 result = zalloc(mapzone);
308 vm_map_init(result, min, max);
309 result->pmap = pmap;
310 return (result);
311}
312
313/*
314 * Initialize an existing vm_map structure
315 * such as that in the vmspace structure.
316 * The pmap is set elsewhere.
317 */
318void
319vm_map_init(map, min, max)
320 struct vm_map *map;
321 vm_offset_t min, max;
322{
323 map->header.next = map->header.prev = &map->header;
324 map->nentries = 0;
325 map->size = 0;
326 map->system_map = 0;
327 map->infork = 0;
328 map->min_offset = min;
329 map->max_offset = max;
330 map->first_free = &map->header;
331 map->hint = &map->header;
332 map->timestamp = 0;
377d4740 333 lockinit(&map->lock, 0, "thrd_sleep", 0, LK_NOPAUSE);
984263bc
MD
334}
335
336/*
8a8d5d85 337 * vm_map_entry_create: [ internal use only ]
984263bc 338 *
8a8d5d85
MD
339 * Allocates a VM map entry for insertion. No entry fields are filled
340 * in. this ruotine may be called from an interrupt.
984263bc 341 */
8a8d5d85
MD
342static vm_map_entry_t
343vm_map_entry_create(map)
984263bc 344 vm_map_t map;
984263bc 345{
8a8d5d85
MD
346 vm_map_entry_t new_entry;
347
984263bc 348 if (map->system_map || !mapentzone)
8a8d5d85 349 new_entry = zalloc(kmapentzone);
984263bc 350 else
8a8d5d85
MD
351 new_entry = zalloc(mapentzone);
352 if (new_entry == NULL)
353 panic("vm_map_entry_create: kernel resources exhausted");
354 return(new_entry);
984263bc
MD
355}
356
357/*
8a8d5d85 358 * vm_map_entry_dispose: [ internal use only ]
984263bc 359 *
8a8d5d85
MD
360 * Dispose of a vm_map_entry that is no longer being referenced. This
361 * function may be called from an interrupt.
984263bc 362 */
8a8d5d85
MD
363static void
364vm_map_entry_dispose(map, entry)
984263bc 365 vm_map_t map;
8a8d5d85 366 vm_map_entry_t entry;
984263bc 367{
984263bc 368 if (map->system_map || !mapentzone)
8a8d5d85 369 zfree(kmapentzone, entry);
984263bc 370 else
8a8d5d85 371 zfree(mapentzone, entry);
984263bc
MD
372}
373
8a8d5d85 374
984263bc
MD
375/*
376 * vm_map_entry_{un,}link:
377 *
378 * Insert/remove entries from maps.
379 */
380static __inline void
381vm_map_entry_link(vm_map_t map,
382 vm_map_entry_t after_where,
383 vm_map_entry_t entry)
384{
385 map->nentries++;
386 entry->prev = after_where;
387 entry->next = after_where->next;
388 entry->next->prev = entry;
389 after_where->next = entry;
390}
391
392static __inline void
393vm_map_entry_unlink(vm_map_t map,
394 vm_map_entry_t entry)
395{
396 vm_map_entry_t prev;
397 vm_map_entry_t next;
398
399 if (entry->eflags & MAP_ENTRY_IN_TRANSITION)
400 panic("vm_map_entry_unlink: attempt to mess with locked entry! %p", entry);
401 prev = entry->prev;
402 next = entry->next;
403 next->prev = prev;
404 prev->next = next;
405 map->nentries--;
406}
407
408/*
409 * SAVE_HINT:
410 *
411 * Saves the specified entry as the hint for
412 * future lookups.
413 */
414#define SAVE_HINT(map,value) \
415 (map)->hint = (value);
416
417/*
418 * vm_map_lookup_entry: [ internal use only ]
419 *
420 * Finds the map entry containing (or
421 * immediately preceding) the specified address
422 * in the given map; the entry is returned
423 * in the "entry" parameter. The boolean
424 * result indicates whether the address is
425 * actually contained in the map.
426 */
427boolean_t
428vm_map_lookup_entry(map, address, entry)
429 vm_map_t map;
430 vm_offset_t address;
431 vm_map_entry_t *entry; /* OUT */
432{
433 vm_map_entry_t cur;
434 vm_map_entry_t last;
435
436 /*
437 * Start looking either from the head of the list, or from the hint.
438 */
439
440 cur = map->hint;
441
442 if (cur == &map->header)
443 cur = cur->next;
444
445 if (address >= cur->start) {
446 /*
447 * Go from hint to end of list.
448 *
449 * But first, make a quick check to see if we are already looking
450 * at the entry we want (which is usually the case). Note also
451 * that we don't need to save the hint here... it is the same
452 * hint (unless we are at the header, in which case the hint
453 * didn't buy us anything anyway).
454 */
455 last = &map->header;
456 if ((cur != last) && (cur->end > address)) {
457 *entry = cur;
458 return (TRUE);
459 }
460 } else {
461 /*
462 * Go from start to hint, *inclusively*
463 */
464 last = cur->next;
465 cur = map->header.next;
466 }
467
468 /*
469 * Search linearly
470 */
471
472 while (cur != last) {
473 if (cur->end > address) {
474 if (address >= cur->start) {
475 /*
476 * Save this lookup for future hints, and
477 * return
478 */
479
480 *entry = cur;
481 SAVE_HINT(map, cur);
482 return (TRUE);
483 }
484 break;
485 }
486 cur = cur->next;
487 }
488 *entry = cur->prev;
489 SAVE_HINT(map, *entry);
490 return (FALSE);
491}
492
493/*
494 * vm_map_insert:
495 *
496 * Inserts the given whole VM object into the target
497 * map at the specified address range. The object's
498 * size should match that of the address range.
499 *
500 * Requires that the map be locked, and leaves it so.
501 *
502 * If object is non-NULL, ref count must be bumped by caller
503 * prior to making call to account for the new entry.
504 */
505int
506vm_map_insert(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
507 vm_offset_t start, vm_offset_t end, vm_prot_t prot, vm_prot_t max,
508 int cow)
509{
510 vm_map_entry_t new_entry;
511 vm_map_entry_t prev_entry;
512 vm_map_entry_t temp_entry;
513 vm_eflags_t protoeflags;
514
515 /*
516 * Check that the start and end points are not bogus.
517 */
518
519 if ((start < map->min_offset) || (end > map->max_offset) ||
520 (start >= end))
521 return (KERN_INVALID_ADDRESS);
522
523 /*
524 * Find the entry prior to the proposed starting address; if it's part
525 * of an existing entry, this range is bogus.
526 */
527
528 if (vm_map_lookup_entry(map, start, &temp_entry))
529 return (KERN_NO_SPACE);
530
531 prev_entry = temp_entry;
532
533 /*
534 * Assert that the next entry doesn't overlap the end point.
535 */
536
537 if ((prev_entry->next != &map->header) &&
538 (prev_entry->next->start < end))
539 return (KERN_NO_SPACE);
540
541 protoeflags = 0;
542
543 if (cow & MAP_COPY_ON_WRITE)
544 protoeflags |= MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY;
545
546 if (cow & MAP_NOFAULT) {
547 protoeflags |= MAP_ENTRY_NOFAULT;
548
549 KASSERT(object == NULL,
550 ("vm_map_insert: paradoxical MAP_NOFAULT request"));
551 }
552 if (cow & MAP_DISABLE_SYNCER)
553 protoeflags |= MAP_ENTRY_NOSYNC;
554 if (cow & MAP_DISABLE_COREDUMP)
555 protoeflags |= MAP_ENTRY_NOCOREDUMP;
556
557 if (object) {
558 /*
559 * When object is non-NULL, it could be shared with another
560 * process. We have to set or clear OBJ_ONEMAPPING
561 * appropriately.
562 */
563 if ((object->ref_count > 1) || (object->shadow_count != 0)) {
564 vm_object_clear_flag(object, OBJ_ONEMAPPING);
565 }
566 }
567 else if ((prev_entry != &map->header) &&
568 (prev_entry->eflags == protoeflags) &&
569 (prev_entry->end == start) &&
570 (prev_entry->wired_count == 0) &&
571 ((prev_entry->object.vm_object == NULL) ||
572 vm_object_coalesce(prev_entry->object.vm_object,
573 OFF_TO_IDX(prev_entry->offset),
574 (vm_size_t)(prev_entry->end - prev_entry->start),
575 (vm_size_t)(end - prev_entry->end)))) {
576 /*
577 * We were able to extend the object. Determine if we
578 * can extend the previous map entry to include the
579 * new range as well.
580 */
581 if ((prev_entry->inheritance == VM_INHERIT_DEFAULT) &&
582 (prev_entry->protection == prot) &&
583 (prev_entry->max_protection == max)) {
584 map->size += (end - prev_entry->end);
585 prev_entry->end = end;
586 vm_map_simplify_entry(map, prev_entry);
587 return (KERN_SUCCESS);
588 }
589
590 /*
591 * If we can extend the object but cannot extend the
592 * map entry, we have to create a new map entry. We
593 * must bump the ref count on the extended object to
594 * account for it. object may be NULL.
595 */
596 object = prev_entry->object.vm_object;
597 offset = prev_entry->offset +
598 (prev_entry->end - prev_entry->start);
599 vm_object_reference(object);
600 }
601
602 /*
603 * NOTE: if conditionals fail, object can be NULL here. This occurs
604 * in things like the buffer map where we manage kva but do not manage
605 * backing objects.
606 */
607
608 /*
609 * Create a new entry
610 */
611
612 new_entry = vm_map_entry_create(map);
613 new_entry->start = start;
614 new_entry->end = end;
615
616 new_entry->eflags = protoeflags;
617 new_entry->object.vm_object = object;
618 new_entry->offset = offset;
619 new_entry->avail_ssize = 0;
620
621 new_entry->inheritance = VM_INHERIT_DEFAULT;
622 new_entry->protection = prot;
623 new_entry->max_protection = max;
624 new_entry->wired_count = 0;
625
626 /*
627 * Insert the new entry into the list
628 */
629
630 vm_map_entry_link(map, prev_entry, new_entry);
631 map->size += new_entry->end - new_entry->start;
632
633 /*
634 * Update the free space hint
635 */
636 if ((map->first_free == prev_entry) &&
637 (prev_entry->end >= new_entry->start)) {
638 map->first_free = new_entry;
639 }
640
641#if 0
642 /*
643 * Temporarily removed to avoid MAP_STACK panic, due to
644 * MAP_STACK being a huge hack. Will be added back in
645 * when MAP_STACK (and the user stack mapping) is fixed.
646 */
647 /*
648 * It may be possible to simplify the entry
649 */
650 vm_map_simplify_entry(map, new_entry);
651#endif
652
653 if (cow & (MAP_PREFAULT|MAP_PREFAULT_PARTIAL)) {
654 pmap_object_init_pt(map->pmap, start,
655 object, OFF_TO_IDX(offset), end - start,
656 cow & MAP_PREFAULT_PARTIAL);
657 }
658
659 return (KERN_SUCCESS);
660}
661
662/*
663 * Find sufficient space for `length' bytes in the given map, starting at
664 * `start'. The map must be locked. Returns 0 on success, 1 on no space.
e9bb90e8
MD
665 *
666 * This function will returned an arbitrarily aligned pointer. If no
667 * particular alignment is required you should pass align as 1. Note that
668 * the map may return PAGE_SIZE aligned pointers if all the lengths used in
669 * the map are a multiple of PAGE_SIZE, even if you pass a smaller align
670 * argument.
671 *
672 * 'align' should be a power of 2 but is not required to be.
984263bc
MD
673 */
674int
e9bb90e8
MD
675vm_map_findspace(
676 vm_map_t map,
677 vm_offset_t start,
678 vm_size_t length,
679 vm_offset_t align,
680 vm_offset_t *addr)
984263bc
MD
681{
682 vm_map_entry_t entry, next;
683 vm_offset_t end;
e9bb90e8 684 vm_offset_t align_mask;
984263bc
MD
685
686 if (start < map->min_offset)
687 start = map->min_offset;
688 if (start > map->max_offset)
689 return (1);
690
e9bb90e8
MD
691 /*
692 * If the alignment is not a power of 2 we will have to use
693 * a mod/division, set align_mask to a special value.
694 */
695 if ((align | (align - 1)) + 1 != (align << 1))
696 align_mask = (vm_offset_t)-1;
697 else
698 align_mask = align - 1;
699
984263bc
MD
700 /*
701 * Look for the first possible address; if there's already something
702 * at this address, we have to start after it.
703 */
704 if (start == map->min_offset) {
705 if ((entry = map->first_free) != &map->header)
706 start = entry->end;
707 } else {
708 vm_map_entry_t tmp;
709
710 if (vm_map_lookup_entry(map, start, &tmp))
711 start = tmp->end;
712 entry = tmp;
713 }
714
715 /*
716 * Look through the rest of the map, trying to fit a new region in the
717 * gap between existing regions, or after the very last region.
718 */
719 for (;; start = (entry = next)->end) {
e9bb90e8
MD
720 /*
721 * Adjust the proposed start by the requested alignment,
722 * be sure that we didn't wrap the address.
723 */
724 if (align_mask == (vm_offset_t)-1)
725 end = ((start + align - 1) / align) * align;
726 else
727 end = (start + align_mask) & ~align_mask;
728 if (end < start)
729 return (1);
730 start = end;
984263bc
MD
731 /*
732 * Find the end of the proposed new region. Be sure we didn't
e9bb90e8
MD
733 * go beyond the end of the map, or wrap around the address.
734 * Then check to see if this is the last entry or if the
735 * proposed end fits in the gap between this and the next
736 * entry.
984263bc
MD
737 */
738 end = start + length;
739 if (end > map->max_offset || end < start)
740 return (1);
741 next = entry->next;
742 if (next == &map->header || next->start >= end)
743 break;
744 }
745 SAVE_HINT(map, entry);
746 *addr = start;
747 if (map == kernel_map) {
748 vm_offset_t ksize;
749 if ((ksize = round_page(start + length)) > kernel_vm_end) {
750 pmap_growkernel(ksize);
751 }
752 }
753 return (0);
754}
755
756/*
757 * vm_map_find finds an unallocated region in the target address
758 * map with the given length. The search is defined to be
759 * first-fit from the specified address; the region found is
760 * returned in the same parameter.
761 *
762 * If object is non-NULL, ref count must be bumped by caller
763 * prior to making call to account for the new entry.
764 */
765int
766vm_map_find(vm_map_t map, vm_object_t object, vm_ooffset_t offset,
767 vm_offset_t *addr, /* IN/OUT */
768 vm_size_t length, boolean_t find_space, vm_prot_t prot,
769 vm_prot_t max, int cow)
770{
771 vm_offset_t start;
772 int result, s = 0;
773
774 start = *addr;
775
776 if (map == kmem_map || map == mb_map)
777 s = splvm();
778
779 vm_map_lock(map);
780 if (find_space) {
e9bb90e8 781 if (vm_map_findspace(map, start, length, 1, addr)) {
984263bc
MD
782 vm_map_unlock(map);
783 if (map == kmem_map || map == mb_map)
784 splx(s);
785 return (KERN_NO_SPACE);
786 }
787 start = *addr;
788 }
789 result = vm_map_insert(map, object, offset,
790 start, start + length, prot, max, cow);
791 vm_map_unlock(map);
792
793 if (map == kmem_map || map == mb_map)
794 splx(s);
795
796 return (result);
797}
798
799/*
800 * vm_map_simplify_entry:
801 *
802 * Simplify the given map entry by merging with either neighbor. This
803 * routine also has the ability to merge with both neighbors.
804 *
805 * The map must be locked.
806 *
807 * This routine guarentees that the passed entry remains valid (though
808 * possibly extended). When merging, this routine may delete one or
809 * both neighbors. No action is taken on entries which have their
810 * in-transition flag set.
811 */
812void
813vm_map_simplify_entry(map, entry)
814 vm_map_t map;
815 vm_map_entry_t entry;
816{
817 vm_map_entry_t next, prev;
818 vm_size_t prevsize, esize;
819
820 if (entry->eflags & (MAP_ENTRY_IN_TRANSITION | MAP_ENTRY_IS_SUB_MAP)) {
12e4aaff 821 ++mycpu->gd_cnt.v_intrans_coll;
984263bc
MD
822 return;
823 }
824
825 prev = entry->prev;
826 if (prev != &map->header) {
827 prevsize = prev->end - prev->start;
828 if ( (prev->end == entry->start) &&
829 (prev->object.vm_object == entry->object.vm_object) &&
830 (!prev->object.vm_object ||
831 (prev->offset + prevsize == entry->offset)) &&
832 (prev->eflags == entry->eflags) &&
833 (prev->protection == entry->protection) &&
834 (prev->max_protection == entry->max_protection) &&
835 (prev->inheritance == entry->inheritance) &&
836 (prev->wired_count == entry->wired_count)) {
837 if (map->first_free == prev)
838 map->first_free = entry;
839 if (map->hint == prev)
840 map->hint = entry;
841 vm_map_entry_unlink(map, prev);
842 entry->start = prev->start;
843 entry->offset = prev->offset;
844 if (prev->object.vm_object)
845 vm_object_deallocate(prev->object.vm_object);
846 vm_map_entry_dispose(map, prev);
847 }
848 }
849
850 next = entry->next;
851 if (next != &map->header) {
852 esize = entry->end - entry->start;
853 if ((entry->end == next->start) &&
854 (next->object.vm_object == entry->object.vm_object) &&
855 (!entry->object.vm_object ||
856 (entry->offset + esize == next->offset)) &&
857 (next->eflags == entry->eflags) &&
858 (next->protection == entry->protection) &&
859 (next->max_protection == entry->max_protection) &&
860 (next->inheritance == entry->inheritance) &&
861 (next->wired_count == entry->wired_count)) {
862 if (map->first_free == next)
863 map->first_free = entry;
864 if (map->hint == next)
865 map->hint = entry;
866 vm_map_entry_unlink(map, next);
867 entry->end = next->end;
868 if (next->object.vm_object)
869 vm_object_deallocate(next->object.vm_object);
870 vm_map_entry_dispose(map, next);
871 }
872 }
873}
874/*
875 * vm_map_clip_start: [ internal use only ]
876 *
877 * Asserts that the given entry begins at or after
878 * the specified address; if necessary,
879 * it splits the entry into two.
880 */
881#define vm_map_clip_start(map, entry, startaddr) \
882{ \
883 if (startaddr > entry->start) \
884 _vm_map_clip_start(map, entry, startaddr); \
885}
886
887/*
888 * This routine is called only when it is known that
889 * the entry must be split.
890 */
891static void
892_vm_map_clip_start(map, entry, start)
893 vm_map_t map;
894 vm_map_entry_t entry;
895 vm_offset_t start;
896{
897 vm_map_entry_t new_entry;
898
899 /*
900 * Split off the front portion -- note that we must insert the new
901 * entry BEFORE this one, so that this entry has the specified
902 * starting address.
903 */
904
905 vm_map_simplify_entry(map, entry);
906
907 /*
908 * If there is no object backing this entry, we might as well create
909 * one now. If we defer it, an object can get created after the map
910 * is clipped, and individual objects will be created for the split-up
911 * map. This is a bit of a hack, but is also about the best place to
912 * put this improvement.
913 */
914
915 if (entry->object.vm_object == NULL && !map->system_map) {
916 vm_object_t object;
917 object = vm_object_allocate(OBJT_DEFAULT,
918 atop(entry->end - entry->start));
919 entry->object.vm_object = object;
920 entry->offset = 0;
921 }
922
923 new_entry = vm_map_entry_create(map);
924 *new_entry = *entry;
925
926 new_entry->end = start;
927 entry->offset += (start - entry->start);
928 entry->start = start;
929
930 vm_map_entry_link(map, entry->prev, new_entry);
931
932 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
933 vm_object_reference(new_entry->object.vm_object);
934 }
935}
936
937/*
938 * vm_map_clip_end: [ internal use only ]
939 *
940 * Asserts that the given entry ends at or before
941 * the specified address; if necessary,
942 * it splits the entry into two.
943 */
944
945#define vm_map_clip_end(map, entry, endaddr) \
946{ \
947 if (endaddr < entry->end) \
948 _vm_map_clip_end(map, entry, endaddr); \
949}
950
951/*
952 * This routine is called only when it is known that
953 * the entry must be split.
954 */
955static void
956_vm_map_clip_end(map, entry, end)
957 vm_map_t map;
958 vm_map_entry_t entry;
959 vm_offset_t end;
960{
961 vm_map_entry_t new_entry;
962
963 /*
964 * If there is no object backing this entry, we might as well create
965 * one now. If we defer it, an object can get created after the map
966 * is clipped, and individual objects will be created for the split-up
967 * map. This is a bit of a hack, but is also about the best place to
968 * put this improvement.
969 */
970
971 if (entry->object.vm_object == NULL && !map->system_map) {
972 vm_object_t object;
973 object = vm_object_allocate(OBJT_DEFAULT,
974 atop(entry->end - entry->start));
975 entry->object.vm_object = object;
976 entry->offset = 0;
977 }
978
979 /*
980 * Create a new entry and insert it AFTER the specified entry
981 */
982
983 new_entry = vm_map_entry_create(map);
984 *new_entry = *entry;
985
986 new_entry->start = entry->end = end;
987 new_entry->offset += (end - entry->start);
988
989 vm_map_entry_link(map, entry, new_entry);
990
991 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
992 vm_object_reference(new_entry->object.vm_object);
993 }
994}
995
996/*
997 * VM_MAP_RANGE_CHECK: [ internal use only ]
998 *
999 * Asserts that the starting and ending region
1000 * addresses fall within the valid range of the map.
1001 */
1002#define VM_MAP_RANGE_CHECK(map, start, end) \
1003 { \
1004 if (start < vm_map_min(map)) \
1005 start = vm_map_min(map); \
1006 if (end > vm_map_max(map)) \
1007 end = vm_map_max(map); \
1008 if (start > end) \
1009 start = end; \
1010 }
1011
1012/*
1013 * vm_map_transition_wait: [ kernel use only ]
1014 *
1015 * Used to block when an in-transition collison occurs. The map
1016 * is unlocked for the sleep and relocked before the return.
1017 */
1018static
1019void
1020vm_map_transition_wait(vm_map_t map)
1021{
1022 vm_map_unlock(map);
377d4740 1023 tsleep(map, 0, "vment", 0);
984263bc
MD
1024 vm_map_lock(map);
1025}
1026
1027/*
1028 * CLIP_CHECK_BACK
1029 * CLIP_CHECK_FWD
1030 *
1031 * When we do blocking operations with the map lock held it is
1032 * possible that a clip might have occured on our in-transit entry,
1033 * requiring an adjustment to the entry in our loop. These macros
1034 * help the pageable and clip_range code deal with the case. The
1035 * conditional costs virtually nothing if no clipping has occured.
1036 */
1037
1038#define CLIP_CHECK_BACK(entry, save_start) \
1039 do { \
1040 while (entry->start != save_start) { \
1041 entry = entry->prev; \
1042 KASSERT(entry != &map->header, ("bad entry clip")); \
1043 } \
1044 } while(0)
1045
1046#define CLIP_CHECK_FWD(entry, save_end) \
1047 do { \
1048 while (entry->end != save_end) { \
1049 entry = entry->next; \
1050 KASSERT(entry != &map->header, ("bad entry clip")); \
1051 } \
1052 } while(0)
1053
1054
1055/*
1056 * vm_map_clip_range: [ kernel use only ]
1057 *
1058 * Clip the specified range and return the base entry. The
1059 * range may cover several entries starting at the returned base
1060 * and the first and last entry in the covering sequence will be
1061 * properly clipped to the requested start and end address.
1062 *
1063 * If no holes are allowed you should pass the MAP_CLIP_NO_HOLES
1064 * flag.
1065 *
1066 * The MAP_ENTRY_IN_TRANSITION flag will be set for the entries
1067 * covered by the requested range.
1068 *
1069 * The map must be exclusively locked on entry and will remain locked
1070 * on return. If no range exists or the range contains holes and you
1071 * specified that no holes were allowed, NULL will be returned. This
1072 * routine may temporarily unlock the map in order avoid a deadlock when
1073 * sleeping.
1074 */
1075static
1076vm_map_entry_t
1077vm_map_clip_range(vm_map_t map, vm_offset_t start, vm_offset_t end, int flags)
1078{
1079 vm_map_entry_t start_entry;
1080 vm_map_entry_t entry;
1081
1082 /*
1083 * Locate the entry and effect initial clipping. The in-transition
1084 * case does not occur very often so do not try to optimize it.
1085 */
1086again:
1087 if (vm_map_lookup_entry(map, start, &start_entry) == FALSE)
1088 return (NULL);
1089 entry = start_entry;
1090 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
1091 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
12e4aaff
MD
1092 ++mycpu->gd_cnt.v_intrans_coll;
1093 ++mycpu->gd_cnt.v_intrans_wait;
984263bc
MD
1094 vm_map_transition_wait(map);
1095 /*
1096 * entry and/or start_entry may have been clipped while
1097 * we slept, or may have gone away entirely. We have
1098 * to restart from the lookup.
1099 */
1100 goto again;
1101 }
1102 /*
1103 * Since we hold an exclusive map lock we do not have to restart
1104 * after clipping, even though clipping may block in zalloc.
1105 */
1106 vm_map_clip_start(map, entry, start);
1107 vm_map_clip_end(map, entry, end);
1108 entry->eflags |= MAP_ENTRY_IN_TRANSITION;
1109
1110 /*
1111 * Scan entries covered by the range. When working on the next
1112 * entry a restart need only re-loop on the current entry which
1113 * we have already locked, since 'next' may have changed. Also,
1114 * even though entry is safe, it may have been clipped so we
1115 * have to iterate forwards through the clip after sleeping.
1116 */
1117 while (entry->next != &map->header && entry->next->start < end) {
1118 vm_map_entry_t next = entry->next;
1119
1120 if (flags & MAP_CLIP_NO_HOLES) {
1121 if (next->start > entry->end) {
1122 vm_map_unclip_range(map, start_entry,
1123 start, entry->end, flags);
1124 return(NULL);
1125 }
1126 }
1127
1128 if (next->eflags & MAP_ENTRY_IN_TRANSITION) {
1129 vm_offset_t save_end = entry->end;
1130 next->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
12e4aaff
MD
1131 ++mycpu->gd_cnt.v_intrans_coll;
1132 ++mycpu->gd_cnt.v_intrans_wait;
984263bc
MD
1133 vm_map_transition_wait(map);
1134
1135 /*
1136 * clips might have occured while we blocked.
1137 */
1138 CLIP_CHECK_FWD(entry, save_end);
1139 CLIP_CHECK_BACK(start_entry, start);
1140 continue;
1141 }
1142 /*
1143 * No restart necessary even though clip_end may block, we
1144 * are holding the map lock.
1145 */
1146 vm_map_clip_end(map, next, end);
1147 next->eflags |= MAP_ENTRY_IN_TRANSITION;
1148 entry = next;
1149 }
1150 if (flags & MAP_CLIP_NO_HOLES) {
1151 if (entry->end != end) {
1152 vm_map_unclip_range(map, start_entry,
1153 start, entry->end, flags);
1154 return(NULL);
1155 }
1156 }
1157 return(start_entry);
1158}
1159
1160/*
1161 * vm_map_unclip_range: [ kernel use only ]
1162 *
1163 * Undo the effect of vm_map_clip_range(). You should pass the same
1164 * flags and the same range that you passed to vm_map_clip_range().
1165 * This code will clear the in-transition flag on the entries and
1166 * wake up anyone waiting. This code will also simplify the sequence
1167 * and attempt to merge it with entries before and after the sequence.
1168 *
1169 * The map must be locked on entry and will remain locked on return.
1170 *
1171 * Note that you should also pass the start_entry returned by
1172 * vm_map_clip_range(). However, if you block between the two calls
1173 * with the map unlocked please be aware that the start_entry may
1174 * have been clipped and you may need to scan it backwards to find
1175 * the entry corresponding with the original start address. You are
1176 * responsible for this, vm_map_unclip_range() expects the correct
1177 * start_entry to be passed to it and will KASSERT otherwise.
1178 */
1179static
1180void
1181vm_map_unclip_range(
1182 vm_map_t map,
1183 vm_map_entry_t start_entry,
1184 vm_offset_t start,
1185 vm_offset_t end,
1186 int flags)
1187{
1188 vm_map_entry_t entry;
1189
1190 entry = start_entry;
1191
1192 KASSERT(entry->start == start, ("unclip_range: illegal base entry"));
1193 while (entry != &map->header && entry->start < end) {
1194 KASSERT(entry->eflags & MAP_ENTRY_IN_TRANSITION, ("in-transition flag not set during unclip on: %p", entry));
1195 KASSERT(entry->end <= end, ("unclip_range: tail wasn't clipped"));
1196 entry->eflags &= ~MAP_ENTRY_IN_TRANSITION;
1197 if (entry->eflags & MAP_ENTRY_NEEDS_WAKEUP) {
1198 entry->eflags &= ~MAP_ENTRY_NEEDS_WAKEUP;
1199 wakeup(map);
1200 }
1201 entry = entry->next;
1202 }
1203
1204 /*
1205 * Simplification does not block so there is no restart case.
1206 */
1207 entry = start_entry;
1208 while (entry != &map->header && entry->start < end) {
1209 vm_map_simplify_entry(map, entry);
1210 entry = entry->next;
1211 }
1212}
1213
1214/*
1215 * vm_map_submap: [ kernel use only ]
1216 *
1217 * Mark the given range as handled by a subordinate map.
1218 *
1219 * This range must have been created with vm_map_find,
1220 * and no other operations may have been performed on this
1221 * range prior to calling vm_map_submap.
1222 *
1223 * Only a limited number of operations can be performed
1224 * within this rage after calling vm_map_submap:
1225 * vm_fault
1226 * [Don't try vm_map_copy!]
1227 *
1228 * To remove a submapping, one must first remove the
1229 * range from the superior map, and then destroy the
1230 * submap (if desired). [Better yet, don't try it.]
1231 */
1232int
1233vm_map_submap(map, start, end, submap)
1234 vm_map_t map;
1235 vm_offset_t start;
1236 vm_offset_t end;
1237 vm_map_t submap;
1238{
1239 vm_map_entry_t entry;
1240 int result = KERN_INVALID_ARGUMENT;
1241
1242 vm_map_lock(map);
1243
1244 VM_MAP_RANGE_CHECK(map, start, end);
1245
1246 if (vm_map_lookup_entry(map, start, &entry)) {
1247 vm_map_clip_start(map, entry, start);
1248 } else {
1249 entry = entry->next;
1250 }
1251
1252 vm_map_clip_end(map, entry, end);
1253
1254 if ((entry->start == start) && (entry->end == end) &&
1255 ((entry->eflags & MAP_ENTRY_COW) == 0) &&
1256 (entry->object.vm_object == NULL)) {
1257 entry->object.sub_map = submap;
1258 entry->eflags |= MAP_ENTRY_IS_SUB_MAP;
1259 result = KERN_SUCCESS;
1260 }
1261 vm_map_unlock(map);
1262
1263 return (result);
1264}
1265
1266/*
1267 * vm_map_protect:
1268 *
1269 * Sets the protection of the specified address
1270 * region in the target map. If "set_max" is
1271 * specified, the maximum protection is to be set;
1272 * otherwise, only the current protection is affected.
1273 */
1274int
1275vm_map_protect(vm_map_t map, vm_offset_t start, vm_offset_t end,
1276 vm_prot_t new_prot, boolean_t set_max)
1277{
1278 vm_map_entry_t current;
1279 vm_map_entry_t entry;
1280
1281 vm_map_lock(map);
1282
1283 VM_MAP_RANGE_CHECK(map, start, end);
1284
1285 if (vm_map_lookup_entry(map, start, &entry)) {
1286 vm_map_clip_start(map, entry, start);
1287 } else {
1288 entry = entry->next;
1289 }
1290
1291 /*
1292 * Make a first pass to check for protection violations.
1293 */
1294
1295 current = entry;
1296 while ((current != &map->header) && (current->start < end)) {
1297 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
1298 vm_map_unlock(map);
1299 return (KERN_INVALID_ARGUMENT);
1300 }
1301 if ((new_prot & current->max_protection) != new_prot) {
1302 vm_map_unlock(map);
1303 return (KERN_PROTECTION_FAILURE);
1304 }
1305 current = current->next;
1306 }
1307
1308 /*
1309 * Go back and fix up protections. [Note that clipping is not
1310 * necessary the second time.]
1311 */
1312
1313 current = entry;
1314
1315 while ((current != &map->header) && (current->start < end)) {
1316 vm_prot_t old_prot;
1317
1318 vm_map_clip_end(map, current, end);
1319
1320 old_prot = current->protection;
1321 if (set_max)
1322 current->protection =
1323 (current->max_protection = new_prot) &
1324 old_prot;
1325 else
1326 current->protection = new_prot;
1327
1328 /*
1329 * Update physical map if necessary. Worry about copy-on-write
1330 * here -- CHECK THIS XXX
1331 */
1332
1333 if (current->protection != old_prot) {
1334#define MASK(entry) (((entry)->eflags & MAP_ENTRY_COW) ? ~VM_PROT_WRITE : \
1335 VM_PROT_ALL)
1336
1337 pmap_protect(map->pmap, current->start,
1338 current->end,
1339 current->protection & MASK(current));
1340#undef MASK
1341 }
1342
1343 vm_map_simplify_entry(map, current);
1344
1345 current = current->next;
1346 }
1347
1348 vm_map_unlock(map);
1349 return (KERN_SUCCESS);
1350}
1351
1352/*
1353 * vm_map_madvise:
1354 *
1355 * This routine traverses a processes map handling the madvise
1356 * system call. Advisories are classified as either those effecting
1357 * the vm_map_entry structure, or those effecting the underlying
1358 * objects.
1359 */
1360
1361int
1362vm_map_madvise(map, start, end, behav)
1363 vm_map_t map;
1364 vm_offset_t start, end;
1365 int behav;
1366{
1367 vm_map_entry_t current, entry;
1368 int modify_map = 0;
1369
1370 /*
1371 * Some madvise calls directly modify the vm_map_entry, in which case
1372 * we need to use an exclusive lock on the map and we need to perform
1373 * various clipping operations. Otherwise we only need a read-lock
1374 * on the map.
1375 */
1376
1377 switch(behav) {
1378 case MADV_NORMAL:
1379 case MADV_SEQUENTIAL:
1380 case MADV_RANDOM:
1381 case MADV_NOSYNC:
1382 case MADV_AUTOSYNC:
1383 case MADV_NOCORE:
1384 case MADV_CORE:
1385 modify_map = 1;
1386 vm_map_lock(map);
1387 break;
1388 case MADV_WILLNEED:
1389 case MADV_DONTNEED:
1390 case MADV_FREE:
1391 vm_map_lock_read(map);
1392 break;
1393 default:
1394 return (KERN_INVALID_ARGUMENT);
1395 }
1396
1397 /*
1398 * Locate starting entry and clip if necessary.
1399 */
1400
1401 VM_MAP_RANGE_CHECK(map, start, end);
1402
1403 if (vm_map_lookup_entry(map, start, &entry)) {
1404 if (modify_map)
1405 vm_map_clip_start(map, entry, start);
1406 } else {
1407 entry = entry->next;
1408 }
1409
1410 if (modify_map) {
1411 /*
1412 * madvise behaviors that are implemented in the vm_map_entry.
1413 *
1414 * We clip the vm_map_entry so that behavioral changes are
1415 * limited to the specified address range.
1416 */
1417 for (current = entry;
1418 (current != &map->header) && (current->start < end);
1419 current = current->next
1420 ) {
1421 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
1422 continue;
1423
1424 vm_map_clip_end(map, current, end);
1425
1426 switch (behav) {
1427 case MADV_NORMAL:
1428 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_NORMAL);
1429 break;
1430 case MADV_SEQUENTIAL:
1431 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_SEQUENTIAL);
1432 break;
1433 case MADV_RANDOM:
1434 vm_map_entry_set_behavior(current, MAP_ENTRY_BEHAV_RANDOM);
1435 break;
1436 case MADV_NOSYNC:
1437 current->eflags |= MAP_ENTRY_NOSYNC;
1438 break;
1439 case MADV_AUTOSYNC:
1440 current->eflags &= ~MAP_ENTRY_NOSYNC;
1441 break;
1442 case MADV_NOCORE:
1443 current->eflags |= MAP_ENTRY_NOCOREDUMP;
1444 break;
1445 case MADV_CORE:
1446 current->eflags &= ~MAP_ENTRY_NOCOREDUMP;
1447 break;
1448 default:
1449 break;
1450 }
1451 vm_map_simplify_entry(map, current);
1452 }
1453 vm_map_unlock(map);
1454 } else {
1455 vm_pindex_t pindex;
1456 int count;
1457
1458 /*
1459 * madvise behaviors that are implemented in the underlying
1460 * vm_object.
1461 *
1462 * Since we don't clip the vm_map_entry, we have to clip
1463 * the vm_object pindex and count.
1464 */
1465 for (current = entry;
1466 (current != &map->header) && (current->start < end);
1467 current = current->next
1468 ) {
1469 vm_offset_t useStart;
1470
1471 if (current->eflags & MAP_ENTRY_IS_SUB_MAP)
1472 continue;
1473
1474 pindex = OFF_TO_IDX(current->offset);
1475 count = atop(current->end - current->start);
1476 useStart = current->start;
1477
1478 if (current->start < start) {
1479 pindex += atop(start - current->start);
1480 count -= atop(start - current->start);
1481 useStart = start;
1482 }
1483 if (current->end > end)
1484 count -= atop(current->end - end);
1485
1486 if (count <= 0)
1487 continue;
1488
1489 vm_object_madvise(current->object.vm_object,
1490 pindex, count, behav);
1491 if (behav == MADV_WILLNEED) {
1492 pmap_object_init_pt(
1493 map->pmap,
1494 useStart,
1495 current->object.vm_object,
1496 pindex,
1497 (count << PAGE_SHIFT),
1498 MAP_PREFAULT_MADVISE
1499 );
1500 }
1501 }
1502 vm_map_unlock_read(map);
1503 }
1504 return(0);
1505}
1506
1507
1508/*
1509 * vm_map_inherit:
1510 *
1511 * Sets the inheritance of the specified address
1512 * range in the target map. Inheritance
1513 * affects how the map will be shared with
1514 * child maps at the time of vm_map_fork.
1515 */
1516int
1517vm_map_inherit(vm_map_t map, vm_offset_t start, vm_offset_t end,
1518 vm_inherit_t new_inheritance)
1519{
1520 vm_map_entry_t entry;
1521 vm_map_entry_t temp_entry;
1522
1523 switch (new_inheritance) {
1524 case VM_INHERIT_NONE:
1525 case VM_INHERIT_COPY:
1526 case VM_INHERIT_SHARE:
1527 break;
1528 default:
1529 return (KERN_INVALID_ARGUMENT);
1530 }
1531
1532 vm_map_lock(map);
1533
1534 VM_MAP_RANGE_CHECK(map, start, end);
1535
1536 if (vm_map_lookup_entry(map, start, &temp_entry)) {
1537 entry = temp_entry;
1538 vm_map_clip_start(map, entry, start);
1539 } else
1540 entry = temp_entry->next;
1541
1542 while ((entry != &map->header) && (entry->start < end)) {
1543 vm_map_clip_end(map, entry, end);
1544
1545 entry->inheritance = new_inheritance;
1546
1547 vm_map_simplify_entry(map, entry);
1548
1549 entry = entry->next;
1550 }
1551
1552 vm_map_unlock(map);
1553 return (KERN_SUCCESS);
1554}
1555
1556/*
1557 * Implement the semantics of mlock
1558 */
1559int
1560vm_map_user_pageable(map, start, real_end, new_pageable)
1561 vm_map_t map;
1562 vm_offset_t start;
1563 vm_offset_t real_end;
1564 boolean_t new_pageable;
1565{
1566 vm_map_entry_t entry;
1567 vm_map_entry_t start_entry;
1568 vm_offset_t end;
1569 int rv = KERN_SUCCESS;
1570
1571 vm_map_lock(map);
1572 VM_MAP_RANGE_CHECK(map, start, real_end);
1573 end = real_end;
1574
1575 start_entry = vm_map_clip_range(map, start, end, MAP_CLIP_NO_HOLES);
1576 if (start_entry == NULL) {
1577 vm_map_unlock(map);
1578 return (KERN_INVALID_ADDRESS);
1579 }
1580
1581 if (new_pageable == 0) {
1582 entry = start_entry;
1583 while ((entry != &map->header) && (entry->start < end)) {
1584 vm_offset_t save_start;
1585 vm_offset_t save_end;
1586
1587 /*
1588 * Already user wired or hard wired (trivial cases)
1589 */
1590 if (entry->eflags & MAP_ENTRY_USER_WIRED) {
1591 entry = entry->next;
1592 continue;
1593 }
1594 if (entry->wired_count != 0) {
1595 entry->wired_count++;
1596 entry->eflags |= MAP_ENTRY_USER_WIRED;
1597 entry = entry->next;
1598 continue;
1599 }
1600
1601 /*
1602 * A new wiring requires instantiation of appropriate
1603 * management structures and the faulting in of the
1604 * page.
1605 */
1606 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1607 int copyflag = entry->eflags & MAP_ENTRY_NEEDS_COPY;
1608 if (copyflag && ((entry->protection & VM_PROT_WRITE) != 0)) {
1609
1610 vm_object_shadow(&entry->object.vm_object,
1611 &entry->offset,
1612 atop(entry->end - entry->start));
1613 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
1614
1615 } else if (entry->object.vm_object == NULL &&
1616 !map->system_map) {
1617
1618 entry->object.vm_object =
1619 vm_object_allocate(OBJT_DEFAULT,
1620 atop(entry->end - entry->start));
1621 entry->offset = (vm_offset_t) 0;
1622
1623 }
1624 }
1625 entry->wired_count++;
1626 entry->eflags |= MAP_ENTRY_USER_WIRED;
1627
1628 /*
1629 * Now fault in the area. The map lock needs to be
1630 * manipulated to avoid deadlocks. The in-transition
1631 * flag protects the entries.
1632 */
1633 save_start = entry->start;
1634 save_end = entry->end;
1635 vm_map_unlock(map);
1636 map->timestamp++;
1637 rv = vm_fault_user_wire(map, save_start, save_end);
1638 vm_map_lock(map);
1639 if (rv) {
1640 CLIP_CHECK_BACK(entry, save_start);
1641 for (;;) {
1642 KASSERT(entry->wired_count == 1, ("bad wired_count on entry"));
1643 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
1644 entry->wired_count = 0;
1645 if (entry->end == save_end)
1646 break;
1647 entry = entry->next;
1648 KASSERT(entry != &map->header, ("bad entry clip during backout"));
1649 }
1650 end = save_start; /* unwire the rest */
1651 break;
1652 }
1653 /*
1654 * note that even though the entry might have been
1655 * clipped, the USER_WIRED flag we set prevents
1656 * duplication so we do not have to do a
1657 * clip check.
1658 */
1659 entry = entry->next;
1660 }
1661
1662 /*
1663 * If we failed fall through to the unwiring section to
1664 * unwire what we had wired so far. 'end' has already
1665 * been adjusted.
1666 */
1667 if (rv)
1668 new_pageable = 1;
1669
1670 /*
1671 * start_entry might have been clipped if we unlocked the
1672 * map and blocked. No matter how clipped it has gotten
1673 * there should be a fragment that is on our start boundary.
1674 */
1675 CLIP_CHECK_BACK(start_entry, start);
1676 }
1677
1678 /*
1679 * Deal with the unwiring case.
1680 */
1681 if (new_pageable) {
1682 /*
1683 * This is the unwiring case. We must first ensure that the
1684 * range to be unwired is really wired down. We know there
1685 * are no holes.
1686 */
1687 entry = start_entry;
1688 while ((entry != &map->header) && (entry->start < end)) {
1689 if ((entry->eflags & MAP_ENTRY_USER_WIRED) == 0) {
1690 rv = KERN_INVALID_ARGUMENT;
1691 goto done;
1692 }
1693 KASSERT(entry->wired_count != 0, ("wired count was 0 with USER_WIRED set! %p", entry));
1694 entry = entry->next;
1695 }
1696
1697 /*
1698 * Now decrement the wiring count for each region. If a region
1699 * becomes completely unwired, unwire its physical pages and
1700 * mappings.
1701 */
1702 while ((entry != &map->header) && (entry->start < end)) {
1703 KASSERT(entry->eflags & MAP_ENTRY_USER_WIRED, ("expected USER_WIRED on entry %p", entry));
1704 entry->eflags &= ~MAP_ENTRY_USER_WIRED;
1705 entry->wired_count--;
1706 if (entry->wired_count == 0)
1707 vm_fault_unwire(map, entry->start, entry->end);
1708 entry = entry->next;
1709 }
1710 }
1711done:
1712 vm_map_unclip_range(map, start_entry, start, real_end,
1713 MAP_CLIP_NO_HOLES);
1714 map->timestamp++;
1715 vm_map_unlock(map);
1716 return (rv);
1717}
1718
1719/*
1720 * vm_map_pageable:
1721 *
1722 * Sets the pageability of the specified address
1723 * range in the target map. Regions specified
1724 * as not pageable require locked-down physical
1725 * memory and physical page maps.
1726 *
1727 * The map must not be locked, but a reference
1728 * must remain to the map throughout the call.
1729 */
1730int
1731vm_map_pageable(map, start, real_end, new_pageable)
1732 vm_map_t map;
1733 vm_offset_t start;
1734 vm_offset_t real_end;
1735 boolean_t new_pageable;
1736{
1737 vm_map_entry_t entry;
1738 vm_map_entry_t start_entry;
1739 vm_offset_t end;
1740 int rv = KERN_SUCCESS;
1741 int s;
1742
1743 vm_map_lock(map);
1744 VM_MAP_RANGE_CHECK(map, start, real_end);
1745 end = real_end;
1746
1747 start_entry = vm_map_clip_range(map, start, end, MAP_CLIP_NO_HOLES);
1748 if (start_entry == NULL) {
1749 vm_map_unlock(map);
1750 return (KERN_INVALID_ADDRESS);
1751 }
1752 if (new_pageable == 0) {
1753 /*
1754 * Wiring.
1755 *
1756 * 1. Holding the write lock, we create any shadow or zero-fill
1757 * objects that need to be created. Then we clip each map
1758 * entry to the region to be wired and increment its wiring
1759 * count. We create objects before clipping the map entries
1760 * to avoid object proliferation.
1761 *
1762 * 2. We downgrade to a read lock, and call vm_fault_wire to
1763 * fault in the pages for any newly wired area (wired_count is
1764 * 1).
1765 *
1766 * Downgrading to a read lock for vm_fault_wire avoids a
1767 * possible deadlock with another process that may have faulted
1768 * on one of the pages to be wired (it would mark the page busy,
1769 * blocking us, then in turn block on the map lock that we
1770 * hold). Because of problems in the recursive lock package,
1771 * we cannot upgrade to a write lock in vm_map_lookup. Thus,
1772 * any actions that require the write lock must be done
1773 * beforehand. Because we keep the read lock on the map, the
1774 * copy-on-write status of the entries we modify here cannot
1775 * change.
1776 */
1777
1778 entry = start_entry;
1779 while ((entry != &map->header) && (entry->start < end)) {
1780 /*
1781 * Trivial case if the entry is already wired
1782 */
1783 if (entry->wired_count) {
1784 entry->wired_count++;
1785 entry = entry->next;
1786 continue;
1787 }
1788
1789 /*
1790 * The entry is being newly wired, we have to setup
1791 * appropriate management structures. A shadow
1792 * object is required for a copy-on-write region,
1793 * or a normal object for a zero-fill region. We
1794 * do not have to do this for entries that point to sub
1795 * maps because we won't hold the lock on the sub map.
1796 */
1797 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
1798 int copyflag = entry->eflags & MAP_ENTRY_NEEDS_COPY;
1799 if (copyflag &&
1800 ((entry->protection & VM_PROT_WRITE) != 0)) {
1801
1802 vm_object_shadow(&entry->object.vm_object,
1803 &entry->offset,
1804 atop(entry->end - entry->start));
1805 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
1806 } else if (entry->object.vm_object == NULL &&
1807 !map->system_map) {
1808 entry->object.vm_object =
1809 vm_object_allocate(OBJT_DEFAULT,
1810 atop(entry->end - entry->start));
1811 entry->offset = (vm_offset_t) 0;
1812 }
1813 }
1814
1815 entry->wired_count++;
1816 entry = entry->next;
1817 }
1818
1819 /*
1820 * Pass 2.
1821 */
1822
1823 /*
1824 * HACK HACK HACK HACK
1825 *
1826 * Unlock the map to avoid deadlocks. The in-transit flag
1827 * protects us from most changes but note that
1828 * clipping may still occur. To prevent clipping from
1829 * occuring after the unlock, except for when we are
1830 * blocking in vm_fault_wire, we must run at splvm().
1831 * Otherwise our accesses to entry->start and entry->end
1832 * could be corrupted. We have to set splvm() prior to
1833 * unlocking so start_entry does not change out from
1834 * under us at the very beginning of the loop.
1835 *
1836 * HACK HACK HACK HACK
1837 */
1838
1839 s = splvm();
1840 vm_map_unlock(map);
1841
1842 entry = start_entry;
1843 while (entry != &map->header && entry->start < end) {
1844 /*
1845 * If vm_fault_wire fails for any page we need to undo
1846 * what has been done. We decrement the wiring count
1847 * for those pages which have not yet been wired (now)
1848 * and unwire those that have (later).
1849 */
1850 vm_offset_t save_start = entry->start;
1851 vm_offset_t save_end = entry->end;
1852
1853 if (entry->wired_count == 1)
1854 rv = vm_fault_wire(map, entry->start, entry->end);
1855 if (rv) {
1856 CLIP_CHECK_BACK(entry, save_start);
1857 for (;;) {
1858 KASSERT(entry->wired_count == 1, ("wired_count changed unexpectedly"));
1859 entry->wired_count = 0;
1860 if (entry->end == save_end)
1861 break;
1862 entry = entry->next;
1863 KASSERT(entry != &map->header, ("bad entry clip during backout"));
1864 }
1865 end = save_start;
1866 break;
1867 }
1868 CLIP_CHECK_FWD(entry, save_end);
1869 entry = entry->next;
1870 }
1871 splx(s);
1872
1873 /*
1874 * relock. start_entry is still IN_TRANSITION and must
1875 * still exist, but may have been clipped (handled just
1876 * below).
1877 */
1878 vm_map_lock(map);
1879
1880 /*
1881 * If a failure occured undo everything by falling through
1882 * to the unwiring code. 'end' has already been adjusted
1883 * appropriately.
1884 */
1885 if (rv)
1886 new_pageable = 1;
1887
1888 /*
1889 * start_entry might have been clipped if we unlocked the
1890 * map and blocked. No matter how clipped it has gotten
1891 * there should be a fragment that is on our start boundary.
1892 */
1893 CLIP_CHECK_BACK(start_entry, start);
1894 }
1895
1896 if (new_pageable) {
1897 /*
1898 * This is the unwiring case. We must first ensure that the
1899 * range to be unwired is really wired down. We know there
1900 * are no holes.
1901 */
1902 entry = start_entry;
1903 while ((entry != &map->header) && (entry->start < end)) {
1904 if (entry->wired_count == 0) {
1905 rv = KERN_INVALID_ARGUMENT;
1906 goto done;
1907 }
1908 entry = entry->next;
1909 }
1910
1911 /*
1912 * Now decrement the wiring count for each region. If a region
1913 * becomes completely unwired, unwire its physical pages and
1914 * mappings.
1915 */
1916 entry = start_entry;
1917 while ((entry != &map->header) && (entry->start < end)) {
1918 entry->wired_count--;
1919 if (entry->wired_count == 0)
1920 vm_fault_unwire(map, entry->start, entry->end);
1921 entry = entry->next;
1922 }
1923 }
1924done:
1925 vm_map_unclip_range(map, start_entry, start, real_end,
1926 MAP_CLIP_NO_HOLES);
1927 map->timestamp++;
1928 vm_map_unlock(map);
1929 return (rv);
1930}
1931
1932/*
1933 * vm_map_clean
1934 *
1935 * Push any dirty cached pages in the address range to their pager.
1936 * If syncio is TRUE, dirty pages are written synchronously.
1937 * If invalidate is TRUE, any cached pages are freed as well.
1938 *
1939 * Returns an error if any part of the specified range is not mapped.
1940 */
1941int
1942vm_map_clean(map, start, end, syncio, invalidate)
1943 vm_map_t map;
1944 vm_offset_t start;
1945 vm_offset_t end;
1946 boolean_t syncio;
1947 boolean_t invalidate;
1948{
1949 vm_map_entry_t current;
1950 vm_map_entry_t entry;
1951 vm_size_t size;
1952 vm_object_t object;
1953 vm_ooffset_t offset;
1954
1955 vm_map_lock_read(map);
1956 VM_MAP_RANGE_CHECK(map, start, end);
1957 if (!vm_map_lookup_entry(map, start, &entry)) {
1958 vm_map_unlock_read(map);
1959 return (KERN_INVALID_ADDRESS);
1960 }
1961 /*
1962 * Make a first pass to check for holes.
1963 */
1964 for (current = entry; current->start < end; current = current->next) {
1965 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
1966 vm_map_unlock_read(map);
1967 return (KERN_INVALID_ARGUMENT);
1968 }
1969 if (end > current->end &&
1970 (current->next == &map->header ||
1971 current->end != current->next->start)) {
1972 vm_map_unlock_read(map);
1973 return (KERN_INVALID_ADDRESS);
1974 }
1975 }
1976
1977 if (invalidate)
1978 pmap_remove(vm_map_pmap(map), start, end);
1979 /*
1980 * Make a second pass, cleaning/uncaching pages from the indicated
1981 * objects as we go.
1982 */
1983 for (current = entry; current->start < end; current = current->next) {
1984 offset = current->offset + (start - current->start);
1985 size = (end <= current->end ? end : current->end) - start;
1986 if (current->eflags & MAP_ENTRY_IS_SUB_MAP) {
1987 vm_map_t smap;
1988 vm_map_entry_t tentry;
1989 vm_size_t tsize;
1990
1991 smap = current->object.sub_map;
1992 vm_map_lock_read(smap);
1993 (void) vm_map_lookup_entry(smap, offset, &tentry);
1994 tsize = tentry->end - offset;
1995 if (tsize < size)
1996 size = tsize;
1997 object = tentry->object.vm_object;
1998 offset = tentry->offset + (offset - tentry->start);
1999 vm_map_unlock_read(smap);
2000 } else {
2001 object = current->object.vm_object;
2002 }
2003 /*
2004 * Note that there is absolutely no sense in writing out
2005 * anonymous objects, so we track down the vnode object
2006 * to write out.
2007 * We invalidate (remove) all pages from the address space
2008 * anyway, for semantic correctness.
2009 *
2010 * note: certain anonymous maps, such as MAP_NOSYNC maps,
2011 * may start out with a NULL object.
2012 */
2013 while (object && object->backing_object) {
2014 object = object->backing_object;
2015 offset += object->backing_object_offset;
2016 if (object->size < OFF_TO_IDX( offset + size))
2017 size = IDX_TO_OFF(object->size) - offset;
2018 }
2019 if (object && (object->type == OBJT_VNODE) &&
2020 (current->protection & VM_PROT_WRITE)) {
2021 /*
2022 * Flush pages if writing is allowed, invalidate them
2023 * if invalidation requested. Pages undergoing I/O
2024 * will be ignored by vm_object_page_remove().
2025 *
2026 * We cannot lock the vnode and then wait for paging
2027 * to complete without deadlocking against vm_fault.
2028 * Instead we simply call vm_object_page_remove() and
2029 * allow it to block internally on a page-by-page
2030 * basis when it encounters pages undergoing async
2031 * I/O.
2032 */
2033 int flags;
2034
2035 vm_object_reference(object);
dadab5e9 2036 vn_lock(object->handle, LK_EXCLUSIVE | LK_RETRY, curthread);
984263bc
MD
2037 flags = (syncio || invalidate) ? OBJPC_SYNC : 0;
2038 flags |= invalidate ? OBJPC_INVAL : 0;
2039 vm_object_page_clean(object,
2040 OFF_TO_IDX(offset),
2041 OFF_TO_IDX(offset + size + PAGE_MASK),
2042 flags);
dadab5e9 2043 VOP_UNLOCK(object->handle, 0, curthread);
984263bc
MD
2044 vm_object_deallocate(object);
2045 }
2046 if (object && invalidate &&
2047 ((object->type == OBJT_VNODE) ||
2048 (object->type == OBJT_DEVICE))) {
2049 vm_object_reference(object);
2050 vm_object_page_remove(object,
2051 OFF_TO_IDX(offset),
2052 OFF_TO_IDX(offset + size + PAGE_MASK),
2053 FALSE);
2054 vm_object_deallocate(object);
2055 }
2056 start += size;
2057 }
2058
2059 vm_map_unlock_read(map);
2060 return (KERN_SUCCESS);
2061}
2062
2063/*
2064 * vm_map_entry_unwire: [ internal use only ]
2065 *
2066 * Make the region specified by this entry pageable.
2067 *
2068 * The map in question should be locked.
2069 * [This is the reason for this routine's existence.]
2070 */
2071static void
2072vm_map_entry_unwire(map, entry)
2073 vm_map_t map;
2074 vm_map_entry_t entry;
2075{
2076 vm_fault_unwire(map, entry->start, entry->end);
2077 entry->wired_count = 0;
2078}
2079
2080/*
2081 * vm_map_entry_delete: [ internal use only ]
2082 *
2083 * Deallocate the given entry from the target map.
2084 */
2085static void
2086vm_map_entry_delete(map, entry)
2087 vm_map_t map;
2088 vm_map_entry_t entry;
2089{
2090 vm_map_entry_unlink(map, entry);
2091 map->size -= entry->end - entry->start;
2092
2093 if ((entry->eflags & MAP_ENTRY_IS_SUB_MAP) == 0) {
2094 vm_object_deallocate(entry->object.vm_object);
2095 }
2096
2097 vm_map_entry_dispose(map, entry);
2098}
2099
2100/*
2101 * vm_map_delete: [ internal use only ]
2102 *
2103 * Deallocates the given address range from the target
2104 * map.
2105 */
2106int
2107vm_map_delete(map, start, end)
2108 vm_map_t map;
2109 vm_offset_t start;
2110 vm_offset_t end;
2111{
2112 vm_object_t object;
2113 vm_map_entry_t entry;
2114 vm_map_entry_t first_entry;
2115
2116 /*
2117 * Find the start of the region, and clip it
2118 */
2119
2120again:
2121 if (!vm_map_lookup_entry(map, start, &first_entry))
2122 entry = first_entry->next;
2123 else {
2124 entry = first_entry;
2125 vm_map_clip_start(map, entry, start);
2126 /*
2127 * Fix the lookup hint now, rather than each time though the
2128 * loop.
2129 */
2130 SAVE_HINT(map, entry->prev);
2131 }
2132
2133 /*
2134 * Save the free space hint
2135 */
2136
2137 if (entry == &map->header) {
2138 map->first_free = &map->header;
2139 } else if (map->first_free->start >= start) {
2140 map->first_free = entry->prev;
2141 }
2142
2143 /*
2144 * Step through all entries in this region
2145 */
2146
2147 while ((entry != &map->header) && (entry->start < end)) {
2148 vm_map_entry_t next;
2149 vm_offset_t s, e;
2150 vm_pindex_t offidxstart, offidxend, count;
2151
2152 /*
2153 * If we hit an in-transition entry we have to sleep and
2154 * retry. It's easier (and not really slower) to just retry
2155 * since this case occurs so rarely and the hint is already
2156 * pointing at the right place. We have to reset the
2157 * start offset so as not to accidently delete an entry
2158 * another process just created in vacated space.
2159 */
2160 if (entry->eflags & MAP_ENTRY_IN_TRANSITION) {
2161 entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
2162 start = entry->start;
12e4aaff
MD
2163 ++mycpu->gd_cnt.v_intrans_coll;
2164 ++mycpu->gd_cnt.v_intrans_wait;
984263bc
MD
2165 vm_map_transition_wait(map);
2166 goto again;
2167 }
2168 vm_map_clip_end(map, entry, end);
2169
2170 s = entry->start;
2171 e = entry->end;
2172 next = entry->next;
2173
2174 offidxstart = OFF_TO_IDX(entry->offset);
2175 count = OFF_TO_IDX(e - s);
2176 object = entry->object.vm_object;
2177
2178 /*
2179 * Unwire before removing addresses from the pmap; otherwise,
2180 * unwiring will put the entries back in the pmap.
2181 */
2182 if (entry->wired_count != 0) {
2183 vm_map_entry_unwire(map, entry);
2184 }
2185
2186 offidxend = offidxstart + count;
2187
2188 if ((object == kernel_object) || (object == kmem_object)) {
2189 vm_object_page_remove(object, offidxstart, offidxend, FALSE);
2190 } else {
2191 pmap_remove(map->pmap, s, e);
2192 if (object != NULL &&
2193 object->ref_count != 1 &&
2194 (object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING &&
2195 (object->type == OBJT_DEFAULT || object->type == OBJT_SWAP)) {
2196 vm_object_collapse(object);
2197 vm_object_page_remove(object, offidxstart, offidxend, FALSE);
2198 if (object->type == OBJT_SWAP) {
2199 swap_pager_freespace(object, offidxstart, count);
2200 }
2201 if (offidxend >= object->size &&
2202 offidxstart < object->size) {
2203 object->size = offidxstart;
2204 }
2205 }
2206 }
2207
2208 /*
2209 * Delete the entry (which may delete the object) only after
2210 * removing all pmap entries pointing to its pages.
2211 * (Otherwise, its page frames may be reallocated, and any
2212 * modify bits will be set in the wrong object!)
2213 */
2214 vm_map_entry_delete(map, entry);
2215 entry = next;
2216 }
2217 return (KERN_SUCCESS);
2218}
2219
2220/*
2221 * vm_map_remove:
2222 *
2223 * Remove the given address range from the target map.
2224 * This is the exported form of vm_map_delete.
2225 */
2226int
2227vm_map_remove(map, start, end)
2228 vm_map_t map;
2229 vm_offset_t start;
2230 vm_offset_t end;
2231{
2232 int result, s = 0;
2233
2234 if (map == kmem_map || map == mb_map)
2235 s = splvm();
2236
2237 vm_map_lock(map);
2238 VM_MAP_RANGE_CHECK(map, start, end);
2239 result = vm_map_delete(map, start, end);
2240 vm_map_unlock(map);
2241
2242 if (map == kmem_map || map == mb_map)
2243 splx(s);
2244
2245 return (result);
2246}
2247
2248/*
2249 * vm_map_check_protection:
2250 *
2251 * Assert that the target map allows the specified
2252 * privilege on the entire address region given.
2253 * The entire region must be allocated.
2254 */
2255boolean_t
2256vm_map_check_protection(vm_map_t map, vm_offset_t start, vm_offset_t end,
2257 vm_prot_t protection)
2258{
2259 vm_map_entry_t entry;
2260 vm_map_entry_t tmp_entry;
2261
2262 if (!vm_map_lookup_entry(map, start, &tmp_entry)) {
2263 return (FALSE);
2264 }
2265 entry = tmp_entry;
2266
2267 while (start < end) {
2268 if (entry == &map->header) {
2269 return (FALSE);
2270 }
2271 /*
2272 * No holes allowed!
2273 */
2274
2275 if (start < entry->start) {
2276 return (FALSE);
2277 }
2278 /*
2279 * Check protection associated with entry.
2280 */
2281
2282 if ((entry->protection & protection) != protection) {
2283 return (FALSE);
2284 }
2285 /* go to next entry */
2286
2287 start = entry->end;
2288 entry = entry->next;
2289 }
2290 return (TRUE);
2291}
2292
2293/*
2294 * Split the pages in a map entry into a new object. This affords
2295 * easier removal of unused pages, and keeps object inheritance from
2296 * being a negative impact on memory usage.
2297 */
2298static void
2299vm_map_split(entry)
2300 vm_map_entry_t entry;
2301{
2302 vm_page_t m;
2303 vm_object_t orig_object, new_object, source;
2304 vm_offset_t s, e;
2305 vm_pindex_t offidxstart, offidxend, idx;
2306 vm_size_t size;
2307 vm_ooffset_t offset;
2308
2309 orig_object = entry->object.vm_object;
2310 if (orig_object->type != OBJT_DEFAULT && orig_object->type != OBJT_SWAP)
2311 return;
2312 if (orig_object->ref_count <= 1)
2313 return;
2314
2315 offset = entry->offset;
2316 s = entry->start;
2317 e = entry->end;
2318
2319 offidxstart = OFF_TO_IDX(offset);
2320 offidxend = offidxstart + OFF_TO_IDX(e - s);
2321 size = offidxend - offidxstart;
2322
2323 new_object = vm_pager_allocate(orig_object->type,
2324 NULL, IDX_TO_OFF(size), VM_PROT_ALL, 0LL);
2325 if (new_object == NULL)
2326 return;
2327
2328 source = orig_object->backing_object;
2329 if (source != NULL) {
2330 vm_object_reference(source); /* Referenced by new_object */
2331 LIST_INSERT_HEAD(&source->shadow_head,
2332 new_object, shadow_list);
2333 vm_object_clear_flag(source, OBJ_ONEMAPPING);
2334 new_object->backing_object_offset =
2335 orig_object->backing_object_offset + IDX_TO_OFF(offidxstart);
2336 new_object->backing_object = source;
2337 source->shadow_count++;
2338 source->generation++;
2339 }
2340
2341 for (idx = 0; idx < size; idx++) {
2342 vm_page_t m;
2343
2344 retry:
2345 m = vm_page_lookup(orig_object, offidxstart + idx);
2346 if (m == NULL)
2347 continue;
2348
2349 /*
2350 * We must wait for pending I/O to complete before we can
2351 * rename the page.
2352 *
2353 * We do not have to VM_PROT_NONE the page as mappings should
2354 * not be changed by this operation.
2355 */
2356 if (vm_page_sleep_busy(m, TRUE, "spltwt"))
2357 goto retry;
2358
2359 vm_page_busy(m);
2360 vm_page_rename(m, new_object, idx);
2361 /* page automatically made dirty by rename and cache handled */
2362 vm_page_busy(m);
2363 }
2364
2365 if (orig_object->type == OBJT_SWAP) {
2366 vm_object_pip_add(orig_object, 1);
2367 /*
2368 * copy orig_object pages into new_object
2369 * and destroy unneeded pages in
2370 * shadow object.
2371 */
2372 swap_pager_copy(orig_object, new_object, offidxstart, 0);
2373 vm_object_pip_wakeup(orig_object);
2374 }
2375
2376 for (idx = 0; idx < size; idx++) {
2377 m = vm_page_lookup(new_object, idx);
2378 if (m) {
2379 vm_page_wakeup(m);
2380 }
2381 }
2382
2383 entry->object.vm_object = new_object;
2384 entry->offset = 0LL;
2385 vm_object_deallocate(orig_object);
2386}
2387
2388/*
2389 * vm_map_copy_entry:
2390 *
2391 * Copies the contents of the source entry to the destination
2392 * entry. The entries *must* be aligned properly.
2393 */
2394static void
2395vm_map_copy_entry(src_map, dst_map, src_entry, dst_entry)
2396 vm_map_t src_map, dst_map;
2397 vm_map_entry_t src_entry, dst_entry;
2398{
2399 vm_object_t src_object;
2400
2401 if ((dst_entry->eflags|src_entry->eflags) & MAP_ENTRY_IS_SUB_MAP)
2402 return;
2403
2404 if (src_entry->wired_count == 0) {
2405
2406 /*
2407 * If the source entry is marked needs_copy, it is already
2408 * write-protected.
2409 */
2410 if ((src_entry->eflags & MAP_ENTRY_NEEDS_COPY) == 0) {
2411 pmap_protect(src_map->pmap,
2412 src_entry->start,
2413 src_entry->end,
2414 src_entry->protection & ~VM_PROT_WRITE);
2415 }
2416
2417 /*
2418 * Make a copy of the object.
2419 */
2420 if ((src_object = src_entry->object.vm_object) != NULL) {
2421
2422 if ((src_object->handle == NULL) &&
2423 (src_object->type == OBJT_DEFAULT ||
2424 src_object->type == OBJT_SWAP)) {
2425 vm_object_collapse(src_object);
2426 if ((src_object->flags & (OBJ_NOSPLIT|OBJ_ONEMAPPING)) == OBJ_ONEMAPPING) {
2427 vm_map_split(src_entry);
2428 src_object = src_entry->object.vm_object;
2429 }
2430 }
2431
2432 vm_object_reference(src_object);
2433 vm_object_clear_flag(src_object, OBJ_ONEMAPPING);
2434 dst_entry->object.vm_object = src_object;
2435 src_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
2436 dst_entry->eflags |= (MAP_ENTRY_COW|MAP_ENTRY_NEEDS_COPY);
2437 dst_entry->offset = src_entry->offset;
2438 } else {
2439 dst_entry->object.vm_object = NULL;
2440 dst_entry->offset = 0;
2441 }
2442
2443 pmap_copy(dst_map->pmap, src_map->pmap, dst_entry->start,
2444 dst_entry->end - dst_entry->start, src_entry->start);
2445 } else {
2446 /*
2447 * Of course, wired down pages can't be set copy-on-write.
2448 * Cause wired pages to be copied into the new map by
2449 * simulating faults (the new pages are pageable)
2450 */
2451 vm_fault_copy_entry(dst_map, src_map, dst_entry, src_entry);
2452 }
2453}
2454
2455/*
2456 * vmspace_fork:
2457 * Create a new process vmspace structure and vm_map
2458 * based on those of an existing process. The new map
2459 * is based on the old map, according to the inheritance
2460 * values on the regions in that map.
2461 *
2462 * The source map must not be locked.
2463 */
2464struct vmspace *
2465vmspace_fork(vm1)
2466 struct vmspace *vm1;
2467{
2468 struct vmspace *vm2;
2469 vm_map_t old_map = &vm1->vm_map;
2470 vm_map_t new_map;
2471 vm_map_entry_t old_entry;
2472 vm_map_entry_t new_entry;
2473 vm_object_t object;
2474
2475 vm_map_lock(old_map);
2476 old_map->infork = 1;
2477
2478 vm2 = vmspace_alloc(old_map->min_offset, old_map->max_offset);
2479 bcopy(&vm1->vm_startcopy, &vm2->vm_startcopy,
2480 (caddr_t) (vm1 + 1) - (caddr_t) &vm1->vm_startcopy);
2481 new_map = &vm2->vm_map; /* XXX */
2482 new_map->timestamp = 1;
2483
2484 old_entry = old_map->header.next;
2485
2486 while (old_entry != &old_map->header) {
2487 if (old_entry->eflags & MAP_ENTRY_IS_SUB_MAP)
2488 panic("vm_map_fork: encountered a submap");
2489
2490 switch (old_entry->inheritance) {
2491 case VM_INHERIT_NONE:
2492 break;
2493
2494 case VM_INHERIT_SHARE:
2495 /*
2496 * Clone the entry, creating the shared object if necessary.
2497 */
2498 object = old_entry->object.vm_object;
2499 if (object == NULL) {
2500 object = vm_object_allocate(OBJT_DEFAULT,
2501 atop(old_entry->end - old_entry->start));
2502 old_entry->object.vm_object = object;
2503 old_entry->offset = (vm_offset_t) 0;
2504 }
2505
2506 /*
2507 * Add the reference before calling vm_object_shadow
2508 * to insure that a shadow object is created.
2509 */
2510 vm_object_reference(object);
2511 if (old_entry->eflags & MAP_ENTRY_NEEDS_COPY) {
2512 vm_object_shadow(&old_entry->object.vm_object,
2513 &old_entry->offset,
2514 atop(old_entry->end - old_entry->start));
2515 old_entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
2516 /* Transfer the second reference too. */
2517 vm_object_reference(
2518 old_entry->object.vm_object);
2519 vm_object_deallocate(object);
2520 object = old_entry->object.vm_object;
2521 }
2522 vm_object_clear_flag(object, OBJ_ONEMAPPING);
2523
2524 /*
2525 * Clone the entry, referencing the shared object.
2526 */
2527 new_entry = vm_map_entry_create(new_map);
2528 *new_entry = *old_entry;
2529 new_entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2530 new_entry->wired_count = 0;
2531
2532 /*
2533 * Insert the entry into the new map -- we know we're
2534 * inserting at the end of the new map.
2535 */
2536
2537 vm_map_entry_link(new_map, new_map->header.prev,
2538 new_entry);
2539
2540 /*
2541 * Update the physical map
2542 */
2543
2544 pmap_copy(new_map->pmap, old_map->pmap,
2545 new_entry->start,
2546 (old_entry->end - old_entry->start),
2547 old_entry->start);
2548 break;
2549
2550 case VM_INHERIT_COPY:
2551 /*
2552 * Clone the entry and link into the map.
2553 */
2554 new_entry = vm_map_entry_create(new_map);
2555 *new_entry = *old_entry;
2556 new_entry->eflags &= ~MAP_ENTRY_USER_WIRED;
2557 new_entry->wired_count = 0;
2558 new_entry->object.vm_object = NULL;
2559 vm_map_entry_link(new_map, new_map->header.prev,
2560 new_entry);
2561 vm_map_copy_entry(old_map, new_map, old_entry,
2562 new_entry);
2563 break;
2564 }
2565 old_entry = old_entry->next;
2566 }
2567
2568 new_map->size = old_map->size;
2569 old_map->infork = 0;
2570 vm_map_unlock(old_map);
2571
2572 return (vm2);
2573}
2574
2575int
2576vm_map_stack (vm_map_t map, vm_offset_t addrbos, vm_size_t max_ssize,
2577 vm_prot_t prot, vm_prot_t max, int cow)
2578{
2579 vm_map_entry_t prev_entry;
2580 vm_map_entry_t new_stack_entry;
2581 vm_size_t init_ssize;
2582 int rv;
2583
2584 if (VM_MIN_ADDRESS > 0 && addrbos < VM_MIN_ADDRESS)
2585 return (KERN_NO_SPACE);
2586
2587 if (max_ssize < sgrowsiz)
2588 init_ssize = max_ssize;
2589 else
2590 init_ssize = sgrowsiz;
2591
2592 vm_map_lock(map);
2593
2594 /* If addr is already mapped, no go */
2595 if (vm_map_lookup_entry(map, addrbos, &prev_entry)) {
2596 vm_map_unlock(map);
2597 return (KERN_NO_SPACE);
2598 }
2599
2600 /* If we would blow our VMEM resource limit, no go */
2601 if (map->size + init_ssize >
2602 curproc->p_rlimit[RLIMIT_VMEM].rlim_cur) {
2603 vm_map_unlock(map);
2604 return (KERN_NO_SPACE);
2605 }
2606
2607 /* If we can't accomodate max_ssize in the current mapping,
2608 * no go. However, we need to be aware that subsequent user
2609 * mappings might map into the space we have reserved for
2610 * stack, and currently this space is not protected.
2611 *
2612 * Hopefully we will at least detect this condition
2613 * when we try to grow the stack.
2614 */
2615 if ((prev_entry->next != &map->header) &&
2616 (prev_entry->next->start < addrbos + max_ssize)) {
2617 vm_map_unlock(map);
2618 return (KERN_NO_SPACE);
2619 }
2620
2621 /* We initially map a stack of only init_ssize. We will
2622 * grow as needed later. Since this is to be a grow
2623 * down stack, we map at the top of the range.
2624 *
2625 * Note: we would normally expect prot and max to be
2626 * VM_PROT_ALL, and cow to be 0. Possibly we should
2627 * eliminate these as input parameters, and just
2628 * pass these values here in the insert call.
2629 */
2630 rv = vm_map_insert(map, NULL, 0, addrbos + max_ssize - init_ssize,
2631 addrbos + max_ssize, prot, max, cow);
2632
2633 /* Now set the avail_ssize amount */
2634 if (rv == KERN_SUCCESS){
2635 if (prev_entry != &map->header)
2636 vm_map_clip_end(map, prev_entry, addrbos + max_ssize - init_ssize);
2637 new_stack_entry = prev_entry->next;
2638 if (new_stack_entry->end != addrbos + max_ssize ||
2639 new_stack_entry->start != addrbos + max_ssize - init_ssize)
2640 panic ("Bad entry start/end for new stack entry");
2641 else
2642 new_stack_entry->avail_ssize = max_ssize - init_ssize;
2643 }
2644
2645 vm_map_unlock(map);
2646 return (rv);
2647}
2648
2649/* Attempts to grow a vm stack entry. Returns KERN_SUCCESS if the
2650 * desired address is already mapped, or if we successfully grow
2651 * the stack. Also returns KERN_SUCCESS if addr is outside the
2652 * stack range (this is strange, but preserves compatibility with
2653 * the grow function in vm_machdep.c).
2654 */
2655int
2656vm_map_growstack (struct proc *p, vm_offset_t addr)
2657{
2658 vm_map_entry_t prev_entry;
2659 vm_map_entry_t stack_entry;
2660 vm_map_entry_t new_stack_entry;
2661 struct vmspace *vm = p->p_vmspace;
2662 vm_map_t map = &vm->vm_map;
2663 vm_offset_t end;
2664 int grow_amount;
2665 int rv = KERN_SUCCESS;
2666 int is_procstack;
2667 int use_read_lock = 1;
2668
2669Retry:
2670 if (use_read_lock)
2671 vm_map_lock_read(map);
2672 else
2673 vm_map_lock(map);
2674
2675 /* If addr is already in the entry range, no need to grow.*/
2676 if (vm_map_lookup_entry(map, addr, &prev_entry))
2677 goto done;
2678
2679 if ((stack_entry = prev_entry->next) == &map->header)
2680 goto done;
2681 if (prev_entry == &map->header)
2682 end = stack_entry->start - stack_entry->avail_ssize;
2683 else
2684 end = prev_entry->end;
2685
2686 /* This next test mimics the old grow function in vm_machdep.c.
2687 * It really doesn't quite make sense, but we do it anyway
2688 * for compatibility.
2689 *
2690 * If not growable stack, return success. This signals the
2691 * caller to proceed as he would normally with normal vm.
2692 */
2693 if (stack_entry->avail_ssize < 1 ||
2694 addr >= stack_entry->start ||
2695 addr < stack_entry->start - stack_entry->avail_ssize) {
2696 goto done;
2697 }
2698
2699 /* Find the minimum grow amount */
2700 grow_amount = roundup (stack_entry->start - addr, PAGE_SIZE);
2701 if (grow_amount > stack_entry->avail_ssize) {
2702 rv = KERN_NO_SPACE;
2703 goto done;
2704 }
2705
2706 /* If there is no longer enough space between the entries
2707 * nogo, and adjust the available space. Note: this
2708 * should only happen if the user has mapped into the
2709 * stack area after the stack was created, and is
2710 * probably an error.
2711 *
2712 * This also effectively destroys any guard page the user
2713 * might have intended by limiting the stack size.
2714 */
2715 if (grow_amount > stack_entry->start - end) {
2716 if (use_read_lock && vm_map_lock_upgrade(map)) {
2717 use_read_lock = 0;
2718 goto Retry;
2719 }
2720 use_read_lock = 0;
2721 stack_entry->avail_ssize = stack_entry->start - end;
2722 rv = KERN_NO_SPACE;
2723 goto done;
2724 }
2725
2726 is_procstack = addr >= (vm_offset_t)vm->vm_maxsaddr;
2727
2728 /* If this is the main process stack, see if we're over the
2729 * stack limit.
2730 */
2731 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount >
2732 p->p_rlimit[RLIMIT_STACK].rlim_cur)) {
2733 rv = KERN_NO_SPACE;
2734 goto done;
2735 }
2736
2737 /* Round up the grow amount modulo SGROWSIZ */
2738 grow_amount = roundup (grow_amount, sgrowsiz);
2739 if (grow_amount > stack_entry->avail_ssize) {
2740 grow_amount = stack_entry->avail_ssize;
2741 }
2742 if (is_procstack && (ctob(vm->vm_ssize) + grow_amount >
2743 p->p_rlimit[RLIMIT_STACK].rlim_cur)) {
2744 grow_amount = p->p_rlimit[RLIMIT_STACK].rlim_cur -
2745 ctob(vm->vm_ssize);
2746 }
2747
2748 /* If we would blow our VMEM resource limit, no go */
2749 if (map->size + grow_amount >
2750 curproc->p_rlimit[RLIMIT_VMEM].rlim_cur) {
2751 rv = KERN_NO_SPACE;
2752 goto done;
2753 }
2754
2755 if (use_read_lock && vm_map_lock_upgrade(map)) {
2756 use_read_lock = 0;
2757 goto Retry;
2758 }
2759 use_read_lock = 0;
2760
2761 /* Get the preliminary new entry start value */
2762 addr = stack_entry->start - grow_amount;
2763
2764 /* If this puts us into the previous entry, cut back our growth
2765 * to the available space. Also, see the note above.
2766 */
2767 if (addr < end) {
2768 stack_entry->avail_ssize = stack_entry->start - end;
2769 addr = end;
2770 }
2771
2772 rv = vm_map_insert(map, NULL, 0, addr, stack_entry->start,
2773 VM_PROT_ALL,
2774 VM_PROT_ALL,
2775 0);
2776
2777 /* Adjust the available stack space by the amount we grew. */
2778 if (rv == KERN_SUCCESS) {
2779 if (prev_entry != &map->header)
2780 vm_map_clip_end(map, prev_entry, addr);
2781 new_stack_entry = prev_entry->next;
2782 if (new_stack_entry->end != stack_entry->start ||
2783 new_stack_entry->start != addr)
2784 panic ("Bad stack grow start/end in new stack entry");
2785 else {
2786 new_stack_entry->avail_ssize = stack_entry->avail_ssize -
2787 (new_stack_entry->end -
2788 new_stack_entry->start);
2789 if (is_procstack)
2790 vm->vm_ssize += btoc(new_stack_entry->end -
2791 new_stack_entry->start);
2792 }
2793 }
2794
2795done:
2796 if (use_read_lock)
2797 vm_map_unlock_read(map);
2798 else
2799 vm_map_unlock(map);
2800 return (rv);
2801}
2802
2803/*
2804 * Unshare the specified VM space for exec. If other processes are
2805 * mapped to it, then create a new one. The new vmspace is null.
2806 */
2807
2808void
2809vmspace_exec(struct proc *p) {
2810 struct vmspace *oldvmspace = p->p_vmspace;
2811 struct vmspace *newvmspace;
2812 vm_map_t map = &p->p_vmspace->vm_map;
2813
2814 newvmspace = vmspace_alloc(map->min_offset, map->max_offset);
2815 bcopy(&oldvmspace->vm_startcopy, &newvmspace->vm_startcopy,
2816 (caddr_t) (newvmspace + 1) - (caddr_t) &newvmspace->vm_startcopy);
2817 /*
2818 * This code is written like this for prototype purposes. The
2819 * goal is to avoid running down the vmspace here, but let the
2820 * other process's that are still using the vmspace to finally
2821 * run it down. Even though there is little or no chance of blocking
2822 * here, it is a good idea to keep this form for future mods.
2823 */
2824 vmspace_free(oldvmspace);
2825 p->p_vmspace = newvmspace;
2826 pmap_pinit2(vmspace_pmap(newvmspace));
2827 if (p == curproc)
2828 pmap_activate(p);
2829}
2830
2831/*
2832 * Unshare the specified VM space for forcing COW. This
2833 * is called by rfork, for the (RFMEM|RFPROC) == 0 case.
2834 */
2835
2836void
2837vmspace_unshare(struct proc *p) {
2838 struct vmspace *oldvmspace = p->p_vmspace;
2839 struct vmspace *newvmspace;
2840
2841 if (oldvmspace->vm_refcnt == 1)
2842 return;
2843 newvmspace = vmspace_fork(oldvmspace);
2844 vmspace_free(oldvmspace);
2845 p->p_vmspace = newvmspace;
2846 pmap_pinit2(vmspace_pmap(newvmspace));
2847 if (p == curproc)
2848 pmap_activate(p);
2849}
2850
2851
2852/*
2853 * vm_map_lookup:
2854 *
2855 * Finds the VM object, offset, and
2856 * protection for a given virtual address in the
2857 * specified map, assuming a page fault of the
2858 * type specified.
2859 *
2860 * Leaves the map in question locked for read; return
2861 * values are guaranteed until a vm_map_lookup_done
2862 * call is performed. Note that the map argument
2863 * is in/out; the returned map must be used in
2864 * the call to vm_map_lookup_done.
2865 *
2866 * A handle (out_entry) is returned for use in
2867 * vm_map_lookup_done, to make that fast.
2868 *
2869 * If a lookup is requested with "write protection"
2870 * specified, the map may be changed to perform virtual
2871 * copying operations, although the data referenced will
2872 * remain the same.
2873 */
2874int
2875vm_map_lookup(vm_map_t *var_map, /* IN/OUT */
2876 vm_offset_t vaddr,
2877 vm_prot_t fault_typea,
2878 vm_map_entry_t *out_entry, /* OUT */
2879 vm_object_t *object, /* OUT */
2880 vm_pindex_t *pindex, /* OUT */
2881 vm_prot_t *out_prot, /* OUT */
2882 boolean_t *wired) /* OUT */
2883{
2884 vm_map_entry_t entry;
2885 vm_map_t map = *var_map;
2886 vm_prot_t prot;
2887 vm_prot_t fault_type = fault_typea;
2888 int use_read_lock = 1;
2889 int rv = KERN_SUCCESS;
2890
2891RetryLookup:
2892 if (use_read_lock)
2893 vm_map_lock_read(map);
2894 else
2895 vm_map_lock(map);
2896
2897 /*
2898 * If the map has an interesting hint, try it before calling full
2899 * blown lookup routine.
2900 */
2901 entry = map->hint;
2902 *out_entry = entry;
2903
2904 if ((entry == &map->header) ||
2905 (vaddr < entry->start) || (vaddr >= entry->end)) {
2906 vm_map_entry_t tmp_entry;
2907
2908 /*
2909 * Entry was either not a valid hint, or the vaddr was not
2910 * contained in the entry, so do a full lookup.
2911 */
2912 if (!vm_map_lookup_entry(map, vaddr, &tmp_entry)) {
2913 rv = KERN_INVALID_ADDRESS;
2914 goto done;
2915 }
2916
2917 entry = tmp_entry;
2918 *out_entry = entry;
2919 }
2920
2921 /*
2922 * Handle submaps.
2923 */
2924
2925 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
2926 vm_map_t old_map = map;
2927
2928 *var_map = map = entry->object.sub_map;
2929 if (use_read_lock)
2930 vm_map_unlock_read(old_map);
2931 else
2932 vm_map_unlock(old_map);
2933 use_read_lock = 1;
2934 goto RetryLookup;
2935 }
2936
2937 /*
2938 * Check whether this task is allowed to have this page.
2939 * Note the special case for MAP_ENTRY_COW
2940 * pages with an override. This is to implement a forced
2941 * COW for debuggers.
2942 */
2943
2944 if (fault_type & VM_PROT_OVERRIDE_WRITE)
2945 prot = entry->max_protection;
2946 else
2947 prot = entry->protection;
2948
2949 fault_type &= (VM_PROT_READ|VM_PROT_WRITE|VM_PROT_EXECUTE);
2950 if ((fault_type & prot) != fault_type) {
2951 rv = KERN_PROTECTION_FAILURE;
2952 goto done;
2953 }
2954
2955 if ((entry->eflags & MAP_ENTRY_USER_WIRED) &&
2956 (entry->eflags & MAP_ENTRY_COW) &&
2957 (fault_type & VM_PROT_WRITE) &&
2958 (fault_typea & VM_PROT_OVERRIDE_WRITE) == 0) {
2959 rv = KERN_PROTECTION_FAILURE;
2960 goto done;
2961 }
2962
2963 /*
2964 * If this page is not pageable, we have to get it for all possible
2965 * accesses.
2966 */
2967
2968 *wired = (entry->wired_count != 0);
2969 if (*wired)
2970 prot = fault_type = entry->protection;
2971
2972 /*
2973 * If the entry was copy-on-write, we either ...
2974 */
2975
2976 if (entry->eflags & MAP_ENTRY_NEEDS_COPY) {
2977 /*
2978 * If we want to write the page, we may as well handle that
2979 * now since we've got the map locked.
2980 *
2981 * If we don't need to write the page, we just demote the
2982 * permissions allowed.
2983 */
2984
2985 if (fault_type & VM_PROT_WRITE) {
2986 /*
2987 * Make a new object, and place it in the object
2988 * chain. Note that no new references have appeared
2989 * -- one just moved from the map to the new
2990 * object.
2991 */
2992
2993 if (use_read_lock && vm_map_lock_upgrade(map)) {
2994 use_read_lock = 0;
2995 goto RetryLookup;
2996 }
2997 use_read_lock = 0;
2998
2999 vm_object_shadow(
3000 &entry->object.vm_object,
3001 &entry->offset,
3002 atop(entry->end - entry->start));
3003
3004 entry->eflags &= ~MAP_ENTRY_NEEDS_COPY;
3005 } else {
3006 /*
3007 * We're attempting to read a copy-on-write page --
3008 * don't allow writes.
3009 */
3010
3011 prot &= ~VM_PROT_WRITE;
3012 }
3013 }
3014
3015 /*
3016 * Create an object if necessary.
3017 */
3018 if (entry->object.vm_object == NULL &&
3019 !map->system_map) {
3020 if (use_read_lock && vm_map_lock_upgrade(map)) {
3021 use_read_lock = 0;
3022 goto RetryLookup;
3023 }
3024 use_read_lock = 0;
3025 entry->object.vm_object = vm_object_allocate(OBJT_DEFAULT,
3026 atop(entry->end - entry->start));
3027 entry->offset = 0;
3028 }
3029
3030 /*
3031 * Return the object/offset from this entry. If the entry was
3032 * copy-on-write or empty, it has been fixed up.
3033 */
3034
3035 *pindex = OFF_TO_IDX((vaddr - entry->start) + entry->offset);
3036 *object = entry->object.vm_object;
3037
3038 /*
3039 * Return whether this is the only map sharing this data. On
3040 * success we return with a read lock held on the map. On failure
3041 * we return with the map unlocked.
3042 */
3043 *out_prot = prot;
3044done:
3045 if (rv == KERN_SUCCESS) {
3046 if (use_read_lock == 0)
3047 vm_map_lock_downgrade(map);
3048 } else if (use_read_lock) {
3049 vm_map_unlock_read(map);
3050 } else {
3051 vm_map_unlock(map);
3052 }
3053 return (rv);
3054}
3055
3056/*
3057 * vm_map_lookup_done:
3058 *
3059 * Releases locks acquired by a vm_map_lookup
3060 * (according to the handle returned by that lookup).
3061 */
3062
3063void
3064vm_map_lookup_done(map, entry)
3065 vm_map_t map;
3066 vm_map_entry_t entry;
3067{
3068 /*
3069 * Unlock the main-level map
3070 */
3071
3072 vm_map_unlock_read(map);
3073}
3074
3075/*
3076 * Implement uiomove with VM operations. This handles (and collateral changes)
3077 * support every combination of source object modification, and COW type
3078 * operations.
3079 */
3080int
3081vm_uiomove(mapa, srcobject, cp, cnta, uaddra, npages)
3082 vm_map_t mapa;
3083 vm_object_t srcobject;
3084 off_t cp;
3085 int cnta;
3086 vm_offset_t uaddra;
3087 int *npages;
3088{
3089 vm_map_t map;
3090 vm_object_t first_object, oldobject, object;
3091 vm_map_entry_t entry;
3092 vm_prot_t prot;
3093 boolean_t wired;
3094 int tcnt, rv;
3095 vm_offset_t uaddr, start, end, tend;
3096 vm_pindex_t first_pindex, osize, oindex;
3097 off_t ooffset;
3098 int cnt;
3099
3100 if (npages)
3101 *npages = 0;
3102
3103 cnt = cnta;
3104 uaddr = uaddra;
3105
3106 while (cnt > 0) {
3107 map = mapa;
3108
3109 if ((vm_map_lookup(&map, uaddr,
3110 VM_PROT_READ, &entry, &first_object,
3111 &first_pindex, &prot, &wired)) != KERN_SUCCESS) {
3112 return EFAULT;
3113 }
3114
3115 vm_map_clip_start(map, entry, uaddr);
3116
3117 tcnt = cnt;
3118 tend = uaddr + tcnt;
3119 if (tend > entry->end) {
3120 tcnt = entry->end - uaddr;
3121 tend = entry->end;
3122 }
3123
3124 vm_map_clip_end(map, entry, tend);
3125
3126 start = entry->start;
3127 end = entry->end;
3128
3129 osize = atop(tcnt);
3130
3131 oindex = OFF_TO_IDX(cp);
3132 if (npages) {
3133 vm_pindex_t idx;
3134 for (idx = 0; idx < osize; idx++) {
3135 vm_page_t m;
3136 if ((m = vm_page_lookup(srcobject, oindex + idx)) == NULL) {
3137 vm_map_lookup_done(map, entry);
3138 return 0;
3139 }
3140 /*
3141 * disallow busy or invalid pages, but allow
3142 * m->busy pages if they are entirely valid.
3143 */
3144 if ((m->flags & PG_BUSY) ||
3145 ((m->valid & VM_PAGE_BITS_ALL) != VM_PAGE_BITS_ALL)) {
3146 vm_map_lookup_done(map, entry);
3147 return 0;
3148 }
3149 }
3150 }
3151
3152/*
3153 * If we are changing an existing map entry, just redirect
3154 * the object, and change mappings.
3155 */
3156 if ((first_object->type == OBJT_VNODE) &&
3157 ((oldobject = entry->object.vm_object) == first_object)) {
3158
3159 if ((entry->offset != cp) || (oldobject != srcobject)) {
3160 /*
3161 * Remove old window into the file
3162 */
3163 pmap_remove (map->pmap, uaddr, tend);
3164
3165 /*
3166 * Force copy on write for mmaped regions
3167 */
3168 vm_object_pmap_copy_1 (srcobject, oindex, oindex + osize);
3169
3170 /*
3171 * Point the object appropriately
3172 */
3173 if (oldobject != srcobject) {
3174
3175 /*
3176 * Set the object optimization hint flag
3177 */
3178 vm_object_set_flag(srcobject, OBJ_OPT);
3179 vm_object_reference(srcobject);
3180 entry->object.vm_object = srcobject;
3181
3182 if (oldobject) {
3183 vm_object_deallocate(oldobject);
3184 }
3185 }
3186
3187 entry->offset = cp;
3188 map->timestamp++;
3189 } else {
3190 pmap_remove (map->pmap, uaddr, tend);
3191 }
3192
3193 } else if ((first_object->ref_count == 1) &&
3194 (first_object->size == osize) &&
3195 ((first_object->type == OBJT_DEFAULT) ||
3196 (first_object->type == OBJT_SWAP)) ) {
3197
3198 oldobject = first_object->backing_object;
3199
3200 if ((first_object->backing_object_offset != cp) ||
3201 (oldobject != srcobject)) {
3202 /*
3203 * Remove old window into the file
3204 */
3205 pmap_remove (map->pmap, uaddr, tend);
3206
3207 /*
3208 * Remove unneeded old pages
3209 */
3210 vm_object_page_remove(first_object, 0, 0, 0);
3211
3212 /*
3213 * Invalidate swap space
3214 */
3215 if (first_object->type == OBJT_SWAP) {
3216 swap_pager_freespace(first_object,
3217 0,
3218 first_object->size);
3219 }
3220
3221 /*
3222 * Force copy on write for mmaped regions
3223 */
3224 vm_object_pmap_copy_1 (srcobject, oindex, oindex + osize);
3225
3226 /*
3227 * Point the object appropriately
3228 */
3229 if (oldobject != srcobject) {
3230
3231 /*
3232 * Set the object optimization hint flag
3233 */
3234 vm_object_set_flag(srcobject, OBJ_OPT);
3235 vm_object_reference(srcobject);
3236
3237 if (oldobject) {
3238 LIST_REMOVE(
3239 first_object, shadow_list);
3240 oldobject->shadow_count--;
3241 /* XXX bump generation? */
3242 vm_object_deallocate(oldobject);
3243 }
3244
3245 LIST_INSERT_HEAD(&srcobject->shadow_head,
3246 first_object, shadow_list);
3247 srcobject->shadow_count++;
3248 /* XXX bump generation? */
3249
3250 first_object->backing_object = srcobject;
3251 }
3252 first_object->backing_object_offset = cp;
3253 map->timestamp++;
3254 } else {
3255 pmap_remove (map->pmap, uaddr, tend);
3256 }
3257/*
3258 * Otherwise, we have to do a logical mmap.
3259 */
3260 } else {
3261
3262 vm_object_set_flag(srcobject, OBJ_OPT);
3263 vm_object_reference(srcobject);
3264
3265 pmap_remove (map->pmap, uaddr, tend);
3266
3267 vm_object_pmap_copy_1 (srcobject, oindex, oindex + osize);
3268 vm_map_lock_upgrade(map);
3269
3270 if (entry == &map->header) {
3271 map->first_free = &map->header;
3272 } else if (map->first_free->start >= start) {
3273 map->first_free = entry->prev;
3274 }
3275
3276 SAVE_HINT(map, entry->prev);
3277 vm_map_entry_delete(map, entry);
3278
3279 object = srcobject;
3280 ooffset = cp;
3281
3282 rv = vm_map_insert(map, object, ooffset, start, tend,
3283 VM_PROT_ALL, VM_PROT_ALL, MAP_COPY_ON_WRITE);
3284
3285 if (rv != KERN_SUCCESS)
3286 panic("vm_uiomove: could not insert new entry: %d", rv);
3287 }
3288
3289/*
3290 * Map the window directly, if it is already in memory
3291 */
3292 pmap_object_init_pt(map->pmap, uaddr,
3293 srcobject, oindex, tcnt, 0);
3294
3295 map->timestamp++;
3296 vm_map_unlock(map);
3297
3298 cnt -= tcnt;
3299 uaddr += tcnt;
3300 cp += tcnt;
3301 if (npages)
3302 *npages += osize;
3303 }
3304 return 0;
3305}
3306
3307/*
3308 * Performs the copy_on_write operations necessary to allow the virtual copies
3309 * into user space to work. This has to be called for write(2) system calls
3310 * from other processes, file unlinking, and file size shrinkage.
3311 */
3312void
3313vm_freeze_copyopts(object, froma, toa)
3314 vm_object_t object;
3315 vm_pindex_t froma, toa;
3316{
3317 int rv;
3318 vm_object_t robject;
3319 vm_pindex_t idx;
3320
3321 if ((object == NULL) ||
3322 ((object->flags & OBJ_OPT) == 0))
3323 return;
3324
3325 if (object->shadow_count > object->ref_count)
3326 panic("vm_freeze_copyopts: sc > rc");
3327
3328 while((robject = LIST_FIRST(&object->shadow_head)) != NULL) {
3329 vm_pindex_t bo_pindex;
3330 vm_page_t m_in, m_out;
3331
3332 bo_pindex = OFF_TO_IDX(robject->backing_object_offset);
3333
3334 vm_object_reference(robject);
3335
3336 vm_object_pip_wait(robject, "objfrz");
3337
3338 if (robject->ref_count == 1) {
3339 vm_object_deallocate(robject);
3340 continue;
3341 }
3342
3343 vm_object_pip_add(robject, 1);
3344
3345 for (idx = 0; idx < robject->size; idx++) {
3346
3347 m_out = vm_page_grab(robject, idx,
3348 VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
3349
3350 if (m_out->valid == 0) {
3351 m_in = vm_page_grab(object, bo_pindex + idx,
3352 VM_ALLOC_NORMAL | VM_ALLOC_RETRY);
3353 if (m_in->valid == 0) {
3354 rv = vm_pager_get_pages(object, &m_in, 1, 0);
3355 if (rv != VM_PAGER_OK) {
3356 printf("vm_freeze_copyopts: cannot read page from file: %lx\n", (long)m_in->pindex);
3357 continue;
3358 }
3359 vm_page_deactivate(m_in);
3360 }
3361
3362 vm_page_protect(m_in, VM_PROT_NONE);
3363 pmap_copy_page(VM_PAGE_TO_PHYS(m_in), VM_PAGE_TO_PHYS(m_out));
3364 m_out->valid = m_in->valid;
3365 vm_page_dirty(m_out);
3366 vm_page_activate(m_out);
3367 vm_page_wakeup(m_in);
3368 }
3369 vm_page_wakeup(m_out);
3370 }
3371
3372 object->shadow_count--;
3373 object->ref_count--;
3374 LIST_REMOVE(robject, shadow_list);
3375 robject->backing_object = NULL;
3376 robject->backing_object_offset = 0;
3377
3378 vm_object_pip_wakeup(robject);
3379 vm_object_deallocate(robject);
3380 }
3381
3382 vm_object_clear_flag(object, OBJ_OPT);
3383}
3384
3385#include "opt_ddb.h"
3386#ifdef DDB
3387#include <sys/kernel.h>
3388
3389#include <ddb/ddb.h>
3390
3391/*
3392 * vm_map_print: [ debug ]
3393 */
3394DB_SHOW_COMMAND(map, vm_map_print)
3395{
3396 static int nlines;
3397 /* XXX convert args. */
3398 vm_map_t map = (vm_map_t)addr;
3399 boolean_t full = have_addr;
3400
3401 vm_map_entry_t entry;
3402
3403 db_iprintf("Task map %p: pmap=%p, nentries=%d, version=%u\n",
3404 (void *)map,
3405 (void *)map->pmap, map->nentries, map->timestamp);
3406 nlines++;
3407
3408 if (!full && db_indent)
3409 return;
3410
3411 db_indent += 2;
3412 for (entry = map->header.next; entry != &map->header;
3413 entry = entry->next) {
3414 db_iprintf("map entry %p: start=%p, end=%p\n",
3415 (void *)entry, (void *)entry->start, (void *)entry->end);
3416 nlines++;
3417 {
3418 static char *inheritance_name[4] =
3419 {"share", "copy", "none", "donate_copy"};
3420
3421 db_iprintf(" prot=%x/%x/%s",
3422 entry->protection,
3423 entry->max_protection,
3424 inheritance_name[(int)(unsigned char)entry->inheritance]);
3425 if (entry->wired_count != 0)
3426 db_printf(", wired");
3427 }
3428 if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
3429 /* XXX no %qd in kernel. Truncate entry->offset. */
3430 db_printf(", share=%p, offset=0x%lx\n",
3431 (void *)entry->object.sub_map,
3432 (long)entry->offset);
3433 nlines++;
3434 if ((entry->prev == &map->header) ||
3435 (entry->prev->object.sub_map !=
3436 entry->object.sub_map)) {
3437 db_indent += 2;
3438 vm_map_print((db_expr_t)(intptr_t)
3439 entry->object.sub_map,
3440 full, 0, (char *)0);
3441 db_indent -= 2;
3442 }
3443 } else {
3444 /* XXX no %qd in kernel. Truncate entry->offset. */
3445 db_printf(", object=%p, offset=0x%lx",
3446 (void *)entry->object.vm_object,
3447 (long)entry->offset);
3448 if (entry->eflags & MAP_ENTRY_COW)
3449 db_printf(", copy (%s)",
3450 (entry->eflags & MAP_ENTRY_NEEDS_COPY) ? "needed" : "done");
3451 db_printf("\n");
3452 nlines++;
3453
3454 if ((entry->prev == &map->header) ||
3455 (entry->prev->object.vm_object !=
3456 entry->object.vm_object)) {
3457 db_indent += 2;
3458 vm_object_print((db_expr_t)(intptr_t)
3459 entry->object.vm_object,
3460 full, 0, (char *)0);
3461 nlines += 4;
3462 db_indent -= 2;
3463 }
3464 }
3465 }
3466 db_indent -= 2;
3467 if (db_indent == 0)
3468 nlines = 0;
3469}
3470
3471
3472DB_SHOW_COMMAND(procvm, procvm)
3473{
3474 struct proc *p;
3475
3476 if (have_addr) {
3477 p = (struct proc *) addr;
3478 } else {
3479 p = curproc;
3480 }
3481
3482 db_printf("p = %p, vmspace = %p, map = %p, pmap = %p\n",
3483 (void *)p, (void *)p->p_vmspace, (void *)&p->p_vmspace->vm_map,
3484 (void *)vmspace_pmap(p->p_vmspace));
3485
3486 vm_map_print((db_expr_t)(intptr_t)&p->p_vmspace->vm_map, 1, 0, NULL);
3487}
3488
3489#endif /* DDB */